混合神经网络算法在GIS局部放电模式识别中的应用研究  

Application of GA-BP Neural Network on GIS Partial Discharge Pattern Recognition

作　　者：梁基重[1] 杨亚奇[2] 张平[1] 

机构地区：[1]山西电力科学研究院,山西太原030001 [2]东北电力大学,吉林吉林132012

出　　处：《山西电力》2013年第3期11-14,共4页Shanxi Electric Power

摘　　要：为了提高基于超声法GIS局部放电模式识别的正确率,在实验室中对GIS典型缺陷局部放电的超声波进行了重复性测量,从43个能够表征缺陷特征的参数中提取了34个稳定的特征参数,然后采用后向序贯算法筛选出了24个有效特征参数作为神经网络输入参数。针对神经网络的局限性,提出了改进的GA-BP混合神经网络算法。训练结果表明,GA-BP神经网络的应用有效地提高了识别的准确率。In order to improve the accuracy of pattern recognition based on partial discharge detected by ultrasonic method, the repetitiveness of partial discharge （PD） under different defects is measured and 34 steady characteristic parameters are extracted from 43 parameters that can characterize defects. Then, 24 effective characteristic parameters are filtered as input of neural network. At last, an improved GA-BP neural network is proposed. After training, the result shows that the application of GA-BP neural network effectively improves the accuracy of pattern recognition.

关 键 词：超声波法 混合神经网络算法 局部放电 模式识别 特征参数提取 

分 类 号：TM7[电气工程—电力系统及自动化]