Optimization of Hyperparameters for Accurate State of Charge Estimation using Deep Neural Network for Electric Vehicles

The safety and effectiveness of Lithium-Ion (Liion) batteries in Electric Vehicles (EVs) depend on the estimation of State of Charge (SOC). The accurate SOC of Liion batteries cannot be easily determined with the available technology because of their significant time-varying and nonlinear features. This research investigates the optimization of hyperparameters for Deep Neural Networks (DNNs) for SOC estimation for EV applications. This article discusses different optimization techniques and is evaluated over a variety of drive cycles under various operating conditions. The study demonstrates that a two-layer DNN model trained with Adaptive Moment Estimation (Adam) optimizer consistently exceeds other optimizers with lowest Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) at -10º, 0º, 10º, and 25ºC temperatures. The outcomes emphasize the relevance of hyperparameter adjustment in enhancing SOC prediction accuracy and offer guidance for constructing robust DNN-based SOC estimation models for real-world Battery Management Systems (BMS).