Objective Addressing the demands for load status monitoring and early warning in metro power supply systems, as well as the inadequate equipment operation and maintenance efficiency under the planned maintenance mode, it is necessary to provide early warnings for the heavy load status of core electrical equipment. It is aimed to promote the transformation of the metro operation and maintenance mode from planned maintenance to condition-based maintenance.

Method Taking Liuliqiao Station on Beijing Subway Line 9 as an example, the load characteristics of three types of core electrical equipment, 10 kV switchgear, traction transformers, and incoming line switchgear of the traction system, are selected for comparison. Based on the current time-series data collected by the SCADA （supervisory control and data acquisition） system, a quantitative calculation of load status is established to form a load status matrix. Combined with expert experience, the load levels are classified into light, medium, and heavy, the LSTM （long short-term memory） network model is employed to conduct short-term predictions. A hyperparameter optimization model training process is reasonably configured, and a multi-indicator comprehensive evaluation is adopted to assess the model prediction performance.

Result & Conclusion  The prediction trend of the LSTM model for the load status of core equipment highly aligns with the actual data, fully validating the robustness of the proposed model. This model can timely predict heavy load levels, achieve predictive early warnings, and provide reliable data support for condition-based maintenance of equipment.