Ultra-Short-Term Load Power Forecasting Method of Urban Rail TPSS Based on Local Information Spatial-Temporal Features

Load forecasting for the traction power supply system (TPSS) in urban rail transit (URT) presents new opportunities for energy management of regenerative energy utilization devices (REUDs), such as supercapacitor energy storage system (ESS) and energy feedback system (EFS). It also facilitates energy efficiency analysis for subway operators and supports the formulation of optimized power supply strategies. To address the challenge of limited information availability in practical engineering scenarios, this study proposes a method for predicting the load power of the entire adjacent power supply section only based on the local station information. By mapping local original features into a clustered space of operating conditions, spatial features that effectively characterize load power are extracted. Based on the composite spatial-temporal features and deep learning, online ultra-short-term load power prediction achieved. Experiments conducted on real-world subway lines, supported by comparative analyses of feature engineering and model engineering, demonstrate that the method significantly enhances prediction accuracy and generalization.