摘要
针对变压器油中溶解气体浓度预测中存在的输入变量选择结果受噪声影响的问题,提出了改进的互信息变量选择和支持向量回归机的油中溶解气体浓度预测方法。首先,对油中溶解气体各变量进行相空间重构,利用独立成分分析方法进行信噪分离;然后,提出改进的标准化互信息方法进行输入变量选择,以降低噪声对互信息变量选择的影响;最后,采用支持向量回归机作为预测器对变压器油中溶解气体浓度进行预测。实验结果表明,改进的标准化互信息的输入变量选择结果吻合油劣化热动力学研究结果,具有较优的预测精度和泛化能力。
Aiming at the problem that the input variable selection result is influenced by noise in the dissolved gas concentration prediction in transformer oil, a new prediction method of dissolved gas concentration in transformer oil based on improved normalized mutual information feature selection （INMIFS） and support vector regression machine is proposed. Firstly, the variables of the dissolved gas in transformer oil are reconstructed in the phase reconstruction space;and the independent component analysis is adopted to separate the signal from noise. Secondly, INMIFS method is proposed to select the input variables and reduce the influence of noise on mutual information variable selection;At last, the support vector regression machine is adopted as the predictor to forecast the dissolved gas concentration in transformer oil. Experiment results show that the improved normalized mutual information input variable selection results are consistent with the study results of mineral oil inferior thermal dynamics, and the proposed prediction method has good prediction accuracy and generalization ability.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第7期1492-1498,共7页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61134006
61273169)资助项目
关键词
油中溶解气体
预测
互信息
支持向量回归机
变量选择
dissolved gas in oil
prediction
mutual information
support vector regression machine
variable selection
