A Data-Driven-Based SOC Estimation Method Using Newly Constructed Features for Different Application Environments

Data-driven-based State of Charge (SOC) estimation methods have garnered extensive attention as they do not need to consider the complex reaction mechanism inside batteries. In the case of the data-driven-based SOC estimation, there are only limited measurable variables to be used for modeling. So, it is challenging for conventional data-driven-based SOC estimation methods to obtain accurate estimation results in different application environments without having a larger amount of modeling data. This work proposes a novel data-driven-based SOC modeling method by constructing some input features related to open circuit voltage and terminal voltage. The objective is to offer more valuable information to conventional data-driven-based SOC modeling methods and enhance their interpretability. Then, by following the changing trends of the constructed features, a battery online SOC estimation method based on feature point sampling is designed to ensure the stability of SOC estimation results. Experimental results validate the effectiveness of these newly constructed features. Besides, the proposed method can obtain accurate SOC estimation results in different application environments and is robust against current noise, inaccurate SOC initial values, etc.