Specialization: Transport Engineering and Logistics Report ...
Transcript of Specialization: Transport Engineering and Logistics Report ...
Delft University of Technology
FACULTY MECHANICAL, MARITIME AND MATERIALS ENGINEERING Department Marine and Transport Technology
Mekelweg 2 2628 CD Delft the Netherlands Phone +31 (0)15-2782889 Fax +31 (0)15-2781397 www.mtt.tudelft.nl
This report consists of 55 pages and 6 appendices. It may only be reproduced literally and as a whole. For commercial purposes only with written authorization of Delft University of Technology. Requests for consult are only taken into consideration under the condition that the applicant denies all legal rights on liabilities concerning the contents of the advice.
Specialization: Transport Engineering and Logistics Report number: 2019.TEL.8323 Title: Improvement of road traffic
sustainability by implementation of priority weights for trucks in predictive signalized intersection control
Author: L. Haanstra
Title (in Dutch) Verbetering van de duurzaamheid van het wegverkeer door de invoering van
prioriteitsgewichten voor vrachtwagens in voorspellende verkeersregelingen op kruispunten
Assignment: Masters thesis
Confidential: Yes (until May 31, 2019)
Chair graduation committee (university): Dr. ir. H. Polinder
Supervisor (university): Dr. ir. X. Jiang
Supervisor (company): ing. A.P. Verhoeven (Siemens Mobility B.V.,
Zoetermeer)
Date: March, 2019
Leroy Haanstra
Improvement of road traffic sustainability by implementation of priority weights for trucks in predictive signalized intersection control
Improvement of road traffic sustainability by implementation of priority weights for trucks in predictive signalized intersection
control
By
Leroy Haanstra
in partial fulfilment of the requirements for the degree of
Master of Science
in Mechanical Engineering
at the Delft University of Technology, to be defended publicly on Wednesday March 20, 2019 at 1:00 PM.
Student number: 4248007
Report number: 2019.TEL.8323 Thesis committee: Dr. ir. H. Polinder TU Delft
Dr. ir. X. Jiang TU Deflt Dr. ir. H. Taale TU Delft
ing. A.P. Verhoeven Siemens
This thesis is confidential and cannot be made public until May 31, 2019.
iii
Abstract In the European Union Road freight transport volume is expected to grow 78% between 2000 and
2030, which results in more trucks on the road network. The worldwide estimated trend shows an
increase of 150 million freight vehicles and an increase of 240 million passenger vehicles. The growth
of both vehicle classes will have a major impact on the road network and the roads will become
congested. Especially in dense urban environments with many intersections. Further, trucks have a
detrimental impact on traffic flows, especially at intersections, because of their slow dynamics and large
size. In addition, a stopping truck results in higher emissions and fuel consumption compared to a car.
However, today’s traffic controllers are not capable of optimizing traffic flow at intersections based on
classification of different vehicles. Therefore, it would be beneficial to all vehicles involved if the number
of stops for trucks would be reduced to a minimum, by servicing each vehicle class in a different way.
Throughout this research the focus is to develop a signalized intersection controller which is able to
reduce the number of stops for trucks, while maintaining an equal level of service for the other modes
of transport. Extensive literature studies provided important insights into the development of signalized
intersection controllers. A selection of the techniques found in the literature is used to develop a new
truck signal priority controller design. The basic idea is to use a weighted traffic light schedule in
combination with priority weights to enable truck signal priority. The design is evaluated in a case study
and simulated for multiple configurations. This leads to an overview of the performance in terms of the
number of stops and vehicle delay.
From the results several conclusions can be drawn on the optimal weight configuration, which is
compared to a state-of-the-art model predictive intersection controller. The results showed an reduction
of the total number of stops by 751 and total vehicle delay by 304 minutes for trucks over a week
(26,6% and 20,4% respectively). While, the total number of stops and total vehicle delay for cars
increased, by 155 stops and 201 minutes (0,1% and 0,3% respectively). However, the overall total
number of stops and total vehicle delay were reduced by 596 stops and 103 minutes (0,42% and 0,16%
respectively). It can be concluded that the proposed truck signal priority controller design can reduce
the number of stops for trucks at a signalized intersection, while maintaining the overall traffic flow at
least as good as a state-of-the-art model predictive intersection controller.
v
Contents Abstract ............................................................................................................................... iii
Contents ............................................................................................................................... v
List of abbreviations ........................................................................................................... vii
List of Figures ...................................................................................................................... ix
List of Tables ....................................................................................................................... xi
1. Introduction .................................................................................................................. 1
1.1. Problem statement ......................................................................................................... 1
1.2. Research objective ......................................................................................................... 3
1.3. Research question .......................................................................................................... 3
1.4. Research scope .............................................................................................................. 3
1.5. Approach ....................................................................................................................... 4
1.6. Report structure ............................................................................................................. 5
2. Literature review ........................................................................................................... 7
2.1. Signalized intersection controllers .................................................................................... 7
2.1.1. Basics..................................................................................................................... 8
2.1.2. Fixed-time control ................................................................................................... 9
2.1.3. Traffic-actuated control ........................................................................................... 9
2.1.4. Traffic-adaptive control ......................................................................................... 10
2.1.5. Model Predictive Control ........................................................................................ 10
2.2. Vehicle detector data approach ..................................................................................... 11
2.2.1. Inductive-loop detector ......................................................................................... 12
2.2.2. Magnetic detectors ................................................................................................ 12
2.2.3. Microwave radar detector ...................................................................................... 13
2.2.4. Infrared detector .................................................................................................. 14
2.2.5. Ultrasonic detector ................................................................................................ 15
2.2.6. Acoustic detector .................................................................................................. 15
2.2.7. Video Image Processor .......................................................................................... 15
2.2.8. Floating Car Data .................................................................................................. 15
2.3. Priority strategy............................................................................................................ 18
2.3.1. Emergency Vehicle Signal Pre-emption ................................................................... 19
2.3.2. Transit Signal Priority ............................................................................................ 19
2.3.3. Truck Signal Priority .............................................................................................. 21
2.4. Conclusions .................................................................................................................. 22
3. DIRECTOR ................................................................................................................... 25
3.1. Cumulative travel time delay ......................................................................................... 26
3.2. Uniform distribution of arrivals ...................................................................................... 26
3.3. Schedule decision ......................................................................................................... 27
vi Contents
3.4. Fixed schedule ahead of time ........................................................................................ 28
3.5. Conclusions .................................................................................................................. 28
4. Truck Signal Priority controller design ....................................................................... 29
4.1. Truck arrivals ............................................................................................................... 29
4.2. Priority weights ............................................................................................................ 30
4.3. Truck weights .............................................................................................................. 32
4.4. Schedule decision ......................................................................................................... 33
4.5. Conclusions .................................................................................................................. 33
5. Experimental setup ..................................................................................................... 35
5.1. Case study ................................................................................................................... 36
5.2. Simulation tools ........................................................................................................... 36
5.2.1. PTV Vissim ........................................................................................................... 37
5.2.2. Python ................................................................................................................. 37
5.3. Simulation data ............................................................................................................ 37
5.3.1. Historic data ......................................................................................................... 37
5.3.2. Truck data ............................................................................................................ 38
5.4. Simulation model ......................................................................................................... 38
5.5. Conclusions .................................................................................................................. 40
6. Simulation results ....................................................................................................... 41
6.1. Truck Signal Priority performance .................................................................................. 41
6.2. Priority weights sensitivity analysis ................................................................................ 44
6.3. Robustness check ......................................................................................................... 47
6.4. Emission analysis ......................................................................................................... 49
6.5. Discussion.................................................................................................................... 50
6.5.1. Truck Signal Priority performance .......................................................................... 50
6.5.2. Priority weights sensitivity analysis ......................................................................... 50
6.5.3. Robustness check ................................................................................................. 51
6.5.4. Emission analysis .................................................................................................. 51
7. Conclusions and recommendations ............................................................................ 53
7.1. Conclusions .................................................................................................................. 53
7.2. Recommendations ........................................................................................................ 54
References ......................................................................................................................... 57
A. Scientific paper ........................................................................................................... 63
B. Python modules and packages .................................................................................... 73
C. Vehicle arrivals ............................................................................................................ 75
D. Simulation results ....................................................................................................... 79
E. Simulation data ........................................................................................................... 85
F. Emission data ............................................................................................................ 103
vii
List of abbreviations ADAS Advanced Driver Assistance Systems
ADVANCE Advanced Driver and Vehicle Advisory Navigation Concept
BPTT Backpropagation Through Time
C-ITS Cooperative Intelligent Transport Systems
CO Carbon monoxide
CO2 Carbon dioxide
COM Component Object Model
DIRECTOR Data-driven Intersection and Road Environment Controller for Traffic Optimization in Real-time
DoE Design of Experiments
EVSP Emergency Vehicle Signal Pre-emption
FCD Floating Car Data
GLOSA Green Light Optimal Speed Advice
GPS Global Positioning System
I2V Infrastructure-to-Vehicle
ITS-G5 IEEE 802.11p, WIFI-P
KPI Key Performance Indicator
LSTM Long Short-Term Memory
LTE Cellular
MPC Model Predictive Control
NOx Nitrogen oxide
OD Origin-Destination
PG Phase Group
RADAR RAdio Detection And Ranging
RNN Recurrent Neural Network
SOIC Self-Organizing Intersection Controller
TkSP Truck Signal Priority
TSP Transit Signal Priority
V2I Vehicle-to-Infrastructure
V2V Vehicle-to-Vehicle
V2x Vehicle-to-Everything
VIP Video Image Processor
VOC Volatile Organic Compounds
xFCD Extended Floating Car Data
ix
List of Figures Figure 1 - Estimated trend of worldwide number of freight vehicles [8] ............................................ 2
Figure 2 - Flowchart of the approach .............................................................................................. 4
Figure 3 - Report structure ............................................................................................................ 5
Figure 4 - Layout chapter 2: Literature review ................................................................................ 7
Figure 5 - Conflict area [14] ........................................................................................................... 8
Figure 6 – Three phase group examples ......................................................................................... 8
Figure 7 - Magnetic anomaly induced in the Earth’s magnetic field by a magnetic dipole [46] .......... 13
Figure 8 - Perturbation of Earth’s magnetic field by a ferrous metal vehicle [46] ............................. 13
Figure 9 - Microwave radar operation [46] .................................................................................... 14
Figure 10 - Extended FCD examples [55] ...................................................................................... 16
Figure 11 - Comparison between IEEE 802.11p and cellular connectivity pipes to the car [57] ......... 16
Figure 12 - 5G the development compared to 4G [59] ................................................................... 17
Figure 13 - Layout chapter 3: DIRECTOR ...................................................................................... 25
Figure 14 - Arrival of vehicles over 10 seconds [93] ...................................................................... 26
Figure 15 - Transition from ρ(t) to ρ[T] [93] ................................................................................. 27
Figure 16 – Illustration of the switching penalty [93] ..................................................................... 28
Figure 17 - Layout chapter 4: Truck Signal Priority controller design .............................................. 29
Figure 18 - Point of detection ...................................................................................................... 30
Figure 19 - Truck arrivals per time bin .......................................................................................... 30
Figure 20 - Layout chapter 5: Experimental setup ......................................................................... 35
Figure 21 - Schematic overview of the case study intersection. ...................................................... 36
Figure 22 - Truck detections (marked with red dots) ..................................................................... 38
Figure 23 - PTV Vissim network layout of the intersection .............................................................. 39
Figure 24 – Layout chapter 6: Simulation results ........................................................................... 41
Figure 25 - Comparison: All ......................................................................................................... 43
Figure 26 - Comparison: Cars ...................................................................................................... 43
Figure 27 - Comparison: Trucks ................................................................................................... 43
Figure 28 – Sensitivity analysis: 24 hours (1/2) ............................................................................. 44
Figure 29 – Sensitivity analysis: 24 hours (2/2) ............................................................................. 45
Figure 30 - Sensitivity analysis comparison: All ............................................................................. 46
Figure 31 - Sensitivity analysis comparison: Cars .......................................................................... 46
Figure 32 - Sensitivity analysis comparison: Trucks ....................................................................... 46
Figure 33 – Sensitivity analysis: Week (1/2).................................................................................. 47
Figure 34 – Sensitivity analysis: Week (2/2).................................................................................. 48
Figure 35 - Carbon monoxide emissions ....................................................................................... 49
Figure 36 – Nitrogen oxide emissions ........................................................................................... 49
Figure 37 - Volatile Organic Compounds emissions ........................................................................ 49
x List of Figures
Figure 38 - Fuel consumption ...................................................................................................... 49
Figure 39 – Week ........................................................................................................................ 80
Figure 40 – Tuesday ................................................................................................................... 80
Figure 41 – Wednesday ............................................................................................................... 81
Figure 42 - Thursday ................................................................................................................... 81
Figure 43 - Friday ....................................................................................................................... 82
Figure 44 - Saturday ................................................................................................................... 82
Figure 45 - Sunday ...................................................................................................................... 83
Figure 46 - Monday ..................................................................................................................... 83
xi
List of Tables Table 1 - Growth in global freight transport [6] ............................................................................... 2
Table 2 - Detector types .............................................................................................................. 12
Table 3 - Traffic signal control treatments for different traffic modes [64] ...................................... 18
Table 4 - Key characteristic of different transit signal priority controls [76] ..................................... 21
Table 5 - Priority weight index to corresponding time bin .............................................................. 31
Table 6 - Priority weights example 4.3.1. ...................................................................................... 32
Table 7 - Truck arrivals example 4.3.1. ......................................................................................... 32
Table 8 - Results example 4.3.1. .................................................................................................. 32
Table 9 - Number of vehicles: 24 hours ........................................................................................ 42
Table 10 – Total stops ................................................................................................................. 42
Table 11 – Total vehicle delay ...................................................................................................... 42
Table 12 – Priority weight data set ............................................................................................... 44
Table 13 – Total stops: Sensitivity analysis ................................................................................... 45
Table 14 - Total vehicle delay: Sensitivity analysis ......................................................................... 45
Table 15 - Number of vehicles: Week ........................................................................................... 48
Table 16 – Total stops: Week ...................................................................................................... 48
Table 17 - Total vehicle delay: Week ............................................................................................ 48
Table 18 - Car arrivals ................................................................................................................. 76
Table 19 - Truck arrivals .............................................................................................................. 77
Table 20 - Simulation data: Week number of stops ....................................................................... 86
Table 21 - Simulation data: Week vehicle delay (s) ....................................................................... 87
Table 22 - Simulation data: Tuesday number of stops ................................................................... 88
Table 23 - Simulation data: Tuesday vehicle delay (s) ................................................................... 89
Table 24 - Simulation data: Wednesday number of stops .............................................................. 90
Table 25 - Simulation data: Wednesday vehicle delay (s) .............................................................. 91
Table 26 - Simulation data: Thursday number of stops .................................................................. 92
Table 27 - Simulation data: Thursday vehicle delay (s) .................................................................. 93
Table 28 - Simulation data: Friday number of stops ...................................................................... 94
Table 29 - Simulation data: Friday vehicle delay (s) ...................................................................... 95
Table 30 - Simulation data: Saturday number of stops .................................................................. 96
Table 31 - Simulation data: Saturday vehicle delay (s) .................................................................. 97
Table 32 - Simulation data: Sunday number of stops ..................................................................... 98
Table 33 - Simulation data: Sunday vehicle delay (s) ..................................................................... 99
Table 34 - Simulation data: Monday number of stops .................................................................. 100
Table 35 - Simulation data: Monday vehicle delay (s) .................................................................. 101
Table 36 - Emission data ........................................................................................................... 104
1
1 1. Introduction
Signalized intersections play an important role in modern society. The introduction of signalized
intersection controllers gave structure and a way of automatic traffic handling at intersections. Only
with the economic growth that developed countries faced an exponentially increasing demand for
personal mobility occurred [1]. It quickly resulted in congestions at signalized intersections in urban
environments. Congestion involves queuing, lower speeds, and increased travel times, which impose
costs on the economy and generate multiple impacts on urban regions and their inhabitants [2]. To
eliminate the congestion, signalized intersections could be replaced with bridges, tunnels or non-
signalized roundabouts. This option is, however, in many cases, both economically and spatially, not
feasible [1]. One alternative option to improve efficiency of urban intersections would be the innovation
of the on-street traffic controllers. However, traffic inefficiencies will still occur, because of the
disruption of traffic flow caused by a red light, even with the latest innovations of traffic controllers. In
addition, a reaction to an unanticipated switch from green to amber causes safety concerns, as drivers
may suddenly stop or quickly accelerate. Apart from the disruption of traffic flow and safety concerns,
the acceleration and deceleration behaviour cause the largest amount of fuel consumption and CO2
emissions [3]. These stops result in over 50% of the fuel consumption of a vehicles trip [4]. Moreover,
one stop of a vehicle could create a backward moving shockwave that induces a cyclic driving state of
acceleration and idling. This behaviour is responsible for up to two thirds of the total energy loss at
intersections [3], [4]. Additionally, once the traffic light turns green, the inability for drivers to anticipate
when they should accelerate from stop, and the time it takes to accelerate to free-flow speed, results
in a queue discharge rate that can be as low as 75% of the road’s capacity [5]. It is therefore the goal
to reduce the number of stops for vehicles in future signalized intersection controllers.
1.1. Problem statement
Worldwide freight transport volumes are increasing and the strongest expected growth is in road freight
transport, as shown in Table 1 [6]. Road freight transport volume in the European Union is expected to
grow 78% between 2000 and 2030 [7]. This will pair with an increasing volume of trucks driving on
the road, displayed in Figure 1 [8].
2 1. Introduction
Table 1 - Growth in global freight transport [6]
Figure 1 - Estimated trend of worldwide number of freight vehicles [8]
Aside from the increase in truck, the number of passenger vehicles is also expected to increase [9].
The growth of both vehicle classes will have a major impact on the road network and the roads will
become congested. Especially in dense urban environments with many intersections. Further, trucks
have a detrimental impact on traffic flows, especially at intersections, because of their slow dynamics
and large size [10]. The time for a heavy truck to respond to a traffic light, accelerate and cross the
intersection is much higher than that of normal passenger cars [11]. In addition, a stopping truck also
results in much higher emissions and fuel consumption. Despite the different vehicle characteristics
between trucks and passenger cars, current traffic controllers do not service them differently. Instead,
the traffic controllers will service both as equal. Hence, the problem could be described as:
“Today’s traffic controllers are not capable of optimizing traffic flow at intersections based on
classification of different vehicles.”
Therefore, it would be beneficial to all vehicles involved if the number of stops for trucks would be
reduced to a minimum, by servicing each vehicle class in a different way. For example, the cars stopping
behind a truck at a red light would be limited to the lower acceleration of the truck driving in front when
the light turns green. This would increase the total time for clearing the queue and less vehicles could
be serviced during green time. Next to the traffic delay due to slow dynamics and less vehicles serviced
during green, reducing the number of stops for truck has also other benefits. These benefits are related
to the environmental impact of a signalized intersection. The high fuel consumption and high emissions
resulting from a stop and go for a truck would be reduced. But also, the high pavement wear and
particulate matter emissions due to the braking of a heavy truck would be minimized. A possible solution
is the implementation of priority weights for trucks approaching the intersection. By making a distinction
1.2. Research objective 3
between vehicles classes and weighing trucks higher than cars would reduce the number of stops for
trucks, as they get more priority for green light.
1.2. Research objective
The goal of this research is to reduce the number of stops for trucks at signalized intersections by
granting priority to trucks via implementation of priority weights for different vehicle classes in
DIRECTOR (Data-driven Intersection and Road Environment Controller for Traffic Optimization in Real-
time). This research focusses on two important Key Performance Indicators (KPI’s): the traffic flow and
the number of stops at signalised intersections for trucks. In this research the traffic flow is indicated
with the vehicle delay. The implementation of priority weights for different vehicle classes provides a
possible improvement to the traffic flow and/or reduction the number of stops for trucks. It is coherent
that this can also reduce the fuel consumption, CO2 emission and safety concerns. Accordingly, finding
the means with which to develop and evaluate the implementation is the research objective.
1.3. Research question
The main research objective is to investigate how the implementation of priority weights, in a model
predictive control-based signalized intersection controller, could lead to an reduction in the number of
stops for trucks. However, the implementation of priority weights should not decrease the traffic flow
of other vehicles. Therefore, the main research question is stated as:
“Can a model predictive control-based signalized intersection controller with priority weights lead to
a reduction in the number of stops for trucks at a signalized intersection, while maintaining
traffic flow at least as good as a state-of-the-art model predictive controller?’’
