ERMN: An enhanced meta-learning approach for state of health estimation of lithium-ion batteries

Abstract

To ensure the safety of lithium-ion batteries (LIBs), accurately estimating the state of health (SOH) of LIBs is crucial for end-users. Challenged by new working conditions and limited training samples, conventional machine learning-based methods have performed unsatisfactorily on SOH estimation of LIBs. To this end, this article introduces an enhanced regressive matching network (ERMN) method for the accurate SOH estimation of LIBs. The method extends the matching network from classification tasks to regression tasks by filtering a certain number of support samples, which can therefore be used for the SOH estimation—a typical regression task. A new loss function is designed to further enhance the distribution of the filtered support SOHs, ensuring their distinguishability and concentration. Additionally, the particle swarm optimization algorithm is incorporated into the ERMN method to automate the selection of hyperparameters. Experimental results from four case studies, considering different charge policies and sample sizes, demonstrate the effectiveness and generalization of the proposed ERMN approach.