Identificação de Perdas Não Técnicas Através de Método de Classificação e Otimização de Hiperparâmetros, Baseado em Dados Endógenos e Exógenos
Keywords: Non-Technical Losses, Outliers Identification, Decision Tree Method, Hyperparameter Optimization, Exogenous Data, Machine Learning
AbstractNon-Technical Losses (NTLs) are problems commonly found in electric power systems distribution, caused by theft or fraud of energy. These problems result in financial losses for distribution utilities as well as consumers, who partially bear the costs involved in the NTLs. In view of this, practices for the identification of consumer units that are committing some type of irregularity in their facilities must be applied. In this scenario data classification models emerge, which based on supervised learning based on endogenous and exogenous historical data, are able to interpret information and label them. One of these models is the Decision Tree, which associated with Machine Learning (ML) techniques for the optimization of hyperparameters can obtain results with high precision in the outliers identification. Thus, this study aims to implement the Decision Tree model to identify consumers with NTLs, and to propose the optimization of Decision Tree hyperparameters following three ML techniques, Bayes Search, Grid Search and Randomized Search. Finally, the results are discussed and analyzed, and considerations are performed on the models.