A new SOH estimation method for Lithium-ion batteries based on model-data-fusion

A new method for the estimation of the state-of-health (SOH) of lithium-ion batteries (LIBs) is proposed. The approach combines a LIB equivalent circuit model (ECM) and a deep learning network. Firstly, correlation analysis is performed between the LIB data and SOH and suitable portions are selected as health features (HFs). Simultaneously, a fractional-order RC ECM of the LIB is derived and a hybrid fractional particle swarm optimization with crisscross learning (FPSO-CL) strategy is used to identify the model parameters. Secondly, correlation analysis between the model parameters and SOH is conducted and those that best represent the battery health are selected as additional HFs. Thirdly, an improved vision transformer network (VIT) is designed by including a dimension transformation layer, multilayer perceptron and a trainable regression token. Finally, the VIT is trained with all determined HFs, yielding a compete framework for predicting the SOH of LIBs. Experimental verification is carried out on real LIBs data and the results show that the proposed scheme can achieve higher prediction accuracy than other alternative methods.