基于LFOA-GRNN模型的矿用锂电池SOC预测  

Mining lithium battery SOC prediction based on LFOA-GRNN model

作　　者：陈德海[1] 丁博文 潘韦驰 CHEN Dehai;DING Bowen;PAN Weichi(Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学,江西赣州341000

出　　处：《现代电子技术》2020年第6期115-118,共4页Modern Electronics Technique

基　　金：国家自然科学基金(61463020);江西省自然科学基金项目(20151BAB206034)。

摘　　要：针对矿用电动汽车锂电池SOC预测易受到工况环境影响、建模复杂、预测误差大等问题,该文将电池端电压、放电电流、环境温度、湿度作为SOC的表征因子,构成样本集以训练广义回归神经网络(GRNN),再引入具有Levy飞行特征的双子群果蝇优化算法(LFOA)优化GRNN的平滑因子σ。LFOA结合了Levy飞行搜索和果蝇优化算法的优点,全局搜索能力更强,收敛速度更快。仿真结果表明,经LFOA优化的GRNN能更快地搜索到合适的σ,并有效预测电池任一充放电状态下的SOC,与FOA-GRNN模型比较,LFOA-GRNN模型预测精度更高、时间更短,最大绝对误差不超过0.03,具有较好的工程应用价值。In allusion to the problem that the state of charge(SOC)prediction of the lithium battery of the mine electric vehicles,which is susceptible to the environmental impact of the working conditions,has complex modeling,large prediction error,etc.,in this paper,the battery terminal voltage,discharge current,ambient temperature and humidity are taken as the characterization factors of SOC to form a sample set to train the generalized regression neural network(GRNN).The double subgroup drosophila optimization algorithm(LFOA)with Levy flight characteristics is introduced to optimize the smoothing factorσof GRNN.In combination with the advantages of Levy flight search and drosophila optimization algorithm,LFOA has stronger global search ability and faster convergence speed.The simulation results show that the LFOA-optimized GRNN can search for the appropriateσmore quickly and predict the SOC of the battery under any states of charging-discharging effectively.In comparison with the FOA-GRNN model,the LFOA-GRNN model has higher prediction accuracy and shorter prediction time,and its maximum absolute error is less than 0.03.It has a certain engineering application value.

关 键 词：矿用锂电池 SOC预测 GRNN LFOA 模型建立 仿真分析 

分 类 号：TN86-34[电子电信—信息与通信工程] TM912.9[电气工程—电力电子与电力传动]