Artificial Intelligence for the Detection of Electricity Theft and Irregular Power Usage in Emerging Markets

Power grids are critical infrastructure assets that face non-technical losses (NTL), which include, but are not limited to, electricity theft, broken or malfunctioning meters and arranged false meter readings. In emerging markets, NTL are a prime concern and often range up to 40% of the total electricity distributed. The annual world-wide costs for utilities due to NTL are estimated to be around USD 100 billion. Reducing NTL in order to increase revenue, profit and reliability of the grid is therefore of vital interest to utilities and authorities. In the beginning of this thesis, we provide an in-depth discussion of the causes of NTL and the economic effects thereof. Industrial NTL detection systems are still largely based on expert knowledge when deciding whether to carry out costly on-site inspections of customers. Electric utilities are reluctant to move to large-scale deployments of automated systems that learn NTL profiles from data. This is due to the latter's propensity to suggest a large number of unnecessary inspections. In this thesis, we compare expert knowledge-based decision making systems to automated statistical decision making. We then branch out our research into different directions: First, in order to allow human experts to feed their knowledge in the decision process, we propose a method for visualizing prediction results at various granularity levels in a spatial hologram. Our approach allows domain experts to put the classification results into the context of the data and to incorporate their knowledge for making the final decisions of which customers to inspect. Second, we propose a machine learning framework that classifies customers into NTL or non-NTL using a variety of features derived from the customers' consumption data as well as a selection of master data. The methodology used is specifically tailored to the level of noise in the data. Last, we discuss the issue of biases in data sets. A bias occurs whenever training sets are not representative of the test data, which results in unreliable models. We show how quantifying and reducing these biases leads to an increased accuracy of the trained NTL detectors. This thesis has resulted in appreciable results on real-world big data sets of millions customers. Our systems are being deployed in a commercial NTL detection software. We also provide suggestions on how to further reduce NTL by not only carrying out inspections, but by implementing market reforms, increasing efficiency in the organization of utilities and improving communication between utilities, authorities and customers.