Predicting the electric energy consumption is a very prominent research topic in the energy industry, since forecast errors can lead to substantial financial losses, being also crucial to establish a more reliable planning horizon for energy utilization by several consumer centers. In fact, the discussion on the importance of load forecasting has significantly increased in the last few decades, especially after the 2001 energy crisis in Brazil, which prompted the energy companies to face the issue of carrying out a more realistic forecast of the energy demand in Brazil. In this context, the use of computational approaches bas become an important ally, making feasible strategic plans for energy efficiency and public policies for energy security while handling exploratory data analysis tools and load forecasting models. This research project aims at investigating the problem of electrical load forecasting by applying Machine Learning (ML) approaches so as to aid decision-making procedures made by energy companies and government agencies. For this purpose, the following ML methodologies are taken: Classic Regression Models, Artificial Neural Networks and Decision Trees + Bagging. The techniques are designed and properly validated by taking distinct datasets of electricity load which are publicly available, in addition to a pilot dataset taken from a specific energy company, therefore allowing us to measure the performance of the evaluated methods in a practical context of use.
News published in Agência FAPESP Newsletter about the scholarship: