Electricity theft detection in smart grids based on deep neural network
Lepolesa, Leloko James L
Electricity theft is a global problem that negatively affects both utility companies and electricity users. It destabilizes the economic development of utility companies, causes electric hazards and impacts the high cost of energy for users. Development of and adherence to smart grids play an important role in the development of theft detection measures. Smart grids generate massive data that includes customer consumption data which, through machine learning and deep learning techniques, can be utilized to detect electricity theft. In this work, statistical analyses are undertaken to investigate the difference in consumption patterns between faithful and unfaithful electricity users. Dataset weaknesses such as missing data and class imbalance problems are addressed through data interpolation and synthetic data generation processes. Comprehensive features in time and frequency domains are extracted and used in a fully connected feedforward deep neural network classifier. The minimum redundancy maximum relevance scheme is used to analyse individual features’ contribution to successful classification, thereby validating frequency-domain features’ dominance over time-domain features. The principal component analysis is employed to reduce the dimensionality of the classifier input while keeping the results satisfactory to simplify the training process. Electricity theft detection performance is improved by optimizing hyperparameters using a Bayesian optimizer. An adaptive moment estimation optimizer is employed to carry out experiments using different values of key parameters to determine the optimal settings that achieve the best accuracy. The classifier achieves 97% area under the curve (AUC), which is 1% higher than the best AUC in existing works evaluated on the same dataset, and 91.8% accuracy, which is the second-best on the benchmark.
A dissertation submitted in fulfilment of the requirements for the degree of Master of Science in Engineering to the Faculty of Engineering and the Built Environment, School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg, 2022