Prediction of lithium-ion battery SOC based on IGA-GRU and the fusion of multi-head attention mechanism

Abstract

It is necessary to establish a sufficiently advanced Battery Management System (BMS) for safe driving of electric vehicles. Lithium-ion batteries have been widely used in electric vehicles due to their advantages of high specific energy and low-temperature resistance, so this paper takes lithium-ion batteries as the research object. BMS can monitor various status information of lithium-ion batteries in real-time, and the State of Charge (SOC) of lithium-ion batteries is a key parameter among them. Accurate SOC estimation is crucial for ensuring the safety and reliability of energy storage applications and new energy vehicles. However, the value of SOC cannot be directly measured. In order to more accurately estimate the SOC, this paper proposes a prediction method that combines an immune genetic algorithm, gated recurrent unit, and multi-head attention mechanism (MHA), using battery experimental data from the University of Maryland as the dataset. Compared with the traditional parameter optimization approach, this paper uses the immune genetic algorithm to find the optimal hyperparameters of the model, which on the one hand has a wider choice of parameters, and on the other hand has been improved for the genetic algorithm is easy to fall into the local optimal solution, so as to improve the SOC estimation accuracy of the GRU model. The model also incorporates a multi-attention mechanism to capture different levels of information, which enhances the expressive power of the model. The data preprocessing part adopts the sliding window technique, through which the original time series data is converted into several different training samples when training the machine learning model, as a way to increase the diversity of the dataset and improve the robustness of the model. Finally, the prediction performance of the fusion model proposed in this paper is verified by Pycharm simulation, and the average absolute error, root mean square error and maximum prediction error of the model are 1.62%, 1.55% and 0.5%, respectively, which proves that the model can accurately predict the SOC of lithium-ion battery. It is shown that the model can significantly improve the accuracy and robustness of SOC estimation, enhance the intelligence, real-time and interpretability of the battery management system, and bring a more efficient, safe and long-lasting battery management solution to the fields of electric vehicles and energy storage systems.