基于深度残差网络的GIS局部放电模式识别技术研究

Research on GIS Partial Discharge Pattern Recognition Based on Deep Residual Network

气体绝缘电器(gas insulated switchgear,GIS)内部绝缘缺陷产生的局部放电(partial discharge,PD)特征表现较复杂,传统局部放电模式识别方法因特征选取具有较强的主观性,其放电类型分类的准确性和鲁棒性均较差。针对这一问题,文中将局部放电的PRPD数据转化为局部放电灰度图,构建深度残差网络对局部放电灰度图进行自适应特征提取,深度挖掘放电灰度图中不同局部放电类型的特征模式,实现局部放电模式的识别,实际实验表明,文中所提方法对局部放电类型的识别同时具有较高的准确率和鲁棒性。

Partial discharge （PD）characteristics caused by internal insulation defects of gas insulated switchgear （GIS）are complicated, and traditional partial discharge pattern recognition method has strong subjectivity on feature selection, so the accuracy and robustness of its classification on partial discharge pattern are unsatisfactory. In re sponse to this problem, the PRPD data of partial discharge is transformed into a partial discharge grayscale image, and a selfadaptive feature extraction of the partial discharge grayscale image is constructed by a network of deep re siduals, in order to deeply extract the different partial discharge patterns in the discharge grayscale image and real ize the recognition of partial discharge pattern. In fact, experiments show that the method proposed in this paper has high accuracy and robustness to the recognition of partial discharge types at the same time.