Composite Travel Generative Adversarial Networks for Tabular and Sequential Population Synthesis

Agent-based microsimulation has become the standard to analyze intelligent transportation systems, using disaggregate travel demand data for entire populations, data that are not typically readily available. Population synthesis approaches are thus needed. We present Composite Travel Generative Adversarial Network (CTGAN), a novel deep generative model to estimate the underlying joint distribution of a population, that is capable of reconstructing composite synthetic agents having tabular (e.g. age and sex) as well as sequential mobility data (e.g. trip trajectory and sequence). The CTGAN model is compared with other recently proposed methods such as the Variational Autoencoders (VAE) method, which has shown success in high dimensional tabular population synthesis. We evaluate the performance of the synthesized outputs based on distribution similarity, multi-variate correlations and spatio-temporal metrics. The results show the consistent and accurate generation of synthetic populations and their tabular and spatially sequential attributes, generated over varying spatial scales and dimensions.

I. Introduction

Agent-based transportation microsimulation models study the interactions between the mobility of travelling agents and how urban systems, for instance, Intelligent Transportation Systems (ITS) operate and evolve through an individual’s daily activities [1]. While the data collection technologies are advancing, the availability of microdata still remains relatively limited owing to the high cost of acquiring reliable data and also the threat to privacy of the collection of spatially- and temporally-detailed information on individuals. In practice, government bodies (e.g. census agencies) conduct travel surveys on a sample of a population whose statistical characteristics are used to represent the behaviour of the entire population. Using sample data and other information (i.e. partial views) as base population information, researchers can reconstruct representative members of a population using synthesis techniques such as Iterative Proportional Fitting (IPF) [2], combinatorial optimization (CO) [3], or Markov chain Monte Carlo (MCMC) simulation [1].