锂离子电池充电策略优化：基于异构集成代理模型的先进多目标优化算法

Abstract Reducing charging time (CT) while maintaining thermal safe and health management of lithium-ion batteries (LIBs) is essential for enhancing the portability of electric vehicles. However, the substantial computational burden, high-dimensional search spaces and multi-objective conflicts traditional mechanism-based charging strategy optimization methods introduce bottlenecks across a wide range of applications. To overcome these issues, a mechanism-based electrothermal-aging model is first developed for dataset generation. Subsequently, a meta -features-based heterogeneous ensemble surrogate model (MetaHES) is proposed to better accommodate the diverse charge–discharge performance characteristics under State-of-Charge multi-stage constant current charging (SMCC) strategy. Additionally, an improved constrained multi-objective Harris Hawks Optimization algorithm, combined with dual-population and utopian point-based evolutionary environmental selection (CDMHUE), is introduced. Finally, verification and application on the LCO18650 Battery revealed several important findings: Experimental data confirm the accuracy of the mechanism model, while comprehensive comparison experiments validate both the effectiveness of each component and the overall superiority of MetaHES. Its rational design enables it to efficiently address challenges in LIB datasets, such as diversity, small-sample, and non-temporal characteristics. Additionally, the MetaHES-based CDMHUE optimizes SMCC strategies, improving computational efficiency by over two orders of magnitude compared to the mechanism-based CDMHUE and outperforming five other algorithms. The three optimized strategies reduce state of health attenuation, CT, and maximum temperature rise compared to corresponding constant current and constant voltage charging strategy. This work provides a novel perspective for prognostics and health management in battery management systems.