CAJ | 학술논문

核极限学习机（kernel-based extreme learning machine，KELM）在分类性能方面优于支持向量机（SVM），但仍存在参数敏感性的缺陷。针对这一缺陷，提出一种结合K 折交叉验证（k-fold cross validation，K-CV）与粒子群优化（particle swarm optimization，PSO）的KELM分类器参数优化方法，将CV训练所得多个模型的平均准确率作为PSO的适应度评价函数，为KELM的参数优化提供评价标准。将该方法应用于变压器故障诊断中，充分利用数量有限的样本数据，提高KELM的泛化性能。实验结果表明，相比结合网格搜索（grid）的KELM、结合CV和Grid的KELM以及结合PSO的KELM，结合PSO的CV参数优化方法具有更好的性能。
핵겁한학습궤（kernel-based extreme learning machine，KELM）재분류성능방면우우지지향량궤（SVM），단잉존재삼수민감성적결함。침대저일결함，제출일충결합K 절교차험증（k-fold cross validation，K-CV）여입자군우화（particle swarm optimization，PSO）적KELM분류기삼수우화방법，장CV훈련소득다개모형적평균준학솔작위PSO적괄응도평개함수，위KELM적삼수우화제공평개표준。장해방법응용우변압기고장진단중，충분이용수량유한적양본수거，제고KELM적범화성능。실험결과표명，상비결합망격수색（grid）적KELM、결합CV화Grid적KELM이급결합PSO적KELM，결합PSO적CV삼수우화방법구유경호적성능。
The kernel-based extreme learning machine (KELM)has better classification performance than the SVM,but it still has the drawback of parameter sensitivity.For this defect,a method combining K-fold cross validation and particle swarm optimization (PSO)was proposed to optimize the parameter of KELM classi-fier,the average accuracy rate of the multi-ple models generated using the CV method was used as the fitness function of PSO to provide an evaluation criteria of KELM classifier.And this proposed method was used in the transformer fault diagnosis to make full use of the limited number of date samples and improve the generalization performance of KELM.Experimental result show that comparing with the method of KELM based on grid search,KELM based on CV and grid search and KELM based PSO,the proposed method has better per-formance.