中文摘要：

针对已有输电线路故障分类方法在样本中存在噪声时准确率有所降低情况,研究了基于Hilbert-Huang变换和模糊支持向量机（fuzzy support vector machines,FSVM）的输电线故障模糊分类方法,以提高输电线路故障分类的准确率。采用HHT变换获得故障时刻,提取故障后A、B、C三相及零序电流的特征能量函数值,组成FSVM的4维输入向量。利用网格优化方法对FSVM二分类器的惩罚参数C、核函数宽度σ进行优化。构造了FSVM的高维空间带状分段隶属度函数,求取样本的模糊决策函数值。构造多FSVM分类器。将故障时刻后特征向量送入多FSVM分类器,得到样本的故障分类初始标签。构造支持向量回归机（support vector regression,SVR）,优化获得测试样本的最终故障隶属度,再对FSVM的分类标签进行修正。研究采用主成分分析法对样本高保真的降维处理方法、在3维坐标系中显示降维后3维向量及其故障模糊分类结果。为了测试算法,做了784组仿真实验,实验结果表明：FSVM＋SVR的输电线路模糊故障分类方法不受故障点、故障类型、过渡电阻影响,故障识别率达到99.4%。在训练集1/5故障数据中加入5 d B Gauss白噪声,故障识别率仍保持不变。研究表明基于模糊支持向量机的分类方法适用于线路故障分类。

英文摘要：

The transmission line fault classification accuracy may decrease with the existence of noise. To improve the accuracy of fault location and accident analysis, we studied the fault classification method based on fuzzy support vector machine. Firstly, the Hilbert-Huang transform （HHT） is used to extract the intrinsic energy function value of A, B, C-phase and zero-sequence post-fault current to constitute the 4-dimensional inputting feature vector. Then an optimiza- tion method of grid search algorithm is used to optimize the penalty parameter C and kernel width tr. The ribbon segmented membership fimction is constructed, as well as the optimal classification function of FSVM. The combined multiple FSVM classifiers are given to get classification label. The support vector regression （SVR） model is constructed after the optimal classification function of FSVM regression function. The fuzzy degrees of membership are obtained lastly. The method based on the principal components analysis （PCA） is studied to display the classification result in three-dimensional space. The simulation experiments with 784 signals are made to test the algorithm. The simulation re- suits illustrate that proposed fault classification method of FSVM＋SVR is not affected by the fault position and transition resistance, its fault classification rate is 99.4%. When Gaussian white noises of 5 dB are introduced to partial training data, the fault classification rate doesn＇t decrease. The research shows that proposed classification method based on fuzzy support vector machine is applicable to the line faults classification.