A Novel Dual-Layer Genetic Algorithm With Parameter Interaction Framework for Battery Parameter Identification

Equivalent circuit models are widely adopted for battery modeling, yet their parameters require frequent updates due to aging-induced variations. While unit data segment (UDS)-based methods leverage operational data for parameter identification, existing approaches fail to address two critical issues: (1) the sensitivity of model accuracy to historical data utilization strategies and (2) parameter discontinuity at adjacent segment boundaries. To overcome these limitations, this study proposes a novel dual-layer genetic algorithm (GA) with a parameter interaction framework. The upper-layer GA autonomously optimizes historical data selection and initializes parameters for the first segment, while the lower-layer GA identifies parameters for subsequent segments. A boundary matrix iteration mechanism enforces parameter continuity across segments by propagating constraints iteratively. Experimental validation on Urban Dynamometer Driving Schedule (UDDS) under 25°C datasets demonstrates superior performance: Under UDDS conditions, the maximum error, mean absolute error, and RMSE are 38.6, 4.7, and 6.1 mV, respectively. These values represent improvements of 8.7%, 29.8%, and 31.4% compared to the UDS-based method; and 45.5%, 42.6%, and 45.0% compared to the Recursive Least Squares-based method. The multi-temperature validation results confirm the strong robustness of the proposed approach under disparate operating temperatures. This work advances data-driven battery modeling by resolving boundary discontinuity and reducing expert dependency in parameter identification, offering a scalable solution for cloud-based battery management systems.