基于PNN的GIS局部放电模式识别方法

GIS Partial Discharge Pattern Recognition Based on PNN

气体绝缘开关设备(GIS)因绝缘缺陷引起的局部放电特性具有复杂性和分散性,其特征量的选取易产生数据的丢失和冗余,导致故障类型的识别效果不佳。据此,提出了采用线性判别分析(LDA)方法和遗传算法优化概率神经网络结合的局部放电模式识别方法。通过GIS局部放电实验平台模拟了5类典型的GIS局部放电模型,并建立相应的超高频图谱,提取了相关的特征参量;经过线性判别分析降维得到低维的样本空间,并送入到遗传算法优化后的概率神经网络中进行模式识别;分别采用BP神经网络、SVM、概率神经网络、优化概率神经网络4种分类器进行模式识别,实验结果表明,样本空间经过LDA降维,并经过遗传算法优化概率神经网络进行模式识别,具有较优的识别效果和识别时长。

The partial discharge(PD)characteristics of gas insulated switchgear(GIS)due to insulation defects are complex and dispersive,and the selection of its characteristic quantity is easy to produce data loss and redundancy,which leads to the poor effect of fault type recognition.Therefore,a method of partial discharge pattern recognition based on linear discriminant analysis(LDA)and genetic algorithm-optimized probabilistic neural network(GA_PNN)was proposed.Five kinds of typical GIS partial discharge models were simulated through the GIS partial discharge experimental platform,and the corresponding ultra-high frequency map was established,and relevant characteristic parameters were extracted.After dimension reduction by LDA,a low-dimensional sample space was obtained,and it can be sent to the probabilistic neural network optimized by genetic algorithm for pattern recognition.Four classifiers,BPNN,SVM,PNN and GA_PNN,were used for pattern recognition respectively.The experimental results show that the sample space is dimensionally reduced by LDA,and the probability neural network optimized by genetic algorithm is used for pattern recognition,which has better recognition effect and recognition time.