Comparison of SOC Estimation Performance with Different Training Functions Using Neural Network

Abstract

The estimation of State Of Charge (SOC) of battery pack attracts wide attention in battery manufacture and application, which is a key issue in Battery Management System (BMS). A practical three-layer BP neural network is proposed and used to estimate the SOC of LiFePO4 lithium-ion battery pack, which consists of three series groups with each group of 8 series modules. Sample data are obtained with different discharging scenarios to train the network with different training functions. And the trained neural networks are used to estimate the SOC. Results of experiments show that the performances of neural networks trained by different training functions differ in estimation accuracy and training speed. The Levenberg-Marquardt (L-M) algorithm achieves the best performance compared with the other two algorithms.