Traffic Congestion Prediction Based on Long-Short Term Memory Neural Network Models

Abstract

Predicting urban network congestion and exploring congestion mechanisms are vital for both transportation researchers and practitioners. The state-of-the-art studies rely on either mathematical equations or simulation techniques to depict the traffic congestion evolution. However, most of the existing studies tend to make simplified assumptions since transportation activities involve complex human factors which are difficult to represent or model accurately using mathematics-driven approaches. In this paper, long-short term memory neural networks (LSTM NN) are employed to interpret traffic congestion in terms of traffic speed. Traffic speed predictions are also made by considering both temporal and spatial correlation information. The proposed approach is tested on different links in one road network in Beijing, China. The results demonstrate the advantage of LSTM NN for analyzing the complex non-linear variations of traffic speeds as well as its promising prediction accuracy.