SOC Estimation of Lithium Battery Based on BP Neural Network with Forgetting Factor

Abstract

AbstractAccurate estimation of state of charge (SOC) of lithium battery for new energy vehicle power is very important to improve the dynamic performance and energy utilization efficiency of the battery. Lithium battery has time-varying characteristics in the process of use, and the accuracy of SOC prediction using fixed parameter offline static network model will decline. In order to accurately estimate the SOC of power lithium battery, it is required that the parameters of the prediction model can be online self-learning and can change with the movement of the battery system. By analyzing the current popular SOC estimation algorithm based on circuit model, this paper introduces BP neural network model to estimate the SOC of lithium battery. In order to realize parameter self-learning ability, time-varying forgetting factor is added to the cost function. A lithium battery test platform was built to test the SOC-OCV relationship under different magnification and temperature. SOC is estimated by ampere hour integral method, extended Kalman filter and improved BP neural network. Experiments show that the improved BP neural network method has faster tracking ability and higher estimation accuracy for fast time-varying parameters, and is a very practical SOC real-time prediction method.KeywordsLithium batterySOCBP neural networkForgetting factor