A health monitoring and fault early warning system based on a Long Short-Term Memory (LSTM) model and nuclear density estimation is proposed. The system is composed of two parts. The first part is the health assessment module based on nuclear density estimation. This can periodically inspect all energy storage batteries in an energy storage station. Based on the historical data collected by BMS, the health status of battery cells can be checked in good time using the nuclear density estimation method. The other part is the fault early warning module based on the LSTM model. LSTM can predict dynamic changes of monitoring variables over a long period, combined with the control limit estimated by the nuclear density, so as to give early fault warning. The experimental results show that by analyzing the historical data of the actual power grid operation, the effectiveness and practicability of the above method are verified. It is proved that the method can effectively reduce the rate of missed faults or false alarms caused by an inaccurate control limit. This work is supported by the National Natural Science Foundation of China (No. 61973123) and the Fundamental Funds for the Central Universities (No. JKH012016024).