用多个对应的后向神经网络进行同杆双回线故障识别及测距的模式

电力系统高压同杆双回输电线的应用日益增多，但其故障识别与测距的问题尚未完全解决，同杆双回线因存在回路间耦合等因素，使得用单一的神经网络进行故障识别与测距的结果并不理想。作者比较分析了BP网络与Kohonen网络在同杆双回线测距方面的优缺点，提出了将故障识别与测距任务分配到多个网络的方法即将同杆双回线的每种故障模式各与一个BP人工神经网络对应，在线路上取一些固定点作为标志点，训练成功的BP网络输出的模糊值代表了标志点上发生故障的可能性。用模糊值构成插值曲线，根据曲线的相对位置确定故障模式，并由曲线的最小值求得故障距离。大量仿真表明该法可以准确可靠地确定故障模式并能测得较高的测距精度。

IDENTIFICATION AND LOCATION OF FAULT ON DOUBLE CIRCUIT TOWER BY MULTI-CORRESPONDING BP ANN METHOD

In high voltage power transmission the double circuit tower is increasingly applied, but the problem of fault identification and location under this mode are still not well solved. Because of the coupling between the two circuits on the same tower, the results of fault identification and location by only single neural network are not satisfactory. After comparing the merit and demerit of applying BP network and Kohonen network to fault identification and location under this mode, a new method of fault identification and location under this mode is put forward. In this method the task of fault identification and location is distributed to multi-BP network, i.e., each pattern of the faults under this mode corresponds to a BP network respectively. Taking some fixed points in transmission line as marked points, the fuzzy value outputted from the trained BP network represents the possibility of occurring fault at marked point. Constituting the interpolating curves by use of fuzzy values, according to the relative position of the interpolating curves the fault pattern can be determined and the faulty position is equal to the minimum of the curve from the determined fault pattern. A lot of simulation results show that with the presented method the fault pattern can be exactly and reliably determined and the accuracy of fault location is satisfactory.