BP神经网络预估锂离子电池SOC训练数据选择  

Training data selection of BP neural network for state-of-charge estimation of Li-ion battery

作　　者：封进[1,2] 

机构地区：[1]桂林航天工业学院汽车工程系,广西桂林541004 [2]北京航空航天大学交通科学与工程学院,北京100191

出　　处：《电源技术》2016年第2期283-286,共4页Chinese Journal of Power Sources

摘　　要：采用BP神经网络对电动汽车用动力锂离子电池荷电状态(SOC)预估进行研究,分析了BP神经网络的模型原理及锂离子电池极化现象。对比采用恒流实验数据训练BP神经网络,提出改进BP神经网络训练数据选择方法,以适应变电流的实际循环中,锂离子电池因极化现象而产生的动态非线性,并进行了电池SOC值的预估。实验表明,采用改进训练数据训练的BP神经网络,在电流剧烈变化的实际工况环境下具有更高的SOC预估精度。The state of charge（SOC） of Li-ion battery on electric vehicle（EV） was estimated by BP neural network.The principle of BP network and the polarization phenomenon of Li-ion battery were analyzed. In contrast with the BP neural network trained by constant current experimental sample, an improving method for selection of training data was proposed in order to compensate for the influence of the dynamic nonlinear of Li-ion battery cause by electrodes polarization when its current varying in actual cycle. The BP neural network model was used to estimate SOC of Li-ion battery. Experiments show that the BP neural network trained by improving training data has higher accuracy on SOC estimation in the actual cycle when the current varying dramatically.

关 键 词：BP神经网络 SOC预估 极化现象 训练数据 

分 类 号：TM912[电气工程—电力电子与电力传动]