State of health estimation of lithium-ion batteries based on the Kepler optimization algorithm-multilayer-convolutional neural network

Lithium-ion batteries are widely used in electric vehicles (EVs), and accurate SOH estimation is essential for ensuring EV safety. This paper proposes a novel SOH estimation method based on the Kepler optimization algorithm-multilayer-convolutional neural network. Firstly, the extracted health indicators (HIs) are filtered, and highly correlated, continuous HIs are identified using Pearson correlation coefficient and scatter plots. Subsequently, the ReliefF algorithm is employed for further dimensionality reduction. Subsequently, a multilayerconvolutional neural network is constructed for SOH estimation, with the Kepler optimization algorithm (KOA) for hyperparameter optimization, a novel application according to the authors' knowledge. The SOH estimation results demonstrate that, a deeper CNN does not necessarily yield better results and the KOA-2-layerCNN performs the best. Additionally, compared with the 2-layer-CNN without hyperparameters optimization, the mean absolute error (MAE), the root mean square error (RMSE), maximum absolute error (Max-AE) of the KOA2-layer-CNN are decreased by 58.97 %, 53.33 %, 39.05 %, respectively. Moreover, compared with commonly used SOH estimation methods based on feature engineering, the KOA-2-layer-CNN also achieves accurate SOH estimation results, with significantly smaller MAE, RMSE, and Max-AE.