A Method for Predicting Traction Load of Electrified Railways Considering Spatiotemporal Correlation Characteristics

作者：Qiang Gao, Hongbo Cheng, Shaohua Zheng, Shouxing Wan, Wuzhao Li · 发表于：IET Electrical Systems in Transportation · 年份：2025 · DOI：10.1049/els2/5516562

In the prediction of traction loads for electrified railways, conventional forecasting methods often focus exclusively on temporal correlations within historical data from individual substations. However, traction loads are profoundly affected by train schedules and exhibit substantial spatial interdependence across different substations. To address this limitation, this study proposes a hybrid model integrating a graph convolutional network (GCN) and a bidirectional long short‐term memory (BiLSTM) network, which comprehensively incorporates both spatial and temporal dependencies to significantly improve ultrashort‐term prediction accuracy. The proposed framework operates in several stages. First, spatial correlations among regional substations are captured using a GCN. To mitigate the risk of including spurious connections—often referred to as “pseudo‐adjacency” relationships—the adjacency matrix is refined using Pearson correlation coefficients, thereby strengthening the model’s representation of meaningful spatial interactions. The spatial features extracted by the GCN at consecutive time steps are then organized into a temporal sequence and input into the BiLSTM module. To further enhance temporal modeling, an attention mechanism is incorporated to adaptively weigh the importance of hidden states, enabling the model to focus on the most relevant temporal information. This integrated approach results in a notable improvement in the accuracy of traction load power forecasti...