In order to answer the main question, several sub questions need to be answered:
1. What are the state-of-the-art signalized intersection control techniques?
2. What are the basic principles of the state-of-the-art model predictive controller?
3. How should the priority weights be assigned to different vehicle types?
4. How could the intersection controller with priority weights be evaluated?
5. Does a change in priority weights affect the performance in terms of traffic flow?
6. Can a model predictive controller with priority weights improve the performance of a state-of-
the-art model predictive controller in terms of reducing the number of stops?
7. How does the controller perform at different traffic demands?
1.4. Research scope
According to the research objectives, the research is bounded by a scope. This is used to set limitations
which are required to ensure that realistic solutions are found and to narrow the size of the research.
Within the search for an optimal state of operation, the practical restrictions of legal and road safety
boundary conditions must be considered. As well as, industry standards should be used to meet the
requirements of a practical application. The possible inclusion of other data sources, such as Floating
4 1. Introduction
Car Data (FCD) should be explored. However, the means of receiving the floating car data form the
road users is not part of the study and it is assumed that all floating car data from trucks is available
for use. Furthermore, the intersection geometry is simplified and the impact of pedestrians, bicycles,
public transport vehicles and emergency vehicles (police, ambulance and fire trucks) is not investigated,
although the traffic light controller should be able to deal with these types of road users.
1.5. Approach
The research approach is visualized in a flowchart, displayed in Figure 2. The first step will be to study
the current model available. This step is very important as it will form the baseline of the proceeding
work. A literature study will be performed to examine the related work and to determine the
requirements. Based on the requirements a method to implement the priority weight will be proposed.
The next step is to define the proposed method in a mathematical form. Following the mathematical
form, a design of the control algorithm is made. Afterwards, the designed control algorithm should be
implemented in the current model. The new model should now be evaluated based on the key
performance indicators of section 1.2. Evaluation in a real environment results in high implementation
and start-up costs. Also, there are safety concerns to be considered. Therefore, a simulation
environment will be used as reality. This means the intelligent traffic controller will be evaluated within
the simulation environment and not on a physical intersection. The simulation environment used in this
study is PTV Vissim, an advanced simulation and visualisation tool. Based on the results of the
evaluation, the model should be adjusted to answer the research questions and enhance the
performance of the model.
Figure 2 - Flowchart of the approach
1.6. Report structure 5
1.6. Report structure
Based on the approach discussed in the previous section, this section describes the structure of this
research. The report starts with an overview of the state-of-the-art signalized intersection control
techniques and a selection of the most promising methods and techniques (chapter 2). Thereafter, the
basic principles in the design of the model predictive controller DIRECTOR are described (chapter 3),
which are required as preliminary knowledge in this thesis. Subsequently, the development of the
design of the Truck Signal Priority controller elaborated (chapter 4). Next, the experimental setup and
a case study, used to evaluate the performance of the proposed truck signal priority controller design,
are defined (chapter 5). Afterwards, the results of the evaluation of the case study are presented and
discussed (chapter 6). Finally, the report is concluded and recommendations for future work are
presented (chapter 7). The relationship between the different chapters is shown in Figure 3.
Figure 3 - Report structure
7
2 2. Literature review
This chapter presents a literature review of relevant topics to this research. The goal of this chapter is
to explore the state-of-the-art signalized intersection control techniques. Starting with the basics of
signalized intersection controllers and continues to review different control methods (Section 2.1).
Afterwards, different types of vehicle detectors are presented (section 2.2). Thereafter, the priority
approaches of signalized intersections are explored (section 2.3). Finally, the literature review is
concludes with a selection of the most promising methods and techniques (section 2.4). The relationship
between the different sections is shown in Figure 4.
Figure 4 - Layout chapter 2: Literature review
This chapter will answer the first sub question, presented in section 1.3, with the following questions:
(1) What are the state-of-the-art signalized intersection control techniques?
• What is the state-state-of-the-art signalized intersection controller?
• How could the intersection controller identify different vehicle type?
• Which priority strategy could be applied for trucks?
2.1. Signalized intersection controllers
Traffic control on an intersection can be summed up as a mechanism for permitting and disallowing the
right of way for different traffic streams with sequences of traffic light indications [12]. Traffic signals
prevent accidents and influence the throughput of intersections [13]. The disallowance of traffic streams
in a direction results in a negative impact on local traffic flow. It is therefore the challenge to design a
signalized intersection controller with the least negative impact on the traffic flow. This section starts
with the basics of signalized intersection controllers. Afterwards, four main types of signalized
intersection control are discussed:
1. Fixed-time control
2. Traffic-actuated control
3. Traffic-adaptive control
4. Model predictive control
8 2. Literature review
2.1.1. Basics
The basics of signalized intersection controllers are discussed, to create a better understanding of the
terminology used in this research. An intersection consists of a number of approaches and a crossing
area. The crossing area connects the approaches to the intersection. Each approach may have one or
more lanes, but has a unique and independent queue. An approach could also be defined as a link. The
intersection provides safe access through the crossing area from an origin link to a destination link. A
connected origin link and destination link are called an Origin-Destination pair (OD). Each Origin-
Destination has one or more lanes and each lane has its own signal. The purpose of a signal is to grant
or prohibit access to the crossing area via the different signal lights, respectively green and red. The
term for granting access is servicing. A signal group is a combination of signals which are always
serviced together and have the same Origin-Destination. Multiple signal groups could be serviced
together as well. However, one must be aware of possible conflict areas. A conflict area is the place
where two Origin-Destination paths cross each other, as shown in Figure 5. A combination of signal
groups which can be serviced simultaneously is defined as a Phase Group (PG), multiple examples are
displayed in Figure 6. The phase is the period a Phase Group is serviced, i.e. the green phase. The
sequence of phases is defined as the schedule of the intersection controller. The goal is to optimize the
schedule according to the desired optimization criteria. Originally the most used criteria are the
cumulative or average travel time delay. Other criteria that are often used are the number of stops or
the queue lengths at an intersection. Today, the importance of reducing the environmental impact
shifted the focus of the optimization criteria to fuel efficiency and emissions. In practice a combination
of criteria is often used, resulting in a trade-off at the scheduling decision.
Figure 5 - Conflict area [14]
Figure 6 – Three phase group examples
2.1. Signalized intersection controllers 9
2.1.2. Fixed-time control
The most straightforward design of a traffic light controller is fixed-time control. In a fixed-time control
strategy is the duration of each phase determined offline and switched in a pre-set cycle. This approach
assumes that the traffic patterns can be predicted accurately based on historical data [14]. The fixed-
time coordinated strategies are applicable to under-saturated traffic conditions [15]. As fixed-time
controllers can operate without traffic detectors installed at the intersection, the construction cost is
much lower than with traffic actuated and traffic-adaptive control [14].
The fixed-time controllers most found in literature are MAXBAND [16], [17], MULTIBAND [18], [19] and
TRANSYT [20]. The first, MAXBAND, solves a binary-mixed-integer-linear-programming problem using
the branch and bound method to reach an optimal schedule. The second, MULTIBAND, is an extended
version of MAXBAND. A variety of new aspects are incorporated such as: time clearance of existing
queues, left-turn movements, and multiple signal groups. The third, TRANSYT, is the most famous
fixed-time and is widely used in the United States. TRANSYT is based on a traffic model that is used to
estimate the queue sizes with time. The objective function is to minimize the sum of average queues,
based on the historical data of the network. To solve the objective function a heuristic hill-climb
optimization algorithm is used. This algorithm attempts to find a better solution by introducing an
incremental change to the decision variables and repeats the process until a local minimum is found.
The main drawback of fixed-time control is that it is not able to adapt itself as it is based on historical
rather than on real-time data. [14]. The traffic demand is unpredictable and changes over time [21],
[22]. As a result, the performance decreases over time, since the optimized settings do not match the
current traffic demand.
2.1.3. Traffic-actuated control
Traffic-actuated controllers schedule the next signal phase based on detector information instead of
historical data. Detectors provide the required information in order to respond to the real-time traffic
situation. The detectors that are used most frequently are inductive loop detectors [14]. The controller
could adjust the planned schedule, according to the real-time traffic demand, to seconds before
switching signal phase. The green phase remains, until traffic is detected on a direction outside the
phase group. Additionally, most traffic-actuated controllers have the ability to extend the green time to
ensure that the green phase is terminated comfortably and safely [14]. However, to prevent infinite
green time extension, controllers have a set of pre-defined static parameters such as the extension
time, minimum green time and maximum green time [23], [24]. The main benefit of the controller is
the reduction in wasted green time on empty lane and a faster response on traffic demand. In summary,
the traffic-actuated controllers are more efficient, but also have higher cost compared to a fixed-time
controller [25].
10 2. Literature review
2.1.4. Traffic-adaptive control
Adaptive traffic control systems continuously sense and monitor traffic conditions and adjust the timing
of traffic signals accordingly [14]. The monitoring of the traffic is performed in the same way as the
actuated controller, via detectors. A key difference between traffic-actuated and traffic-adaptive control,
is that a traffic-actuated controller has no foresight on incoming vehicle flows. The traffic-adaptive
controller adds an upstream detector loop to count incoming vehicles on a direction, this creates a
foresight on incoming vehicle and green time durations can be optimized accordingly [26]. Additionally,
a traffic-adaptive controller decision, to either continue the current green phase or to switch to a
different phase, is based on the entire intersection [14]. Whereas the decision at a traffic-actuated
controller is solely based on the presence of demand on the active green phase. A drawback to this
method is that if one or more loop detectors are not operating, the performance of the traffic-adaptive
signal control system might be notably degraded [27].
2.1.5. Model Predictive Control
Model predictive control is in the basics an extended version of the adaptive controller. The key
difference is similar to the difference between actuated and adaptive, the extension of the vision
horizon. The predictive controller has the vision horizon extended to at least the exit flows of the
previous intersection. First implementations of an extended vision horizon are found in SCOOT [28]
and SCATS [29]. These controllers modify the cycles, splits and offsets according to estimations of the
arrivals, based the upstream measurements. More recent developments of predictive controllers are
for example UTOPIA [30], OPAC [31], PRODYN [32], RHODES [33], CRONOS [34] and SURTRAC [35].
These controllers do not explicitly modify the cycles, splits and offsets. Instead, they calculate optimal
switching schedules, based on pre-specified plans for the traffic light phases and the predicted vehicle
arrivals, over a future rolling time horizon [25]. The purpose of using longer horizons for traffic flow
predictions and schedule optimizations is to estimate a full cycle and avoid controller fluctuations due
to short-sighted control outputs [15]. Consequently, the controller must be able to handle the large
uncertainties associated with forecasting traffic flow progression. Three main analytical models for
traffic flow progression are Lighthill & Witham’s fluid dynamic traffic model [36], Pacey’s diffusion
model [37] and Robertson’s platoon dispersion model [20]. Respectively, each model is a modification
of the predecessor. However, these analytical models are limited by a major assumption, which is the
conservation of vehicles from upstream to downstream. As a result, very few locations are suitable for
application. An alternative to the analytical models are data-driven methods. These methods proved
more suitable for traffic flow prediction, as shown by the recurrent neural network (RNN) design
proposed by Helmy [38]. This model returns the arrival flows downstream as an output and uses the
upstream departure flows, the presence of a queue downstream, the traffic split between downstream
lanes, the states of the downstream signals and information about the day of the week and the current
time as inputs. The RNN design is trained with backpropagation through time (BPTT). Helmy evaluated
the model in a case study with real data and outperformed Robertson’s model in terms of prediction
accuracy. However, conventional RNNs trained with BPTT are unable to learn long-term time
2.2. Vehicle detector data approach 11
dependencies [39]. The backwards propagation error accumulates over time, which results in an
unstable network. A solution to this problem was proposed by Hochreiter & Schmidhuber, adding a
long short-term memory (LSTM) [40]. The intuition behind the LSTM is to control the memory in a
structured way. Therefore, the LSTM can determine whether the content of the memory should be
remembered, updated or forgotten. Long-term dependencies can be recognized in the LSTM network
by training this memory. Evaluation of the proposed LSTM network showed it outperforms the
conventional RNNs in terms of prediction accuracy and required training time.
The architecture of a model predictive controller could be divided in a centralized or a distributed
approach. The first, relies on a central computer to make control decisions and direct the actions of
individual controllers, whereas at the second the intersection controller is responsible for operation
decisions [14]. The centralized control approach is often not feasible due to computational complexity,
communication overhead and lack of scalability [14]. However, distributed approaches are relatively
easy to expand and could reduce the computational complexity. The reduction of the computational
complexity is achieved by a concept developed in SURTRAC [35]. The intention behind this concept is
to model vehicle flows based on a simple model and then aggregate arrivals to create clusters of arriving
vehicles. After, the aggregated arrivals are handled as indivisible jobs in a forward-recursive scheduling
algorithm. The concept resulted in a reduction of the computational complexity that is polynomial in
the prediction horizon.
A variant of the distributed approach is the self-organizing traffic light controller [41]. The objective is
relatively simple, the controller gives preference to cars that have been waiting longer and to larger
groups of cars [15]. In other words, the controller optimizes the traffic light control according to the
cumulative travel time delay. The cumulative travel time delay is calculated for each approaching lane
of the intersection, based on the predicted short-term arrival flows [42]. These arrival flows are
predicted following the Lighthill and Whitham’s fluid dynamic traffic model [36]. The intersection uses
a non-periodic optimization technique to create optimal schedules, which can lead to instability [43].
Therefore, a stabilization mechanism is applied to ensure servicing of each direction as least as good
as a fixed-time strategy [44]. Initially, the controller without a stabilization mechanism was compared
to a fixed-time controller, showing a significant reduction in terms of average queue length and average
travel time delay [42]. In later work, the controller with the stabilization mechanism is compared the
previous controller without the stabilization mechanism, which resulted again in a reduction of the
average travel time delay [44].
2.2. Vehicle detector data approach
It is understood that no matter which control method is used, vehicle data is required as an input to
the traffic light controller. The vehicle data is acquired by a vehicle detector. Any traffic-responsive
control system depends on its ability to sense traffic for local intersection control and / or system-wide
adjustment of timing plans [45]. The detectors are categorized in two different categories: Pavement
12 2. Literature review
Invasive Detector and Non-Pavement Invasive Detector. The first, an in-roadway sensor is one that is
placed as one of the following ways [46]:
• Embedded in the pavement of the roadway.
• Embedded in the subgrade of the roadway.
• Taped or otherwise attached to the surface of the roadway.
The second, an over-roadway sensor is one that is mounted above the surface of the roadway in one
of the following two ways [46]:
• Above the roadway itself.
• Alongside the roadway, offset from the nearest traffic lane by some distance.
This section will cover the detector types presented in Table 2.
Table 2 - Detector types
2.2.1. Inductive-loop detector
Induction loops are the most widely used technology for vehicle detection worldwide [47]. An Inductive-
loop detector consists of one or more turns of wire embedded in the pavement and is connected to a
control box. An inductive-loop detector senses the presence of a conductive metal object by inducing
currents in the object, which reduce the loop inductance [46]. When a vehicle passes over or stops
above the loop, the inductance of the loop is reduced. This causes a detection signal in the control box.
The shape and size of the loop detector depends on the area to be detected, the types of vehicles to
be detected and the objective, such as queue detection, vehicle counting or speed measurements [45].
2.2.2. Magnetic detectors
Magnetic sensors are passive devices that detect the presence of a ferrous metal object through the
perturbation they cause in the Earth’s magnetic field [46]. The perturbation is also known as a magnetic
anomaly. Figure 7 and Figure 8 show the magnetic anomaly created by a ferrous metal vehicle. The
first, indicates how the vector addition of the dipole magnetic field to the quiescent Earth’s magnetic
field produces the magnetic anomaly [46]. The second, shows several dipoles on a vehicle and their
effect on compass readings and sensor output [46].
Pavement Invasive Detector
Inductive loop Microwave Radar Acoustic
Magnetic Detectors Infrared Video Image Processor
Ultrasonic Floating Car Data
Non-Pavement Invasive Detector
2.2. Vehicle detector data approach 13
Figure 7 - Magnetic anomaly induced in the Earth’s magnetic field by a magnetic dipole [46]
Figure 8 - Perturbation of Earth’s magnetic field by a ferrous metal vehicle [46]
Two types of magnetic field sensors are used to detect traffic: The magnetometer and the magnetic
detector.
Magnetometer
A magnetometer measures the changes in both the horizontal and vertical components of the earth’s
magnetic field caused by a passing vehicle. This is similar to the inductive loop detector, except that it
consists of a coil of wire wrapped around a magnetic core. The magnetometer can detect passing
vehicles, as well as the presence of a vehicle. Magnetometers are useful on bridge decks and viaducts,
where the steel support structure interferes with loop detectors, and loops can weaken the existing
structure [45].
Magnetic detector
The magnetic detector detects the vehicle signature by measuring the distortion in the magnetic flux
lines induced by the change in the Earth’s magnetic field produced by a moving ferrous metal vehicle
[46]. These detectors can only detect moving vehicles, consequently they cannot be used as a presence
detector [45]. However, multiple units of some magnetic detectors can be installed and utilized with
specialized signal processing software to generate vehicle presence data [46].
2.2.3. Microwave radar detector
Microwave radar was developed for detecting objects in the period before and during World War II
[46]. The term radar is an acronym for RAdio Detection And Ranging (RADAR). Originating from military
applications, the radar technology is now also used for traffic data collection. The microware radar
transmitting electromagnetic signals and measures the energy reflected from a vehicle. With this
14 2. Literature review
technique the reflected signal from vehicles can be used to determine presence, passage, volume, lane
occupancy, speed, and vehicle length depending on the waveform transmitted by the radar sensor [46].
The two waveforms used are continuous wave and frequency modulated continuous wave. The first is
known as Doppler radar and can only detect flow and speed [45]. The second can also detect the
presence of a vehicle [45]. Figure 9 shows the transmission of energy by an overhead-mounted
microwave radar toward an area of roadway.
Figure 9 - Microwave radar operation [46]
2.2.4. Infrared detector
The infrared detector could be used in an active or passive state for traffic flow monitoring applications.
The energy captured by active and passive infrared sensors is focused by an optical system onto an
infrared-sensitive material mounted at the focal plane of the optics [46]. The measured data is
processed and analysed to extract the information used utilized for signal control.
Active infrared
The active infrared detector transmits energy and detects the wave that are reflected [45]. This type
provides vehicle count, presence, speed, occupancy, vehicle classification and the detection of
pedestrians.
Passive infrared
The passive infrared detector does not transmit energy, instead they detect energy from two sources
[46]:
• Energy emitted from vehicles, road surfaces, and other objects in their field-of-view.
• Energy emitted by the atmosphere and reflected by vehicles, road surfaces, or other objects
into the sensor aperture.
Passive infrared detectors provide less information compared to the active type. It detects vehicle count,
presence and occupancy.
Vehicle
Microwave
Radar Antenna
Sign bridge,
overpass, pole, or mast arm mounting
Reflected signal from vehicle can be used
to determine presence, passage, volume, lane
occupancy, speed, and vehicle length depending on the waveform transmitted by the radar sensor
Controller
cabinet
Power and
data cables
2.2. Vehicle detector data approach 15
2.2.5. Ultrasonic detector
The ultrasonic detector is similar to the microwave radar, except it transmits ultrasonic sound energy
waves instead of electromagnetic signals. It measures the distance that the reflected wave travels [45].
These measurements are processed to obtain data of vehicle presence, speed and occupancy.
2.2.6. Acoustic detector
The acoustic detector uses an array of microphones to determine vehicle passage, presence and speed
by measuring acoustic energy or audible sounds [46]. The sounds are produced by vehicular and from
the interaction of vehicle’s tires with the road [45].
2.2.7. Video Image Processor
Video cameras were introduced to traffic management for roadway surveillance based on their ability
to transmit closed-circuit television imagery to a human operator for interpretation [46]. Today a Video
Image Processor (VIP) is used to automatically analyse traffic and extract information. The video
camera can detect traffic and the images of the camera are digitized, processed and converted into
traffic data [45]. A video image processor can provide vehicle count, presence, speed, occupancy and
vehicle class. Video image processors that track vehicles may also have the capability to register turning
movements and lane changes [46].
2.2.8. Floating Car Data
The emergence of floating car systems was made possible by using wireless communication methods,
especially the Global Positioning System (GPS) [48]. The first Floating Car Data (FCD) based application
has evolved in the early 1990’s [48], [49]. Advanced Driver and Vehicle Advisory Navigation Concept
(ADVANCE) was launched in 1991 as a major test of a dynamic, in vehicle route guidance system in
the United States [50]. ADVANCE was first practically implemented in 1997 in the Gary-Chicago-
Milwaukee Corridor transport system, which is considered the first intelligent transport system [51].
