A novel static model prediction method based on machine learning for Li-ion batteries operated at different temperatures

Abstract

In electric vehicles (EVs) and microgrid systems, lithium-ion (Li-ion) batteries are the most realistic choice for energy storage and utilization applications. The precise prediction of the static and dynamic properties of the battery is crucial for these applications. Traditional methods use system identification and optimization concepts to predict the static characteristics; these strategies require both higher-order terms and model re-tuning for each operating temperature. In order to create a unique static model of the battery, this work suggests a machine learning (ML)-based prediction method that uses data-driven concepts to develop the prediction model between the input and output variables with more accuracy. In order to create an accurate prediction model based on the given dataset, a complete dataset of the entire process is fed to the training algorithm. In the conventional ML approach, for each operating temperature a prediction model is developed to predict the static model. But in the proposed technique, a unique ML based model is suggested to determine the relationship between the battery's open circuit voltage (OCV) and State of Charge (SoC) at various operating temperatures. To identify the best prediction model, performance metrics such as root mean square (RMSE) and R2 are used. From the results, it is identified that the neural network-based SoC-OCV prediction is the best among the strategies that have been proposed. Since the predicted model is unique, it can be used at any operating temperature without requiring the prediction model to be adjusted. The proposed model can be used to estimate battery parameters and battery state such as SoC in EV applications.