Lithium-ion Battery State of Energy Estimation Using Deep Neural Network and Support Vector Regression

For the advancement of an effective energy management system for an electric vehicle (EV) application, it is constantly needed to utilize a precise battery state of energy (SOE) estimation method. In this study, two different data- driven SOE estimation methods using deep neural network (DNN) and support vector regression (SVR) are compared. The electric vehicle drive cycles dataset is utilized for training, validation, and testing. Three drive cycle data sets such as DST, FUDS, US06 are utilized for training and validation. Whereas, the WLTP drive cycle is considered for testing. The optimum hyperparameters are obtained by gradient search CV optimization method for both DNN and SVR. Two different testing datasets (e.g., known, and unknown) are considered for the evaluation of SOE estimation using the DNN and SVR models. The SOE estimation results demonstrated the high accuracy of DNN over SVR under the same dynamic operating conditions. For WLTP drive cycle datasets, the recorded value of the estimated SOE RMSE using DNN and SVR are 2.0527 and 9.0688, respectively. The value of the estimated SOE MAE using DNN and SVR are 0.00421 and 0.0822, respectively.