State of Health Estimation with Incrementally Integratable Data-Driven Methods in Battery Energy Storage Applications

State of health holds critical importance in lithium-ion battery storage systems, providing indispensable insights for lifespan management. Traditional data-driven models for battery state of health estimation rely on extracting features from various signals. However, these methods face significant challenges, including the need for extensive battery aging data, limited model generalizability, and a lack of continuous updates. Here, we present an innovative approach called incrementally integratable long short-term memory networks to address these issues during health state estimation. First, the data is partitioned into sub-datasets with a defined step size, which is used to train the long short-term memory network-based weak learners. Transfer learning technique is employed among these weak learners to facilitate efficient knowledge sharing, accelerate training, and reduce time consumption. Afterward, conducted weak learners are filtered and weighted based on estimation error to form strong learners iteratively. Furthermore, newly acquired data is applied to train additional weak learners. By combining transfer and incremental learning methods on the long short-term memory network, the proposed method can effectively utilize a small amount of data to estimate the battery state of health. Experimental results demonstrate the impressive performance and robustness of our method.