Date of Award

Fall 2002

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational Analysis and Modeling

First Advisor

Steven Rovnyak

Abstract

The objective of this research is to demonstrate pattern recognition tools such as decision trees (DTs) and neural networks that will improve and automate the design of relay protection functions in electric power systems. Protection functions that will benefit from the research include relay algorithms for high voltage transformer protection (TP) and for high impedance fault (HIF) detection. A methodology, which uses DTs and wavelet analysis to distinguish transformer internal faults from other conditions that are easily mistaken for internal faults, has been developed. Also, a DT based solution is proposed to discriminate HIFs from normal operations that may confuse relays. Both methods have been verified with simulation data generated by the Electromagnetic Transients Program. Compared with traditional methods, both show better performance. After being trained with a large number of carefully selected features, the desired DTs can obtain an accuracy of greater than 95%. Further, no special equipment is necessary; the DT-based controller only needs the standard relay input signals sampled at 1920 Hz. So far, no one has applied the same methodologies to solve these problems. Even though some future work with experimental data is needed to make the methods more convincing for utilities, the research has already shown that pattern recognition is a promising direction in developing power system protection algorithms.

Download

Share

COinS