Generating synthetic population with activity chains as agent-based model input using statistical raster census data
Publication from Life
Kompetenzgruppe Urban Living Lab
Felbermair Samuel, Lammer Florian, Trausinger-Binder Eva, Hebenstreit Cornelia
Procedia Computer Science Volume 170, 2020, Pages 273-280, 4/2020
Agent-based transport modelling needs more detail on the synthetic population compared to conventional transport models, as activity chains are required. In many cases, however the sample size of travel surveys from which to gain activity chains is small. Using Bayesian networks and Markov Chain Monte Carlo as well as stratified sampling, we show how a population with activities plans can be generated using limited survey data.
Moreover, this paper presents a method for using statistical raster (250 m) census data for all activities and facilities, which guarantees a high spatial resolution. The synthetic population was developed for the predominantly rural to intermediately urban state of Carinthia in Austria. Realistic travel plans were assigned to each agent, considering trip dependencies between household members as well as correlations between socio-demographic attributes and travel behaviour. The resulting synthetic population includes agents with a sequence of activities for 24 hours. The activities and trip length distributions of the simulated population fit the survey data well. The simulation results fit the traffic counts.
Keywords: agent-based modelling, transport modelling, MATSim, population synthesis, Markov Chain Monte Carlo, Bayesian Networks