AT&C and non-technical loss reduction in smart grid using smart metering with AI techniques

Abstract

Aggregate Technical and Commercial (AT&C) damage are a serious issue for electricity distribution companies globally, hindering economic growth and sustainability.  Among them, non-technical losses (NTLs), such as electricity theft, fraud, and non-payment, contribute to substantial financial losses and may jeopardize power quality and grid stability. Growing usage of smart grids and Advanced Metering Infrastructure (AMI) opens new ways of effective management of energy, as well as sophisticated approaches to electricity theft, creating demands on cutting-edge methods of detection. This research aims to enhance NTL detection by introducing a hybrid approach that integrates Temporal Convolutional Networks (TCN) and LightGBM, or Light Gradient Boosting Machine.  TCNs are used in order to detect complex temporal features in smart meter consumption records, recognizing sequential patterns characteristic of fraudulent behaviour. LightGBM, which is an extremely effective gradient boosting architecture, which is then applied to classify consumption behaviour correctly as normal or suspicious.  An real dataset is used to train and evaluate the suggested model of smart meter records, demonstrating its ability to discriminate between normal and potentially fraudulent consumption patterns. Results present promising effectiveness in identifying usual use; however, the research indicates challenges to achieving high accuracy and memory in detecting energy theft.   This emphasizes the necessity of further research and model refinement to enhance its effectiveness in real-world applications and to counteract the negative impacts of NTLs on electricity utilities and consumers.