Abstract:

For the prediction of the battery state of health (SOH), an SOH estimation method that integrates the categorical boosting (CatBoost) algorithm and the autoregressive integrated moving average (ARIMA) models is proposed. The features are extracted through time series analysis, and the residuals of the ARIMA model are obtained. These residuals are used as additional features and processed by the CatBoost algorithm to construct an enhanced dataset for prediction.The experimental results demonstrate that the prediction performance is significantly enhanced by the proposed method.The optimal root mean square error (RMSE) reaches 0.004 6, the mean absolute error (MAE) reaches 0.003 4, and the coefficient of determination (R²) reaches 0.999 4. Compared with the model using only the initial data, it has higher accuracy and stability.

Key words: battery state of health, CatBoost algorithm, ARIMA model, residuals, enhanced dataset