Multi-objective optimization design of lithium-ion batteries based on deep learning and electrochemical-thermal-hydraulic-mechanical model

Optimizing the performance and lifespan of lithium-ion batteries (LIBs) is critical. Existing multiphysics coupling models typically lack a more comprehensive representation of coupled physical mechanisms, which reduces simulation accuracy. This study proposes a hybrid multi-objective optimization framework that combines physics-based modeling with data-driven methods. An electrochemical–thermal–hydraulic–mechanical (ETHM) model is first introduced to accurately simulate electrochemical behavior, temperature rise, and the evolution of anode maximum principal stress. Based on orthogonal experimental design, cathode thickness, the porosity of both the anode and cathode, and anode initial lithium-ion concentration are selected as key design variables, and further validated through correlation analysis. A deep neural network (DNN), identified as the best-performing model among several machine learning (ML) approaches, is employed as a surrogate and coupled with the non-dominated sorting genetic algorithm II (NSGA-II) to optimize energy density, power density, anode maximum principal stress, and maximum temperature. Compared with the baseline, the maximum energy and power density are increased by approximately 19% and 11%, respectively, while the minimum anode maximum principal stress and maximum temperature are reduced by 47% and 20%. This study confirms the feasibility of the hybrid framework, providing theoretical insights and practical guidance for designing next-generation high-performance LIBs.