APPLICATION OF LEARNING MACHINE TECHNIQUES IN THE DETECTION OF NON-TECHNICAL LOSSES.
Machine Learning, Non-technical Losses, Commercial Losses, K-Means, Multilayer Perceptron, Hierarchical clustering.
The high rate of non-technical losses (PNT), also known as non-technical losses (PC), directly affects the billing of energy distributors electricity in Brazil, where, in the year 2020 alone, they subtracted around 7.5% of the amount of energy purchased by the concessionaires. The loss is passed on to consumers up to the regulatory limit, the difference between the regulatory limit and the total of PNT is paid for by the distributors. Actions to combat PNT are not always assertive, causing even more damage to companies. in order to direct inspections have been using ML algorithms to detect customers who possibly they are diverting energy or have a fault in their system of metering and billing. This work simulates this situation using a base of real data and performing tests with the Multilayer Perceptron, K-means and Hierarchical grouping, applying data processing techniques. was used the Jupyter IDE, available in Anaconda software. Each model was tested using the pure database, using the normalized database and also with the of standardized data. The model that obtained the best result was the MLP, with a accuracy of 69%, however the implementation of other techniques of data processing to obtain a better processing time.