Abstract: A state of health(SOH)evaluation method for LiFePO₄ (LFP) energy storage system based on cell-to-module transfer was proposed in this paper in order to solve the problems of slow data processing and poor evaluation results when traditional machine learning battery SOH evaluation methods were applied in the scenario of large-scale electrochemical energy storage power station. Aging datasets of LFP battery cells and modules were acquired through experiments in this paper. The SOH evaluation model framework of transfer learning was constructed and verified to evaluate the effect by few-shot sample retraining. The evaluation effects of models of LSTM(long short-term memory networks) and GRU(gated recurrent unit) were tested. The effect of different segments of short feature sample data on the results was compared. The research results verify that the battery cell model optimized by few-shot sample data can realize the evaluation of the SOH of the battery module. The transfer learning model using GRU as the main network has the best overall performance in evaluating the SOH. The model adopting a short feature data set with a voltage range of 24.5~30 V segments can further improve the evaluation accuracy and speed. The mean square error of the model's SOH evaluation of the battery pack can be reduced to 0.1%, which meet the needs of large-scale energy storage power station scenarios. The research results can provide possible technical reference and data support for the evaluation method of the operating status of electrochemical energy storage power plants.