Transport Modelling for managers 2014 willumsen

29
ROLE OF TRANSPORT MODELLING Luis Willumsen How to use Transport Models to assist decision making; assumptions and constraints Managing Transport Modelling

description

Short introduction to Transport Modelling for Managers.

Transcript of Transport Modelling for managers 2014 willumsen

Page 1: Transport Modelling for managers 2014 willumsen

ROLE OF TRANSPORT MODELLING

Luis Willumsen

How to use Transport Models to assist decision making; assumptions and constraints

Managing Transport Modelling

Page 2: Transport Modelling for managers 2014 willumsen

BETTER FORECASTS

We use models to improve decision making

How to use transport models to assist decision making.

• Better Models -> Better Forecasts • Better Forecasts -> Better Decisions

• Short Term Decisions -> Better Accuracy

• Long Term Decisions -> Better basis for decision making

2

Page 3: Transport Modelling for managers 2014 willumsen

3

• We cannot do experiments, in most cases

• Transport investments takes a long time so we must envisage future conditions that do not yet exist

• We can test alternative solutions in this modelled future

Why are models required?

Page 4: Transport Modelling for managers 2014 willumsen

4

Formulation of the problem

Test solutions in the model

Forecast underlying(planning) variables

Generate solution packagesto test

Build, calibrate andvalidate analytical model

Data collation andcollection

Model specification

Implement solutions

Evaluate solutions andrecommendations

Sensitivity and risk tests

How do we use models?

Page 5: Transport Modelling for managers 2014 willumsen

MO

DELS AN

D REALITY

The nature of Analytical Models

• Models are a simplification of reality based on some useful theoretical assumptions and sufficient data to estimate them

• We can deliver forecasts based on our models provided the theoretical assumptions remain reasonable and data about the future is reliable

• These conditions are never 100% met, so we need to interpret the results from our models

• Even perfect models will not be able to deliver totally accurate forecasts; some uncertainties cannot be eliminated

5

Page 6: Transport Modelling for managers 2014 willumsen

Accuracy and precision

• Sometimes pursuing precision may lead to inaccuracy

6

Page 7: Transport Modelling for managers 2014 willumsen

Accuracy and precision

• Sometimes pursuing precision may lead to inaccuracy• It is better to be approximately right than precisely wrong

ONE SHOT is not enough!

7

Page 8: Transport Modelling for managers 2014 willumsen

8

MO

DEL O

BJECTIVES

Assessing options• Scheme testing

• Policy testing

Assessing impacts• Traffic and congestion

• Economic (user, non-user benefits)

• Financial (revenue)

• Environmental (CO2, NOX emissions), Safety (accidents)

• Performance

• Equity: winners and losers

Forecasting usage (patronage, traffic levels etc.)

Where models can help

Page 9: Transport Modelling for managers 2014 willumsen

9

MO

DEL SPECIFICATIO

N

Need to understand and define the problem Have an idea of the possible solutions What are the main behavioural responses to these

solutions Resources Available: Time and Budgets, Data, Software,

Skills The costs of “getting it wrong”

Important to identify what the model should be able to do

And what the model will NOT be able to do Interpretation and post modelling adjustments are fair

How to specify a Model

Page 10: Transport Modelling for managers 2014 willumsen

10

MO

DEL D

ESIGN

STEPS

Scope of the model Area to be covered Behavioural responses to be included Government standards (WebTag, etc.) Data available Data to be collected Calibration/validation to be conducted of various stages Sensitivity Tests to be carried out

The resources spent in modelling and analysis should be related to the cost of a wrong decision

Model design steps

Page 11: Transport Modelling for managers 2014 willumsen

11

Source: ptv

AREA TO BE CO

VERED

Directly related to the scope of the model

Also related to the understanding travel patterns

Type of study, strategic or tactical

How much detail is needed for route choice and other responses

Area to be covered and level of detail

Page 12: Transport Modelling for managers 2014 willumsen

12

RESPON

SES

• Typical 4/5 stage model response. Change in:

