A Model for estimating the state of health of retired lithium-ion EV batteries based on machine learning

Abstract

The electric vehicle market is growing rapidly and with it comes subsequent growth in the number of lithium-ion batteries that reach end of life in electric vehicle applications. Instead of being discarded in landfills, these batteries can be used in other applications such as energy storage since they still retain about 70% to 80% of their original capacity. This is known as battery repurposing, and it helps to manage battery waste. To repurpose batteries, their state of health must be tested to determine if they are adequately safe and reliable to use in second life applications. Current testing methods are time-consuming. Long testing times inhibit the scalability of repurposing operations to match the rapidly increasing number of electric vehicles, hence retired electric vehicle batteries. In this study, a machine learning model was developed to determine the SOH of used batteries. The model was based on quantum particle swarm optimization-support vector regression (QPSO-SVR) and used partial discharge data from differential capacity curves to estimate SOH. It was trained on data obtained from cycling used battery cells. The model achieved best MAE of 0.6139, RMSE of 0.7875, and R2 of 0.8481.