Optimizing city-scale traffic flows through modeling isolated observations of vehicle movements

Fan Yang, Alina Vereshchaka, Bruno Lepri, Wen Dong

Mobile phones and the Internet of Things provide unprecedented opportunities for transportation researchers and computational social scientists to observe city-scale human dynamics in terms of millions of vehicles or people moving around. They also enable policy researchers to identify the best strategies to influence the individuals in order for the complex system to achieve the best utility. However, the mobility data become sparse at the individual level and it is non-trivial to stitch together the isolated observations with high fidelity models to infer the macroscopic dynamics. In this paper, we introduce a discrete-event decision process to capture the high fidelity dynamics of a complex system at the individual level in terms of a collection of microscopic events where each one brings minimum changes but together induce complex behaviors. We further derive a particle filter algorithm to connect the dots of isolated observations through driving the discrete-event decision process in agreement with these observations. Finally, we solve a partially observable Markov decision process problem through reducing it into a learning and inference task. Evaluation with one synthesized dataset (SynthTown), one partly real and partly synthesized dataset (Berlin), and three real world datasets (Santiago de Chile, Dakar, and NYC) show that the discrete-event decision process gives an accurate estimation of complex system dynamics due to its better integration of high-fidelity dynamics and human mobility data.

Knowledge Graph



Sign up or login to leave a comment