Aplicação de redes neurais para previsão de demanda de energia elétrica

Abstract

In the early 2000s, the Brazilian electrical system was less developed than it is today and depended much more on hydroelectric generation, as 80% came from it. A study conducted by the Ministry of Mines and Energy revealed that the possibility of an electricity deficit at the end of the 1990s was at levels above acceptable levels and the balance between supply and demand was precarious. For this reason, a program to rationalize the use of electricity was created, divided into two stages. The first was rationalization, and if the risk of demand greater than supply was not reduced, the government would trigger the second stage, rationing. Therefore, in order for better planning and more efficient production to occur, avoiding excessive production or production below demand, it is essential to use and improve methodologies for forecasting electricity demand. That said, the current work aims to develop a computational support tool for strategic planning in electricity generation and distribution systems. In the proposed methodology, a short/medium term forecasting system will be implemented for the state of São Paulo, using computational techniques of artificial intelligence based on artificial neural networks (ANN). For this purpose, electricity consumption data on the grid (MWh) was used, acquired through the electronic portals of the National Electric Energy Agency (ANEEL) and Energy Research Company (EPE).