基于贝叶斯正则化算法BP神经网络钒电池SOC预测  

Application of BP neural network improved by Bayesian regularization algorithm in VRB SOC prediction

作　　者：杨春生[1] 牛红涛[1] 隋良红[1] 李明兴[2] 

机构地区：[1]中国测试技术研究院,四川成都610021 [2]四川大学制造科学与工程学院,四川成都610064

出　　处：《现代电子技术》2016年第8期158-161,共4页Modern Electronics Technique

基　　金：四川省科技支撑计划项目(2014GZ0085;2014GZ0001)

摘　　要：电池荷电状态(SOC)用于表征电池的剩余电量,是全钒液流电池的一个重要参数。在此介绍常用的钒电池SOC预测方法,并对比其优缺点。基于电池SOC的非线性特征,提出采用BP神经网络预测钒电池的SOC,并采用L-M优化算法以及贝叶斯正则化算法对网络进行优化。使用贝叶斯正则化改进的神经网络在对项目中全钒液流电池测试过程实时预测SOC。实验结果表明,采用贝叶斯正则化算法改进的神经网络能够提高SOC的实时预测精度,具有很好的实用前景。The state of charge（SOC）is an important parameter of VRB to character the remaining capacity of the battery.In this paper,the common prediction methods of VRB SOC are introduced,and the advantages and disadvantages are compared. Based on the nonlinear characteristic of VRB SOC,the method of using BP neural network to predict VRB SOC is proposed. The BP neural network was optimized with Levenberg-Marquardt optimization algorithm and Bayesian regularization algorithm. The neural network improved with Bayesian regularization can predict SOC in real time in VRB testing process. The experimental results show that the neural network improved by Bayesian regularization algorithm can improve the real-time prediction accuracy of SOC,and has good application prospect.

关 键 词：钒电池 荷电状态 BP神经网络 贝叶斯正则化算法 

分 类 号：TN915-34[电子电信—通信与信息系统] TP301.6[电子电信—信息与通信工程]