中文摘要：

提出了一种基于粗神经网络（roughmembershipneuralnetwork，RMNN）的高压输电线路故障分类方法，用10个独立的粗神经网络来分类识别输电线路的10种故障类型。粗神经网络利用粗神经元和模糊神经元代替普通神经元，有效地提高了神经网络的训练速度，并能减少网络的训练样本。基于对大量故障数据的分析，综合利用故障发生后5ms内故障电流时域和时频域的13种不同特征量作为故障分类的依据，以提高故障分类的正确率。PSCAD／EMTDC仿真实验结果表明：该故障分类方法能快速准确地分类识别各类故障，并且不易受故障时刻、过渡电阻、故障位置等因素的影响，具有较好的适应性。

英文摘要：

A novel approach to fault classification of transmission lines based on rough membership neural networks （RMNN） is presented. Ten RMNNs are designed to classify ten kinds of fault type such as single phase to ground faults （Ag, Bg, Cg）, phase to phase to ground faults （ABg, BCg, CAg）, phase to phase faults （AB, BC, CA）, and three phase fault （ABC）. In order to reduce the training time and cases of artificial neural network, the input layer of rough membership neural networks consist of rough neurons while the hidden layer and output layer consist of fuzzy neurons. The distinctive time domain features and time-frequency domain features are extracted from only quarter period of post-fault current signals and fed into the rough membership neural networks for classifying faults. Extensive computer simulation has been conducted using PSCAD/EMTDC. Results show that the approach provides an accepted degree of accuracy in fault classification under various fault conditions.