State-of-Health Estimation for Lithium-Ion Battery With Interpretable Analysis

The safe and sustainable development of battery energy storage systems (BESS) hinges on addressing safety monitoring and predictive maintenance. Solving these problems is an urgent priority and relies heavily on accurate battery state of health (SOH) estimation. However, the complex working environment affects the consistency and stability of operating data, which increase the difficulty of estimating battery SOH. To this end, a data-driven method based on improved gaussian process regression with matern automatic relevance determination (MARD) is proposed in this paper. The training of the model depends on feature selection; however, the redundant information between features is often ignored, thereby directly affecting the effectiveness of the model training process. Therefore, by using MARD kernel function, it can calculate the correlation degree between feature and battery SOH, while avoiding mutual redundancy between feature information. In addition, the top-ranked 10 relevant features with correlation degree are selected, and then count the probability of these 10 features in each working subinterval. Using these probability values to quantify the importance of the features and complete the interpretable analysis of the features on the estimation results. Finally, the accuracy of the proposed method is verified by using three performance metrics, which includes the mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). Compared with other data-driven methods, the proposed method has the highest estimation accuracy, and the maximum MAE, RMSE and MAPE value are below 0.007, 0.0082 and 0.9% under single temperature mode, while these values also does not exceed 0.0061, 0.0084 and 0.7% under the cross temperature mode.