An intelligent decision making system for detecting high impedance faults

Abstract

The clearing of distribution line faults is usually accomplished by devices which can sense the overcurrent produced by a fault and react to disconnect the faulted section of the feeder from the healthy section. However, high impedance faults do not draw sufficient fault current to be detected by such a conventional protective scheme. Such faults may be caused by a conductor on the ground. Arcing is usually associated with these faults, which may result in a fire hazard. The harmonic currents characterized by an arc are variable, transitory, and random in their behavior. The relative amplitude increase of harmonic currents is very large in some high impedance faults, but sometimes it is very low, and other times very similar to the level of the normal state. While a few techniques to detect high impedance faults have been proposed, and some progress has been made, a complete solution has not been found. This research concentrates on designing an intelligent decision making system which uses multiple detection techniques incorporated with an appropriate detection reasoning method and a learning ability to provide a more effective solution for high impedance fault detection. Major parts of this system are a technique selection, a technique combination, and an induction process. The method of decision making under incomplete knowledge is used to select appropriate techniques because the information on the performance of techniques are available but not complete. With these selected techniques, the Dempster-Shafer theory is adopted to find a final belief about the system status by combining the belief from each technique. Inductive reasoning with minimum entropy is applied to find decision rules and thus to adjust the technique selection process. A learning detection system which combines all three major parts is proposed to realize this intelligent decision making system. The learning detection system synthesizes the final belief of the combined techniques, the status output of a decision tree from the inductive reasoning process, and an event detector output to detect and identify the system status. The intelligent decision making system makes a smart decision on an example execution with a complex test set of sample data.