Battery health state prediction based on lightweight neural networks: A review

Abstract

Due to their superior properties, lithium-ion batteries (LIBs) have become the primary energy storage medium for electric vehicles (EVs), driven by widespread adoption. Nevertheless, a significant barrier hindering EV uptake lies in accurately assessing power LIBs’ health status and lifespan under prolonged demanding conditions. The neural network–based prediction method can increase the model’s prediction accuracy. However, because of the model’s complexity and abundance of features, current data-driven prediction technology frequently requires a lot of processing power to predict the battery’s health state. This study discusses recent approaches to life prediction using lightweight neural networks, with an emphasis on the aforementioned issues. The LIB’s aging mechanism and state of health (SOH) definition are first explained. A number of neural network models are then presented, followed by a summary of the available lightweight neural network prediction techniques and the machine learning framework for the prediction model, which aims to produce a more flexible and accurate model. This research provides references for predicting the health condition of LIBs and posits that in the future, more creative lightweight neural network models will become the standard in SOH prediction.