Detection of High Impedance Faults Using Artificial Neural Networks

High impedance faults (HIF) on electric power distribution feeders are of a great concern to electric power utilities. These faults often occur when a bare conductor breaks and falls to ground through a high impedance current path [1]. Some HIFs draw sufficient current to be cleared by overcurrent protection, however, many HIFs draw only small amount of current, which makes them difficult to detect by conventional overcurrent relays. To date, there is no detection technique that can identify all HIF and achieve a high degree of security or dependability for distinguishing them from HIF like events, such as load or capacitor switching. This is practically impossible because of the probabilistic nature of HIF detection.. Several methods for detection of HIF on power distribution feeders have been developed using various techniques and methods [2-12]. In 1990, a strategy for HIF detection was developed using ANNs [9]. Twenty parameters are chosen to represent the feeder operation over one cycle. This strategy showed the great potential of ANNs as a more effective method for HIF detection. A. F. Sultan et al proposed a method that uses a three-layer feed-forward ANN to differentiate between HIF and fault-like loads [10]. Another promising method based on ANN is proposed to detect single phase to ground faults on power distribution systems, with compensated neutral grounding, by comparing the residual and the phase currents [21]. K. L. Butler and et al described another approach for fault diagnosis on power distribution systems [11]. A.M. Sharat, and et al presented a new FFT-based relaying scheme for HIF detection on radial distribution feeders [12]. The objective of this work is to design an ANN that can detect HIF with high accuracy. In addition, the ANN shall be able to locate the fault, distinguish HIF from other normal switching events and identify the faulty phase.