State of Charge Prediction Study of Vanadium Redox-Flow Battery with BP Neural Network

Real-time capacity of a battery is normally indicated by the state of charge (SOC). In this paper, the SOC prediction methods of vanadium redox-flow battery (VRB) are introduced and the advantages and disadvantages of those are compared. Based on the nonlinear characteristic of SOC, the method of using BP neural network to predict SOC of VRB is proposed. The BP neural network is optimized with Levenberg-Marquardt optimization algorithm and Bayesian regulation algorithm, respectively. The neural network improved with Bayesian regulation can predict SOC in real time during the VRB testing process. The experimental results show that the neural network improved by Bayesian regulation algorithm can improve the real-time prediction accuracy of SOC and has a good application prospect.