Aplicação de métodos de aprendizado de máquina para identificação de perdas não-técnicas de energia elétrica

Abstract

An adversity that has been established since the beginning of the energy distribution systems is the confrontation between electric utilities and consumers that continuously searching methods for their own privilege, consuming energy and not paying for it correctly, representing one of numerous forms of non-technical losses of electricity. While the agents are still evaluating the cost-benefit ratio of installing smart meters, which can provide real-time control of customer consumption, statistical techniques associated with the recent enthusiasm caused by the development of artificial intelligence have been widely discussed in the literature, with several works and different applications. For this reason, this work models two methods of machine learning in Python, Neural Networks and Random Forests, in order to obtain a classifier of customers in regular or irregular, based on supervised learning. For this, real samples of historical data with return codes for inspections and consumption characteristics were pre-processed, data provided by a concessionaire in the south of the country. A total of 104 tests were performed to train the algorithms, ending in the best percentages, with an accuracy of 99.624%, specificity of 100% and reliability of 99.253%, criteria used to evaluate the performance of the classifiers. [resumo fornecido pelo autor]