PLSTM: Long Short-Term Memory Neural Networks for Propagatable Traffic Congested States Prediction

Abstract

The accurate prediction of traffic congested states in major cities is indispensable for urban traffic management and public traveling routes planning. However, the understanding of traffic congestion propagation has not raised much concern. Traffic congestion propagation reflects how the current congested roads will affect their connected roads, which is vital to improve prediction accuracy of traffic conditions. In this paper, we propose a novel method named PLSTM to further explore the characteristics of traffic congestion propagation and predict short-term traffic congested states, which is a long short-term memory (LSTM) neural network for modeling traffic propagation. Firstly, we consider local spatial-temporal correlation of congestion and integrate the data into input series. Secondly, the PLSTM component that comprises multi-LSTM layers is trained with the input series. Finally, we conduct various contrast experiments with state-of-the-art predictors to evaluate the performance of PLSTM. The experimental results have validated the rationality of input series on improving prediction accuracy and the effectiveness of PLSTM.