Modelagem estatístico-computacional do modelo de negócio da CEMIG-D utilizando bases de dados e conhecimento técnico

Abstract

The Brazilian electricity sector differs from other more traditional service sectors. It is charac- terized as a natural monopoly and therefore must be regulated to guarantee its efficiency. The agency responsible for regulation in Brazil is the National Electric Energy Agency (ANEEL). Among its various attributions, it can be highlighted the acting as a mediating agent between the generators, transmission and distribution companies of electric energy and the customers. In the case of distributors, there are other details that make it difficult to understand how the distribution sector works. As the contracts are signed with ANEEL, the distribution companies do not own the electrical assets they manage, they are only the owner since the owner is the Union. This condition, together with economic and political crises in the last few years, makes the management of distribution companies a challenge. Among the distributors in Brazil, the one with the largest operation (number of customers and revenue) is Companhia Energética de Minas Gerais (CEMIG). This condition makes the company the benchmark for others in the sector. Furthermore, carrying out an R&D project with the company allowed for an approach and easier access to its data. Thus, in this work, the application of different modeling, statistical and computational tools is proposed in order to support the management of different areas of CEMIG. For this, first frameworks are used to model, understand and evaluate the business model of the distribution companies. The results found allow us to visualize the functioning of the sector and serve as a basis for the construction of other statistical and mathematical models. Then, an analysis of the sector was carried out from the point of view of sustainability as defined by ANEEL. The study used comparative and geographic analyses and ordinal logistic regression models. The results allowed us to visualize the state of sustainability of the sector between 2011 and 2018, in addition to evaluating the impact on sustainability of factors such as type of control and region. Then, a study was carried out on the Index of Equivalent Outage Duration per Consumer Unit (DEC) using SEM and multilayer hybrid models. This index is used by ANEEL as the main indicator of the quality level of the operations of the distribution companies. The objective in this step was twofold: 1) to identify the main variables that impact the DEC; 2) find the best possible predictive model for DEC. The first objective allows the company to identify the best way to diversify its investments and make decisions on a quantitative basis. The second objective aims to enable the company to carry out simulations and analyses of future or uncertain scenarios. The results found allowed us to identify the most relevant variables when analysing the DEC. A model with considerable predictive power was also achieved using accounting, operational, climatic and geographic variables as input. This research work is characterized as a proposal for a doctoral degree in Production Engineering as it was not found in the national and international scientific literature a computational statistical model developed specifically to assist the management of a Brazilian electricity distribution company. The results obtained so far are unprecedented and have already been awarded at a national event in the energy sector. In addition, an innovative computational tool was developed that allows the adjustment and simulation of the developed methodologies in a user-friendly interface. The proposed model for the DEC indicator aggregates Bayesian models for regionalization, multiple regression models and machine learning models (machine learning) and is, therefore, characterized as a hybrid model. This model can be easily applied to other contexts such as financial compensation and annual revenue of CEMIG-D or any other Brazilian electricity distribution company.