Uso de machine learning para otimização da renovação do parque de medidores em distribuidora de energia

Abstract

The present study proposes a new approach to the renovation of the electric energy meter park considered obsolete from a regulatory point of view. THE significant number of old meters still in use, combined with to the approaches currently employed, has generated high costs operational and exorbitant investments. It therefore becomes It is essential to seek more efficient alternatives. In that regard, The objective of the study is to improve cost efficiency operational, through increased energy aggregation, made possible by the precise identification of equipment that require replacement. The approach will act as a complement to the metrics already used, integrating Machine Learning to identify opportunities in this process renovation of the meter park. The models incorporated a series of input data and select more precisely and strategic the meters to be replaced, improving decision making. The results obtained demonstrate improvements in resource management and an increase in the efficiency of aggregating energy. This approach was implemented using a base historical period of four years (2020 – 2023), allowing the realization of a comparative analysis between learning algorithms machine: Random Forest, Logistic Regression and Gradient Boosting, compared to Multilayer Neural Network. Each one of these methods was analyzed in terms of performance, precision and generalization capacity. In the results comparative studies, the Artificial Neural Network showed an effectiveness superior, achieving an accuracy of approximately 75% and an AUC of 73%. This performance significantly surpasses the methodology currently used, which does not exceed 39% of accuracy. This will make it possible to optimize update processes technology, driving operational excellence and promoting more efficient and sustainable use of energy electrical.