Research on Short-term Traffic Flow Prediction Based on KNN-GRU

Abstract

Traffic flow prediction is all-important to the intelligent transportation system (ITS). However, the traffic flow data is complex and varied, there is a strong connection between nodes of the urban network, while the road traffic volume changes dynamically over time. For the sake of obtain the spatio-temporal dependence of traffic volume, this paper raise a traffic flow prediction method based upon the k-order neighbor algorithm and gated recurrent unit. Using the Euclidean distance to figure the spacial correlation between traffic networks and the gated recurrent neural network obtains the temporal dependency of traffic volume. Using the California highway data set verified the accuracy of our model. The experimental result expresses that the accuracy of the K-Nearest Neighbor-Gated Recurrent Unit (KNN-GRU) prediction model pass beyond that of the traditional simplex prediction method. Compared with Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM)model, the prediction accuracy of the KNN-GRU model is about 3% exceed in that of the GRU and LSTM model; and 29.31% higher than that of Graph Convolutional Network (GCN). The results show that the proposed model has better predictive performance.