Multi-category health indicator fusion-based state of health forecast of lithium-ion batteries with high generality to partial discharging profiles

Abstract

Recent years have shown the adaptability and precision of data-driven approaches in state of health (SOH) forecast of lithium-ion batteries (LIBs). However, the complete discharging process that encompass the state of charge (SOC) region from 0 to 100% is not common in real-world, which provides a barrier for selecting health indicators (HIs). In this article, a multifaceted HIs fusion approach with broad applications for various cell degradation scenarios is presented to characterize the aging state by combining discrete curvature HIs and morphological features of discharge voltage that are extracted from the partial discharging data with different SOC ranges. To use these two kinds of aging features, a multiresolution temporal convolutional network model (MTCN) is specifically presented. The MTCN not only fits the relation between the feature vectors and SOH but also is able to process raw sensor data without artificially preprocessing steps, which is important for the battery management system (BMS) where only limited memory and computing power are available to establish a SOH estimation model. In order to comprehensively validate the suggested strategy, we performed experiments using aging data collected on LiCoO2/LiNiCoAlO2, LiFePO4, and LiCoO2 cells using a variety of testing methodologies. The results suggest that the proposed method has high generality to the scenarios with diverse discharging characteristics. The suggested approach also has a number of other benefits, such as excellent resilience to various discharge conditions and satisfied adaptability to various cell types.