Publication

Abstract : Accurate parameter estimation of lithium-ion battery Equivalent Circuit Models (ECMs) is essential for precise State-of-Charge (SOC) estimation and optimized control in Battery Management Systems (BMS). Conventional estimation methods face limitations when modeling nonlinear and time-varying battery characteristics, and they require extensive experimental data. To address these challenges, this paper presents a novel hybrid deep-learning framework that integrates a Long Short-Term Memory–Recurrent Neural Network (Hybrid LSTM-RNN) with Transfer Learning (TL) for enhanced parameter estimation of the Second-Order (2-RC) ECM. The proposed architecture leverages LSTM capabilities to learn long-term temporal dependencies in current–voltage sequences, while TL improves generalization across different battery domains with minimal target data. Experimental validation using Hybrid Pulse Power Characterization (HPPC) data demonstrates significantly improved estimation performance across multiple C-rates, achieving MAE - 0.04315 and RMSE - 0.06633. Results confirm the robustness, high accuracy, and rapid convergence capability of the proposed approach in extracting dynamic ECM parameters. The framework provides a practical, scalable, and data-efficient solution for advanced lithium-ion battery modeling in electric-vehicle applications.