DETECTION OF NON-TECHNICAL LOSSES IN DISTRIBUTION SYSTEMS: A MACHINE LEARNING .
Random Forest, Support Vector Machine, Multilayer Perceptron.
Energy losses occur in any electrical system, whether due to physical phenomena or human action. Technical losses are those inherent to the system, while non-technical losses, also known as commercial losses, are generally associated with some type of fraud committed by the consumer. Machine Learning Algorithms can be used to detect patterns of electricity consumption, in order to indicate whether a particular consumer is committing fraud. A real electricity consumption database was used, which was already labeled between honest and fraudulent consumers. As this base is unbalanced, it was necessary to apply the SMOTE oversampling technique to correct this problem. The detection was performed using three different models, Random Forest, Support Vector Machine and Multilayer Perceptron, in order to compare their performance. It was found that the Random Forest model provided both the best classification performance and the lowest computational cost among the three models. The simulations were performed in the IDE Spyder through the programming language Python.