Detection of electrical theft and fault meter in power distribution system using long short-term memory on automatic meter reading

Abstract

This thesis presents the method to detect power theft and measurement equipment error based on the real data of meters in power distribution. That the real data included the parameter of voltage and current and was recorded for 15 minutes. The data is a large record that demonstrates the attribute of meter onsite. In this study, the real data of power theft and measurement equipment faults in electricity distribution are collected. It has extracted features by using EDL standards and the experience of technicians to classify. Then, the dataset has divided for training and testing are 80%, and 20% followed by the amount of each class, respectively. The main objective of the research is to use LSTM algorithms for classifier event types in meter data. In addition, the datasets have been applied with algorithms of Ada boost, neural network, and support vector machine (SVM) used in past research to compare our research. From the comparison and accuracy models of this research. The accuracy of LSTM is more than the other algorithm and achieved 99%. it has illustrated the model’s significance classification and can be used in the practice of work.