论文部分内容阅读
根据电机绝缘中的主要放电形式，设计了模拟电机放电的６种试验模型，进行模型在不同电压或电流下的放电试验。应用采样率为１２８Ｍｓ／ｓ的数字化测量装置，在双层屏蔽试验室内，取得了各种模型的放电电流脉冲。采用自回归模型来提取脉冲波形特征，并用人工神经网络来识别不同的放电类型。研究了人工神经网络输入特征矢量的构成方式及自回归模型阶次对放电识别的影响。将放电脉冲波形比较接近的模型归并为一种类型，可提高识别的可靠率。
According to the main discharge form in the motor insulation, six kinds of test models for simulating the motor discharge were designed and the discharge test of the model under different voltage or current was carried out. Using a digital measuring device with a sampling rate of 128Ms / s, various models of discharge current pulses were obtained in a double-shelled laboratory. Autoregressive models were used to extract the pulse waveform characteristics and to identify different discharge types using artificial neural networks. The composition of input feature vector of artificial neural network and the influence of autoregressive model order on discharge recognition are studied. The discharge pulse waveform closer to the model merged into a type, can improve the reliability of recognition.