Neural network based techniques for distribution line condition monitoring

Because a power distribution line is spread over a significant area, it is exposed to a variety of hazards. Causes of line abnormal conditions include lightning, wind, ice, snow, salt spray, birds etc. These make it extremely difficult to design an accurate condition monitoring system for distribution lines by using conventional techniques. In this paper, a neural network technique is proposed to develop a novel condition monitoring scheme for a distribution system. The paper starts with description of the modelling techniques for some common abnormal conditions in distribution systems, followed by a presentation of digital simulation of some typical situations such as high impedance, arcing and solid faults. The spectrum technique is employed to analyze the features associated with different conditions. Then special emphasis is placed on the neural network, including the determination of network input, network size and its training. The validation results demonstrate the feasibility of this approach. >