Real-Time Adaptive Traffic Flow Prediction Based on a GE-GRU-KNN Model

Traffic flow prediction is an important part of urban intelligent transportation systems. However, due to strong nonlinear characteristics and spatiotemporal correlations of the traffic within the network, traffic flow prediction has been a challenging task. In order to capture the spatiotemporal correlation, and improve the traditional methods of using predefined adjacency matrices that cannot effectively characterise the dynamic correlation of traffic flow, a GE-GRU-KNN model for predicting the road traffic flow is proposed. Specifically, the spatial representation of the road network learned by GE is used to automatically extract the spatial features of the network; GRU is used to learn the nonlinear characteristics of the time series to capture the temporal correlation of the traffic flow; finally, the KNN algorithm is introduced to combine real-time traffic flow and historical data and adaptively update the fusion weights of predicted values for different road sections. The method enables the model to effectively characterise the dynamic correlation of traffic flow. An experiment using traffic flow data from 22 detectors on California freeways is conducted. The results show that compared with traditional methods, the prediction error of this method is reduced by 1.08%–14.71%, indicating that the hybrid GE-GRU-KNN model exhibits good performance.