Intelligent techniques for fault location in distribution systems using artificial neural networks

Abstract

Power Distribution System forms a vital part of any Electric Power System. It acts as a final power delivery stage to the consumers. Hence, its study is important for the development of an efficient and reliable Power System. As utilities world over are competing for improvement in service to consumers, the area of Fault Diagnosis in distribution systems has attracted many researchers in the last two decades. Techniques involving algorithmic approaches, Artificial Intelligence (AI) was reported the most. AI techniques involving Artificial Neural Networks (ANNs) have proved their ability in recent years, in a variety of applications. Locating Faults on a Distribution System poses a major challenge to utility operators, due to the unique characteristics of the distribution system. These characteristics are worthy of discussion in the present work. Most of the previous work on Fault Location concentrated on estimatmg the faulty line segments through statuses of the circuit breakers and relays with the aid of algorithmic approaches. In this thesis, a new approach to fault location is adopted, in the general absence of circuit breaker and relay statuses in each section of the distribution feeder. Methods for estimation of fault location are developed using ANN techniques. Voltage and current measurements at the substation are used to estimate the location of fault imder various practical conditions of the distribution systems. Initially, a general three-phase fault detection algorithm is described with results of a practical distribution system, for detecting faulty line segments. Also, an efficient method of Short Circuit Analysis is described, which is used to simulate all types of fault conditions on a distribution systems. The work concentrates mostly on the fault pattern processing techniques and their right applicability in a Fault Locating System. While Function Approximation (FA) is the main theory used in the technique, Classification techniques take on a major supportive role to the FA problem. Fault Location in distribution systems is explained as a function approximation problem, which is complex to solve due to the various practical constraints particular to distribution systems. Incorporating Classification techniquesreduces this complex FA problem to simpler ones, which then gives efficient fault location estimations. The function that is approximated is the relation between threephase voltage and current measurements at the substation during fault, and the line reactances o f the fault points jfrom the substation. This function is approximated by FeedForward Neural Networks (FFNN). The voltage and current measurements are the inputs and the line reactances o f fault points are the outputs o f the neural network. Similarly, for solving the classification problems such as fault type classification, source short circuit level classification, and fault bus classification. Support Vector Machines (SVM) are employed. The work presented in this thesis consists o f using the FFNNs and the SVMs in a combination, for locating faults in radial distribution systems. Four schemes employing such a combination of FFNNs and SVMs are proposed for estimation o f fault location, along with the estimation o f the fault type and the source short circuit level during fault condition. Levenberg Marquardt learning strategy, which is robust and fast, is used for training FeedForward Neural Networks. A very fast and simple to implement algorithm. Sequential Minimal Optimization is used as the learning strategy for the case o f Support Vector Classification. These learning methods have the advantage that they can be implemented in practice with much ease, and at the same time provide very efficient results. Practical distribution systems of 52 nodes and 133 nodes are considered for studies. The results presented show that the proposed approach of fault location gives accurate results in terms of the estimated fault location. Practical situations in distribution systems, such as protective devices placed only at the substation, all types o f faults, a wide range of varying source short circuit levels, varying loading patterns, and long feeders with multiple laterals, are considered for studies. The results demonstrate the feasibility of applying the proposed method in practical distribution system fault diagnosis.