SOH Estimation of Power Battery Based on the Real Vehicle Multi-feature and LightGBM Algorithm

Abstract

State of Health (SOH) estimation of power batteries is the core of ensuring the safe driving of electric vehicles. By using the data-driven method, considering the difference between the driving data of the real vehicle and the experimental data, the aging characteristics suitable for the real vehicle and which can characterize the battery aging trend are extracted, including a total of 9 features, such as capacity increment curve characteristics, voltage, accumulated mileage, and current. LightGBM, a machine learning algorithm using grid search optimization, is combined to estimate the SOH of real vehicles, and the results show that for the data samples provided by the National New Energy Big Data Alliance, the root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) of the method in the No. 9 vehicle test set reach 2.39% and 1.86%, respectively. The fitted curve R2 of the predicted value and the true value reached 0.88, and the RMSE and MAPE reached 0.25% and 0.21%, respectively.