摘要
针对变压器油中溶解气体浓度的检测问题,提出基于二次维数约简的油中溶解气体浓度预测模型。首先,采用互信息变量选择方法选取预测模型的输入变量;然后,对输入变量进行相空间重构,采用核主元分析对重构相空间进行特征提取,达到数据降维、滤除数据噪声、消除变量间相关性的目的,并用Renyi熵信息测度确定核主元分析的模型参数;最后,将核主元分析提取的主元变量作为核极限学习机的输入,建立变压器油中溶解气体浓度的预测模型。与灰色预测模型、仅变量选择的预测模型、仅特征提取的预测模型的对比实验结果表明,所提出的基于二次维数约简的油中溶解气体浓度预测模型具有较优的预测精度和泛化能力。
Aiming at the testing problem of dissolved gases content in transformer oil,a new prediction model based on twice dimensionality reduction was proposed. Firstly, mutual information variable selection method was used to select relevant input variables of the prediction model; Secondly,the relevant variables were reconstructed in the phase space where feature extraction was carried out by using kernel principal component analysis( KPCA) for the purpose of data dimension reduction,denoising and eliminating relativity of variables,meanwhile,the parameters of KPCA were determined by Renyi information entropy. At last, kernel extreme learning machine( KELM) was employed to forecast dissolved gases content in transformer oil, and kernel principal components were used as the inputs of KELM. Compared with gray model and the prediction model which only adopt variable selection method or feature extraction method, experimental results show that the proposed prediction model has a better prediction and generalization.
出处
《电工技术学报》
EI
CSCD
北大核心
2017年第21期104-112,共9页
Transactions of China Electrotechnical Society
基金
国家自然科学基金(51366013)
江西省教育厅科技项目(GJJ161015)资助
关键词
油中溶解气体
核极限学习机
核主元分析
RENYI熵
变量选择
特征提取
Dissolved gases in oil
kernel extreme learning machine
kernel principal component analysis
Renyi entropy
variable selection
feature extraction