The main benefit of floating car data is that every vehicle acts as moving sensor, therefore no additional
hardware is required on the roadway [49], [52]. Furthermore, floating car data benefits from maximum
flexibility as it can be extended over large areas with only a marginal increase in variable costs [53].
In addition, developments in communication and sensor technology create the possibility to send
additional vehicle information next to the existing floating car data. This is the second generation of
floating car data and labelled as Extended Floating Car Data (xFCD) [48]. Possible information xFCD
could provide is shown in Figure 10. xFCD is considered a promising development for future dynamic
traffic management [54].
16 2. Literature review
Figure 10 - Extended FCD examples [55]
Floating car data is part of the Cooperative Intelligent Transport Systems (C-ITS). C-ITS is a concept
in which mobile road users such as vehicles and the road side infrastructure get engaged in mutual
information exchange to align their behaviours and intentions such that traffic conditions can be
optimized [56]. A key enabling technology of C-ITS is wireless vehicle communication, covering vehicle-
to-vehicle (V2V) communication, vehicle-to-infrastructure (V2I) communication, infrastructure-to-
vehicle (I2V) communication and could collectively be referred to as V2x communication [57].
The C-ITS environment could be achieved via two connectivity options, Cellular (LTE) and IEEE 802.11p,
also known as WIFI-P (ITS-G5). A third option could be a hybrid concept, a combination of both LTE
and ITS-G5. The communications of both connectivity options are displayed in Figure 11. The key
difference is the direct communications among 802.11p equipped devices and the reliance on the
presence of the network for cellular based [57]. The three options for an C-ITS environment are
discussed in more detail below.
Figure 11 - Comparison between IEEE 802.11p and cellular connectivity pipes to the car [57]
LTE
The first option is LTE. The connectivity over the cellular network, currently 3G and 4G, is already widely
adopted in most countries in the world. Additionally, car manufacturers are today offering connected
car solutions as more and more cars are connected via mobile networks (3G/4G) [56]. These services
2.2. Vehicle detector data approach 17
are an internet connection for in-car infotainment services, over-the-air updates or connected
navigation. 5G communications will expand the possibilities of what mobile networks can do and extend
applications which are more demanding than can currently be supported by 4G [58]. Figure 12 shows
the development of 5G compared to 4G.
Figure 12 - 5G the development compared to 4G [59]
ITS-G5
The second option is ITS-G5. ITS-G5 is based on the IEEE 802.11p wireless access protocol, which is
an adjusted version of the widespread 802.11 technology, branded as WiFi-p [60], [61]. It defines the
overall vehicular communication protocol stack and matches the requirements for V2X communications
in C-ITS applications [62]. This means that the communication is based on broadcast ad hoc networks
to support direct V2V communication without the need of a communication network infrastructure of
third parties [56]. ITS-G5 is expected to be implemented in new production cars starting in 2019 [63].
Therefore, a large deployment is not realistic in the near future. Hence, smaller scale ITS-G5
deployments targeting specific well-defined groups and business cases are the first focus, for example
[56]:
• V2V communications supporting C-ACC and truck platooning. Such intervehicle applications do
not specifically require the PKI Security framework to be set up. For example, fleet owners can
apply alternative security solutions to allow platooning at least between vehicles of their own
fleet;
• Interactions between intelligent traffic control installations and vehicles belonging to specific
categories (e.g. busses and trams, emergency services, taxi’s, value transport vehicles, etc);
• Traffic light installations can be equipped with a camera and so facilitating ‘around the corner’
warnings also at low ITS-G5 penetration rates;
• Renewal of street light furniture could be combined with the installation of ITS-G5 in these
systems. This could support C-ITS services on traffic junctions where traffic lights are not
used.
18 2. Literature review
Combination of LTE and ITS-G5
Finally, the third option is a combination LTE and ITS-G5. The postponement of introduction of ITS-G5
into new cars to 2019 is an immediate delay factor for large scale deployment of C-ITS [56]. In addition,
the International standards for 5G are announced to be ready in 2020 [56]. The situation as developed
to date and the uncertainties which apply, has initiated discussions about a hybrid approach, leveraging
both ITS-G5 and cellular communications as supporting information channels for C-ITS applications
[56]. A hybrid approach would also allow to combine the strengths of both options. The advantages of
a combined approach are the following [56]:
• The service provider achieves a higher penetration / coverage as it invokes multiple
communication channels via cellular and ITS-G5;
• The cellular connection can provide a fast take-up of suitable services due to the redundant
availability and high adoption of mobile communication services. Hence the dependency on
ITS-G5 is strongly reduced;
• With cellular, the in-car terminal has either this as a primary channel during the low-penetration
stages or has a backup channel in case of interruptions in ITS-G5 reception;
• The cellular connection can be used to accommodate better coverage to support C-ITS security
functions (e.g. exchange of certificate information);
• Multiple uplink channels via cellular and ITS-G5 connectivity are available to drop probe data
which is collected by cars.
2.3. Priority strategy
A traffic light controller equipped to with a priority strategy requires vehicle detections as an input to
grant priority to specific vehicles. Traffic signal priority is in principle the concept of improving service
or reducing delay for specific traffic modes at signalized intersections. Traffic signal control systems
traditionally treat either the aggregated flow of traffic or each mode separately, as summarized in Table
3.
Table 3 - Traffic signal control treatments for different traffic modes [64]
The two most widely implemented traffic signal priority control systems are Emergency Vehicle Signal
Pre-emption (EVSP) and Transit Signal Priority (TSP). A third traffic signal priority control system is
freight signal priority, which can also be described as truck signal priority. Before discussing the
strategies in more detail, it is important clarify the difference of pre-emption and priority. In general,
both methods have the same goal. However, the difference is found in the way they achieve it. Signal
2.3. Priority strategy 19
priority modifies the normal signal operation process to better accommodate transit vehicles, while pre-
emption interrupts the normal process for special events [65]. First, section 2.3.1 describes the
emergency vehicle pre-emption systems. Second, section 2.3.2 elaborates the transit signal priority
methods. Finally, section 2.3.3 explores the truck signal priority.
2.3.1. Emergency Vehicle Signal Pre-emption
Emergency vehicles are considered as a special vehicle type, including ambulances, police vehicles and
fire trucks. These vehicles are the first responders at emergency’s and for this reason are required to
reach their destination as quickly as possible. One of the most time-consuming delays is the travel time
to the emergency location. Especially manoeuvring through signalized intersections causes delays and
involves major safety risks, both for the emergency vehicles and other vehicles at the intersection.
Hence, it is the challenge to ensure safe passage of an emergency vehicle and at the same time to
maintain safe and smooth traffic flow in the road network [66].
Instead of allowing emergency vehicles to pass intersections without following the traffic lights, it is
preferable to provide a green light at intersections. The most used method is Emergency Vehicle Signal
Pre-emption (EVSP) at the signalized intersection [67], [68]. Pre-emption generally involves a control
strategy that immediately switches from the current phase to a pre-selected phase for the first request
received [64]. emergency vehicles could request a signal pre-emption treatment via the following
methods: optical systems [69], acoustic systems [70], inductive loop technology, radio controlled
systems [71] or Global Positioning System (GPS) [72], [73].
In addition, several attempts have been made to reduce the response time of emergency vehicles and
minimize the impact on general traffic. Wang et al. [74] proposed a degree-of-priority based control
strategy for emergency vehicle pre-emption operation. The paper of Qin and Khan [66] reports two
control strategies. First, a real-time control strategy is that enables signal transitioning from normal
operation to emergency vehicle signal pre-emption, in order to provide the approaching emergency
vehicles a safe crossing over the intersection safely at its operating speed. Second, an optimal two-
phase control algorithm, consisting of a relaxation method and a stepwise search strategy, is used for
the signal transitioning back to normal operation. A different approach is proposed by Viriyasitavat and
Tonguz [75], a self-organized traffic control scheme using a different set of local rules at intersections.
As a result, the priority management of emergency vehicles is controlled is a self-organized manner.
2.3.2. Transit Signal Priority
Transit signal priority (TSP) control is widely used at signalized intersections has been recognized as a
practical strategy to improve the efficiency and reliability of transit operations [76]. In contrast to the
emergency vehicle pre-emption strategy, transit signal priority could provide a priority service treatment
within the normal signal operation. Consequently, a transit signal priority strategy has a significantly
lower impact on other traffic modes at an intersection, compared to the emergency vehicle pre-emption
strategy. In addition, transit service is typically much more frequent than emergency vehicle service,
20 2. Literature review
use of priority rather than pre-emption allows the system to maintain a higher level of performance
[65]. Based on different control methods, transit signal priority can be divided into two categories:
Passive priority and active priority.
Passive priority control operates without explicitly recognizing the actual transit vehicle presence and
uses predetermined signal timings to provide benefits to transit vehicles [76]. The timings are based
on historical data and an arrival distribution. Therefore, the success of passive priority strategies
depends on low fluctuation of the traffic volumes and deterministic dwell times of buses at stops [77].
Compared with passive priority methods, active priority systems use upstream detection systems and
can respond to transit vehicles in real time [78]. Approaching transit vehicles are detected and the
signal controller would provide the designated signal control methods. Furthermore, based on different
computational complexities, active priority can be categorized into rule-based priority and model-based
priority.
In a rule-based priority strategy the controller grants priority based on a series of constraints, for
example the presence of a transit vehicle and if the transit vehicle is behind schedule. The rule-based
priority could again be subdivided in unconditional and conditional priority. The general idea behind
unconditional priority is that a transit vehicle should always receive priority when approaching an
intersection, without consideration if the vehicle is behind schedule. In contrast, conditional priority
should consider more conditions to grant priority. In addition, to the actual presence of the transit
vehicle, a conditional priority request is only granted if the vehicle is behind schedule [78]. Additional
rules could be applied to reduce the impact on other traffic modes. First attempts involve limiting the
red truncation frequency, it is only allowed if the previous green was not extended [79]. A similar
concept is to consider the allowance of priority based on granted priority in previous cycles [80], [81].
Development of advanced signal control and traffic detection techniques led to more sophisticated
conditional control to account for the overall benefits of passenger cars and transit vehicles [76].
Model based priority control refers to a relatively new generation of priority schemes, which attempts
to achieve advanced operational objectives by means of adaptive signal control [82]. Possible objectives
are reducing total vehicle delay, total transit vehicle delay and total person delay. These objectives are
optimized in real-time according to performance criteria, which may include person delay, transit delay,
vehicle delay or a combination of the previous [65]. This method requires the controller to collect real-
time data of the intersection, examples are the transit location data and traffic conditions. Based on
the actuated signal control system, a real-time multi-objective transit signal priority controller was
developed [83]. The required input information is collected by the inductive loop detectors and
according to the objective function, which is the weighted summation of transit passenger delay,
passenger car delay and transit schedule delay, the transit signal priority controller is able to reduce the
total person delay of the entire intersection. Another approach is based on the model predictive control
system RHODES [33]. A hierarchical optimization strategy determines the durations of the phases
2.3. Priority strategy 21
considering the total vehicle delay, total transit delay and the transit schedule conditions while providing
transit priority [84].
A summarize of the key characteristics of different transit signal priority controls is found in Table 4.
Table 4 - Key characteristic of different transit signal priority controls [76]
2.3.3. Truck Signal Priority
In contrast to transit signal priority, which has the objective is to reduce transit delays, the objective of
truck signal priority (TkSP) is to reduce the overall traffic delay and environmental impact. Since truck
signal priority aims to provide a green light for approaching trucks, a reduction in the number of stops
could be achieved. As a result, improving the safety at signalized intersections and possibly stimulate
truck drivers to use specific routes.
Similar concepts as the transit signal priority could be applied for truck priority. However, there are
three key differences [85]. First, the arrival frequency of trucks is much higher. Second, the trucks do
not have relatively fixed schedules as transit vehicles. Third, the priority level of trucks is lower.
Consequently, trucks require a different priority approach to avoid a large negative impact on other
traffic modes. Early implementations of truck priority strategies have been specifically designed for high
speed rural intersections [86], [87]. These strategies had the objective to minimize the truck stops and
solely relied on loop detectors for truck detections. Recent developments of new detection solutions
and communication methods provide opportunities for active priority strategies, which require a real-
time detection of approaching vehicles [88]. Examples are truck detections using video cameras [89],
a priority system based on GPS and wireless communication [90], [91] and truck priority using
connected vehicle technology [92].
22 2. Literature review
In recent years, there has been an increasing amount of literature on the development of new truck
priority strategies. Ioannou [85] proposed two different controller strategies for truck signal priority: a
neural network-based controller and one based on integrated priority strategies. The first, uses a neural
network approach to model the vehicle delays based on different classes of vehicles and adaptively
controls the traffic signals. The second, uses a combined passive and active strategy, which grants
priority to truck under predetermined conditions. In contrast to previous mathematical prediction
models, Zhao [88] developed a truck priority system that uses a co-simulation-based optimization
control approach for traffic light control by using real time simulators to predict the traffic state.
Suthaputchakun and Sun [3] propose an adaptive traffic light scheduling scheme via two-way traffic-
light-to-vehicle communication for fuel consumption and carbon dioxide (CO2) emission reduction. In
addition, a priority framework to optimize a weighted traffic light schedule is proposed, by assuming
the weight of a truck two-times higher than a normal vehicle.
2.4. Conclusions
In chapter 2, extensive literature studies have been conducted about signalized intersection controllers
in order to present an overview of the state-of-the-art of signalized intersection controllers. The chapter
started with the different signalized intersection controller approaches. Afterwards, multiple ways of
detecting vehicles were discussed. Finally, the priority strategies were explored. Together these studies
provide important insights into the development of signalized intersections controllers. These will be
highlighted and presented alongside the design choices for each section below.
Section 2.1 presents four main types of signalized intersection controllers. The most promising findings
in the literature on signalized intersection controllers is the development of model predictive control
(MPC). Especially, the Self-Organizing Intersection Controller (SOIC) approach. Moreover, the
combination with traffic flow predictions using Long Short-Term Memory (LSTM) approach showed
major potential. Therefore, for this research an MPC is proposed, more specific the SOIC approach, in
combination with LSTM traffic flow predictions.
Section 2.2 provides an overview of the available vehicle detector data approaches. Literature on this
topic showed a rapid development of connected vehicle techniques. These techniques enable vehicle
detections based on Floating Car Data (FCD). In addition, extensive information could be shared with
the intersections controller by Extended Floating Car Data (xFCD), such as direction, speed and vehicle
type. Accordingly, it is suggested that floating car data will eventually replace conventional detection
technique, such as the inductive-loop detector. For this reason, floating car data will be included in this
research.
Section 2.3 presents three different priority strategies. Most of the research found on the topic of
priority strategy was focused on Emergency Vehicle Signal Pre-emption (EVSP) and Transit Signal
Priority (TSP). However, the research found on truck priority often used EVSP and TSP strategies as a
baseline in the development of their approach. An interesting approach used for truck signal priority in
2.4. Conclusions 23
a traffic-adaptive controller, is a weighted traffic light schedule. This approach allows a traffic light
controller to optimize the light schedule based on the vehicle type. For this reason, it is proposed to
use a weighted traffic light schedule as a basis of the truck signal priority controller in this research.
The selection of the proposed techniques will be used to develop a truck signal priority controller. To
summarize, the proposed truck priority strategy is a MPC. The MPC will have a SOIC approach and will
use a LSTM approach for traffic flow predictions. A floating car data vehicle detection technique will be
used in order to detect different vehicle types. Finally, a weighted traffic light schedule will be used to
enable truck signal priority.
25
3 3. DIRECTOR
This chapter describes the basic principles in the design of the state-of-the-art model predictive
controller DIRECTOR, which are required as preliminary knowledge in this thesis. DIRECTOR has a
Self-Organizing Intersection Controller approach and uses Long Short-Term Memory traffic flow
predictions, which are the proposed techniques defined in the literature review of the previous chapter.
Further, DIRECTOR uses available floating car data to improve the traffic flow predictions. The floating
car data could be extended to provide additional information, e.g. the vehicle type and the Origin-
Destination, to the controller. Given the above, DIRECTOR includes the design choices made in the
previous chapter and could be used as the baseline throughout this research.
The main goal of DIRECTOR is to minimize the average travel time delay per vehicle, while still being
able to provide predictable and stable information on signal changes. Accordingly, the intuition behind
DIRECTOR is to use the cumulative travel time delay, which is based on the predictions of the arriving
traffic flows and the current queue length, to optimize the traffic light schedule and fix the schedule
ahead of time. Ultimately, enabling Advanced Driver Assistance Systems (ADAS), such as Green Light
Optimal Speed Advice (GLOSA). A more detailed description is found in the work of Van Senden [93].
The following sections describe the basic principles of DIRECTOR in an intuitive manner. First, the
cumulative travel time delay is introduced (section 3.1). Subsequently, a uniform distribution of arrivals
is assumed (section 3.2). Thereafter, the mathematical formulation of the schedule decision is
presented and extended with a switching penalty (section 3.3). Finally, the schedule will be fixed ahead
of time (section 3.4). The last section concludes the basic principles in the design of DIRECTOR (section
3.5). The relationship between the different sections is shown in Figure 13.
Figure 13 - Layout chapter 3: DIRECTOR
This chapter will answer the following sub question, presented in section 1.3:
(2) What are the basic principles of the state-of-the-art model predictive controller?
26 3. DIRECTOR
3.1. Cumulative travel time delay
The overall concept of DIRECTOR is based on the cumulative travel time delay, which is calculated by
multiplying the queue length with the waiting time. To illustrate, Figure 14 displays an example of how
a queue evolves in practice for one Origin-Destination. It shows the arrival of vehicles over ten seconds
and the relation between the queued vehicles and the cumulative travel time delay. The blue line, 𝜌(𝑡),
represents the number of queued vehicles and the blue shaded area represents the cumulative travel
time delay of the interval zero to ten, 𝐷𝑂𝐷(0,10). A mathematical form is given by (3.1).
𝐷𝑂𝐷(𝑡𝑠𝑡𝑎𝑟𝑡, 𝑡𝑒𝑛𝑑) = ∫ 𝜌𝑂𝐷
(𝑡)𝑡𝑒𝑛𝑑
𝑡𝑠𝑡𝑎𝑟𝑡
(3.1)
Figure 14 - Arrival of vehicles over 10 seconds [93]
3.2. Uniform distribution of arrivals
However, the future vehicle arrivals are unknown. It is therefore the challenge to estimate the future
arrivals. These vehicles can be predicted within a chosen time period according to a Long Short-Term
Memory traffic flow prediction model, using detector information of surrounding traffic controllers. The
chosen time period is considered as a time bin. It should be noted that the Long Short-Term Memory
does not predict the exact arrival times within the time bin. For this reason and without further regarding
the arrival patterns of vehicles, DIRECTOR assumes a uniform distribution of arrivals. Van Senden [93]
introduces an average queue length over a ten second interval, since it will reduce the computational
complexity required to calculate the travel time delay. This method is similar to the concept of SURTRAC
[35], which reduced the computational complexity by aggregating the arrivals to create clusters of
arriving vehicles. The average queue length will be denoted as 𝜌𝑂𝐷[𝑇], where the index T is a time bin
of a ten second interval. The example of Figure 14 is extended to 20 seconds in Figure 15 and illustrates
the transition from 𝜌(𝑡) to 𝜌[𝑇] at two time bins. For example, 𝜌[0] and 𝜌[1] in Figure 15 correspond
to the average queue length value of 4 and 7.5 respectively.
3.3. Schedule decision 27
Figure 15 - Transition from ρ(t) to ρ[T] [93]
3.3. Schedule decision
According to the arrivals described in the previous section, a schedule decision is made. The scheduled
Origin-Destination at time bin T, denoted as 𝜒[𝑇], is based on the Origin-Destination with the highest
priority at time bin T, denoted as 𝜋[𝑇]. The mathematical formulation of the schedule decision is given
by (3.2).
𝜒[𝑇] = arg max𝑂𝐷∈𝑂𝐷𝑠
{𝜋[𝑇]} = arg max𝑂𝐷∈𝑂𝐷𝑠
{ 𝜌𝑂𝐷[𝑇]} (3.2)
However, an intersection often has several non-confliction Origin-Destinations which can be serviced
simultaneously as a Phase Group. Therefore, the cumulative travel time delays of the Origin-
Destinations within a Phase Group are added together. The representation of the schedule decision
based on the Phase Groups is given by (3.3).
