Prediction of Lithium battery State of Health (SOH) based on a Nonlinear Autoregressive Exogenous model

Abstract

The state of health (SOH) of a battery is a crucial indicator reflecting its current condition, and forecasting SOH is essential in battery management systems to ensure efficient and safe operation. Data-driven forecasting methods have demonstrated higher effectiveness compared to model-based approaches. This paper employs a nonlinear autoregressive model with exogenous inputs (NARX) to predict battery SOH for both single-step and multi-step forecasting. Features extracted during the charging process from the NASA battery dataset are used as inputs for the model. The results indicate that the NARX model improves error values for single-step forecasting. Additionally, the error values in multi-step forecasting also demonstrate high accuracy.