SSA(BBO)-optimized neural networks for remaining useful life estimation and health monitoring of lithium-ion batteries.

Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for advancing battery health management, enhancing safety, and improving the efficiency of electric vehicles and energy storage systems. This study introduces a novel hybrid deep learning framework that integrates a deep neural network (DNN) with advanced optimization algorithms, including Biogeography-Based Optimization (BBO) and the Improved Sparrow Search Algorithm (ISSA). The framework further employs a competitive adversarial learning mechanism utilizing a linear Support Vector Machine (SVM) as a discriminator. Unlike conventional approaches that primarily rely on past cycle data under constant conditions, the proposed model incorporates real operating parameters-such as variable charge/discharge C-rates, load profiles, ambient temperature, depth of discharge (DoD), and driving cycle statistics-as input features. This significantly reduces uncertainty and enhances the model's generalizability across diverse operating conditions. Within this framework, the DNN acts as the RUL predictor (generator), while the SVM formulates a min-max optimization problem to penalize physically implausible predictions. Validation on NASA-based datasets augmented with realistic variable operating profiles demonstrates that the proposed ISSA-optimized DNN-SVM model achieves excellent performance with RMSE of 2.25 ± 0.12 cycles, MAE of 1.68 ± 0.09 cycles, and R² of 0.995 ± 0.002 (mean ± std. dev. over 10 independent runs). These results represent a substantial improvement of 58-72% in RMSE compared to standard methods (including the baseline Autoencoder-DNN, LSTM, and CNN-LSTM). The framework also significantly outperforms the BBO-optimized variant, confirming the effectiveness of ISSA in global parameter optimization. These findings underscore the proposed approach's high robustness under real-world variable operating conditions and its strong potential for real-time deployment in battery management systems (BMS) after offline training, marking a significant step toward extending the lifespan and reliability of lithium-ion batteries.