HYBRID MODELING APPROACH FOR STATE-OF-CHARGE AND STATE-OF-HEALTH ESTIMATION IN BATTERIES UNDER VARIABLE LOAD CONDITIONS

Abstract

Electric Vehicle (EV) effectiveness critically depends on advanced Battery Management Systems (BMS), where accurate Lithium-ion battery State of Health (SOH) estimation is paramount. Traditional SOH indicators (internal resistance, capacity) are impractical for online monitoring under real-world EV conditions, which preclude complete charge-discharge cycles. Consequently, State of Charge (SOC) becomes the primary parameter for energy management, though its estimation is challenged by measurement noise and model limitations. Leveraging AI advancements, this work proposes a hybrid Long Short-Term Memory (LSTM) and Artificial Neural Network (ANN) model for simultaneous SOC and SOH prediction under variable loads. Rigorous evaluation on benchmark datasets demonstrates superior performance: For SOC, the model achieved 98.7% accuracy, with a low MAE of 0.8% and RMSE of 1.1%, significantly outperforming standalone ANN (MAE: 1.8%, RMSE: 2.4%) and equivalent circuit model (MAE: 2.5%, RMSE: 3.2%) baselines, especially during transients. For SOH, it achieved an F1-score of 0.96, capacity MAE of 1.2%, and RMSE of 1.5%, confirming reliable degradation tracking using only operational data. These metrics validate the hybrid model as a highly accurate, reliable solution for real-time BMS, enhancing EV performance, safety, and longevity