• Number of trips• Destination• Mode used• Time of travel• Route used

• Generally• Small area models with

network issues –route choice and assignment only

• PT models – Mode Choice very important

• Strategic Models – All responses

Activity SystemNetworks and

services

Trip Generation/Attraction,trip frequency choice

Destination choice

Mode choice

Route choice andassignment

Flows by link and modeTravel costs, especially

time

Time of departurechoice

Responses to include

Page 13: Transport Modelling for managers 2014 willumsen

13

DATA AVAILABILITY

Data is the key in any model development taskEach response model requires different data; always look for what is already available

Previous studies, Census data, National Database, etc.

If no other study specific data is available then essential to carry out surveys, most commonly:

Traffic Counts, Bus, Rail, Metro occupancy surveys/boarding-alighting counts, Intercept surveys: Road Side Interviews and PT user OD surveys, Travel time and service quality surveys

A rich source of travel user behaviour is a detailed Household Interviews (HHI) or Household Travel Surveys (HTS) including a Travel Diary; generally a 1% to 2% sample.

Data availability

Page 14: Transport Modelling for managers 2014 willumsen

New sensors and probes:• Radar, loops

• CCTV

• ANPR

• SmartCards

• Transponders/Tags

• Bluetooth

• WiFi & Others

• GPS

• Mobile phones

The future is BRIGHT with new sources of mobility data

14

Page 15: Transport Modelling for managers 2014 willumsen

15

ESTIMATIO

N, CALIBRATIO

N AN

D VALIDATIO

N O

F MO

DEL

Three distinct aspects of model design:• Calibration of a model is finding the best values for its

parameters • Model estimation reflects the fact that the functional form is

not decided before hand in some cases• Model Validation follows post estimation and tests the

model against data not used in calibration; backcasting is ideal but seldom possible

Means different things for different response modelsDemand model elements: total demand, Trip Length Distribution, mode

shares: compared against observed totalsAssignment model: Traffic flows, Boarding Alighting, Patronage

Calibration, Estimation and Validation

Page 16: Transport Modelling for managers 2014 willumsen

DATA ERRORS

Sources of error: Data and ……. Models• Base Year Data errors: Poor and small sample, limited data

for calibration/validation; • Better methods and QA would reduce these errors

• Future Data errors in the input variables are unavoidable; typically

• GDP, Population, employment Growth• Key inputs like fuel prices, fares of competing modes• Parameter changes: willingness to pay, Values of Time; fixed

tastes assumptions

• And also future Scenarios (not entirely under control)• New Developments that do/do not materialise• Competing facilities that DO materialise• New disruptive technologies

16

Page 17: Transport Modelling for managers 2014 willumsen

NZ G

DP

New Zealand GDP forecasts

17

Page 18: Transport Modelling for managers 2014 willumsen

SPECIFICATION

ERRORS

Sources of error: Model specification

• Bad modelling• Poor sensitivity to price, too aggregate Values of Time• Poor models of delay, crowding• Ignoring relevant behavioural responses• Including irrelevant behavioural responses

• Overconfidence in the model and its implicit assumptions (that are never 100% right!):

• Users have perfect information and can process it efficiently

• Users respond instantly to changes

• Etc….

• Over specifying the model: unrealistic precision

18

Page 19: Transport Modelling for managers 2014 willumsen

Consider a modelling effort with 30 years planning horizon

Base year data is obtained the usual way (2-3 % HH survey plus some intercept, travel time and service surveys plus counts);

Model is calibrated/estimated for year 0; …and then applied for future periods, say every 10 years; Using projected planning data (GDP, population,

employment, etc. allocated over time and space); ..and some assumptions about changes in the technologies,

networks, competitors, economic regeneration, etc.