𝜒[𝑇] = arg max𝑃𝐺∈𝑃𝐺𝑠
{𝜋[𝑇]} = arg max𝑃𝐺∈𝑃𝐺𝑠
{ ∑ 𝜌𝑂𝐷
[𝑇]
𝑂𝐷∈𝑃𝐺
} (3.3)
The previous schedule decision (3.3) assumes the possibility of an instant switch between Phase
Groups. However, an instant switch is not allowed for safety reasons. It is required to have a period
between termination of a green signal and the next signal turning green. This period is referred to as
intergreen time. During this time no vehicles can be serviced by the intersection. Hence, switching
phase groups leads to potential inefficiencies at the intersection. Therefore, a switching penalty is
introduced, denoted as σ𝑃𝐺. The idea is that this penalty reduces the priority of the phase groups
currently not being serviced and therefore increases the relative importance of the phase group
currently being serviced. An illustration of the switching penalty is found in Figure 16. The red marked
area (first 4,5 sec) represents the time where no vehicles can be serviced.
28 3. DIRECTOR
Figure 16 – Illustration of the switching penalty [93]
The scheduling decision including the switching penalty is given by (3.4).
𝜒[𝑇] = arg max𝑃𝐺∈𝑃𝐺𝑠
{𝜋[𝑇]} = arg max𝑃𝐺∈𝑃𝐺𝑠
{( ∑ 𝜌𝑂𝐷
[𝑇]
𝑂𝐷∈𝑃𝐺
) − σ𝑃𝐺[𝑇])} (3.4)
3.4. Fixed schedule ahead of time
The schedule decision equation of the previous section responds ad hoc on the predicted arrivals of the
next time bin. However, DIRECTOR aims to provide predictable and stable information on signal
changes, which is not guaranteed in the ad hoc situation. For this reason, DIRECTOR will fix the
schedule ahead of time by using the predicted arrivals in future time bins and the current queue length.
The fixed schedule enables DIRECTOR to provide predictable and stable information on signal changes.
Accordingly, the schedule decision is given by (3.5).
𝜒[𝑇 + 1] = arg max𝑃𝐺∈𝑃𝐺𝑠
{𝜋[𝑇 + 1]} = arg max𝑃𝐺∈𝑃𝐺𝑠
{( ∑ 𝜌𝑂𝐷
[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
) − σ𝑃𝐺[𝑇 + 1]} (3.5)
3.5. Conclusions
This chapter served as the basis for the development of the truck signal priority controller design. The
basic principles and terminology of DIRECTOR are described and will be used in the remainder of this
thesis. DIRECTOR is a state-of-the-art model predictive controller, which is able to provide predictable
and stable information on signal changes. This is accomplished by fixing the schedule ahead of time,
based on the predicted arrivals and the current queue length. To reduce the computational complexity,
an average queue length over a ten second interval is used. Further, a switching penalty is included to
account for the inefficiencies related to switching phase groups. However, DIRECTOR is not able to
make a schedule decision based on different vehicle types. This is considered as an opportunity to
improve the controller. For this reason, the next chapter will propose a new design based on DIRECTOR
with a truck signal priority strategy.
29
4 4. Truck Signal Priority
controller design This chapter describes the design of the Truck Signal Priority (TkSP) controller. The work of this
research is based on the predictive controller DIRECTOR developed by Van Senden [93] described in
chapter 3. The control algorithm is modified and extended with the ability to make a schedule decision
based on different vehicle types. In this research, the main focus is the truck signal priority. For this
reason, only cars and trucks will be part of the study.
The following sections elaborate on the proposed design of DIRECTOR TkSP, starting with a single
Origin-Destination and step by step extend to multiple Origin-Destinations in a Phase Group. First, the
truck arrivals are described for a single Origin-Destination (section 4.1). Subsequently, a set of priority
weights is proposed (section 4.2). Next, the equation to calculate the truck weight for one Origin-
Destination is defined (section 4.3). Thereafter, the new schedule decision including the truck weight
is described (section 4.4). Finally, the Truck Signal Priority controller design is concluded in the last
section (section 4.5). The relationship between the different sections is shown in Figure 17.
Figure 17 - Layout chapter 4: Truck Signal Priority controller design
This chapter will answer the third sub question, presented in section 1.3:
(3) How should the priority weights be assigned to different vehicle types?
4.1. Truck arrivals
Prioritization of specific vehicle types requires the signalized intersection controller to detect different
vehicle types. Literature described multiple detection techniques, which are able to detect different
vehicle types. Especially, floating car data showed potential. The main benefit of floating car data is
that every vehicle acts as moving sensor, therefore no additional hardware is required on the roadway
30 4. Truck Signal Priority controller design
[49], [52]. In addition, floating car data has the possibility to send additional information, such as
direction, speed and vehicle type. Therefore, the method used to detect an approaching truck is based
on floating car data, as proposed in chapter 2. The range of the detection is set to a distance of 330
m, based on the minimum distance between the intersection under control and an upstream intersection
in the network. Alternatively, the selected distance could also be extended to a larger distance, which
enables earlier knowledge of an approaching truck. However, an increased detection distance would
increase the uncertainty of the estimated time of arrival. If a larger detection distance is required, a
truck could update its floating car data more frequently to reduces the uncertainty of the estimated
time of arrival. For simplification is the floating car data of the truck limited to one update throughout
this research. Accordingly, the estimated time of arrival is calculated, which is based on the distance to
the intersection and a speed limit of 50 km/h. Resulting in an estimated time of arrival of approximately
24 seconds. For this reason, three 10 second time bins are proposed, the index of the time bin is
denoted as T. Figure 18 illustrates the point of detection and the corresponding time bins in an approach
of the intersections.
Figure 18 - Point of detection
It should be noted that the count for the time bins starts at zero on the stop line of an intersection and
increases in steps of one moving upstream. The truck arrivals, denoted 𝜏, will be assigned to a time
bin, as shown in Figure 19. Equation (4.1) gives the result of the combined truck arrivals of a single
Origin-Destination.
Figure 19 - Truck arrivals per time bin
4.2. Priority weights
According to the truck arrival time bins, discussed in the previous section, a set of priority weights is
proposed. This set will correspond to the first three time bins, since the trucks have an estimated
arriving time at the intersection within 30 seconds. In the future this set could be extended to more
𝜏[𝑇] = [0, 0, 1] (4.1)
4.2. Priority weights 31
time bins. However, it should be noted that if more time bins are used the uncertainty would increase
as well. The set of priority weights will be denoted as 𝜔. Table 5 illustrates the index corresponding to
the time bin, which follows the same structure as defined for the truck arrivals in the previous section.
Table 5 - Priority weight index to corresponding time bin
𝜔 [0] 𝜔 [1] 𝜔 [2]
Time [s] (0,10) (11,20) (21,30)
The assigned weights are based on the differences in vehicle characteristics. Suthaputchakun and Sun
[3] make a connection between different types of vehicles based on their actual weight. For example,
heavily loaded vehicles normally have higher emissions and consume more fuel. For this reason,
Suthaputchakun and Sun assumed that the weight of a heavily loaded vehicle is two-times higher than
that of the small vehicles. However, more vehicle characteristics are found in the literature. For instance,
the length of a truck is around 1.5 to 4 times the length of a standard car [10]. Further, significant
differences are found in the vehicle dynamics, especially in the acceleration rates after a complete stop.
A typical truck has an acceleration rate around five times lower compared to a passenger car, when
accelerating to 50 km/h [11]. Combining the vehicle characteristics above, an assumption is made on
the impact of a truck at an intersection compared to a standard car. The following characteristics are
taken into consideration to determine the impact: weight (2), length (4) and acceleration rate (5). The
values correspond to the number of times a truck has more impact compared to a standard car.
However, not all defined vehicle characteristics are taken as equal. It is assumed that the acceleration
rate of a truck has two times more impact on the intersection. Since, all cars stopped behind a truck
are limited to the acceleration rate of the truck in front. The weighted average of the vehicle
characteristics above result that a truck has four times more impact at the intersection compared to a
standard car. Hence, an extra weight of four is given to an arriving truck at a specific time bin. A truck
is required to anticipate on red light earlier than a regular car, due to their slow dynamics. Therefore,
it is assumed that a truck will experience delays, if the signal is not green in the ten seconds before a
truck arrives. The extra weight will be added to the second time bin to ensure a smooth passage for a
truck crossing the intersection. Consequently, the set of priority weights is given in (4.2).
𝜔[𝑇] = [1, 4, 1] (4.2)
However, queues could already exist in front of an arriving truck. In the same way as stopping for a
red light, a truck has to anticipate earlier on a queue. To account for the possible queues and clear
them before a truck is arriving, a second weight is added in the set of priority weights. However, this
second priority weight should not be equal to the value of the second time bin. Since, it is intended to
clear a queue in advance and not give early green. The value of the additional weight is therefore
proposed to be half of the value of the second time bin. It follows that the priority weight is added in
the third time bin, because this is the farthest time bin away from the intersection. Finally, the proposed
set of priority weights is found in (4.3).
32 4. Truck Signal Priority controller design
𝜔[𝑇] = [1, 4, 2] (4.3)
4.3. Truck weights
Following the previous sections, all the required is available to calculate the truck priority weight,
denoted as 𝛿. The basic idea is to multiply the truck arrival by the assigned priority weight, given by
(4.4).
𝛿 = 𝜏 ∗ 𝜔 (4.4)
However, one or more approaches of the intersection could have multiple Origin-Destinations. For this
reason, it is required to receive the Origin-Destination from the floating car data of the truck. In
addition, multiple trucks could approach the intersection on the same Origin-Destination. To find the
truck weight of one Origin-Destination, the weight of the approaching trucks should be summed.
Accordingly, (4.4) is rewritten to add the weights of three time bins together. The calculation of the
truck priority weight of a single Origin-Destination is described by (4.5).
𝛿𝑂𝐷[T] = ∑ 𝜏[𝑇] ∗ 𝜔[𝑇]
2
𝑇=0
(4.5)
Example 4.3.1.
According to (4.5) an example is presented. In this example, three Origin-Destinations will be reviewed
and the 𝛿𝑂𝐷[T] will be calculated based on the inputs given by Table 6 and Table 7. The results of the
example are presented in Table 8. As can be seen in the results, the highest truck priority weight is
found for Origin-Destination number two. It should be noted that Origin-Destination number three
receives a lower truck priority weight, despite having three instead of two trucks approaching the
intersection. This is because of the arrival times of the trucks. The first truck arrives in the first time
bin and does not receive additional priority weights, since it already experienced delays. For this reason,
the results are as expected.
Table 6 - Priority weights example 4.3.1.
𝜔[𝑇]
[1, 4, 2]
Table 7 - Truck arrivals example 4.3.1.
OD 𝜏[𝑇]
1 [0, 0, 1]
2 [0, 1, 1]
3 [1, 0, 2]
Table 8 - Results example 4.3.1.
OD 𝛿𝑂𝐷[T]
1 2
2 6
3 5
4.4. Schedule decision 33
4.4. Schedule decision
The previous sections limits the calculation of the truck weight to a single Origin-Destination. However,
multiple Origin-Destinations are often not-conflicting and can be serviced simultaneously. The set of
combined Origin-Destinations is described as a Phase Group. To find the truck weight of a Phase Group,
denoted as 𝛿𝑃𝐺, all Origin-Destinations within the Phase Group are summed up following (4.6).
𝛿𝑃𝐺[T] = ∑ ∑ 𝜏[𝑇] ∗ 𝜔[𝑇]
2
𝑇=0𝑂𝐷∈𝑃𝐺
= ∑ 𝛿𝑂𝐷[T]
𝑂𝐷∈𝑃𝐺
(4.6)
Finally, the schedule decision described by (3.5) in chapter 3 could be extended by (4.6) to find the
new schedule decision, including the extra weights for approaching trucks. The mathematical
formulation of the final schedule decision is described by (4.7).
𝜒[𝑇 + 1] = arg max𝑃𝐺∈𝑃𝐺𝑠
{ ∑ 𝜌𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
− σ𝑃𝐺[𝑇 + 1] + 𝛿𝑃𝐺[𝑇 + 1]}
= arg max𝑃𝐺∈𝑃𝐺𝑠
{ ∑ 𝜌𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
− σ𝑃𝐺[𝑇 + 1] + ∑ 𝛿𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
}
= arg max𝑃𝐺∈𝑃𝐺𝑠
{( ∑ 𝜌𝑂𝐷[𝑇 + 1] + 𝛿𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
) − σ𝑃𝐺[𝑇 + 1]}
(4.7)
4.5. Conclusions
This chapter proposed a new controller design with a Truck Signal Priority (TkSP) strategy. The goal
was to develop a controller design that is able to make a predicted schedule decision based on different
vehicle types and give priority to trucks. In order to reach the goal, different steps have been taken.
First, the truck arrivals have been specified to corresponding time bins for a single Origin-Destination.
Second, the priority weights were assigned to the defined time bins individually. The value of the
weights was assumed based on the impact of a truck at an intersection compared to a standard car.
To determine the impact the following vehicle characteristics were taken into consideration: the weight,
the length and the acceleration rate. However, not all defined vehicle characteristics were taken as
equal. It was assumed that the acceleration rate had two times more impact compared to the weight
and length. Consequently, the value was determine by a weighted average. A second value was
introduced to account for possible queues in front of an arriving truck, which was proposed to be half
of the value of the previous time bin. Subsequently, the truck weights were formulated for a single
Origin-Destination. Finally, the truck weights were determined for a Phase Group and combined with
the schedule decision of chapter 3. This resulted in the new schedule decision with the truck signal
priority strategy and completed the design of DIRECTOR TkSP.
35
5 5. Experimental setup
This chapter describes the experimental setup used to evaluate the performance of the proposed
controller in Chapter 4. There are two options to evaluate the performance of the proposed controller.
The first options is to evaluate the performance of the proposed controller with live traffic on the street.
However, this could lead to unsafe situations at the intersection. In addition, evaluation with live traffic
is not reproducible. The results obtained would therefore not give an equal comparison between
different controller configurations. The second option is to use a simulation environment. The simulation
environment provides a controlled environment, which allows to reproduce traffic flows and give an
equal comparison between different controller configurations. Further, the simulation environment
could create the simplified intersection geometry defined in section 1.4. Given the above, it is clear that
a simulation environment is the best option in this research and will be used to evaluate the
performance of the proposed controller.
The chapter starts by introducing a case study (section 5.1). Thereafter, the simulation tools required
to create, run and evaluate the performance of the controller are discussed (section 5.2). Subsequently,
the data required to run a simulation will be identified (section 5.3). Afterwards, the simulation model
is described in detail, including the intersection geometry, the communication to the controller and the
implemented measurements (section 5.4). Finally, the experimental setup is concluded in the last
section (section 5.5). The relationship between the different sections is shown in Figure 20.
Figure 20 - Layout chapter 5: Experimental setup
This chapter will answer the fourth sub question, presented in section 1.3:
(4) How could the intersection controller with priority weights be evaluated?
36 5. Experimental setup
5.1. Case study
The proposed controller of Chapter 4 is tested within a case study. The case study will use an
intersection that is located in the Netherlands near Hoofddorp in the province of Noord-Holland. The
name of the intersection is N201 – Verbindingsweg N205 and the number is 201234. The intersection
has a T-shape geometry and connects the N205 and N201. It has a feed of three surrounding
intersections. Each intersection is at driving distance of approximately thirty seconds. These
intersections are Spaarnepoort, Leenderbos and Verbindingsweg, and correspond to the numbers
201231, 201209 and 205195 respectively. Further, a simplification of the geometry is used in this
research. The lanes for pedestrians and cyclist are left out the network. Therefore, the intersection
under control has six Origin-Destinations and could serve multiple vehicle types. However, only two are
currently under investigation: cars and trucks. Each Origin-Destination has an own identification number
according to Dutch traffic conventions, which is why the identification numbers 1, 5 and 9 are not
present, as seen in Figure 21. This particular intersection was chosen by Helmy [38] based on the
following reasons: First, the availability of all required detectors and the lack of influential sources and
sinks. Second, representatives from the Province of Noord-Holland were quite supportive of this project
and were willing to provide any resources needed. Another reason to continue with the location is that
it is the first site where the province of Noord-Holland has agreed to test DIRECTOR on the street.
Figure 21 - Schematic overview of the case study intersection.
5.2. Simulation tools
The case study presented in the previous section is modelled in a simulation environment. Multiple
simulation tools are required to create, run and evaluate the performance of the controller. This section
will introduce the tools used for the simulation. First, the simulation software program PTV Vissim,
described in section 5.2.1. Second, in section 5.2.2 is the programming language Python introduced,
which is used to write the controller algorithm.
5.3. Simulation data 37
5.2.1. PTV Vissim
PTV Vissim is a powerful tool for traffic simulation and the main program used by Dutch Road owners
[94]. Therefore, one reason for use is that most intersections in the Netherlands are already created in
PTV Vissim. Another reason is the possibility of extensive evaluation and visualization methods. Finally,
a large part of the case study is available for evaluation in PTV Vissim and requires minor adjustments
for the implementation of the new controller.
PTV Vissim is mainly used through its graphical interface for simulation, evaluation and visualization
purposes. Additionally, PTV Vissim also provides a COM (Component Object Model) interface, which
enables control of functions in PTV Vissim via programming. A few examples of these functions are
adjusting simulation settings, vehicle behaviour, evaluation during or after a simulation run and signal
control. There are several programming languages that could be used in the COM interface e.g. C++,
Visual Basic or Python. The version of PTV Vissim used in this research is 10.00-12 (64 bit).
5.2.2. Python
The programming language used for this research is Python, which has been used in the previous work
of DIRECTOR. Continuing to use Python made it possible to implement the new design, without
rewriting the complete control algorithm. Python in combination with the COM interface has been the
main method to control the simulations in PTV Vissim and test the controller.
Python is an interpreted, object-oriented, high-level programming language with dynamic semantics
[95]. A major benefit of Python is that both the standard library and the interpreter are available free
of charge. Python supports the use of modules and packages, which enables program modularity and
code reuse. The modules and packages used in this research are found in Appendix B.
5.3. Simulation data
The simulation tools defined in section 5.2 require data of vehicle arrivals to run a simulation and
evaluate the controller performance. The aim of the simulation environment is to reflect the reality as
close as possible. For this reason, historic data of the intersection is used in the simulation to evaluate
the controller in a close to real scenario. However, the historic data does not contain information on
the vehicle type. Therefore, additional information on the truck arrivals is required. First, section 5.3.1
will elaborate on the historic data. After, section 5.3.2 describes how the additional information on the
truck arrivals is obtained.
5.3.1. Historic data
The historic data is received from the province of Noord-Holland and is provided in a V-Log data format.
A format in which Dutch traffic light controllers log their data [93]. The V-Log data only records changes
in the data stream, which contains the detectors states, signals states and the internal system state of
the traffic light controller. The historic data used in the simulation dates from January 2017 to May
38 5. Experimental setup
2017 and is compiled from the V-Log format to numerical data that can be used for machine learning
and simulation purposes [93]. The data set of vehicle arrivals is found in Appendix C.
5.3.2. Truck data
The data compiled from the V-Log data does not contain information on the vehicle type. Therefore, it
is required to create a realistic data set of truck arrivals. The truck arrival data is estimated from the
original detector data. This data contains the timestamps of when a detector becomes occupied and
when the detector is free again. These timestamps are compared to find the occupation time of a
detector. In this occupied time data is a three-point median search conducted to find expected truck
detections. If the median is equal or larger than nine seconds, then the point is marked as a truck
detection. Figure 22 shows the truck detections, marked with red dots, for a single Origin-Destination
in 24 hours. These truck detections are saved to a new data file, in the same structure as the original
detections, to use as an input for the truck simulations. The data set of truck arrivals is found in
Appendix C.
Figure 22 - Truck detections (marked with red dots)
5.4. Simulation model
The PTV Vissim model for the intersection of the case study presented in section 5.1 is available and
provided by the province of Noord-Holland. However, this model contains a larger network of multiple
intersections, which are not used in this research. Although they are not actively used in the simulation,
they use computational power of the computer to simulate the vehicles in the network. Therefore, all
unnecessary elements in the PTV Vissim network are removed to have an optimal network for a smooth
simulation. Subsequently, the model is adjusted to be able to replay historical data of vehicle arrivals.
