Destination Choice Modeling of Discretionary Activities in Transport Microsimulations Andreas Horni

destination choice modeling for transport microsimulations

This Thesis problem: implementation of a MATSim destination choice module for shopping and leisure activities efficiently applicable for large-scale scenarios and easily adoptable by other simulation models consistent and efficient computation of quenched randomness destination choice utility function estimation agent interactions infrastructure competition modeling CA cruising-for-parking simulation results variabilityanalysis of temporal variability and aggregation and variability choice sets specification analysis contribute to microsimulation destination choice modeling efficiency and consistency

Basic Procedure instantiation microsimulation core Output input feedback U max (day chains) population situation (e.g. season, weather) situation (e.g. season, weather) choice model generalized costs census travel surveys infrastructure data estimatione.g., network constraints, opening hours e.g., socio- demographcis network load simulation constraints

Basic Procedure microsimulation core feedback choice model network load simulation  (usually non-linear) system of equations fixed point problem (== UE)

Evolutionary algorithm optimized plans Initial plans scoring replanning execution agent 1..n optimized plans initial plans scoring replanning execution MATSim agent 0 interaction species 1..n optimized population initial population recombination mutation survivor selection parent selection parents offsprings fitness evaluation fitness evaluation species 0 optimized population initial population recombination mutation survivor selection parent selection parents offsprings fitness evaluation fitness evaluation interaction Co-

Destination Choice & Other Frameworks TRANSIMS ALBATROSS PCATS search space draw from discrete choice model hierarchical destination choice (zone and intra-zonal choice) various constraints draw from decision trees time geography draw from discrete choice model

MATSim Destination Choice Approaches time-geographic space-time prismshollow prisms PPA time space t1t1 destination t0t0 origin distance r in,out = f(act dur) min (c travel ) min (c travel ) with  r < c travel <   r i i 

Unobserved Heterogeneity adding heterogeneity: conceptually easy, full compatibility with DCM framework MATSim: discrete choice modeling: but: technically tricky for large-scale application

Repeated Draws: Quenched Randomness fixed initial random seed freezing the generating order of  ij storing all  ij destinations persons  00  nn  10  ij i person i (act q ) store seed k i store seed k j regenerate  ij on the fly with random seed f(k i,k j ) one additional random number can destroy «quench» i,j ~ O(10 6 ) -> 4x10 12 Byte (4TByte) alternative j

Search for U max global optimum local optimum space travel disutility → restrain search spaceexhaustive search  i,j U

Search for U max : Search Space Boundary approximate by distance realized utilities with Gumbel distribution pre-process once for every person  max –  t travel = 0 search space boundary d max := ? d max := distance to destination with  max A 0 = πr 2 A 1 = π(2r) 2 - πr 2 = 3πr 2 A 2 = π(3r) 2 - 4πr 2 = 5πr 2 A 3 = π(4r) 2 - 9πr 2 = 7πr 2 A r

Search for U max in Search Space t departure t arrival Dijkstra forwards 1-nDijkstra backwards 1-n approximation probabilistic choice search space workhome shopping exact calculation of tt for choice

Results 10% Zurich Scenario shopping leisure 70K agents iteration: 10 days  5 minutes link volumes

Conclusions ZH scenario: 10 days  5 minutes (iteration) but: module still needs to be faster for CH scenario improve sampling, sample correction factor more validation data with more degrees of freedom procedure for quenched randomness important in all iterative stochastic frameworks