Abstract:

Abstract Electricity theft becomes a major concern for utilities in this new era of high tech, self-sufficient dwellings. Finding and reducing energy losses or theft has proven challenging due to insufficient inspection methods. In terms of energy, both technical and non-technical losses (NTL) are included in distribution. Energy theft is a significant factor in NTL that can strain the finances of service providers. Wireless data transmission is used in modern smart metres. It follows that hi-tech dwellings can be easily hacked to steal power. Many new technologies have been implemented into Advance Metering Infrastructure (AMI) to combat energy theft. It is necessary to derive the consumption pattern in order to identify illegal energy customers. Using data mining methods, a computational system is designed for examining and identifying energy consumption patterns. Through the use of machine learning, we are able to improve our customers' energy consumption statistics and provide them with early warning of any irregularities. Multiple supervised learning techniques are examined and contrasted in relation to their predictive accuracy, recall, precision, AUC as well as F1 score. These include the decision tree (DT), ANN, Deep ANN, Modified ANN and AdaBoost. Based on the results of the study, MDANN is superior to alternative classifiers for supervised learning including ANN AdaBoost as well as DT according to recall, F1 Score along with AUC. The upcoming research should focus on testing different supervised learning algorithms using various datasets and including appropriate pre-processing procedures to boost performance.

Keywords: