Deep neural network-enabled battery open-circuit voltage estimation based on partial charging data

Abstract

Battery management systems (BMSs) play a vital role in ensuring efficient and reliable operations of lithium-ion batteries. The main function of the BMSs is to estimate battery states and diagnose battery health using battery open-circuit voltage (OCV). However, acquiring the complete OCV data online can be a challenging endeavor due to the time-consuming measurement process or the need for specific operating conditions required by OCV estimation models. In addressing these concerns, this study introduces a deep neural network-combined framework for accurate and robust OCV estimation, utilizing partial daily charging data. We incorporate a generative deep learning model to extract aging-related features from data and generate high-fidelity OCV curves. Correlation analysis is employed to identify the optimal partial charging data, optimizing the OCV estimation precision while preserving exceptional flexibility. The validation results, using data from nickel-cobalt-magnesium (NCM) batteries, illustrate the accurate estimation of the complete OCV-capacity curve, with an average root mean square errors (RMSE) of less than 3 mAh. Achieving this level of precision for OCV estimation requires only around 50 s collection of partial charging data. Further validations on diverse battery types operating under various conditions confirm the effectiveness of our proposed method. Additional cases of precise health diagnosis based on OCV highlight the significance of conducting online OCV estimation. Our method provides a flexible approach to achieve complete OCV estimation and holds promise for generalization to other tasks in BMSs.