Detection of electricity theft in Chinese power utility state grid corporation using hybrid deep learning model

Energy conservation and carbon reduction have become the top objectives for all countries due to the necessity of tackling the global climate problem. The cost of gathering data on residential energy usage has greatly decreased, thanks to improvements in metering infrastructure and the widespread use of smart meters. Nevertheless, the ongoing issue of energy theft continues to be a major obstacle for power providers throughout the world and results in high expenses for the smart grid. Traditional energy theft detection (ETD) models struggle with a number of problems, such as the unequal distribution of data on power use, the difficulty of managing high-dimensional data, and the interference of unintentional variables that hinder detection attempts. This work offers a novel ETD strategy designed to address these issues while taking into account the specifics of smart grids. An Autoencoder-based Deep Belief Network (AE-DBN) model created for theft estimate forms the basis of this strategy. The AE-DBN deep learning model uses the DBN’s ability to extract characteristics that are pertinent to electricity theft (ET) and takes use of its capacity to identify long-range correlations in the data. The study uses Fire Hawk Optimization (FHO) to fine-tune configurable parameters to improve the AE’s performance and increase its capacity to spot anomalies and deviations in power consumption data. The ultimate goal of theft detection is to reduce the incidence of false positives. This proposed model is put through testing utilizing actual electricity consumption (EC) profiles obtained from the State Grid Corporation of China (SGCC), a significant Chinese power utility, in order to confirm its efficacy. The findings show an outstanding accuracy rate of 98.17%, outperforming the capabilities of existing models in successfully handling the challenge of detecting ET in smart grids.