基于核熵成分分析的油中溶解气体浓度预测  

Concentration Prediction of Dissolved Gas in Oil Based on Kernel Entropy Component Analysis

作　　者：江风云[1] 唐勇波[1] JIANG Feng-yun;TANG Yong-bo(School of Physical Science and Engineering,Yichun University,Yichun 336000,China)

机构地区：[1]宜春学院物理科学与工程技术学院,江西宜春336000

出　　处：《控制工程》2020年第8期1419-1424,共6页Control Engineering of China

基　　金：江西省教育厅科技研究项目(GJJ161015)。

摘　　要：针对变压器油中溶解气体浓度预测中信息利用不完善问题,提出基于核熵成分分析(Kernel Entropy Component Analysis,KECA)的油中溶解气体浓度预测建模方法。首先用灰关联分析方法选取预测模型的输入变量;然后对选取的输入变量进行相空间重构;最后采用Renyi熵信息测度确定KECA核参数,用KECA对重构相空间提取核熵成分作为支持向量机(Support Vector Machine,SVM)的输入,建立变压器油中溶解气体浓度预测模型。用本文方法、单变量时间序列方法、多元变量时间序列方法测试60例样本,本文方法具有最小的均方根误差,为0.1607。实验结果表明,本文提出的方法具有较优的预测精度和泛化能力。In order to deal with the insufficient utilizing of dissolved gases information in transformer oil,a new prediction modeling method based on kernel entropy component analysis(KECA)is proposed.Firstly,grey relational analysis(GRA)method was used to select input variables of the prediction model.Then,the relevant variables were reconstructed in the phase reconstruction space where feature extraction was carried out by using KECA,meanwhile,the kernel parameter of KECA was determined by Renyi information entropy.At last,support vector machine(SVM)was employed to forecast dissolved gases content in which kernel entropy components extracted by KECA were used as the inputs.Compared with time series and multivariate time series methods,the proposed method has minimum root mean square error which was 0.1607 by testing 60 samples.Experimental results show that the proposed method has a better prediction accuracy and generalization ability.

关 键 词：变压器 油中溶解气体 核熵成分分析 RENYI熵 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TM855[自动化与计算机技术—控制科学与工程]