19

Page 20: Transport Modelling for managers 2014 willumsen

ERRORS IN

FORECASTS

Sources of errors in forecasts

The model. It is simplified, ignores some behavioural responses, network is not perfect, assumes constant parameters, etc. It is likely to be even more uncertain as time progresses

Base year data: small sample, on certain days, blended with distribution/destination choice models to produce semi-synthetic matrices: declining influence

Future year data: uncertainty grows over time horizon; gets more uncertain the more granular (disaggregated) it is: increasing influence

Scenario uncertainty: the elements we do not influence and cannot fully predict: competitors in a PPP, success of land use planning and control, policy changes, disruptive technologies: increases over time

20

Page 21: Transport Modelling for managers 2014 willumsen

FORECASTIN

G ERRO

RSSources of errors in forecasts

21

Page 22: Transport Modelling for managers 2014 willumsen

BETTER DATA AND

MO

DEL

With improved Base Year Data and Model

22

Page 23: Transport Modelling for managers 2014 willumsen

IMPACT O

F DATA REQU

IREMEN

TSModel with more disaggregated data requirements

23

Page 24: Transport Modelling for managers 2014 willumsen

MO

DELS AN

D EQ

UILIBRIU

MA model is always a simplification of reality

Some simplifications make our life easier, others are essentialTake Route Choice and Assignment

We model it as a steady state phenomenon: all vehicles clear the network during the modelled (peak) period.The software finds a solution with an iterative process until “equilibrium” is reached:Wardrop’s Equilibrium:Under equilibrium conditions traffic arranges itself in congested networks in such a way that no individual trip maker can reduce

his path costs by switching routes

Certain conditions with respect of the relationship between flow and travel time must be met to achieve this equilibrium

24

Page 25: Transport Modelling for managers 2014 willumsen

Issues

• If equilibrium of routes and costs is not reached the flows (and costs) depend on when do we stop the iterative process

• This may affect our choice of scheme

• However, reality is never in “equilibrium”• Noise in traffic (empty taxis, delivery vehicles, cruisers, lost)• Noise in the network: stalled vehicles, rain, poor light, roadworks• Day-to-day variability in the trip matrices

• Moreover, the better we represent delays at junctions, the more difficult (and sometimes impossible) it is to ensure “equilibrium”.

25

Page 26: Transport Modelling for managers 2014 willumsen

EQU

ILIBRIUM

Equilibrium Activity SystemNetworks and

services

Trip Generation/Attraction,trip frequency choice

Destination choice

Mode choice

Route choice andassignment

Flows by link and modeTravel costs, especially

time

Time of departurechoice

There is a risk of oscillations in the complete model for future runs:1. A lot of congestion, shift

to PT2. Less congestion, back to

car3. A lot of congestion, shift

to PT.4. Etc….

26

Page 27: Transport Modelling for managers 2014 willumsen

10 MESSAG

ES Ten messages to take with you

1. You may outsource the Model but not the Planning/Decision Making

2. Engage in Model Specification, Develop/retain capacity to interpret results, protect and transfer acquired knowledge

3. In specifying the model focus on What Matters; do not aim for the academic ideal

4. Decide how will you deal with the inevitable uncertainties

5. What assumptions were made in the model build? What assumptions are made about the future in the application?

6. Do results make sense? Check model results by other means (e.g. back of envelope calculations)

7. The model will be audited: good documentation & transparency essential; better if peer-reviewed throughout

8. Model results will be used in a wider context and by other disciplines; avoid intellectual imperialism

9. Is the model WebTAG compliant? If not, does it matter and how does it impact its reliability?

10. Has the model been validated through alternative techniques to WebTAG, e.g. benchmarking?

27

Page 28: Transport Modelling for managers 2014 willumsen

ROU

ND

UP

Round up

a. Models can be very useful tools to support decision makingb. Resources devoted to analysis should be aligned to the cost

of poor decisionsc. Government guidance often drive model specificationd. Models and their application, however validated, are not

error freee. There are significant risks in over-specifying a model: delays,

spurious accuracy and excessive faithf. But they can be very useful to compare solutions on a

common basis and eventually under alternative scenariosg. The interpretation of model results requires understanding

what they include and what they leave outh. …and judgment based on experience

28

Page 29: Transport Modelling for managers 2014 willumsen

THANK YOU

29