As energy storage scales up, overcharge-induced thermal runaway (TR) in large-capacity batteries poses critical risks. This study develops a 3D multiphysics model for 280 Ah prismatic LFP batteries, coupling electrochemistry, heat conduction, and side reactions while incorporating gas-pressure-structural response dynamics. To resolve high-dimensional parameter identification challenges, a staged strategy using Bayesian optimization and cross-dimensional mapping is proposed, ensuring consistency between 1D and 3D models while minimizing computational costs. Validated against 0.5C and 1C overcharge tests, the model accurately predicts voltage and temperature evolution, with TR onset and peak temperature errors under 5% and 3%, respectively. Mechanistic analysis reveals a transition from electrochemical to side-reaction-dominated heat generation, with electrolyte oxidation being the primary driver of heat and gas evolution during deep overcharge. Furthermore, results indicate that battery expansion is primarily driven by internal pressure accumulation, characterizing overcharge TR as a coupled thermal-gas-mechanical failure process. This framework provides a robust tool for safety characterization and early warning in large-scale energy storage systems.
8 July 2026 in Journal of the Electrochemical Society
Electrochemical models are widely used in battery design and optimization due to their capability to systematically characterize the internal physicochemical evolution of batteries. However, such models typically involve a large number of parameters with high-dimensional coupling and multi-timescale characteristics, which makes conventional parameter identification methods prone to ill-posed optimization or convergence to local optima. To address these challenges, this paper proposes a multi-stage parameter identification method for the pseudo-two-dimensional (P2D) model. First, a sensitivity analysis and parameter sensitivity ranking were conducted for the 31 uncertain parameters in the model. By incorporating physical properties and sensitivity intervals, parameter classification and dimensionality reduction are achieved. Subsequently, a staged identification framework is constructed based on the classification results, where tailored excitation conditions and optimization algorithms are assigned to different parameter subsets. This approach enhances parameter identifiability while preserving physical consistency, enabling high-precision identification of 19 key electrochemical parameters. Finally, the proposed method is validated through experiments and simulations under real dynamic grid operating conditions. The results demonstrate that the proposed method achieves relatively high accuracy and applicability, with the root mean square error of terminal voltage maintained within 12 mV under dynamic conditions.
14 September 2026 in Journal of the Electrochemical Society
The pseudo-two-dimensional (P2D) model provides a physics-based description of lithium-ion battery behavior, but its numerous strongly coupled electrochemical and thermal parameters make simultaneous voltage and temperature prediction challenging. This study proposes a multi-stage parameter identification method for an electrochemical–thermal coupled P2D model. Sobol global sensitivity analysis with Saltelli sampling is first used to quantify the effects of model parameters on voltage and temperature responses under different operating conditions. Based on the sensitivity results, a stage-wise multi-objective identification strategy is developed, with adaptive weights determined from normalized sensitivity indices. A hybrid Whale Optimization Algorithm–Differential Evolution (WOA–DE) method is then employed for parameter optimization. Compared with the original model, voltage prediction errors are reduced by 25.68%, 42.13%, and 47.48% under 0.5 C, HPPC, and DST conditions, respectively, while temperature errors decrease by 37.97%, 45.16%, and 30.96%. Under dynamic conditions, the model achieves RMSEs of 10.21 mV for voltage and 0.115 °C for temperature. The results demonstrate improved prediction accuracy and robustness for coupled voltage–temperature responses.