Feature selection and data-driven model for predicting the remaining useful life of lithium-ion batteries

Abstract

To ensure long and reliable operation of lithium-ion battery storage workstations, accurate, fast, and stable lifetime prediction is crucial. However, due to the complex and interrelated ageing mechanisms of Li-ion batteries, using physical model-based methods for accurate description is challenging. Therefore, building data-driven models based on direct measurement data (voltage, current, capacity, etc.) during battery operation may be a more effective approach. This paper employs a time series analysis of discharge capacity/voltage curves to perform feature predication. The goal is to predict the state of health using a short-term model and the remaining useful life of batteries using a long-term iterative model. The validity of this method is verified using the open-source MIT battery dataset. Comparisons with models reported in the literature demonstrate that this method is generalisable and ensures accuracy across a wider range of predictions.