State of charge estimation for lithium battery based on Levenberg-marquardt back-propagation neural network with momentum term

Abstract

Abstract: State of charge (SOC) can reflect the residual charge of the battery, which helps the driver of a new energy vehicle to infer the battery endurance, playing a significant role. However, when using back-propagation(BP) neural network to estimate SOC, the results have some problems such as slow convergence. In this paper, the momentum term is introduced into the Levenberg-Marquardt back-propagation (LM-BP) neural network structure for the prediction of lithium battery SOC. To be specific, a momentum term with fixed momentum factor is added to the LM algorithm to replace the gradient descent method used in the standard BP neural network. The improved weight correction formula is used to update the weight to obtain a higher convergence speed of the network, and the neural network structure with double hidden layers is utilized to improve the prediction accuracy. It can be seen from the results that the mean absolute error of the proposed method is reduced by 1.422% compared with the standard BP algorithm, and the estimation performance is significantly improved.