On-line Multi-step Prediction of Short Term Traffic Flow Based on GRU Neural Network

Abstract

Strengthened road traffic flow monitoring and forecasting can ease road traffic congestion and facilitate road traffic safety planning. Multi-step ahead of the ability to predict the traffic flow is particularly important. The monitoring data of road traffic flow is characterized by uncertainty and non-linearity. And using the existing methods to carry out multi-step prediction error will be very large. In this paper, based on these feature, we propose GRU neural network and autocorrelation analysis for multi-step prediction. We make this model dynamically update the network with the input of the measured real-time data, namely on-line prediction, to work effectively and constantly. Through the theoretical derivation and simulation analysis, it is shown that the prediction accuracy of the proposed GRU prediction model is improved. The model can be used as an effective method for multi-step traffic prediction.