Ultra-short-term forecasting of high-speed railway traction load integrating train operation diagrams and deep learning

Abstract Accurate forecasting of high-speed railway traction loads is of great significance for the power regulation and control of electrified railways. However, most existing forecasting methods fail to take into account the impacts of traffic organization and load characteristics on traction load variations. To address this issue, an ultra-short-term traction load forecasting method integrating deep learning and train operation diagrams is proposed in this paper. First, the characteristics of traction loads are analyzed, and the process of traction load forecasting is elaborated. Second, a model library of active power curves is established based on load characteristics, which serves as the data foundation for traction load forecasting. Third, a monitoring model is constructed based on the self-attention mechanism and bidirectional long short-term memory (BiLSTM). This model can effectively identify the moment when electric multiple units (EMUs) enter the power supply section. Finally, taking the data of a traction substation as an example, the proposed method is verified. The results demonstrate that the proposed method meets the requirements of forecasting accuracy, with a forecasting time horizon ranging from 6 to 20 min. Compared with existing traction load forecasting methods, the proposed method has stronger interpretability and generalization capability.