Research on the Health Status Evaluation Model of Real Vehicle Power Battery Based on Attention-GRU Algorithm

There are issues with poor data quality, difficulty in obtaining initial State of Health (SOH), and poor comparability of calculated SOH under different operating conditions in evaluating the health status of power batteries in actual vehicles. This article establishes an SOH evaluation model suitable for step charging of actual vehicle batteries. Firstly, we need to preprocess the actual vehicle data by cleaning abnormal data through box plots, cutting charging segments, and linear filling; Secondly, SOH is calculated within different step currents, and temperature compensation for SOH is achieved by analyzing and fitting the relationship between capacity and temperature; Secondly, we need to extract battery health factors from two dimensions: current operating condition characteristics and historical cumulative characteristics, and conduct Spearman correlation analysis; Finally, attention mechanism was introduced to optimize the GRU network, and an Attention-GRU model was established to validate a total of 309 sets of actual vehicle charging data. Compared with the LSTM model, the RMSE of the Attention-GRU model did not exceed 2.14%, MAPE did not exceed 2.2%, and R2 was not less than 0.91. Moreover, it has significant advantages in reducing training time and adapting to SOH accelerated decay.