Previsão de carga para consumidores de baixa renda no estado do Ceará

Abstract

The development of predictive models of electrical energy demand is a fundamental component in the operation of the electrical sector, being very useful in the planning and operation of an electrification system. Electricity concessionaires need to predict, as best as possible, the demand of their consumer units, so that in auctions in the Regulated Contracting Environment (ACR), they can avoid under or over contracting within their concession area. Given this context, the objective of this dissertation is to test models for forecasting the electrical energy demand of residential consumers in the low-income subclass, in the State of Ceará, using as data for training the models, those referring to the years 2012 to 2020 and the year 2021 for their validation. To this end, this research uses three predictive models, in order to estimate the electricity demand of residential consumers in the low-income subclass. The models are Exponential Smoothing: ARIMA (Auto-Regressive Integrated Moving Average) and SARIMAX (Seasonal Auto-Regressive Integrated Moving Average with and Xogenous factors), using the minimum wage and tariff as exogenous variables. And the mean absolute percentage error MAPE (Mean Absolute Percentage Error) as a metric for evaluating the models. The databases analyzed are from the Energy Research Company (EPE) and programming tools in Python language, used to simulate the models. From this, the results indicate that the Average Absolute Percentage Errors of the three predictive models used are: 2.91% for Exponential Smoothing, 3.40% for ARIMA, 2.99% for SARIMAX, when the exogenous variable is the minimum wage and 2.58% for SARIMAX, when the exogenous variable used is the tariff. By observing patterns, errors can be shown with good accuracy in forecasting demand.