Battery Pack SOC Estimation Based on Representative Cells and Support Vector Regression

To use a battery pack in a safe and rational manner, it is crucial to exactly predict its state of charge (SOC). However, inconsistencies between batteries are inevitable, which results in accurate prediction of the battery pack SOC remaining challenging. Therefore, this article proposes a battery pack SOC estimation method based on representative cells and support vector regression (SVR). Firstly, since the battery pack SOC is determined by the battery with the minimum rechargeable and dischargeable capacities, this article assigns the initial values of the battery pack SOC obeying normal and Weibull distributions and calculates the dischargeable and rechargeable capacities of the batteries, and selects the battery that can represent the state of the battery pack. Secondly, use the trained SVR to predict the representative cells SOC. Since the SOC predicted by the model has certain fluctuations, the article uses the ampere-hour integration with constraint factor for smoothing and correction processing, and employs the processed results to determine the battery pack SOC. Finally, this article uses the Worldwide Harmonized Light Vehicles Test Cycle (WLTC) working condition to validate the method at -15°C, 25°C, and 45°C. The results show that the SOC estimation method based on representative cells and SVR proposed in this article has high estimation accuracy, with RMSE of 0.398%, 0.151%, and 0.195% at -15°C, 25°C, and 45°C, and MAE of 0.335%, 0.111%, and 0.156%, respectively.