【摘要】 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。更多还原

【Abstract】 ccording to the main discharge types in insulation of electrical machines ,6 kinds of experimentalmodels simulating discharges in electrical machines were designed and model experiments under variousvoltages or currents were performed .Using digital measuring device with sampling rate of 128 Ms/s ,in2- coat shielding room ,pulse shapes of discharge current of 6 models were obtained .According to theauto -regression model the characteristic of discharge current pulse was extracted and the artificial neuralnetwork was used to recognize the type of discharge .The influence of the constiute methods of the input characteristic vector used forartificial neural network and the order of the auto-regression model on therecognition were investigated .Classifying the discharge types with similar current pulse shapes togetheras same type ,the recognition reliability could be raised.更多还原