Application of artificial neural networks to distance protection

Artificial neural network (ANN) strategy was developed as a method of using a large number of simple parallel processors to recognize preprogrammed, or "learned", patterns. This approach can be adapted to recognizing learned patterns of behavior in electric power-systems where exact functional relationships are neither well defined nor easily computable, and is able to compute the answer quickly by using associations learned from previous experience. Certain problems in power systems, with their inherent nonlinear and complex nature, seem amenable to solutions through trained ANNs.; A distance relay is an important protective relay with its excellent performance for transmission line protection. However, the suitability of conventional distance relays to adapt to change in source impedance and to the effect of remote infeed and nonlinear arcing fault resistance is still unsatisfied. Utilization of artificial neural networks is a good strategy for those problems, using pattern recognition, a basic function of distance relays.; The goal of this thesis is concentrated on creating more selective ground fault detection by using artificial neural networks. Two applications of artificial neural networks to distance protection are presented in this thesis, one for non-linear arcing fault resistance and another for remote infeed. At the current stage of research, only single-line-to-ground faults are considered because most faults in power system transmission lines are line-to-ground faults.; In the case concerning the effect of remote source infeed, research was focused on creating more sensitive ground fault detection in spite of pre-fault loading in either direction, variable source impedance and variable ground fault resistance. A matured power system simulator named Electromagnetic Transients Simulation Program (EMTDC), was utilized to create the training and testing cases with varying system parameters. The proposed neural network was trained using many load and fault cases, tested using cases with different system conditions and run using more detailed fault cases along the whole transmission line.; In the case concerning the nonlinear nature of arcing fault resistance, research was focused on creating more sensitive arcing fault detection, especially for radial distribution lines where arc resistance can be a significant part of the zero sequence impedance. A neural network was trained, tested and run by three sets of pattern vectors with different system conditions. A simple power system model and a nonlinear arcing fault resistance model were used to collect training, testing and running patterns for the proposed neural network. A new operating characteristic based on fault voltage instead of fault resistance was devised.; The prospective ANN distance relays showed very good performance in detecting a single-line-to-ground fault with the effect of remote source infeed, or with nonlinear arcing resistance along the whole transmission line. Basic principles learned from this investigation of application of ANN's to power system protection will be of value to future advances in this direction.