The locations of the vehicle inputs are shown in Figure 23. Further, multiple vehicle detectors are used
in the model. These are displayed as blue rectangles in Figure 23. During the simulation the controller
reads the states of the detectors every 100 milliseconds. The change between a state of the detector
is used to count vehicles passing a detector. As each lane has two detectors, i.e. the arrival detector
5.4. Simulation model 39
and stop line detector, the controller can count arriving and departing vehicles. Combining information
of two detectors on the same lane enables the controller to calculate current queue length of each
Origin-Destination pair. Further, the controller is able to change the state of the signal heads in PTV
Vissim, which are shown as red strip in Figure 23. Due to limitations of the simulation software the
changes are limited to once every second. Subsequently, measurement points are added for each
individual Origin-Destination, in order to evaluate the controller on the following outputs:
- Number of stops per vehicle - Vehicle delay [s]
The measurements points are vehicle travel time detectors in PTV Vissim. This detector type measures
the time it takes a vehicle to travel from one point to the next. The locations of the travel time detectors
are displayed in Figure 23, where the pink line is the start of the measurement and the green line is
the end of the measurement. The following outputs can be calculated with these measurements: the
vehicle travel delay, the number of stops and the number of vehicles passing. A delay of a vehicle is
calculated when the actual travel time is compared to the travel time it would need under free flow
conditions. Free flow conditions are considered as for example when the vehicle can maintain its desired
speed, i.e. without reacting on another vehicle or a red signal. In addition, each individual Origin-
Destination of the measurement could be specified for the different vehicle types. Hence, evaluations
could be reviewed for three configurations: all vehicles, only cars and only trucks.
Figure 23 - PTV Vissim network layout of the intersection
Further, node evaluations are used to determine the exhaust emissions. The basis for the emission
calculations are formed by standard formulas for consumption values, as well as data on emissions
incorporated in PTV Vissim. The data refers to a typical North American vehicle fleet and does not
differentiate between individual vehicle types. As can be seen in Figure 23, one rectangular node is
used and includes the entire intersections. Four outputs will be obtained through the evaluations:
carbon monoxide (CO) emissions, nitrogen oxide (NOx) emissions, Volatile Organic Compounds (VOC)
emissions and fuel consumption. The first three are measured in grams and the fuel consumption is
measured in US liquid gallon.
40 5. Experimental setup
5.5. Conclusions
This chapter presented an experimental setup, which is developed to evaluate the performance of the
proposed controller with priority weights. A case study is used based on an existing intersection in the
province of Noord-Holland, the Netherlands. Historic data of vehicle detections enabled the case study
to evaluate the controller in a close to real scenario. However, the truck arrivals were extracted from
the original historic data, due to the unavailability of historic truck data. The layout of the intersection
was modelled in the traffic simulation program PTV Vissim. Via the COM interface could the controller,
written in the programming language Python, control the signal states in the simulation environment.
Several measurement points were added to evaluate the performance, based on the number of stops
per vehicle and the vehicle delay. Further, node evaluations were used to determine the exhaust
emissions and fuel consumption. Finally, the experimental setup is complete and able to evaluate the
performance of the proposed controller with priority weights.
41
6 6. Simulation results
This chapter describes and discusses the results from the simulations of the case study presented in
the previous chapter. Multiple weight configurations will be simulated to evaluate the performance of
the Truck Signal Priority controller. The following sections present the results in an intuitive manner.
First, a comparison is made between the baseline and the proposed weights of chapter 4 (section 6.1).
Subsequently, a sensitivity analysis is performed to determine an optimal configuration of priority
weights (section 6.2). Thereafter, the results are compared for a full week (section 6.3). Afterwards,
an emission analysis is described for three different priority weights configurations (section 6.4). Finally,
the obtained results are discussed (section 6.5). The relationship between the different sections is
shown in Figure 24.
Figure 24 – Layout chapter 6: Simulation results
This chapter will answer the following sub questions, presented in section 1.3:
(5) Does a change in priority weights affect the performance in terms of traffic flow?
(6) Can a model predictive controller with priority weights improve the performance of a state-of-
the-art model predictive controller in terms of reducing the number of stops?
(7) How does the controller perform at different traffic demands?
6.1. Truck Signal Priority performance
To evaluate the performance of the new design for truck signal priority, simulations following the case
study of the previous chapter are conducted. These simulations include two sets of different weights
for trucks. The first, is the baseline of [1, 1, 1]. In this configuration no extra weight is added to a truck
in the schedule decision. The second, is the proposed set of [1, 4, 2]. In this configuration, extra weight
is added to a truck in the schedule decision. Here, a truck is equal to four cars in the second time bin
and equal to two cars in the third time bin. Both configurations are simulated with a data set of 24
hours and have the same vehicle inputs, presented in Table 9. The simulations are evaluated according
42 6. Simulation results
the Key Performance Indicators (KPI), described in section 1.2. These are the number of stops and the
vehicle delay. Both can be further specified, by dividing the key performance indicators in three vehicle
groups. Each group is a different configuration of vehicle types. The first contains all the vehicles, which
is a mix of cars and trucks. The second contains only cars and likewise the third contains only trucks.
The results of the two simulations are found in Figure 25, Figure 26 and Figure 27. The graphs present
the results per Origin-Destination for each vehicle group, which are: All, Cars and Trucks respectively.
Further, the key performance indicators are found in the graphs. The left axis is the percentage of stops
and the right axis is the average vehicle delay in seconds. The total number of vehicles per vehicle
group, as well as the number of vehicles per Origin-Destination, are presented in Table 9.
As can be seen in Figure 25 and Figure 26 the stop percentages of the vehicle groups All and Car have
a minor difference. However, a more distinct difference, with a range of 2% to 19%, is noticed in Figure
27 for the different Origin-Destinations. This indicates that the priority weights have a different impact
on the stop percentage for each Origin-Destination. A similar trend is found for the average vehicle
delay. The results of the simulation with the priority weights show a decrease in average vehicle delay
in comparison to the baseline. Finally, the total stops and the total vehicle delay, for both simulations,
are shown in Table 10 and Table 11 respectively. The lowest values for each vehicle group are
highlighted yellow, which quickly shows similar results to the comparison per Origin-Destination. The
simulation with priority weights has 87 stops less for trucks and 265 stops more for cars. Further, the
total vehicle delay for trucks is reduced by 45 minutes (2673 seconds) and increased by 111 minutes
(6642 seconds) for cars.
Given the above, the introduction of priority weights for trucks has a positive influence on the key
performance indicators for the trucks. However, it has a minor negative impact for the cars.
Table 9 - Number of vehicles: 24 hours
Origin-Destination Total
2 3 4 6 7 8
All 12405 4979 5648 6463 7054 11505 48054
Cars 12304 4865 5458 6226 6983 11381 47217
Trucks 101 114 190 237 71 124 837
Table 10 – Total stops
Total stops
All Car Truck
[1, 1, 1] 23239 22750 489
[1, 4, 2] 23417 23015 402
Table 11 – Total vehicle delay
Total vehicle delay [s]
All Car Truck
[1, 1, 1] 653197 637582 15615
[1, 4, 2] 657166 644224 12942
6.1. Truck Signal Priority performance 43
Figure 25 - Comparison: All
Figure 26 - Comparison: Cars
Figure 27 - Comparison: Trucks
44 6. Simulation results
6.2. Priority weights sensitivity analysis
As can be seen in the previous section, the use of priority weights has an positive influence on the
performance. Hence, it is expected that a change in the priority weights would change the performances
as well. To analyse the sensitivity of the priority weights a Design of Experiments (DoE) is proposed.
This method aims to describe the results under a variation of conditions. By introducing one change in
the set of priority weights it is expected to result in a change in one or more output variables. Following,
the priority weights are chosen as input variables. The output variables contain the key performance
indicators, as described in section 6.1 and include the number of stops and vehicle delay. Based on the
results described in the previous section, a design of experiments is proposed with the following
changes of input variables: The set of priority weights will start at [1, 1, 1] and will change the variables
one by one. The first time bin is always one, as described in section 4.2. The other two will vary in a
range of eight values, starting at one. The used weight data set is illustrated in Table 12.
Table 12 – Priority weight data set
[1, 1, 1] [1, 2, 1] [1, 3, 1] [1, 4, 1] [1, 5, 1] [1, 6, 1] [1, 7, 1] [1, 8, 1]
[1, 1, 2] [1, 2, 2] [1, 3, 2] [1, 4, 2] [1, 5, 2] [1, 6, 2] [1, 7, 2] [1, 8, 2]
[1, 1, 3] [1, 2, 3] [1, 3, 3] [1, 4, 3] [1, 5, 3] [1, 6, 3] [1, 7, 3] [1, 8, 3]
[1, 1, 4] [1, 2, 4] [1, 3, 4] [1, 4, 4] [1, 5, 4] [1, 6, 4] [1, 7, 4] [1, 8, 4]
[1, 1, 5] [1, 2, 5] [1, 3, 5] [1, 4, 5] [1, 5, 5] [1, 6, 5] [1, 7, 5] [1, 8, 5]
[1, 1, 6] [1, 2, 6] [1, 3, 6] [1, 4, 6] [1, 5, 6] [1, 6, 6] [1, 7, 6] [1, 8, 6]
[1, 1, 7] [1, 2, 7] [1, 3, 7] [1, 4, 7] [1, 5, 7] [1, 6, 7] [1, 7, 7] [1, 8, 7]
[1, 1, 8] [1, 2, 8] [1, 3, 8] [1, 4, 8] [1, 5, 8] [1, 6, 8] [1, 7, 8] [1, 8, 8]
The results of the simulations are visualized in Figure 28 and Figure 29. The graphs show the difference,
between the baseline and the weight configuration, for the key performance indicators. The left axis is
the number of stops and the right axis is the vehicle delay in seconds. To clarify, a positive number of
stops indicates an increase in the number of stops compared to the baseline. More detailed information
on the number of stops and the total vehicle delay is found in Appendix E.
Figure 28 – Sensitivity analysis: 24 hours (1/2)
6.2. Priority weights sensitivity analysis 45
Figure 29 – Sensitivity analysis: 24 hours (2/2)
The overall objective is to reduce the number of stops for trucks. However, the implementation of
priority weights should have a minimal impact on the traffic flow of other vehicles. Figure 29 shows the
higher weight configurations, which reduces the number of stops for trucks compared to the baseline.
On the other hand, the number of stops for cars are increasing for higher weight configurations. Similar
results are found for the vehicle delay. For this reason, the results presented in Figure 29 do not meet
the objective. The lower weight configurations, presented in Figure 28, show promising results. Multiple
weight configurations show a decrease in the vehicle delay, compared to the baseline for both cars and
trucks. In addition, three weight configurations also show a reduction in the number of stops. Especially
the weight configuration [1, 3, 4]. This is also noticed Table 13 and Table 14, which present the total
stops and vehicle delay for three weight configurations: the baseline, the first proposed and the weight
configuration [1, 3, 4]. Compared to the first proposed weights, the new weight configuration reduced
the number of stops for a truck even more, to 344 total stops and the total vehicle delay is decreased
to 206 minutes (12362 seconds). Moreover, the number of stops and total vehicle delay for the other
vehicle groups are also decreasing. This concludes that the weight configuration [1, 3, 4] has an overall
better performance than the first suggested weights. Subsequently, the new weight configuration is
compared to the baseline. Based on the previous comparison it was clear it would outperform the
baseline in terms of the number of stops for trucks. In fact, the results showed that the weight
configuration [1, 3, 4] also outperforms the baseline for the other vehicle groups. The new weight
configuration reduced the total stops by 183 for cars and 143 for trucks. Further, the total vehicle delay
for cars is decreased by 59 minutes (3558 seconds) and 54 minutes (3253 seconds) for trucks.
Table 13 – Total stops: Sensitivity analysis
Total stops All Car Truck
[1, 1, 1] 23239 22750 489
[1, 4, 2] 23417 23015 402
[1, 3, 4] 22911 22567 344
Table 14 - Total vehicle delay: Sensitivity analysis
Total vehicle delay [s] All Car Truck
[1, 1, 1] 653197 637582 15615
[1, 4, 2] 657166 644224 12942
[1, 3, 4] 646386 634024 12362
46 6. Simulation results
Following the same approach as section 6.1, the new weight configuration is compared in more detail.
Accordingly, the weight configuration [1, 3, 4] is compared per Origin-Destination with the baseline
configuration [1, 1, 1] and the first proposed configuration [1, 4, 2]. The comparisons for the vehicle
groups All, Cars and Trucks are found in Figure 30, Figure 31 and Figure 32 respectively.
Figure 30 - Sensitivity analysis comparison: All
Figure 31 - Sensitivity analysis comparison: Cars
Figure 32 - Sensitivity analysis comparison: Trucks
6.3. Robustness check 47
Again, comparisons of the vehicle groups All and Car, Figure 30 and Figure 31 respectively, show a
minor difference. While the comparison in Figure 32 shows a more distinct difference in the key
performance indicators. Especially in the decrease in the percentage of stops. Further, the average
vehicle delay is lower for both priority weight configurations compared to the baseline. Although, the
decrease in average vehicle delay is less evident between weight configurations [1, 4, 2] and [1, 3, 4].
Finally, it can be concluded the weight configuration [1, 3, 4] has an overall better performance, in
terms of the key performance indicators, compared to the baseline and other weight configurations.
However, the simulations are limited to 24 hours and one set of vehicle inputs.
6.3. Robustness check
The previous section performed a sensitivity analysis for the priority weights. However, it was limited
to 24 hours and one set of vehicle inputs. Different sets of vehicle inputs should be evaluated to check
the robustness of the weight configuration [1, 3, 4], found in the previous section. Therefore, six
additional days have been simulated, each having a different set of vehicle inputs of 24 hours. The sets
vehicle inputs can be found in Appendix C. The simulation results of the seven days are added together,
in order to evaluate the performance over a longer period. The number of vehicles over a week are
presented in Table 15. Subsequently, the results could be evaluated for a complete week, as shown in
Figure 33 and Figure 34. The results for the individual days can be found in Appendix D. The overall
objective is similar to the previous section, which quickly reduces the number of suitable weight
configurations. In fact, only one weight configuration meets the objective, [1, 2, 1]. Compared to the
baseline it reduces the total stops by 77 for cars and 88 for trucks, and decreases the total vehicle
delays for cars by 61 minutes (3673 seconds) and 66 minutes (3957 seconds) for trucks. Although, the
weight configuration [1, 2, 1] meets the objective, the results are not impressive. Since it only reduced
the total stops for trucks by 88 over a week. However, the weight configuration [1, 3, 4] shows a more
significant reduction in the total number of stops for trucks over a week, while the increase in the total
number of stops for cars remains minimal.
Figure 33 – Sensitivity analysis: Week (1/2)
48 6. Simulation results
Figure 34 – Sensitivity analysis: Week (2/2)
Alternatively, the priority weight configuration [1, 3, 4] could also be evaluated for the individual days.
For instance, Figure 40 and Figure 42, found in Appendix D, show an decrease in total vehicle delay for
cars, while Figure 43 and Figure 44 show an increase in vehicle delay for cars. In the same way, an
increase in the total stops for cars is found in Figure 41 and Figure 43. Further, Figure 44 shows only
a minor deviation to the baseline, compared to the other simulations. However, it should be noted that
this simulation has the least vehicle inputs. The above suggest that the performance of the priority
weight configuration, in terms of total stops and vehicle delay, is depended on the traffic demand.
Finally, it can be concluded that the weight configuration [1, 3, 4] reduces the total of stops for trucks
by 751 over a week, compared to the baseline [1, 1, 1]. However, the total stops for cars increases by
155. As can be seen in Table 17, the total vehicle delay over a week shows a similar trend as the total
stops. Namely, the total vehicle delay for trucks is decreased by 304 minutes (18227 seconds) and
increased by 201 minutes (12033 seconds) for cars. Given the above, it is noticed that both the total
stops and the total vehicle delay decreases more for trucks than it increases for cars. As a result, the
vehicle group All shows an improvement of the key performance indicators for the weight configuration
[1, 3, 4], compared to the baseline.
Table 15 - Number of vehicles: Week
Origin-Destination Total 2 3 4 6 7 8
All 81954 30138 35030 41641 44135 73159 306057
Cars 81405 29582 34007 40300 43781 72499 301574
Trucks 549 556 1023 1341 354 660 4483
Table 16 – Total stops: Week
Total stops All Car Truck
[1, 1, 1] 142225 139404 2821
[1, 3, 4] 141629 139559 2070
Table 17 - Total vehicle delay: Week
Total vehicle delay [s] All Car Truck
[1, 1, 1] 3987494 3897966 89528
[1, 3, 4] 3981301 3909999 71301
6.4. Emission analysis 49
6.4. Emission analysis
In addition to the key performance indicators, it is also interesting to evaluate the performance of the
truck signal priority controller based on emissions and fuel consumption. Therefore, an emission
analysis is conducted for the following priority weights configurations: the baseline [1, 1, 1], the first
proposed configuration [1, 4, 2] and the overall best performing configuration [1, 3, 4]. Similar to the
section 6.3, a complete week will be evaluated based on seven individual days. Further, the emission
analysis will focus on the following outputs of the simulations: carbon monoxide (CO), nitrogen oxide
(NOx), Volatile Organic Compounds (VOC) and fuel consumption. These outputs are compared for each
weight configuration and presented in Figure 35, Figure 36, Figure 37 and Figure 38 respectively. The
data of the emission analysis can be found in Appendix F.
Figure 35 - Carbon monoxide emissions
Figure 36 – Nitrogen oxide emissions
Figure 37 - Volatile Organic Compounds emissions
Figure 38 - Fuel consumption
The figures above show a decrease in emissions and fuel consumption for the priority weight
configuration of [1, 3, 4], which was expected since the total stops decreased as well. In the same
way, an increase in the total stops for the weight configuration [1, 4, 2], shows an increase in the
emissions and fuel consumption. Further, it should be noted that the differences in the results between
the priority weight configurations are relatively small. Since, it was expected that the reduction in the
total stops for trucks would have a more distinct impact, by reducing the emissions and fuel
consumption.
50 6. Simulation results
6.5. Discussion
The following section will discuss the results of the previous sections; the truck signal priority
performance evaluation, the priority weights sensitivity analysis, robustness check and the emission
analysis.
6.5.1. Truck Signal Priority performance
The expected results for adding priority weights are confirmed by the results. Compared to the baseline
[1, 1, 1], the weight configuration [1, 4, 2] shows a decrease in stops and vehicle delay for trucks.
However, cars experience a minor increase in the total stops and the vehicle delay. Therefore, the
overall performance is negatively impacted in terms of total stops and total vehicle delay. This can be
explained because there are more cars in relation to trucks in the simulation. To find if it is possible to
reduce the number of stops and vehicle delay for trucks, without negatively impacting other vehicles,
more priority weight configurations should be evaluated.
6.5.2. Priority weights sensitivity analysis
By performing a design of experiments to analyse the sensitivity of the priority weights, it was expected
to find a relation between the priority weights configuration and the key performance indicators. Figure
29 shows the higher weight configurations, which reduces the number of stops for trucks compared to
the baseline. On the other hand, the number of stops for cars are increasing for higher weight
configurations. Similar results are found for the vehicle delay. It follows that the higher priority weight
configurations do not have the desired results, since they negatively impact the traffic flow.
Another interesting result is noticed in Figure 28 and Figure 29, which shows that it is more beneficial
to have a higher value for the third time bin in combination with a lower value for the second time bin.
This finding indicates the importance to clear a possible queue in front of the arriving truck. In order
to cross the intersection without a stop. However, the weight values 6, 7 and 8 in the third time bin
also show a major increase in the totals stops and total vehicle delay for cars. This may indicate that
the trucks are simply given too much priority weight before they arrive at the intersection. On the other
hand, it could also indicate that a weight is added to clear a queue, while no queue is present. A possible
solution would be to first check the current queue and only add the priority weights if a queue is
present.
Further, Figure 28 shows a clear reduction for the priority weight configuration [1, 3, 4] based the key
performance indicators, for both cars and trucks. For this reason, the weight configuration [1, 3, 4] is
added in the comparison of section 6.1, which showed an improvement for all vehicle groups compared
to the first proposed priority weight configuration [1, 4, 2]. Moreover, a decrease in the total stops and
total vehicle delay was also noticed compared to the baseline [1, 1, 1]. Finally, it can be concluded the
weight configuration [1, 3, 4] has an overall better performance, in terms of the key performance
indicators, compared to the baseline and other weight configurations. However, the simulations are
6.5. Discussion 51
limited to 24 hours and one set of vehicle inputs. Therefore, different sets of vehicle inputs should be
evaluated to check the robustness of the priority weight configuration [1, 3, 4].
