基于稀疏表示的绝缘子紫外图谱闪络状态分类评估方法

Method for Evaluating Flashover State of Insulator Ultraviolet Image Based on Sparse Representation

绝缘子运行状态的检测和故障诊断对于维持电力系统安全稳定运行至关重要。针对目前存在输变电绝缘子的紫外检测图像故障特征不明显、诊断准确率不高的问题,提出了一种基于稀疏表示法的绝缘子紫外图谱的闪络状态分类评估方法。通过字典学习构建图谱信号自适应的过完备字典,采用加速近邻梯度算法和正交匹配追踪算法对待测紫外图像进行稀疏求解,依据稀疏矩阵的非零项进行分类诊断。结果表明,该方法的检测准确率较高,最高可达98%,其中正交匹配追踪算法依赖于字典的健全程度,当训练样本充足时算法识别时间仅为0.000 8 s。而加速近邻梯度算法则选取多个较优参量,适用于样本量较小的分类评估。此外,稀疏度参数敏感度较低,具有较好的鲁棒性。该算法同多分类支持向量机（M-SVM）算法相比,具有更好的表现性能,在绝缘子紫外检测分级预警和故障检测方面具有良好的应用前景。

It is essential for monitoring and diagnosing defects of insulators to guarantee safe and stable operation of power systems. In view of the problem that the deterioration characteristic of the UV detection image of the power transmission and transformation insulator is unapparent and the diagnostic accuracy is low, we put forward a novel method for evaluating flashover grade of insulator ultraviolet image based on the sparse representation method. Firstly, a adaptive dictionary is constructed by a dictionary learning method. Then, the accelerated proximal gradient algorithm and the orthogonal matching pursuit algorithm were employed to solve the sparse solution, and the classification was diagnosed according to the nonzero term of the sparse matrix. The results show that the method has the advantages of high detection accuracy, up to 98%. The accuracy of the orthogonal matching pursuit algorithm depends on the completeness of the dictionary. When the training samples are sufficient, the algorithm recognition time is only 0.000 8 s. While the accelerated neighbor gradient algorithm selects multiple preferred parameters and is suitable for classification evaluation of smaller data sets. In addition, the sparse parameters are insensitive and robust. Compared with multi-classification support vector machine algorithm, the proposed method has a better performance and application prospect in UV layered pre-warning and fault detection of insulators.