Research on Regional Traffic Flow Prediction Based on MGCN-WOALSTM

Regional traffic flow forecasting is the key to the realization of intelligent transportation system. The existing traffic flow forecasting methods have problems such as insufficient Spatio-temporal Correlation modeling and ignoring the impact of weather factors, which lead to high prediction errors. Therefore, in this study, a Multi-channel Graph Convolutional Neural network (MGCN) was first established to analyze and express the spatial correlation of traffic flow on different dimensions, and a self-attention mechanism was used to weight the spatial correlation features of MGCN output. Next, an LSTM is built after the MGCN network layer to obtain temporal features, an Embedding layer is added to embed traffic flow temporal periodic features, and the Whale Optimization Algorithm (WOA) is introduced to find the global optimal LSTM network parameter combination, which is then applied to the prediction model. The performance of the model was tested using the public dataset PeMSD4 and corresponding weather data. Compared with prediction models that were not optimized by WOA and did not consider the influence of weather factors (GCN, LSTM, ASTGCN, etc.), the prediction errors RMSE, MAE, and MAPE of the final constructed prediction model were reduced, indicating that the MGCN-WOALSTM model has better traffic flow prediction performance.