绝缘子污秽度预测特征量提取与神经网络模型

为了给输电线路防污闪提供科学指导，优化绝缘子污秽度预测特征量从而提高污秽度预测时间裕度和可靠性是关键。在人工雾室模拟35 kV线路3片串普通型和防污型染污绝缘子，测取了相同运行状态下相同采样时段内的泄漏电流，提取泄漏电流有效值均值、有效值最大值和有效值标准差作为污秽度预测的特征量，建立了3个特征量与污秽度的回归拟合关系，对比分析2种绝缘子泄漏电流3个特征量与污秽度的关系，推导验证了防污型绝缘子泄漏电流3个特征量小于普通型的原因。提出在污秽沿面放电安全区段内，主要利用泄漏电流有效值的3个特征量作为污秽预测神经网络模型的输入量。仿真和试验结果对比表明，所建立的预测模型误差在4%以下。所给出的研究结果不仅为输电线路防污预警系统特征量的优化选择提供了理论依据，对线路清扫选线决策也有指导作用。

Characteristics Extraction for Contamination Forecast of Insulators and the Neural Network Model

In order to provide insulators with scientific guide for the prevention of contamination flashover,the key is to optimize the input of the forecast model and to increase the time margin of pre-warning.The leakage currents(LCs) of insulators with different pollution levels are measured in the same period after the pollution layers are saturated moist in the fog chamber.Three characteristics based on the effective value of LCs and the fitting equations of three characteristics and pollution levels are established.In addition,the relation between the three characteristics and pollution levels are compared for two kinds of insulators with different leakage distances,namely,the general suspension insulator(GSI) and the anti-pollution suspension insulator(ASI).The reason the three characteristics of ASI are less than those of GSI is deduced.Furthermore,the neural network model with the input of the three characteristics of LC is put forward at the safe stage of LC to forecast the pollution levels of insulators.The error with this model is less than 4%.Finally the influence of the three characteristics on the forecast results is analyzed.The research findings have provided not only a theoretical basis for the characteristic selection of the pre-warning system of transmission lines,but also the decision on clearing polluted insulators.