Forecasting Electricity Consumption Using Function Fitting Artificial Neural Networks and Regression Methods

Abstract

With the growth of smart grids, consumers now have access to new technologies that enable improvements in the quality of service provided and allow new levels of energy efficiency. Much of this increase in energy efficiency is directly related to changes in consumption habits due to the quantity and quality of information made available by new technologies. At this point, short-term consumption forecasting can be considered an effective information tool in the search for better consumption patterns and energy efficiency. This paper presents prediction tests combining the result obtained from an artificial neural network and regression methods. The artificial neural network used was the Multilayer Perceptron (MLP), and its results were compared with polynomial regression techniques (first, second, and third degree), demonstrating the superiority of the network. The neural network has proven to be a highly effective tool for forecasting future data, demonstrating its ability to capture complex patterns in input data and produce accurate estimates. Additionally, the flexibility of neural networks in handling large volumes of data and their continuous adjustment capability further enhance their suitability as a robust tool for future predictions. The results corroborate the capacity of the methodology presented for short-term consumption forecasting.