Residual Capacity Estimation with Incremental Learning for Lithium-Ion Batteries

Accurately estimating the state of health (SOH) for lithium-ion batteries based on their residual capacity is essential for future intelligent battery management systems. However, the absence of complete charge-discharge data presents significant challenges in developing an accurate and reliable SOH estimation model. To address this issue, we propose a closedloop battery residual capacity (BRC) estimation method using an incremental learning-based model that can be re-trained during most practical charging operations. Initially, an optimal voltage range is selected using a multi-objective particle swarm optimization algorithm based on the correlation between partial charging capacity (PCC) and BRC, as well as the initial voltage data distribution during the charging process. Unlike traditional BRC estimation methods, support vector regression quickly characterizes the mapping using degradation features and PCC. A linear BRC estimation model is then established based on the relationship between PCC and BRC. Using continuous updates in online monitoring data, adaptive training with online learning is used to enhance the prediction performance of the BRC estimation model. By implementing incremental learning technology, we construct a closed-loop model that improves as more observation data is collected. An analysis of an open-source dataset demonstrates that the proposed SOH estimation method achieves satisfying and robust results in estimating the BRC.