In order to monitor the operating status of the power equipment, remove a hidden danger in the power equipment to avoid accidents and give early warning, reduce maintenance costs and ensure unpausing operation pf the power equipment, a fault prediction method of rail transit power equipment based on deep learning is proposed. Based on the historical data and real-time monitoring data in the fault case database, fault prediction analysis of the power equipment is realized through random forest and LSTM prediction model: (1) taking the exploratory data analysis (EDA) to improve missing and abnormal values, and correct the data; (2) Based on random forests, performing feature selection and simplifying the power equipment data; (3) The power equipment data is converted into three-dimensional data composed of samples, time steps, and features. LSTM (Long Short-Term Memory) prediction model is constructed to perform equipment failure analysis and prediction. This method can prevent maintenance of power equipment early, and provide a basis for scientific maintenance management, which greatly improves the reliability of the power equipment, reduces the management cost of the power equipment, and also avoid operating losses from the failure of the power equipment.