Bidirectional Spatial–Temporal Graph Convolutional Model: Traffic Flow Forecasting With Enhanced Extended Capabilities

Traffic flow forecasting, as a crucial component of intelligent transportation systems (ITS), enables the prediction of future traffic conditions based on historical traffic data, thereby optimizing travel strategies and achieving the goal of reducing traffic congestion. Considering the limited nature of specific road network spatial structures, specific road network datasets often overlook the influence of surrounding networks on the network itself, motivating the need for a framework that captures boundary interactions. This paper introduces the bidirectional spatial–temporal expanded graph convolutional model (Bi‐STEGCM) to traffic flow forecasting. This addresses the limitations of conventional models, particularly in capturing spatial features and managing missing or anomalous data. The Bi‐STEGCM reconstructs and aggregates traffic data while preserving the temporal dynamics of traffic flow. This offers a more nuanced representation of the spatiotemporal dynamics within road networks. The model utilizes causal convolution for temporal feature extraction and an auto‐regressive moving average (ARMA) filter for spatial feature extraction. It integrates these with bidirectional graph convolution to aggregate spatial features across various layers. Validation using real‐world traffic datasets PEMS03, PEMS04, PEMS07, and PEMS08 demonstrates that the Bi‐STEGCM outperforms state‐of‐the‐art models, including spatial–temporal synchronous graph convolutional networks (STSGCN) and spatial–temporal fusion graph neural networks (STFGNN), across three key evaluation metrics. Notably, the Bi‐STEGCM requires significantly fewer parameters and less training time than its counterparts, rendering it a more efficient and effective solution for traffic flow forecasting tasks.