Traffic Flow Forecasting Based on Complex Gray Relational Degree Network and Multi-Period Exattention-GRU Model

Traffic flow forecasting is vital but challenging in traffic management. This study proposes a complex gray relational degree-based network to cluster traffic flow observation nodes using the Louvain algorithm. The method effectively distinguishes the nodes, providing accurate clustering for equivalent training sets and traffic flow prediction. To capture the characteristics of time series data in traffic flow data, the gated recurring unit (GRU) is the backbone network. For improved accuracy, a multi-period Exattention-GRU model is developed. GRU enhances network performance as the main component with an external attention mechanism and a residual structure. Stacking and fusing the improved GRU network captures various traffic information, which contributes to improving accuracy. The model, trained on an equivalent set using the RAdam algorithm, demonstrates effectiveness with favorable prediction results across various datasets and traffic conditions. Taking into account 36 data sets, the average metrics for RMSE, MAPE and MAE are recorded as 60.6363, 7.40%, and 41.4897.