基于概率神经网络的油纸绝缘局部放电发展阶段识别

Development Stage Recognition of Partial Discharge in Oil-paper Insulation Based on Probabilistic Neural Network

采用阶梯升压法进行油纸绝缘沿面放电与气隙放电实验,根据放电发展过程的差异将局部放电划分成4个阶段:放电初始阶段、放电发展阶段、放电稳定阶段和临近击穿阶段,分析两种模式下放电特征参量的发展规律。提取放电相位分布(PRPD)图谱的29个统计特征参量,通过局部线性嵌入算法降维得到新的六维特征参量,采用概率神经网络(PNN)算法对两种放电模式下油纸绝缘放电发展阶段进行识别,并与广义回归神经网络(GRNN)模型以及反向传播神经网络(BPNN)模型进行比较,发现其识别结果更加准确。

The step-up method is used to perform the experiments of surface discharge and air-gap discharge in oilpaper insulation.According to the difference of discharge development process,the partial discharge is divided into four stages:initial stage,development stage,stabilization stage and near breakdown stage of discharge.And the development rules of discharge characteristic parameters under the two modes are analyzed.Twenty-nine statistical characteristic parameters of phase resolved partial discharge(PPRD)are extractedand new six-dimensional characteristic parameter is obtained by using the dimensionality reduction of local linear embedding algorithm.The probabilistic neural network(PNN)algorithm is adopted to identify discharge development stages of oil-paper insulation under two discharge modes and is compared with the generalized regression neural network(GRNN)model and the back-propagation neural network(BPNN)model,finding its recognition resultis more accurate.