Análise da demanda elétrica brasileira baseada em métodos de previsão ; Analysis of brazilian electrical demand based on forecast methods

Monografia (especialização) - Universidade Federal de Santa Maria, Centro de Ciências Naturais e Exatas, Curso de Especialização em Estatística e Modelagem Quantitativa, RS, 2022. ; The great technological advance in the last decades has presented a significant change in humanity's relations with the planet, energy consumption, as energy sources and the destiny of this energy that accompanies this profound change. In this more efficient process, it is an objective increasingly sought in most productive and social processes. Brazil is a major supplier of different energy sources and, at the same time, a major consumer. Therefore, an energy forecast is essential so that those linked to global sustainability will be close to energy demand. Forecast models play a fundamental role in the study of energy demand behavior, the main influential factors and for making more assertive decisions in any process. The present work analyzes the performance of Brazilian demand forecast models over the period from 1970 to 2020 considering the influence of variables from energy and social sectors of the country. Classical time series models such as the trend-corrected Exponential Smoothing model (Holt method) and the ARIMA model, as well as multivariate fuzzy and long-and short-memory recurrent neural time series models were used in this work. The univariate model obtained satisfactory results from the RMSE model, the ARIMA(2,0,1) model presented the best value of performance measurements for univariate models (Recurrent Neural Networks = 0.288), the second Recurrent Neural Networks (LSTM) model presents the best performance (RMSE = 0.830), in the third position, we have the Fuzzy Time Series model (RMSE = 4.271), in the fourth position, we have the Classical Holt Exponential Smoothing model (RMSE = 4.420) followed by the Smoothing model Damped Holt Exponential (RMSE = 4.497), finally, the Simple Exponential Smoothing model (RMSE = 5.4144). Analyzing the parametric models, the ARIMA model (2,0,1) also presented the best result for ...