State of health estimation of lithium-ion battery with improved radial basis function neural network

Accurate state of health (SOH) estimation for lithium-ion batteries is crucial to ensure the safety and reliability of electric vehicles. However, traditional neural network algorithms to estimate SOH often focus on fitting nonlinear fluctuation and is weak in the overall tracking trend. This paper thus proposes an improved radial basis function neural network (IRBFNN) to estimate the SOH with the simultaneous fitting of general trends and local fluctuations. A polynomial is provided to describe the overall trend of SOH. Meanwhile, the hidden layer of the IRBFNN converts the features nonlinearly to simulate the local battery capacity regeneration. Moreover, the initial parameters of the IRBFNN are obtained after training and then optimized by the improved gray wolf optimization algorithm. Two different datasets are utilized to verify the effectiveness of the presented method by comparing it with several other algorithms. Experimental results show that the IRBFNN-based method can accurately estimate the SOH, and the maximum estimation errors are within ±4%. Therefore, the results imply that the proposed method can effectively alleviate the problem of the poor estimation performance of traditional neural network-based algorithms in the later stage of battery aging. • A linear polynomial is proposed to show the global linear change of SOH. • The local nonlinear change of SOH is shown by the RBF neural network. • The improved gray wolf optimization is used to optimizing the network parameters. • The four features are extracted in the partial constant current charging process. • Experimental results highlight that the proposed method is more accurate.