Agent- and activity-based large-scale simulation of enroute travel, enroute refuelling and parking behaviours in Beijing, China

Abstract

This paper develops an agent- and activity-based large-scale simulation model for Beijing, China (MATSim-Beijing) to explicitly simulate enroute travel, enroute refuelling and parking behaviours, as well as the associated vehicular energy consumption and emissions, based on MATSim (Multi-Agent Transport Simulation), which is a typical integrated activity-based model. In order to take into account heterogeneous parking and refuelling behaviours, the MATSim-Beijing model incorporates several Multinomial Logit (MNL) models to predict individual choices about the maximum acceptable times of walking from trip destination to parking lot, of diverting to a refuelling station and of queuing at a station, using the data collected in a paper-based questionnaire survey in Beijing. A Sensitivity Analysis (SA) -based calibration method was used to estimate the model parameters by searching for an optimal parameter combination with the objective of minimize the gap between simulated and observed traffic flow data, exhibiting a relatively good performance of decreasing the Mean Absolute Percentage Error (MAPE) by around 23%. Further, the calibrated model was used to investigate whether and how the population scaling and network simplification, which were two commonly used approaches to speeding up large-scale traffic simulations, might influence model accuracy and computing time. The results indicated that both approaches could to some extent influence model outputs, though they could significantly reduce computing time.