6.5.3. Robustness check
Different sets of vehicle inputs were evaluated to check the robustness of the weight configuration
[1, 3, 4]. Therefore, the simulation results of the seven days were added together, in order to evaluate
the performance over a longer period. Subsequently, the results could be evaluated for a complete
week. The results showed one priority weight configuration that met the objective, which was
[1, 2, 1]. However, the results were not impressive. Since it only reduced the total stops for trucks by
88 over a week. On the other hand, the weight configuration [1, 3, 4] showed a more significant
reduction in the total number of stops for trucks over a week, while the increase in the total number of
stops for cars remained minimal. Subsequently, the priority weight configuration [1, 3, 4] was evaluated
for the individual days. The evaluation showed that the performance of the priority weight configuration,
in terms of total stops and vehicle delay, is depended on the traffic demand. This indicates that is would
be beneficial to have an dynamic priority weight configuration, which would adapt to the current traffic
demand. Another approach could be to implement of multiple priority weight configurations for each
Origin-Destination. The priority weight configurations could then be optimized per Origin-Destination,
since the traffic demand is often different for each Origin-Destination.
However, the results for the priority weight configuration [1, 3, 4] showed an overall improvement of
the key performance indicators over a week. Despite, increasing the total stops and total vehicle delay
for cars. It can be concluded that the proposed truck signal priority controller design can reduce the
number of stops for trucks at a signalized intersection, while maintaining the overall traffic flow at least
as good as a state-of-the-art model predictive intersection controller.
6.5.4. Emission analysis
In addition to the key performance indicators, an emission analysis was conducted to evaluate the
performance of DIRECTOR TkSP based on emissions and fuel consumption. The results of the emission
analysis showed a similar trend as the result of the key performance indicators. As can be seen in the
figures of section 6.4. These show a decrease in emissions and fuel consumption for the priority weight
configuration of [1, 3, 4], which was expected since the total stops decreased as well. In the same
way, an increase in the total stops for the weight configuration [1, 4, 2], shows an increase in the
emissions and fuel consumption. Further, it should be noted that the differences in the results between
the priority weight configurations are relatively small. Since, it was expected that the reduction in the
total stops for trucks would have a more distinct impact, by reducing the emissions and fuel
consumption. However, the available version of PTV Vissim, for this research, was not able to calculate
vehicle type specific emissions and fuel consumptions, which resulted in equal emissions for cars and
trucks. Hence, the evaluation can only be used to compare the emissions of different scenarios. For
this reason, the impact of a decrease in the total stops for trucks was not visible in the results.
53
7 7. Conclusions and recommendations
7.1. Conclusions
The research described the development of a new truck signal priority controller, to address the
increasing freight transport and improve the road traffic sustainability. The goal was to reduce the
number of stops for trucks at signalized intersections by granting priority to trucks via implementation
of priority weights for different vehicle classes. The implementation of this controller was expected to
reduce the number of stops for trucks and improve the traffic flow at signalized intersections. An
extensive literature study resulted in an overview of the state-of-the-art of signalized intersection
controllers. Findings in the literature provided the required techniques to develop the controller. The
proposed truck priority strategy was a Model Predictive Controller (MPC) with a Self-Organizing
Intersection Controller (SOIC) approach and used a Long Short-Term Memory (LSTM) approach for
traffic flow predictions. The detection of different vehicle types was based on Floating Car Data (FCD)
and a weighted traffic light schedule in combination with priority weights was used to enable truck
signal priority.
The proposed design was evaluated according to a case study, which was based on an intersection in
the province of Noord-Holland. The case study evaluated different configurations of priority weights
with the traffic simulation tool PTV Vissim. Historic data of the intersection was used in the simulation
to evaluate the controller in a close to real scenario. Each simulation provided detailed results, including
the vehicle delay and the number of stops. Further, node evaluation provided results on the following
emissions: carbon monoxide (CO) emissions, nitrogen oxide (NOx) emissions, Volatile Organic
Compounds (VOC) emissions and fuel consumption.
The results of this research showed that the implementation of priority weights for trucks has an
influence on the performance of the intersection controller, in terms of number of stops and vehicle
delay. Early results presented a comparison of a proposed priority weight configuration to a baseline
with no priority weights. Simulating a full day, it was found that the simulation with priority weights
54 7. Conclusions and recommendations
has 87 stops less for trucks and 265 stops more for cars (17,8% and 1,16% respectively). Further, the
total vehicle delay for trucks is reduced by 45 minutes and increased by 111 minutes for cars (17,1%
and 1,0% respectively). A sensitivity analysis was performed to further study the effects of the priority
weights. The results showed that the priority weight configuration [1, 3, 4] outperforms the baseline.
The total stops were reduced by 143 for trucks and 183 for cars (29,67% and 0,8% respectively), also
the total vehicle delay for truck was decreased by 54 minutes and 59 minutes for cars (0,6% and 20,8%
respectively).
Subsequently, a robustness check was performed to evaluate the performance of the priority weight
configuration over a week, compared to the baseline. The results showed an reduction of the total
stops by 751 and total vehicle delay by 304 minutes for trucks (26,6% and 20,4% respectively).
However, the total stops and total vehicle delay for cars increased, by 155 stops and 201 minutes (0,1%
and 0,3% respectively). The node evaluations, included in PTV Vissim, used for the emission analysis
were not vehicle type specific. Hence, the impact of the reduced total number of stops for trucks was
not visible in the results.
Finally, the results for the priority weight configuration [1, 3, 4] showed an overall improvement of the
key performance indicators over a week. The overall total stops and total vehicle delay were reduced
by 596 stops and 103 minutes (0,42% and 0,16% respectively). Despite, increasing the total stops and
total vehicle delay for cars. To answer the research question, a model predictive control-based
signalized intersection controller with priority weights can lead to a reduction in the number of stops
for trucks at a signalized intersection, while maintaining traffic flow at least as good as a state-of-the-
art model predictive controller.
7.2. Recommendations
Based on the results, several research directions could be indicated to improve the truck signal priority
controller design. However, due to the limited time available in this research these are considered part
of future work. The following recommendations can be made for future work:
- A future study might explore adding a dynamic priority weight configuration. It is shown that
the optimal priority weights configurations is influenced by the current traffic demand.
Therefore, machine learning is suggested to adjust the priority weights according to the current
traffic situation in order to maintain an optimal performance.
- Another approach to improve the performance of the controller could be the implementation
of multiple priority weight configurations for each Origin-Destination. The priority weight
configurations could then be optimized per Origin-Destination, since the traffic demand is often
different for each Origin-Destination.
7.2. Recommendations 55
- Further research can be done with a more advanced emission evaluation program. One that is
recommended is the add-on module EnViVer, which is based on the VERSIT+ exhaust
emissions model from TNO. This module enables to determine pollutant emissions based on
vehicle trajectories and other information from PTV Vissim.
- It would be interesting to see the effects of Green Light Optimal Speed Advice (GLOSA).
Currently, the controller only receives Floating Car Data (FCD) from the vehicles. Based on the
calculated schedule the controller could send GLOSA information back to the vehicles. It is
expected that the implementation of GLOSA will reduce the number of stops even more, for
both cars and trucks.
- This research is limited by using two types of road users. It would be interesting to see how
the controller performs with more types of road users. More research is required to incorporate
other road users into the simulation to further approach the reality.
- A future study might explore adding weights to other road users. For example, it could be
beneficial to give more priority to cyclist in city centres. Hence, it would stimulate the use of a
bicycle and discourages the use of cars.
57
References
[1] Nunzio, G.D., “Trafc eco-management in urban trafc networks,” Université Grenoble Alpes, 2015.
[2] Kaparias, I., Zavitsas, K., & Bell, M.G.H., “State-Of-The-Art of Urban Traffic Management Policies
and Technologies,” CONDUITS, Coordination Of Network Descriptors for Urban Intelligent Transport Systems, Imperial College London, 2010.
[3] Suthaputchakun, C. & Sun, Z., “A Novel Traffic Light Scheduling Based on TLVC and Vehicles’
Priority for Reducing Fuel Consumption and CO2 Emission,” IEEE Systems Journal, vol. 12, no. 2, pp. 1230-1238, 2018.
[4] Wu, L., Yusheng, C., Chu, J. & Zhang, H., “The Influence of Intersections on Fuel Consumption in Urban Arterial Road Traffic: A Single Vehicle Test in Harbin,” PLoS One, China, 2015.
[5] Brilon, W. & Zurlinden, H., “Ueberlastungswahrscheinlichkeiten und Verkehrsleistung als Bemessungskriterium fuer Strassenverkehrsanlagen (Breakdown Probability and Traffic
Efficiency as Design Criteria for Freeways),” Forschung Strassenbau und Strassenverkehrstechnik, Heft 870, 2003.
[6] Azar, C., Lindgren, K. & Andersson, B.A., “Global energy scenarios meeting stringent CO2 constraints—cost-effective fuel choices in the transportation sector,” Energy Policy, Volume 31, Issue 10,, pp. 961-976, 2003.
[7] Essen, Van E., “The Environmental Impacts of Increased International Road and Rail Freight
Transport, Past trends and future perspectives,” in Global Forum on Transport and Environment in a Globalising World 10-12 November 2008, OECD, ITF, Guadalajara, Mexico, 2008.
[8] Metz, N., “World Wide Emission Trend 1950 to 2050 Road Transport and all Sources,”
Eurochamp Workshop in Andechs, Chemistry, Transport and Impacts of Atmospheric Pollutants With Focus on Fine Particulates, EMITRADE-Herrsching, Germany, 2005.
[9] Metz, N., “Estimation of worldwide CO-, NMVOC- NOX- and PM-Emissions,” Proceedings of the
10th International Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, EMITRADE-Herrsching, Germany, 2005.
[10] Garber, N.J. & Hoel, L.A., Traffic & Highway Engineering, 4th edition, Cengage Learning, ch. 3, 2008.
[11] Institute of Transportation Engineers, Traffic Engineering Handbook, Washington DC: 5th Edition, ch. 5, 2000.
[12] Guberinić, S. & Minic, S.M., “Signal group: Definitions and algorithms.,” YUJOR. Yugoslav Journal of Operations Research. 3., 1993.
[13] Berg, M. van den, “Integrated Control of Mixed Traffic Networks using Model Predictive Control,” Ph.D. thesis, Delft University of Technology, Delft, The Netherlands, 2010.
[14] Katwijk, R.V., “Multi-Agent Look-Ahead Traffic-Adaptive Control.,” Delft University of Technology, Delft, (2008).
58 References
[15] Nunzio, G.D., “Traffic Eco-management in Urban Traffic Networks.,” Université Grenoble Alpes, 2015.
[16] Little, J., “The Synchronization of Traffic Signals by Mixed-Integer Linear Programming,” Operations Research, vol. 14, no. 4, pp. 568–594,, 1966.
[17] Little, J., Kelson, M. & Gartner, N., “MAXBAND. A Versatile Program for Setting Signals on Arteries and Triangular Networks,” Transportation Research Record: Journal of the Transportation Research Board, vol. 795,, 1981.
[18] Gartner, N., Assmann, S., Lasaga, F. & Hou, D., “A Multi-Band Approach to Arterial Signal
Optimization Traffic,” Transportation Research Part B: Methodological, vol. 25B, no. 1, pp. 55–74,, 1991.
[19] Gartner, N. & Stamatiadis, C., “MULTIBAND-96: A Program for Variable Bandwidth Progression Optimization of Multiarterial Traffic Networks,” Transportation Research Record, vol. 1554, pp. 9–17,, 1996.
[20] Robertson, D.I., “TRANSYT: A Traffic Network Study Tool, Road Research Laboratory Report, 1969.”.
[21] Maslekar, N., Mouzna, J., Boussedjra, M. & Labiod, H., “CATS: An adaptive traffic signal system based on car-to-car communication. J. Netw. Comput. Appl. 2013, 36, 1308–1315”.
[22] Hoogendoorn, S. & Knoop, V., “Traffic Flow Theory and Modelling; Edward Elgar Publishing Limited: Cheltenham, UK, 2012”.
[23] Zheng, X., Recker, W. & Chu, L., “Optimization of Control Parameters for Adaptive Traffic-Actuated Signal Control,” J. Intell. Transp. Sys, vol. 14, p. 95–108, 2010.
[24] Zheng, X. & Chu, L., “Optimal Parameter Settings for Adaptive Traffic-Actuated Signal Control,”
in In Proceedings of the International IEEE Conference on Intelligent Transportation Systems, Beijing, China, 12–15 October, 2008.
[25] Papageorgiou, M., Diakaki, C., Dinopoulou, V., Kotsialos, A. & Wang, Y., “Review of Road Traffic Control Strategies, Proceedings of the IEEE, vol. 91, no. 12, pp. 2043–2067, 2003.”.
[26] Blokpoel, R. & Niebel, W., “Advantage of Cooperative Traffic Light Control Algorithms,” ITS European Congress., 2016.
[27] Feng, Y., Head, K.L., Khoshmagham, S. & Zamanipour, M., “A real-time adaptive signal control
in a connected vehicle environment,” Transp. Res. Part C Emerg. Technol., vol. 55, no. 55, p. 460–473., 2015.
[28] Hunt, P.B., Robertson, D.I., Bretherton, R.D. & Winton, R.I., “SCOOT A Traffic Responsive Method of Co-Ordinating Signals,” TRRL Laboratory Report 1014, 1981.
[29] Lowrie, P.R., “The Sydney Coordinated Adaptive Traffic System (SCATS) - Principles, Methodology, Algorithms,” in International Conference on Road Traffic Signaling, 1982.
[30] Mauro, V. & Taranto, C., “Utopia,” in IFAC Proceedings Volumes Control, Computers, Communications in Transportations, 1989.
[31] Gartner, N., “OPAC: A Demand Responsive Strategy for Traffic Signal Control,” Transportation Research Record, vol. 906, p. 75–81, 1983.
[32] Henry, J.J., Fargas, J. & Tuffal, J., “The PRODYN Real Time Traffic Algorithm,” 4th IFAC-IFICIFORS Conference on Control in Transportation Systems, p. 305–310, 1984.
References 59
[33] Sen, S. & Head, K. L., “Controlled Optimization of Phases at an Intersection,” Transportation Science, vol. 31, no. 1, pp. 5-17, 1997.
[34] Boillot, F., Blosseville, J.M., Lesort, J.B., Motyka, V., Papageorgiou, M. & Sellam, S., “Optimal Signal Control of Urban Traffic Networks,” Road Traffic Monitoring, 1992.
[35] Xie, X., Smith, S.F., Lu, L. & Barlow, G.J., “Scheduledriven intersection control,” Transportation Research Part C: Emerging Technologies, vol. 24, p. 168–189, 2012.
[36] Lighthill, M.J. & Witham, G.B., “A theory of traffic flow on long crowded roads,” In Proceedings of the Royal Society A. On kinematic waves II, vol. 229, p. 317–345, 1955.
[37] Pacey, G.M., “The Progression of a Bunch of Vehicles Released from a Traffic Signal,” Road Research Laboratory, Berkshiye, 1956.
[38] Helmy, N., “An intelligent traffic flow progression model for predictive control applications,” Msc thesis, Delft University of Technology, Delft, 2017.
[39] Hochreiter, S., “Untersuchungen zu dynamischen neuronalen netzen,” Msc thesis, Technische Universität München, München, 1991.
[40] Hochreiter, S. & Schmidhuber, J., “Long short-term memory,” Neural computation, vol. 9, pp. 1735-1780, 1997.
[41] Lämmer, S., Donner, R. & Helbing, D., “Anticipative control of switched queueing systems,” The European Physical Journal B, vol. 3, no. 63, p. 341, 2008.
[42] Lämmer, S. & Helbing, D., “Self-control of traffic lights and vehicle flows in urban road networks,” Journal of Statistical Mechanics: Theory and Experiment, no. 04, 2008.
[43] Kumar, P.R. & Seidman, T.I., “Dynamic instabilities and stabilization methods in distributed real-
time scheduling of manufacturing systems,” IEEE Transactions on Automatic Control, vol. 3, no. 35, p. 289–298, 1990.
[44] Lämmer, S. & Helbing, D., “Self-stabilizing decentralized signal control of realistic, saturated network traffic,” Santa Fe Institute Working Paper, Santa Fe, 2010.
[45] Gordon, R.L. & Tighe, W., “Traffic Control Systems Handbook,” Transportation Management Federal Highway Administration, Washington, 2005.
[46] Klein, L.A., Mills, M.K. & Gibson, D.R.P., “Traffic Detector Handbook: Third Edition—Volume I,” Federal Highway Administration, Georgetown Pike, 2006.
[47] Manasse, H., “Level of Service estimation with in-vehicle sensor floating car data,” Technische Universiteit Delft, ITS Edulab, Delft, 2013.
[48] Banach, S., “Betrouwbaarheid en toepassingen van Floating Car Data,” Technische Universiteit Eindhoven (TUe), Design & Decision Support Systems (DDSS), Eindhoven, 2013.
[49] Tiedong, W. & Jinging, H., “Applying Floating Car Data in Traffic Monitoring,” 2014 IEEE International Conference on Control Science and Systems Engineering, Yantai, China, 2014.
[50] Turksma, S., “The various uses of floating car data,” in Tenth International Conference on Road Transport Information and Control, (Conf. Publ. No. 472) pp. 51-55, London, UK, 2000.
[51] Sen, A., Condie, H., Sheffey, A., Tian, X. & Zhu, X., “ADVANCE Advanced Driver and Vehicle Advisory,” University of Illinois, Chicago, 1996.
60 References
[52] Fabritiis, C. de, Ragona, R. & Valenti, G., “Traffic Estimation And Prediction Based On Real Time
Floating Car Data,” in Proceedings of the 11th International IEEE, Conference on Intelligent Transportation Systems, Beijing, China, 2008.
[53] Fedra, K., Greppin, H., Haurie, A., Hussy, C., Dao, H. & Kanala, R., “GENIE: An integrated environmental information and decision support system for Geneva. Part I: Air quality,” Archives des Sciences, vol. vol. 49, pp. p. 247-263, 1996.
[54] Böhm, M. & Scheider, T., “Extended Floating Car Data in Co-operative Traffic Management,”
International Series in Operations Research & Management Science, volume 144, 161-170,, New York, NY, 2010.
[55] Huber, W., Lädke, M. & Ogger, R., “Extended floating-car data for the acquisition of traffic information,” BMWgroup, Munich, 1999.
[56] Ende, B. van den, Sambeek, M. van, Berkers, F., Broeck, W. van den, & Sluis, J. van de,
“Assessment of wireless connectivity options in support of ITS,” Nederlandse Organisatie voor toegepast-natuurwetenschappelijk onderzoek (TNO), The Hague, The Netherlands, 2016.
[57] Filippi, A., Moerman, K., Daalderop, G., Alexander, P.D., Schober, F. & Pfliegl, W., “Ready to roll: Why 802.11p beats LTE and 5G for V2x,” NXP Semiconductors, Cohda Wireless, and Siemens, 2016.
[58] Kombate, D. & Wanglina, “The Internet of Vehicles based on 5G Communications,” IEEE
International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData), Chengdu, 2016.
[59] ITU-R , “IMT Vision – Framework and overall objectives of the future development of IMT for
2020 and beyond,” International Telecommunication Union - M Series: Mobile, radiodetermination, amateur and related satellite services, Geneva, 2015.
[60] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE, Std 802.11-2012, 2012.
[61] Intelligent Transport Systems (ITS); European Profile Standard for the Physical and Medium Access Control Layer of Intelligent Transport Systems Operating in the 5 GHz Frequency Band, ETSI, Std ES 202 663, 2009.
[62] Lyamin, N., Deng, Q. & Vinel, A., “Study of the platooning fuel efficiency under ETSI ITS-G5 communications,” 016 IEEE 19th International Conference on Intelligent Transportation Systems (ITSC), pp. 551-556, Rio de Janeiro, 2016.
[63] Ende, B. van den, Sambeek, M. van, Berkers, F., Broeck, W. van den, & Sluis, J. van de,
“Assessment of wireless connectivity options in support of ITS - An investigation into and assessment of ITS-G5 based and cellular communications for ITS infrastructure support,”
Nederlandse Organisatie voor toegepast-natuurwetenschappelijk onderzoek (TNO), The Hague, The Netherlands, 2016.
[64] He, Q., Head, K.L. & Ding, J., “Multi-modal traffic signal control with priority, signal actuation and coordination,” Transportation Research Part C: Emerging Technologies, vol. 46, pp. 65-82, 2014.
[65] Baker, R.J., Collura, J., Dale, J.J., Head, K.L. & Hemily, B., “An Overview of Transit Signal Priority,” ITS America, Washington, DC, 2002.
