Abstract:
This Thesis presents a process for classifying the cause of faults occurring in the Provincial Electricity Authority (PEA) distribution systems, consisting of a group of animals, tree contact, and equipment failure. By using data from the analysis of current and voltage waveforms obtained from relay installed at the substation to find the voltage dip, rate of change of current and voltage, neutral current and voltage, temporary or permanent fault, fault type, evolving fault, fed data to machine learning (ML) compared to the artificial neural network (ANN) to classify the cause of faults in order to reduce the duration to patrol the line of the distribution system after the fault occurs and to use in the management planning for resolving electrical interruption. Accuracy testing of the algorithm by real testing was performed with the recorded data from the distribution system of PEA in the central southern area of Thailand. The test result of the proposed process shows that the ML can classify more accuracy than the ANN with 200 training data, the group of fault events is correctly classified with the ML accuracy of 72.72%, the proposed method can provide the accuracy of identifying fault cause with 81.21% - 88.89% accuracy.