Preferred citation style

24
1 Preferred citation style Axhausen, K.W. and K. Meister (2007) Parameterising the scheduling model, MATSim Workshop 2007, Castasegna, October 2007.

description

Preferred citation style. Axhausen, K.W. and K. Meister (2007) Parameterising the scheduling model, MATSim Workshop 2007, Castasegna, October 2007. Parametrising the scheduling model. KW Axhausen and K Meister IVT ETH Zürich October 2007. Detour: Why social networks ?. - PowerPoint PPT Presentation

Transcript of Preferred citation style

Page 1: Preferred citation style

1

Preferred citation style

Axhausen, K.W. and K. Meister (2007) Parameterising the scheduling model, MATSim Workshop 2007, Castasegna, October 2007.

Page 2: Preferred citation style

Parametrising the scheduling model

KW Axhausen and K Meister

IVTETHZürich

October 2007

Page 3: Preferred citation style

3

Detour: Why social networks ?

Page 4: Preferred citation style

4

Distance distributionF

req

ue

nc

y

250

200

150

100

50

0

Great circle distance [km]100'00010'0001'0001001010

Page 5: Preferred citation style

5

Example of a social network geography

Page 6: Preferred citation style

6

Size of network geometries

95%-confidence ellipse of the social network geography

1.E10

1.E91.E81.E71.E61.E51.E41.E31.E21.E11.E0

Pe

rce

nt

40

30

20

10

0

Page 7: Preferred citation style

7

Contacts and population shares

0.00

0.05

0.10

0.15

0.20

0 10 20 30 40 50 60 70 80 90 100Distance band [km]

Sha

re [%

]

0.0

1.0

2.0

3.0

4.0

Rat

io []

Share of contacts [%]

Share of population [%]

Ratio of contact and population shares

Ratios at 1km: 39; 2km: 9; 3km: 5

Page 8: Preferred citation style

8

Contact frequencies by distance band

10000.001000.00100.0010.001.000.001

100.00

80.00

60.00

40.00

20.00

0.00

Great circle distance (km)

10000.001000.00100.0010.001.000.001

100.00

80.00

60.00

40.00

20.00

0.00

SMS messages/year Great circle distance [km]

Email messages/year Great circle distance [km]

Phone calls/year Great circle distance [km]

Face-to-face visits/year Great circle distance [km]

Page 9: Preferred citation style

9

End of detour – So why parametrisation ?

We use uniform current wisdom values

We need:

• Locally specific values• Heterogenuous values

Page 10: Preferred citation style

10

Degrees of freedom of activity scheduling

• Number (n ≥ 0) and type of activities• Sequence of activities

• Start and duration of activity• Group undertaking the activity (expenditure share)• Location of the activity

• Connection between sequential locations

• Location of access and egress from the mean of transport

• Vehicle/means of transport• Route/service• Group travelling together (expenditure share)

Page 11: Preferred citation style

11

2007: Planomat versus initial demand versus ignored

• Number (n ≥ 0) and type of activities• Sequence of activities

• Start and duration of activity• Group undertaking the activity (expenditure share)• Location of the activity

• Connection between sequential locations

• Location of access and egress from the mean of transport

• Vehicle/means of transport• Route/service• Group travelling together (expenditure share)

Page 12: Preferred citation style

12

Generalised costs of the schedule

Risk and comfort-weighted sum of time and money expenditure:

• Travel time• Late arrival• Duration by activity type• Expenditure

Page 13: Preferred citation style

13

Generalised costs of the schedule

Risk and comfort-weighted sum of time and money expenditure:

• Travel time• By mode (vehicle type)• Idle waiting time• Transfer

• Late arrival by group waiting and activity type• Duration by activity type

• By time of day/group• Minimum durations• By unmet need (priority)

• Expenditure

Page 14: Preferred citation style

14

Generalised costs of the schedule

Risk and comfort-weighted sum of time and money expenditure:

• Travel time• By mode (vehicle type)• Idle waiting time• Transfer

• Late arrival by group waiting and activity type• (Desired arrival time imputation via Kitamura et al.)

• Duration by activity type• By time of day/group• Minimum durations• By unmet need (priority) (Panel data only)

• Expenditure – Thurgau imputation; Mobidrive: observed

Page 15: Preferred citation style

15

Approaches

Name Need for Estimationunchosenalternatives

Discrete choice model Yes MLWork/leisure trade-off No MLW/L & DC (Jara-Diaz) (Yes) ML

Time share replication (Joh) No Ad-hoc

Rule-based systems No CHAID etc.Ad-hoc rule bases No Ad-hoc

Page 16: Preferred citation style

16

Criteria

• How reasonable is the approach ?

• How easily can the objective function by computed ?

• Are standard errors of the parameters easily available ?

• Can all our parameters be identified ? Can we estimate means only ?

• What is the data preparation effort required ?

• Do we need to write the optimiser ourselves ?

Page 17: Preferred citation style

17

Frontier model of prism vertices (Kitamura et al.)

• Idea: Estimate Hägerstrand’s prisms to impute earliest departure and latest arrival times

• Approach: Frontier regression (via directional errors)

• Software: LIMDEP

Page 18: Preferred citation style

18

PCATS (Kitamura, Pendyala)

• Not a scheduling model in our sense

• Idea: Sequence of type, destination/mode, duration models inside the pre-determined prisms

• Target functions: • ML (type, destination/mode, number of activities)• LS (duration)

• Software: Not listed (Possibilities: Biogeme; LIMDEP)

Page 19: Preferred citation style

19

TASHA (Roorda, Miller)

Not quite a scheduling model in our sense

• Idea: Sequence of conditional distributions (draws) by person type:

• Type and number of activities• Start time• Durations

• Rule-based insertion of additional activities

No estimation as such; validation of the rules

Page 20: Preferred citation style

20

AURORA - durations (Joh, Arentze, Timmermans)

• Idea: • Duration of activities as a function of time since last

performance ( time window and amount of discretionary time)

• Marginal utility shifts from growing to decreasing

• Target function: Adjusted OLS of activity duration under marginal utility equality constraint

• Software: Specialised ad-hoc GA

• See also: Recent SP, MNL & non-linear regression (including just decreasing marginal utilities functions)

Page 21: Preferred citation style

21

W/L tradeoff with DCM (Jara-Diaz et al.)

• Idea: Combine W/L with DCM to estimate all elements of the value of time

• Value of time savings in activity i

μ: Marginal value of time

λ: Marginal value of income

μ/λ: Value of time as a resource

i i w iK U T U T U Tw

Page 22: Preferred citation style

22

W/L tradeoff with DCM (Jara-Diaz et al.)

• Idea: Combine W/L with DCM to estimate all elements of the value of time

• Target function:

• Cobb-Douglas for the work/leisure trade-off

• DCM for mode choice

• Estimation: LS for W/L trade-off; ML for DCM

Page 23: Preferred citation style

23

Discrete continuous multivariate: Bhat (Habib & Miller)

• Idea: Expand Logit to MVL and add continuous elements

• Target function: closed form logit

• Estimation: ML

• Example: Activity engagement and time-allocated to each actvity

Page 24: Preferred citation style

24

Issue:

• Various frameworks for activity participation and time allocation

• No joint model including timing