[66] Qin, X. & Khan, A.M., “Control strategies of traffic signal timing transition for emergency vehicle preemption,” Transportation Research Part C: Emerging Technologies, vol. 25, pp. 1-17, 2012.
References 61
[67] Deeter, D., Zarean, H.M. & Register, D., “Rural ITS Toolbox,” 2001.
[68] Federal Highway Administration (FHWA), “Traffic Signal Preemption for Emergency vehicles: A Cross-Cutting Study,” Federal Highway Administration (FHWA), 2006a.
[69] Hunter-Zaworski, K. & Danaher, A., “NE Multnomah street Opticom bus signal priority pilot study,” Final Report TNW97-03, Seattle, Wash, USA,, 1995.
[70] Trafc Technologies LLC, “Sonem 2000 digital siren detector,” Technical Document for Installation, Trafc Technologies, LLC, 2001.
[71] Hashim, Z., “Traffic Light Control System for Emergency Vehicles Using Radio Frequency,” OSR Journal of Engineering, vol. 3, no. 7, pp. 43-52, 2013.
[72] Bullock, D., Morales, J. & Sanderson, B., “Evaluation of emergency vehicle signal preemption on the Route 7 Virginia corridor,” Federal Highway Administration, Richmond, Va, USA, 1999.
[73] Bullock, D. & Nelson, E., “Impact evaluation of emergency vehicle preemption on signalized corridor operation,” Proceedings of the TRB Annual Meeting, Transportation ResearchBoard, Washington, DC, USA, 2000.
[74] Wang, J., Ma, W. & Yang, X., “Development of Degree-of-Priority Based Control Strategy for
Emergency Vehicle Preemption Operation,” Discrete Dynamics in Nature and Society, p. 10, 2013.
[75] Viriyasitavat, W. & Tonguz, O.K., “Priority Management of Emergency Vehicles at Intersections Using Self-Organized Traffic Control,” IEEE Vehicular Technology Conference (VTC Fall), Quebec City, QC, 2012.
[76] Lin, Y., Yang, X., Zou, N. & Franz, M., “Transit Signal Priority Control at Signalized Intersections:
A Comprehensive Review,” Transportation Letters The International Journal of Transportation Research. 7., 2014.
[77] Christofa, E. & Skabardonis, A., “Trafic signal optimization with application of transit signal priority to an isolated intersection,” Transportation Research Record, no. 2259, pp. 192–201, 2011.
[78] Furth, P.G. & Muller, T.H.J., “Conditional bus priority at signalized intersections: better service quality with less traffic disruption,” Transportation Research Record. 1731. 23-30., 2000.
[79] Evans, H. & Skiles, G., “Improving public transit through bus preemption of traffic signals,” Eno Transportation Foundation, 24, (4), 53l–543., 1970.
[80] Allsop, R.E., “Priority for buses at signal-controlled junctions: some implications for signal timings,” Transportation and Traffic Theory, 1977.
[81] Cottinet, M., Breteque, A., Henry, J.J. & Gabard, F., “Assessment by observation and by
simulation studies of the interest of different methods of bus preemption at traffic lights,” in Proc. Int. Symp.on Traffic Systems, Vol.2, pp. 92–l05., Berkeley, CA, 1980.
[82] Shalaby, A., Lee, J., Greenough, J., Hung, S. & Bowie, M., “Development, Evaluation, and Selection of Advanced Transit Signal Priority Concept Directions,” Journal of Public Transportation. 9. 97-120., 2006.
[83] Chang, G.L., Vasudevan, M. & Su, C.C., “Modeling and evaluation of adaptive bus-preemption
control with and without automatic vehicle location system,” Transportation Research Part A: Policy and Practice,, vol. 30, no. 4, pp. 251-268, 1996.
62 References
[84] Mirchandani, P.B., Knyazyan, A., Head, K.L. & Wu, W., “An Approach Towards the Integration
of Bus Priority, Traffic Adaptive Signal Control, and Bus Information/Scheduling Systems,” Computer-Aided Scheduling of Public Transport, Springer-Verlag, Germany, 2001.
[85] Ioannou, P., “Design and Evaluation of Impact of Traffic Light Priority for Trucks on Traffic Flow,” University of Southern California, Electrical Engineering – Systems, Los Angeles, CA, 2015.
[86] Sunkari, S.R., Charara, H. & Urbanik, T., “Minimizing Truck Stops at High-Speed Rural Signalized Intersections,” Institute of Transportation Engineers, Washington, DC, 2001.
[87] Bonneson, J., Middleton, D., Zimmerman, K. & Charara. H., “Intelligent Detection-Control system for High Speed Rural Signalized Intersections,” Texas Transportation Institute, Texas, 2002.
[88] Zhao, Y. & Ioannou, P., “A Traffic Light Signal Control System with Truck Priority,” IFAC (International Federation of Automatic Control), Los Angeles, United States, 2016.
[89] Saunier, N., Sayed, T. & Lim, C., “A prototype system for truck signal priority using video
sensors,” Conference and Exhibition of the Transportation Association of Canada - Transportation in a Climate of Change, 2009.
[90] Liao, C. & Davis, G.A., “Bus signal priority based on GPS and wireless communications phase 1 – simulation study,” University of Minnesota, Department of Civil Engineering, Minnesota, 2006.
[91] Li, F., Wang, D.H., Wang, J. & Jin, S., “An approach of transit passive priority with transit phase
overlapped at intersection of arterial signal progression,” IEEE Conference on Intelligent Transportation Systems, 729-733, 2008.
[92] Kari D., Wu G. & Barth, M.G., “Eco-friendly freight signal priority using connected vehicle technology: a multi-agent systems approach,” IEEE Intelligent Vehicles Symposium, 1187-1192, 2014.
[93] van Senden, J.C., “DIRECTOR: Enabling advanced driver assistance systems with predictive
signalized intersection control using LSTM networks,” Delft University of Technology, Faculty of Electrical Engineering, Mathematics and Computer Science, Embedded Software Section, Delft, The Netherlands, 2018.
[94] PTV GROUP, "Brochure: PTV Vissim," [Online]. Available: http://vision-
traffic.ptvgroup.com/fileadmin/files_ptvvision/Downloads_N/0_General/2_Products/2_PTV_Vissim/BRO_PTV_Vissim_EN.pdf. [Accessed 8 03 2018].
[95] Python, “Python.org,” Python Software Foundation , [Online]. Available: https://www.python.org/doc/essays/blurb/. [Accessed 17 07 2018].
63
A A. Scientific paper
1
Improvement of road traffic sustainability by implementation of priority weights for trucks in
predictive signalized intersection control L. Haanstra¹, ing. A.P. Verhoeven², dr. Ir. X. Jiang¹, dr.ir. H. Polinder¹
¹Delft University of Technology, faculty of 3mE, Department of Transport Engineering and Logistics, Delft,
The Netherlands
²Siemens Mobility BV, Zoetermeer, The Netherlands
Abstract: In the European Union Road freight transport volume is expected to grow 78% between 2000 and
2030, which results in more trucks on the road network. The worldwide estimated trend shows an increase of
150 million freight vehicles and an increase of 240 million passenger vehicles. The growth of both vehicle
classes will have a major impact on the road network and the roads will become congested. Especially in dense
urban environments with many intersections. Further, trucks have a detrimental impact on traffic flows,
especially at intersections, because of their slow dynamics and large size. In addition, a stopping truck results
in higher emissions and fuel consumption compared to a car. However, today’s traffic controllers are not
capable of optimizing traffic flow at intersections based on classification of different vehicles. It would be
beneficial to all vehicles involved if the number of stops for trucks would be reduced to a minimum, by servicing
each vehicle class in a different way. Therefore, a new signalized intersection controller that can reduce the
number of stops for trucks is developed. The controller grants priority to trucks via implementation of priority
weights for different vehicle classes and reduced the number of stops for trucks, while maintaining the overall
traffic flow at least as good as a state-of-the-art model predictive intersection controller.
Keywords: Signalized intersection controller, Truck Signal Priority, Simulation, Optimization.
I. Introduction
Signalized intersections play an important role in
modern society. The introduction of signalized
intersection controllers gave structure and a way of
automatic traffic handling at intersections. Only with
the economic growth that developed countries faced
an exponentially increasing demand for personal
mobility occurred [1]. It quickly resulted in
congestions at signalized intersections in urban
environments. Congestion involves queuing, lower
speeds, and increased travel times, which impose
costs on the economy and generate multiple impacts
on urban regions and their inhabitants [2]. To
eliminate the congestion, signalized intersections
could be replaced with bridges, tunnels or non-
signalized roundabouts. This option is however in
many cases, both economically and spatially, not
feasible [1]. One alternative option to improve
efficiency of urban intersections would be the
innovation of the on-street traffic controllers.
However, traffic inefficiencies will still occur, because
of the disruption of traffic flow caused by a red light,
even with the latest innovations of traffic controllers.
In addition, a reaction to an unanticipated switch
from green to amber causes safety concerns, as
drivers may suddenly stop or quickly accelerate. Apart
from the disruption of traffic flow and safety
concerns, the acceleration and deceleration
behaviour cause the largest amount of fuel
consumption and CO2 emissions [3]. These stops
result in over 50% of the fuel consumption of a
vehicles trip [4]. Moreover, one stop of a vehicle could
create a backward moving shockwave that induces a
cyclic driving state of acceleration and idling. This
behaviour is responsible for up to two thirds of the
total energy loss at intersections [3], [4]. Moreover,
once the traffic light turns green, the inability for
drivers to anticipate when they should accelerate
from stop, and the time it takes to accelerate to free-
flow speed, results in a queue discharge rate that can
be as low as 75% of the road’s capacity [5]. In
addition, the time for a heavy truck to respond to a
traffic light, accelerate and cross the intersection is
much higher than that of normal passenger cars [11].
Despite the different dynamics between trucks and
passenger cars, current traffic controllers do not
service them different. Instead, the traffic controllers
will service both as equal. Hence, the problem could
be described as:
“Today’s traffic controllers are not capable of
optimizing traffic flow at intersections based on
classification of different vehicles.”
2
It is therefore the goal to reduce the number of stops
for trucks, by servicing vehicles based on vehicle type
in future signalized intersection controllers.
Literature
Several techniques are found in the literature to
develop a new signalized intersection controller with
truck signal priority, which are described below. First,
a controller type is elaborated. Second, a detection
method is described. Third, a priority handling
mechanism is presented.
Model predictive control (MPC) is in the basics an
extended version of the adaptive controller. The key
difference is similar to the difference between
actuated and adaptive, the extension of the vision
horizon. The predictive controller has the vision
horizon extended to at least the exit flows of the
previous intersection Consequently, the controller
must be able to handle the large uncertainties
associated with forecasting traffic flow progression.
by Hochreiter & Schmidhuber, adding a Long Short-
Term Memory (LSTM) [6]. The intuition behind the
LSTM is to control the memory in a structured way.
Therefore, the LSTM can determine whether the
content of the memory should be remembered,
updated or forgotten. Long-term dependencies can
be recognized in the LSTM network by training this
memory. A variation of the MPC is the self-organizing
traffic light controller [7]. The objective is relatively
simple, the controller gives preference to cars that
have been waiting longer and to larger groups of cars
[1]. In other words, the controller optimizes the traffic
light control according to the cumulative travel time
delay. The intersection uses a non-periodic
optimization technique to create optimal schedules,
which can lead to instability [8]. Therefore, a
stabilization mechanism is applied to ensure servicing
of each direction as least as good as a fixed-time
strategy [9]. Initially, the controller without a
stabilization mechanism was compared to a fixed-
time controller, showing a significant reduction in
terms of average queue length and average travel
time delay [10]. In later work, the controller with the
stabilization mechanism is compared the previous
controller without the stabilization mechanism, which
resulted again in a reduction of the average travel
time delay [9].
Literature described multiple detection techniques,
which are able to detect different vehicle types.
Especially, Floating Car Data (FCD) showed potential.
The main benefit of FCD is that every vehicle acts as
moving sensor, therefore no additional hardware is
required on the roadway [11], [12]. Furthermore,
FCD benefits from maximum flexibility as a large scale
FCD system can be extended over large areas with
only a marginal increase in variable costs [13]. In
addition, developments in communication and sensor
technology create the possibility to send additional
vehicle information next to the existing FCD. This is
the second generation of FCD and labelled as
Extended Floating Car Data (xFCD) [14].
Suthaputchakun and Sun [3] propose an adaptive
traffic light scheduling scheme via two-way traffic-
light-to-vehicle communication (TLVC) for fuel
consumption and CO2 emission reduction. In
addition, a priority framework to optimize a weighted
traffic light schedule is proposed, by assuming the
weight of a truck two-times higher than a normal
vehicle.
The selection of the proposed techniques will be used
to develop a truck signal priority controller. To
summarize, the proposed truck priority strategy is a
MPC. The MPC will have a SOIC approach and will use
a LSTM approach for traffic flow predictions. In
addition, the FCD vehicle detection technique will be
used in order to detect different vehicle types. Finally,
a weighted traffic light schedule will be used to enable
truck signal priority.
II. Methods
This section describes the design of the Truck Signal
Priority (TkSP) controller and a cast study to evaluate
the performance. The work of this research is based
on the predictive controller DIRECTOR developed by
Van Senden [15]. The control algorithm is modified
and extended with the ability to make a schedule
decision based on different vehicle types. In this
research, the main focus is the truck signal priority.
For this reason, only cars and trucks will be part of
the study.
Prioritization of specific vehicle types requires the
signalized intersection controller to detect different
vehicle types. Literature described multiple detection
techniques, which are able to detect different vehicle
types. Especially, floating car data showed potential.
In addition, floating car data has the possibility to
send additional information, such as direction, speed
and vehicle type. Therefore, the method used to
detect an approaching truck is based on floating car
data. The range of the detection is set to a distance
of 330 m, based on the minimum distance between
the intersection under control and an upstream
intersection in the network. Alternatively, the selected
distance could also be extended to a larger distance,
which enables earlier knowledge of an approaching
truck. However, an increased detection distance
would increase the uncertainty of the estimated time
3
of arrival. If a larger detection distance is required, a
truck could update its floating car data more
frequently to reduces the uncertainty of the estimated
time of arrival. For simplification is the floating car
data of the truck limited to one update throughout
this research. Accordingly, the estimated time of
arrival is calculated, which is based on the distance to
the intersection and a speed limit of 50 km/h.
Resulting in an estimated time of arrival of
approximately 24 seconds. For this reason, three 10
second time bins are proposed, the index of the time
bin is denoted as T. Figure 1 illustrates the point of
detection and the corresponding time bins in an
approach of the intersections.
Figure 1 - Point of detection
It should be noted that the count for the time bins
starts at zero on the stop line of an intersection and
increases in steps of one moving upstream. The truck
arrivals, denoted 𝜏, will be assigned to a time bin. Eq.
(1) gives the result of the combined truck arrivals of
a single Origin-Destination.
According to the truck arrival time bins, discussed in
the previous section, a set of priority weights is
proposed. This set will correspond to the three first
time bins, since the trucks have an estimated arriving
time at the intersection within 30 seconds. In the
future this set could be extended to more time bins.
However, it should be noted that if more time bins
are used the uncertainty would increase as well. The
set of priority weight will be denoted as 𝜔. Table 1
illustrates the index corresponding to the time bin.
Table 1 - Priority weight index to corresponding time bin
𝜔 [0] 𝜔 [1] 𝜔 [2]
Time [s] (0,10) (11,20) (21,30)
The assigned weights are based on the differences in
vehicle characteristics. Suthaputchakun and Sun [3]
make a connection between different types of
vehicles based on their actual weight. For example,
heavily loaded vehicles normally have higher
emissions and consume more fuel. For this reason,
Suthaputchakun and Sun assumed that the weight of
a heavily loaded vehicle is two-times higher than that
of the small vehicles. However, more vehicle
characteristics are found in the literature. For
instance, the length of a truck is around 1.5 to 4 times
the length of a standard car [16]. Further, significant
differences are found in the vehicle dynamics,
especially in the acceleration rates after a complete
stop. A typical truck has an acceleration rate around
five times lower compared to a passenger car, when
accelerating to 50 km/h [17]. Combining the vehicle
characteristics above, an assumption is made on the
impact of a truck at an intersection compared to a
standard car. The following characteristics are taken
into consideration to determine the impact: weight
(2), length (4) and acceleration rate (5). The values
correspond to the number of times a truck has more
impact compared to a standard car. However, not all
defined vehicle characteristics are taken as equal. It
is assumed that the acceleration rate of a truck has
two times more impact on the intersection. Since, all
cars stopped behind a truck are limited to the
acceleration rate of the truck in front. The weighted
average of the vehicle characteristics above result
that a truck has four times more impact at the
intersection compared to a standard car. Hence, an
extra weight of four is given to an arriving truck at a
specific time bin. A truck is required to anticipate on
red light earlier than a regular car, due to their slow
dynamics. Therefore, it is assumed that a truck will
experience delays, if the signal is not green in the ten
seconds before a truck arrives. The extra weight will
be added to the second time bin to ensure a smooth
passage for a truck crossing the intersection.
Consequently, the set of priority weights is given in
eq. (2).
𝜔[𝑇] = [1, 4, 1] (2)
However, queues could already exist in front of an
arriving truck. In the same way as stopping for a red
light, a truck has to anticipate earlier on a queue. To
account for the possible queues and clear them
before a truck is arriving, a second weight is added in
the set of priority weights. However, this second
priority weight should not be equal to the value of the
second time bin. Since, it is intended to clear a queue
in advance and not give early green. The value of the
additional weight is therefore proposed to be half of
the value of the second time bin. It follows that the
priority weight is added in the third time bin, because
this is the farthest time bin away from the
intersection. Finally, the proposed set of priority
weights is found in eq. (3).
𝜔[𝑇] = [1, 4, 2] (3)
The next step is to calculate the truck priority weight,
denoted as 𝛿. The basic idea is to multiply the truck
arrival by the assigned priority weight, given by eq.
(4).
𝛿 = 𝜏 ∗ 𝜔 (4)
𝜏[𝑇] = [0, 0, 1] (1)
4
However, one or more approaches of the intersection
could have multiple Origin-Destinations. For this
reason, it is required to receive the Origin-Destination
from the floating car data of the truck. In addition,
multiple trucks could approach the intersection on the
same Origin-Destination. To find the truck weight of
one Origin-Destination, the weight of the approaching
trucks should be summed. Accordingly, eq. (4) is
rewritten to add the weights of three time bins
together. The calculation of the truck priority weight
of a single Origin-Destination is described by eq. (5).
𝛿𝑂𝐷[T] = ∑𝜏[𝑇] ∗ 𝜔[𝑇]
2
𝑇=0
(5)
However, multiple Origin-Destinations are often not-
conflicting and can be serviced simultaneously. The
set of combined Origin-Destinations is described as a
Phase Group. To find the truck weight of a Phase
Group, denoted as 𝛿𝑃𝐺, all Origin-Destinations within
the Phase Group are summed up following eq. (6).
𝛿𝑃𝐺[T] = ∑ ∑𝜏[𝑇] ∗ 𝜔[𝑇]
2
𝑇=0𝑂𝐷∈𝑃𝐺
= ∑ 𝛿𝑂𝐷[T]
𝑂𝐷∈𝑃𝐺
(6)
Finally, the schedule decision including the priority
weights for approaching trucks could be described by
eq. (7).
𝜒[𝑇 + 1] =
= arg max𝑃𝐺∈𝑃𝐺𝑠
{
∑ 𝜌𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
− σ𝑃𝐺[𝑇 + 1]
+ 𝛿𝑃𝐺[𝑇 + 1] }
= arg max𝑃𝐺∈𝑃𝐺𝑠
{
∑ 𝜌𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
− σ𝑃𝐺[𝑇 + 1]
+ ∑ 𝛿𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺 }
= arg max𝑃𝐺∈𝑃𝐺𝑠
{( ∑ 𝜌𝑂𝐷[𝑇 + 1] + 𝛿𝑂𝐷[𝑇 + 1]
𝑂𝐷∈𝑃𝐺
)
− σ𝑃𝐺[𝑇 + 1]
}
(7)
Case study The proposed controller is evaluated within a case
study. The simulation software PTV Vissim is used to
develop and evaluate the case study. The controller
algorithm is written in Python and connected via a
COM (Component Object Model) interface to Vissim.
