Analyzing electric vehicle battery health performance using supervised machine learning

Lithium-ion batteries having high energy and power densities, fast depleting cost, and multifaceted technological improvement lead to the first choice for electric transportation systems. A noble supervised K nearest neighbors (KNN), support vector regressor (SVR), decision tree (DT), and random forest (RF) regressors machine learning algorithm are developed with different accuracy and usefulness for the state of health estimation from direct measurable indices of voltage, current, and temperature without inherited electrochemical characteristics like voltage hysteresis, aging, degradation level, operational, and environmental effect. Based on the battery dataset from Sandia National Laboratories, the proposed technique is validated. The comparison characteristics are provided with the help of mean absolute percent error (MAPE), mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) techniques. The MAPE error at 15 °C for different regression algorithms SVR, DT, KNN, and RF are 2.791E-02, 1.607E-03, 9.957E-04, and 4.323E-03 respectively. The MAE error at 25 °C for different regression algorithms SVR, DT, KNN, and RF are 5.593E-02, 2.379E-03, 2.429E-03, and 5.073E-03 respectively. The RMSE error at 35 °C for different regression algorithms SVR, DT, KNN, and RF are 5.062E-02, 6.862E-03, 6.333E-03, and 1.275E-02 respectively. The MSE error percentages for the testing sample at 35 °C for different regression algorithms are 0.37 %, 0.01 %, 0.01 %, and 0.02 % respectively. The results indicate that KNN and DT exhibit higher precision when applied to lithium-ion batteries for electric vehicles. Collaborating with hardware manufacturers to exchange data has the potential to foster the creation of innovative battery health monitoring systems.