Toward Fast and Accurate SOH Prediction for Lithium-Ion Batteries

For timely maintenance and replacement in lithium-ion battery system, it is crucial to achieve fast and accurate State of Health (SOH) prediction. SOH is a dynamic status parameter of a battery indicated by its available capacity compared to the initial condition. SOH tends to decrease in the long-term because a battery is aged by its charging and discharging cycles due to the degradation of chemical composition. To ensure a battery has sufficiently good conditions, it is significant to predict the battery SOH precisely so that the depleted or weak battery can be replaced in time. However, the prediction of battery SOH is a difficult and complex task due to the following reasons (1) SOH degradation is a dynamic process because of the time-varying nature of both battery electrochemistry and working condition; (2) SOH prediction model usually involves numerous input attributes that lead to sheer complexity and computational cost for real-time battery management; (3) the balance of overhead-accuracy tradeoff is challenging to adapt to various application scenarios. In this paper, we propose a theoretical framework for fast and accurate SOH prediction based on the non-additive measure and overhead-accuracy balancing. Specifically, we characterize the interactions among battery attributes and their aggregated impacts on SOH prediction. By choosing the most significant subset of battery attributes, the computational complexity of SOH prediction can be substantially reduced while maintaining high accuracy. The tradeoff between accuracy and computational overhead can be balanced by adjusting the number of attributes and volume of training data. The experimental results validate the effectiveness and efficiency of the proposed framework.