摘要
为了提高锂电池健康状态(SOH)的预测精度,该文提出了1种基于遗传算法和支持向量回归(GA-SVR)的联合算法。通过GA解决SVR模型中的超参数优化问题。GA-SVR随机生成1组染色体,每个染色体包含了相应的SVR超参数信息。利用适应度函数计算出每条染色体的适应度值。根据适应度值对染色体进行选择、基因重组和变异等遗传操作,从而更新染色体的超参数信息。经过多次迭代后,找到适应度最大的染色体。从该染色体中提取相应的超参数信息,并训练最终的SVR预测模型。在美国国家航空航天局(NASA)锂电池数据集上的实验结果表明,该文算法优于基于混合像元核函数的高斯过程回归(SMK-GPR)算法、基于多尺度周期协方差函数的高斯过程回归(P-MGPR)算法、基于多尺度平方指数函数的高斯过程回归(SE-MGPR)算法和改进的基于粒子群优化的支持向量回归(IPSO-SVR)算法。
A joint algorithm based on genetic algorithm(GA) and support vector regression(GA-SVR) is proposed to improve the prediction accuracy of state of health(SOH) for lithium-ion batteries. GA is used to optimize the hyper-parameters in SVR model. Several chromosomes are initialized randomly by GA-SVR,each includes the hyper-parameters of SVR. The fitness of each chromosome is calculated by a fitness function. The hyper-parameters information of chromosomes is updated by selection,crossover and mutation according to the fitness. A chromosome with the highest fitness is chosen after multiple iterations. The SVR is trained as a prediction model based on the hyper-parameters of the selected chromosome. The experimental results on batteries datasets of National Aeronautics and Space Administration(NASA) of the USA show that the proposed GA-SVR outperforms the four popular SOH predictors,including spectral mixture kernel-Gaussian process regression(SMK-GPR),periodic covariance function-multiscale Gaussian process regression(P-MGPR),squared exponential function-multiscale Gaussian process regression(SE-MGPR),improved particle swarm optimization-support vector regression(IPSO-SVR).
作者
刘皓
胡明昕
朱一亨
於东军
Liu Hao;Hu Mingxin;Zhu Yiheng;Yu Dongjun(NAIRI Group Corporation(State Grid Electric Power Researeh Institute),Nanjing 210003,China;School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2018年第3期329-334,351,共7页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(61772273
61373062)
关键词
遗传算法
支持向量回归
锂电池
健康状态
超参数优化
genetic algorithm
support vector regression
lithium-ion batteries
state of health
hyper-parameter optimization