一种用于电力变压器故障识别的理论方法研究

A theoretical method for power transformer fault recognition research

支持向量机方法基于小样本的统计学习理论,其本质上是个优化和分类问题。设计了一种使用遗传算法优化多分类支持向量机参数,并将参数优化后的多分类支持向量机用于电力变压器故障识别的方法。该方法对色谱分析法检测到的特征气体含量进行数值预处理,提取出故障识别所需要的n+1个特征量,然后利用数值预处理后得到的数据样本对多分类支持向量机进行训练和识别,通过输出结果判断变压器所处的状态,以达到设备状态监测的目的。

Based on small samples statistical learning theory,the support vector machine(SVM)method appears essentially as the optimization and classification problems.So the multiple classification support vector machine(multi-SVM)with optimized parameters is designed and applied for identification of the power transformer faults.The characteristics gas content detected by chromatography is pre-processed to extract n+1 characteristics as sample data needed by fault recognition,then numerical data samples are processed by multi-SVM for training and recognition,finally the running status is identified according to training and recognition results in order to achieve the purpose of equipment condition monitoring.