Data-Driven Transit Simulation

Abstract

This thesis details the design of a modified MATSim agent-based simulation system that is driven by transit smart card data. A complete set of bus vehicle trajectories are reconstructed from the smart card transactions. The reconstructed trajectories are then used to estimate a number of regression models of the speed at which buses travel between stops. These models drive the simulation dynamics, along with a dwell time model which takes account of bus ridership, in a MATSim simulation where smart card transaction records have been transformed into an agent-based travel demand. Besides an exposition of the various elements of the data-driven system, as well as validation of the simulation output and a test case scenario, the thesis also covers a number of related topics, including a technique to improve simulation times for MATSim simulations in general, a method to distribute MATSim simulations across multiple computer nodes, a useful approach to perform parameter optimisation in MATSim using a Kriging meta-model, and ends with a technique to synthesise surrogate data that could be used to drive the simulation, while eliminating any associated privacy concerns.