Battery pack capacity estimation based on improved cooperative co-evolutionary strategy and LightGBM hybrid models using indirect health features

Abstract

Focuses on the accurate estimation of battery pack capacity under real-world operating conditions, which is critical to improving the reliability of battery-powered systems, extending battery life, and optimizing health management strategies. This paper proposes an innovative CC hybrid strategy combined with LightGBM model estimation hybrid model. By introducing an improved MODECC strategy, the framework uses improved IMDEA and GA as sub-algorithms, and the framework performs parameter optimization and incremental evolution on the main algorithm I-MOEA-D. The efficient optimization of hyperparameters of LightGBM model is realized. The experiment is based on the CS battery data set, using 50 % of the data as the training set and the remaining 50 % as the test set, to verify the effectiveness of the proposed IMODEC-LightGBM hybrid model. The results show that the hybrid model achieves an average decrease of 15.1 %, 16.7 %, and 16.6 % in RMSE, MAE, and MAPE compared with the benchmark model of 1.74 %, 1.06 %, and 2.36 %, respectively, which significantly improves the prediction accuracy and fully proves the high precision and strong robustness of the hybrid model in the estimation of battery pack capacity.