Artificial neural network high impedance arcing fault detection

A High Impedance Fault (HIF) is an abnormal event on electric power distribution feeder which does not draw sufficient fault current to be detected by conventional protective devices. Such faults may be caused by a conductor on the ground, tree contact or insulation failure. Arcing is usually associated with these faults, which is dangerous to human life and could cause fire. HIFs do not usually cause any major problem to the power system; however, the protection against them desirable.; The objective of this research work is to explore the use of Artificial Neural Network (ANN) techniques to develop a reliable HIF detection scheme, which is capable of providing the required relay dependability, and security. A few techniques to detect HIFs have been previously proposed and some progress has been made, yet a complete solution has not been found. Most of the detection methods require extensive computation in the preprocessing stage for feature extraction of the input signals. Then a strategy is applied to obtain detection parameters.; In order to obtain transient waveforms of the fault event, digital simulation were performed using ATP/EMTP for different variations of fault types, fault locations, capacitor and load switching, etc. A commercial distribution feeder with various types of loads was used for simulation purpose.; Fourteen parameters were selected and extracted to represent the transient behavior of switching events. Extensive simulations of various events were studied to generate a large data bank. A feed-forward multi-layer artificial neural network was trained using high impedance arcing faults, load switchings, normal loads, and capacitor switchings. The network was trained by a back-propagation training algorithm. The results indicated that neural network was able to reach the solution of the problem. The detection algorithm was able to identify high impedance arcing faults and fault-like transients with a high percentage of success rates.; As a future work, it was suggested that the network should be trained by real samples of currents from a distribution feeder. Later, this algorithm can be implemented in hardware to be a part of electric power distribution feeder protection systems.