The case study will use an intersection that is located
in the Netherlands near Hoofddorp in the province of
Noord-Holland. The Vissim model for the intersection
of the case study is available and the historic data is
received form the province of Noord-Holland, which
is provided in a V-Log data format. A format in which
Dutch traffic light controllers log their data [15]. The
V-Log data only records changes in the data stream,
which contains the detectors states, signals states
and the internal system state of the traffic light
controller. The historic data used in the simulation
dates from January 2017 to May 2017 and is compiled
from the V-Log format to numerical data that can be
used for machine learning and simulation purposes
[15]. The data compiled from the V-Log data does not
contain information on the vehicle type. Therefore, it
is required to create a realistic data set of truck
arrivals. The truck arrival data is estimated from the
original detector data. This data contains the
timestamps of when a detector becomes occupied
and when the detector is free again. These
timestamps are compared to find the occupation time
of a detector. In this occupied time data is a three-
point median search conducted to find expected truck
detections. If the median is equal or larger than nine
seconds, then the point is marked as a truck
detection. Figure 2 shows the truck detections,
marked with red dots, for a single Origin-Destination
in 24 hours. These truck detections are saved to a
new data file, in the same structure as the original
detections, to use as an input for the truck
simulations.
Figure 2 - Truck detections (marked with red dots)
The PTV Vissim model for the intersection of the case
study is available and provided by the province of
Noord-Holland. However, this model contains a larger
network of multiple intersections, which are not used
in this research. Although they are not actively used
in the simulation, they use computational power of
the computer to simulate the vehicles in the network.
Therefore, all unnecessary elements in the PTV
Vissim network are removed to have an optimal
network for a smooth simulation. Subsequently, the
model is adjusted to be able to replay historical data
of vehicle arrivals. The locations of the vehicle inputs
are shown in Figure 3. Further, multiple vehicle
detectors are used in the model. These are displayed
as blue rectangles in Figure 3. During the simulation
the controller reads the states of the detectors every
5
100 milliseconds. The change between a state of the
detector is used to count vehicles passing a detector.
As each lane has two detectors, i.e. the arrival
detector and stop line detector, the controller can
count arriving and departing vehicles. Combining
information of two detectors on the same lane
enables the controller to calculate current queue
length of each Origin-Destination pair. Further, the
controller is able to change the state of the signal
heads in PTV Vissim, which are shown as red stripe
in Figure 3. Due to limitations of the simulation
software the changes are limited to once every
second. Subsequently, measurement points are
added for each individual Origin-Destination, in order
to evaluate the controller on the following outputs:
the number of stops per vehicle and the vehicle delay.
The measurements points are vehicle travel time
detectors in PTV Vissim. This detector type measures
the time it takes a vehicle to travel from one point to
the next. The locations of the travel time detectors
are displayed in Figure 3, where the pink line is the
start of the measurement and the green line is the
end of the measurement. The following outputs can
be calculated with these measurements: the vehicle
travel delay, the number of stops and the number of
vehicles passing. A delay of a vehicle is calculated
when the actual travel time is compared to the travel
time it would need under free flow conditions. Free
flow conditions are considered as for example when
the vehicle can maintain its desired speed, i.e.
without reacting on another vehicle or a red signal. In
addition, each individual Origin-Destination of the
measurement could be specified for the different
vehicle types. Hence, evaluations could be reviewed
for three configurations: all vehicles, only cars and
only trucks
Figure 3 - Vissim network layout of the intersection
III. Results
To evaluate the performance of the new design for
truck signal priority, simulations following the case
study of the previous section are conducted. These
simulations include two sets of different weights for
trucks. The first, is the baseline of [1, 1, 1]. In this
configuration no extra weight is added to a truck in
the schedule decision. The second, is the proposed
set of [1, 4, 2]. In this configuration, extra weight is
added to a truck in the schedule decision. Here, a
truck is equal to four cars in the second time bin and
equal to two cars in the third time bin. Both
configurations are simulated with a data set of 24
hours and have the same vehicle inputs. The
simulations are evaluated according the Key
Performance Indicators (KPI): the number of stops
and the vehicle delay (s). The results showed that the
simulation with priority weights, compared to the
baseline, has 87 stops less for trucks and 265 stops
more for cars. Further, the total vehicle delay for
trucks is reduced by 45 minutes (2673 seconds) and
increased by 111 minutes (6642 seconds) for cars.
Given the above, the introduction of priority weights
for trucks has a positive influence on the key
performance indicators for the trucks. However, it has
a minor negative impact for the cars.
To analyse the sensitivity of the priority weights a
Design of Experiments (DoE) is proposed. This
method aims to describe the results under a variation
of conditions. By introducing one change in the set of
priority weights it is expected to result in a change in
one or more output variables. Following, the priority
weights are chosen as input variables. The output
variables contain the key performance indicators: the
number of stops and vehicle delay. The set of priority
weights will start at [1,1,1], which is the baseline. In
this configuration no extra weight is added to a truck
in the schedule decision. Subsequently, the variables
are changed one by one. The first time bin is always
one, as described in the previous section. The other
two will vary in a range of eight values, starting a one.
The used weight data set is illustrated in Table 2.
Table 2 - Weight data set
The overall objective is to reduce the number of stops
for trucks. However, the implementation of priority
weights should have a minimal impact on the traffic
flow of other vehicles. Multiple weight configurations
show a decrease in the vehicle delay, compared to
the baseline for both cars and trucks. In addition,
three weight configurations also show a reduction in
6
the number of stops. Especially the weight
configuration [1, 3, 4]. This is also noticed Table 3
and Table 4, which present the total stops and vehicle
delay for three weight configurations: the baseline,
the first proposed and the weight configuration
[1, 3, 4]. The results showed that the weight
configuration [1, 3, 4] also outperforms the baseline
for the other vehicle groups. The new weight
configuration reduced the total stops by 183 for cars
and 143 for trucks. Further, the total vehicle delay for
cars is decreased by 59 minutes (3558 seconds) and
54 minutes (3253 seconds) for trucks. However, the
simulations are limited to 24 hours and one set of
vehicle inputs.
Table 3 – Total stops: Sensitivity analysis
Total stops
All Car Truck
[1, 1, 1] 23239 22750 489
[1, 4, 2] 23417 23015 402
[1, 3, 4] 22911 22567 344
Table 4 - Total vehicle delay: Sensitivity analysis
Total vehicle delay [s]
All Car Truck
[1, 1, 1] 653197 637582 15615
[1, 4, 2] 657166 644224 12942
[1, 3, 4] 646386 634024 12362
Different sets of vehicle inputs should be evaluated to
check the robustness of the weight configuration [1,
3, 4]. Therefore, six additional days have been
simulated, each having a different set of vehicle
inputs of 24 hours. Subsequently, the results could be
evaluated for a complete week, as can be seen in
Figure 4. The number of vehicles over a week are
presented in Table 7. The overall objective is similar
to the previous analysis, which quickly reduces the
number of suitable weight configurations. In fact,
only one weight configuration meets the objective, [1,
2, 1]. Compared to the baseline it reduces the total
stops by 77 for cars and 88 for trucks, and decreases
the total vehicle delays for cars by 61 minutes (3673
seconds) and 66 minutes (3957 seconds) for trucks.
Although, the weight configuration [1, 2, 1] meets the
objective, the results are not impressive. Since it only
reduced the total stops for trucks by 88 over a week.
However, the weight configuration [1, 3, 4] shows a
more significant reduction in the total number of
stops for trucks over a week, while the increase in the
total number of stops for cars remains minimal. The
comparison can be found in Table 5 and Table 6.
Table 5 – Total stops: Week
Total stops
All Car Truck
[1, 1, 1] 142225 139404 2821
[1, 3, 4] 141629 139559 2070
Table 6 - Total vehicle delay: Week
Total vehicle delay [s]
All Car Truck
[1, 1, 1] 3987494 3897966 89528
[1, 3, 4] 3981301 3909999 71301
Table 7 - Number of vehicles: Week
Origin-Destination Total 2 3 4 6 7 8
All 81954 30138 35030 41641 44135 73159 306057
Cars 81405 29582 34007 40300 43781 72499 301574
Trucks 549 556 1023 1341 354 660 4483
Finally, it can be concluded that the weight
configuration [1, 3, 4] reduces the total of stops for
trucks by 751 over a week, compared to the baseline
[1, 1, 1]. However, the total stops for cars increases
by 155. Further, the total vehicle delay over a week
shows a similar trend as the total stops. Namely, the
total vehicle delay for trucks is decreased by 304
minutes (18227 seconds) and increased by 201
minutes (12033 seconds) for cars. Given the above,
it is noticed that both the total stops and the total
vehicle delay decreases more for trucks than it
increases for cars. As a result, the priority weight
configuration [1, 3, 4] showed an overall
improvement of the key performance indicators over
a week.
Figure 4 – Sensitivity analysis: Week
7
IV. Discussion
The expected results for adding priority weights are
confirmed by the results. Compared to the baseline
[1, 1, 1], the weight configuration [1, 4, 2] shows a
decrease in stops and vehicle delay for trucks.
However, cars experience a minor increase in the
total stops and the vehicle delay. Therefore, the
overall performance is negatively impacted in terms
of total stops and total vehicle delay. This can be
explained because there are more cars in relation to
trucks in the simulation. To find if it is possible to
reduce the number of stops and vehicle delay for
trucks, without negatively impacting other vehicles,
more priority weight configurations should be
evaluated.
By performing a design of experiments to analyse the
sensitivity of the priority weights, it was expected to
find a relation between the priority weights
configuration and the key performance indicators.
The higher weight configurations, which reduces the
number of stops for trucks compared to the baseline.
On the other hand, the number of stops for cars are
increasing for higher weight configurations. Similar
results are found for the vehicle delay. It follows that
the higher priority weight configurations do not have
the desired results, since they negatively impact the
traffic flow.
Another interesting result is noticed, it showed that it
is more beneficial to have a higher value for the third
time bin in combination with a lower value for the
second time bin. This finding indicates the importance
to clear a possible queue in front of the arriving truck.
In order to cross the intersection without a stop.
However, the weight values 6, 7 and 8 in the third
time bin also show a major increase in the totals stops
and total vehicle delay for cars. This may indicate that
the trucks are simply given too much priority weight
before they arrive at the intersection. On the other
hand, it could also indicate that a weight is added to
clear a queue, while no queue is present. A possible
solution would be to first check the current queue and
only add the priority weights if a queue is present.
Different sets of vehicle inputs were evaluated to
check the robustness of the weight configuration
[1, 3, 4]. Subsequently, the results could be
evaluated for a complete week. The results showed
one priority weight configuration that met the
objective, which was [1, 2, 1]. However, the results
were not impressive. Since it only reduced the total
stops for trucks by 88 over a week. On the other
hand, the weight configuration [1, 3, 4] showed a
more significant reduction in the total number of
stops for trucks over a week, while the increase in the
total number of stops for cars remained minimal.
Subsequently, the priority weight configuration
[1, 3, 4] was evaluated for the individual days. The
evaluation showed that the performance of the
priority weight configuration, in terms of total stops
and vehicle delay, is depended on the traffic demand.
This indicates that is would be beneficial to have an
dynamic priority weight configuration, which would
adapt to the current traffic demand. Another
approach could be to implement of multiple priority
weight configurations for each Origin-Destination.
The priority weight configurations could then be
optimized per Origin-Destination, since the traffic
demand is often different for each Origin-Destination.
However, the results for the priority weight
configuration [1, 3, 4] showed an overall
improvement of the key performance indicators over
a week. Despite, increasing the total stops and total
vehicle delay for cars. It can be concluded that the
proposed truck signal priority controller design can
reduce the number of stops for trucks at a signalized
intersection, while maintaining the overall traffic flow
at least as good as a state-of-the-art model predictive
intersection controller.
V. Conclusion
The results of this research showed that the
implementation of priority weights for trucks has an
influence on the performance of the intersection
controller, in terms of number of stops and vehicle
delay. Early results presented a comparison of a
proposed priority weight configuration to a baseline
with no priority weights. Simulating a full day, it was
found that the simulation with priority weights has 87
stops less for trucks and 265 stops more for cars
(17,8% and 1,16% respectively). Further, the total
vehicle delay for trucks is reduced by 45 minutes and
increased by 111 minutes for cars (17,1% and 1,0%
respectively). A sensitivity analysis was performed to
further study the effects of the priority weights. The
results showed that the priority weight configuration
[1, 3, 4] outperforms the baseline. The total stops
were reduced by 143 for trucks and 183 for cars
(29,67% and 0,8% respectively), also the total
vehicle delay for truck was decreased by 54 minutes
and 59 minutes for cars (0,6% and 20,8%
respectively). Subsequently, a robustness check was
performed to evaluate the performance of the priority
weight configuration over a week. The results showed
an reduction of the total number of stops by 751 and
total vehicle delay by 304 minutes for trucks over a
week (26,6% and 20,4% respectively). While, the
8
total number of stops and total vehicle delay for cars
increased, by 155 stops and 201 minutes (0,1% and
0,3% respectively). However, the overall total
number of stops and total vehicle delay were reduced
by 596 stops and 103 minutes (0,42% and 0,16%
respectively). It can be concluded that the proposed
truck signal priority controller design can reduce the
number of stops for trucks at a signalized
intersection, while maintaining the overall traffic flow
at least as good as a state-of-the-art model predictive
intersection controller.
References
[1] Nunzio, G. D., “Trafc eco-management in urban trafc
networks,” Université Grenoble Alpes, 2015.
[2] Kaparias, I., Zavitsas, K. & Bell, M.G.H., “State-Of-
The-Art of Urban Traffic Management Policies and
Technologies,” CONDUITS, Coordination Of Network
Descriptors for Urban Intelligent Transport Systems,
Imperial College London, 2010.
[3] Suthaputchakun, C. & Sun, Z., “A Novel Traffic Light
Scheduling Based on TLVC and Vehicles’ Priority for
Reducing Fuel Consumption and CO2 Emission,” IEEE
Systems Journal, vol. 12, no. 2, pp. 1230-1238, 2018.
[4] Wu, L., Yusheng, C., Chu, J. & Zhang, H., “The
Influence of Intersections on Fuel Consumption in
Urban Arterial Road Traffic: A Single Vehicle Test in
Harbin,” PLoS One, China, 2015.
[5] Brilon, W. & Zurlinden, H.,
“Ueberlastungswahrscheinlichkeiten und
Verkehrsleistung als Bemessungskriterium fuer
Strassenverkehrsanlagen (Breakdown Probability and
Traffic Efficiency as Design Criteria for Freeways),”
Forschung Strassenbau und
Strassenverkehrstechnik, Heft 870, 2003.
[6] Hochreiter, S. & Schmidhuber, J., “Long short-term
memory,” Neural computation, vol. 9, pp. 1735-1780,
1997.
[7] Lämmer, S., Donner, R. & Helbing, D., “Anticipative
control of switched queueing systems,” The
European Physical Journal B, vol. 3, no. 63, p. 341,
2008.
[8] Kumar, P.R. & Seidman, T.I., “Dynamic instabilities
and stabilization methods in distributed real-time
scheduling of manufacturing systems,” IEEE
Transactions on Automatic Control, vol. 3, no. 35, p.
289–298, 1990.
[9] Lämmer, S. & Helbing, D., “Self-stabilizing
decentralized signal control of realistic, saturated
network traffic,” Santa Fe Institute Working Paper,
Santa Fe, 2010.
[10] Lämmer, S. & Helbing, D., “Self-control of traffic
lights and vehicle flows in urban road networks,”
Journal of Statistical Mechanics: Theory and
Experiment, no. 04, 2008.
[11] Tiedong, W. & Jinging, H., “Applying Floating Car
Data in Traffic Monitoring,” 2014 IEEE International
Conference on Control Science and Systems
Engineering, Yantai, China, 2014.
[12] Fabritiis, C. de, Ragona, R. & Valenti, G., “Traffic
Estimation And Prediction Based On Real Time
Floating Car Data,” in Proceedings of the 11th
International IEEE, Conference on Intelligent
Transportation Systems, Beijing, China, 2008.
[13] Fedra, K., Greppin, H., Haurie, A., Hussy, C., Dao, H.
& Kanala, R., “GENIE: An integrated environmental
information and decision support system for Geneva.
Part I: Air quality,” Archives des Sciences, vol. vol.
49, pp. p. 247-263, 1996.
[14] Banach, S., “Betrouwbaarheid en toepassingen van
Floating Car Data,” Technische Universiteit
Eindhoven (TUe), Design & Decision Support Systems
(DDSS), Eindhoven, 2013.
[15] van Senden, J.C. , “DIRECTOR: Enabling advanced
driver assistance systems with predictive signalized
intersection control using LSTM networks,” Delft
University of Technology, Faculty of Electrical
Engineering, Mathematics and Computer Science,
Embedded Software Section, Delft, The Netherlands,
2018.
[16] Garber, N.J. & Hoel, L.A., Traffic & Highway
Engineering, 4th edition, Cengage Learning, ch. 3,
2008.
[17] Institute of Transportation Engineers, Traffic
Engineering Handbook, Washington DC: 5th Edition,
ch. 5, 2000.
73
B B. Python modules and
packages
74 B. Python modules and packages
Python version:
- Python 3.6.5 :: Anaconda, Inc.
Spyder version:
- Spyder 3.3.1
The following modules and packages are used for the simulations in this research:
- sys
- pickle
- time
- datetime
- json
- os
- queue
- math
- win32com.client
- numpy
o Version: 1.14.3
o Summary: NumPy: array processing for numbers, strings, records, and objects.
- openpyxl
o Version: 2.5.3
o Summary: A Python library to read/write Excel 2010 xlsx/xlsm files
o Home-page: https://openpyxl.readthedocs.io
- pandas
o Version: 0.23.0
o Summary: Powerful data structures for data analysis, time series, and statistics
o Home-page: http://pandas.pydata.org
- XlsxWriter
o Version: 1.0.4
o Summary: A Python module for creating Excel XLSX files.
o Home-page: https://github.com/jmcnamara/XlsxWriter
75
C C. Vehicle arrivals
76 C. Vehicle arrivals
Table 18 - Car arrivals
77
Table 19 - Truck arrivals
79
D D. Simulation results
80 D. Simulation results
Figure 39 – Week
Figure 40 – Tuesday
81
Figure 41 – Wednesday
Figure 42 - Thursday
82 D. Simulation results
Figure 43 - Friday
Figure 44 - Saturday
83
Figure 45 - Sunday
Figure 46 - Monday
85
E E. Simulation data
86 E. Simulation data
Table 20 - Simulation data: Week number of stops
87
Table 21 - Simulation data: Week vehicle delay (s)
88 E. Simulation data
Table 22 - Simulation data: Tuesday number of stops
89
Table 23 - Simulation data: Tuesday vehicle delay (s)
90 E. Simulation data
Table 24 - Simulation data: Wednesday number of stops
91
Table 25 - Simulation data: Wednesday vehicle delay (s)
92 E. Simulation data
Table 26 - Simulation data: Thursday number of stops
93
Table 27 - Simulation data: Thursday vehicle delay (s)
94 E. Simulation data
Table 28 - Simulation data: Friday number of stops
95
Table 29 - Simulation data: Friday vehicle delay (s)
96 E. Simulation data
Table 30 - Simulation data: Saturday number of stops
97
Table 31 - Simulation data: Saturday vehicle delay (s)
98 E. Simulation data
Table 32 - Simulation data: Sunday number of stops
99
Table 33 - Simulation data: Sunday vehicle delay (s)
100 E. Simulation data
Table 34 - Simulation data: Monday number of stops
101
Table 35 - Simulation data: Monday vehicle delay (s)
103
F F. Emission data
104 F. Emission data
Table 36 - Emission data Tuesday Wednesday Thursday Friday Saturday Sunday Monday Week
[1,
1,
1]
CO [g] 49916 49513 47737 39848 33568 39152 48962 308696
NOx [g] 9712 9634 9288 7753 6531 7618 9526 60061
VOC [g] 11568 11475 11064 9235 7780 9074 11347 71543
Fuel consumption [gal] 714 708 683 570 480 560 700 4416
[1,
4,
2]
CO [g] 50147 49528 47676 39898 33609 39353 49271 309483
NOx [g] 9757 9636 9276 7763 6539 7657 9586 60214
VOC [g] 11622 11479 11049 9247 7789 9120 11419 71726
Fuel consumption [gal] 717 709 682 571 481 563 705 4428
[1,
3,
4]
CO [g] 49409 49496 47515 39905 33568 39045 48871 307808
NOx [g] 9613 9630 9245 7764 6531 7597 9508 59888
VOC [g] 11451 11471 11012 9248 7780 9049 11326 71337
Fuel consumption [gal] 707 708 680 571 480 559 699 4404