Capacity estimation of Lithium-ion batteries through a Machine Learning approach

Lithium-ion Batteries (LiBs) have become of paramount importance due to their employment in application fields, including renewable energy sources and electric vehicles (EVs), which heavily rely on them. This has spurred research efforts to develop battery models capable of predicting and estimating battery behavior to optimize usage and reduce degradation. To this end, key state parameters, including State Of Charge (SOC) and State of Health (SOH), should be accurately estimated. In the literature, many estimation methods are based on the knowledge of the relationship between Open-Circuit Voltage (OCV) and SOC. The latter can be modeled in different ways, organized into three main approaches: table-based, analytical, and artificial intelligence approaches. Among these, Machine Learning approaches have gained popularity and have shown great promise for this purpose. However, previous studies typically require many OCV-SOC data points or entire fragments of the OCV curve, which makes them unsuitable for EV applications. To address this limitation, the present paper develops and validates an ML algorithm to estimate the battery capacity of a LiB using only two experimental OCV points, accounting for different levels of cycle aging. The results demonstrate that the model, when trained on an accelerated-aged battery, can accurately predict the actual capacity of other batteries with similar characteristics but different aging levels.