Objective The traction load of urban rail transit features strong temporal correlation and high-frequency volatility, and is significantly influenced by the coupling of multi-dimensional meteorological factors and the impact of random passenger flows. Traditional forecasting methods struggle to balance prediction accuracy and robustness. To solve this problem, it is necessary to study a high-precision forecasting model that can effectively capture the complex evolution patterns of the load, facilitating the optimal dispatching and intelligent operation and maintenance of the traction power supply system.

Method Based on the measured data of a metro line in Wuhan, the spatio-temporal evolution characteristics of the traction load are analyzed, the offsetting and smoothing effect of the entire network-wide load relative to a single substation is revealed, and key influence mechanisms such as non-linear temperature driving, seasonal evolution, and calendar effects are quantified. An improved LSTM （long- and short-term memory） neural network forecasting model integrating a dual attention mechanism and multi-scale features is proposed, constructing a nested forecasting framework. A dual attention mechanism is introduced to dynamically weight time steps and external features to enhance the extraction ability of key causal variables. Parallel fine-grained and coarse-grained LSTM sub-networks are designed to capture the instantaneous high-frequency volatility and long-term trend characteristics of the load, respectively, and multi-scalar fusion is performed through adaptive weights. The model is trained and validated using historical time-series datasets.

Result & Conclusion  The MAPE （mean absolute percentage error） of the proposed improved LSTM forecasting model on the test sets decreased to 2.75%, and the goodness of fit R² reached 0.99. The model demonstrates excellent generalization ability under complex working conditions such as sudden high-temperature and holidays, providing a high-precision decision-making basis for the energy-saving operation and refined management of urban rail transit traction power supply systems.