Network-Wide Vehicle Trajectory Prediction in Urban Traffic Networks using Deep Learning

This paper proposes a deep learning approach to learning and predicting network-wide vehicle movement patterns in urban networks. Inspired by recent success in predicting sequence data using recurrent neural networks (RNN), specifically in language modeling that predicts the next words in a sentence given previous words, this research aims to apply RNN to predict the next locations in a vehicle's trajectory, given previous locations, by viewing a vehicle trajectory as a sentence and a set of locations in a network as vocabulary in human language. To extract a finite set of "locations,'' this study partitions the network into "cells,'' which represent subregions, and expresses each vehicle trajectory as a sequence of cells. Using large amounts of Bluetooth vehicle trajectory data collected in Brisbane, Australia, this study trains an RNN model to predict cell sequences. It tests the model's performance by computing the probability of correctly predicting the next k consecutive cells. Compared with a base-case model that relies on a simple transition matrix, the proposed RNN model shows substantially better prediction results. Network-level aggregate measures such as total cell visit count and intercell flow are also tested, and the RNN model is observed to be capable of replicating real-world traffic patterns.