State of Charge Estimation of Lithium-ion Battery Based on Recurrent Neural Network

Abstract

State of charge (SOC) is one of the key indices for battery management system (BMS). Accurate estimation of SOC can greatly improve battery safety performance and service life. Due to the complex electrochemical reaction inside the battery, SOC is highly nonlinear and cannot be directly measured. This paper proposes a SOC estimation method based on recurrent neural network (RNN). The input current and terminal voltage are selected as input parameters to predict SOC. The information transfer of RNN is mainly divided into two steps: from the input layer to the hidden layer and from the hidden layer to the output layer. To identify the internal parameters of RNN, RNN is firstly opened the feedback loop, and then closed the feedback loop after the training using the improved gauss-newton backpropagation algorithm. Through the cycle charge and discharge experiment of Lithium-ion battery, this paper trained 8 neural networks based on different training sets and predicted them with corresponding verification sets respectively. It was found that the experimental data were consistent with the predicted data, which verified the effectiveness of the method.