Electric Vehicle State-of-charge Prediction using Deep LSTM Network Model

The successful adoption of electric vehicles (EVs) depends upon two major factors. One is the accurate prediction of the vehicle's driving range, and the other one is the optimal charge-discharge control of the battery by the battery management system. Both factors depend on the vehicle battery's state of charge (SOC). In this paper, a deep long short-term memory (LSTM) network is proposed to predict the battery SOC using the two real-world historical datasets of a Nissan Leaf EV. Battery current, voltage, temperature, vehicle speed, tractive effort, and road elevations are input features for the deep LSTM model. The evaluation of the model is done using training and testing accuracies like root mean square error (RMSE), mean absolute error (MAE), and training loss. The results show that the RMSE and MAE values are 0.0239, 0.0202, 0.0449, and 0.0412, respectively, for dataset-I and dataset-II. Moreover, during the training progress, decrement in the RMSE and loss values with respect to the number of iterations show the effectiveness of the proposed model.