SHORT TERM LOAD FORECASTING USING ARTIFICIAL NEURAL NETWORKS ENSEMBLE
Load forecasting, neural networks, signal processing.
Industrial, residential and commercial buildings depend primarily on an uninterrupted and low cost energy supply. Brazil is a country rich in natural resources and renewable energy, but energy prices across the country are high. Despite the
excellent levels of solar irradiation, the prevailing wind characteristics (constant and unidirectional) and also the generation of energy through sugarcane (biomass), the Brazilian energy matrix is predominantly dependent on hydroelectric plants. Due to
high energy costs in the country, many companies are looking for alternative sources and systems that manage the demand and consumption of electricity. In this context, one of the most important prerequisites for power management is the power forecast
required by the facility. The demand curves of a given location are greatly influenced by factors such as weather, human activities, and installed load. In this way, the appearance of load curves may vary greatly on a given day, making the forecasting task difficult. A proper forecast combined with a management system leads to the efficient use of energy by the consumer at the lowest possible cost. In this work, statistical and Artificial Intelligence methods were used to forecast short-term power in the interval of 15 and 30 minutes ahead using data from the power demanded by the electrical installation of the State University of Santa Cruz (Ilhéus, Bahia, Brazil) from March 2014 to March 2015. Only active power data was used to predict the active power demanded by the electrical system and three case studies were proposed to establish how to handle with time information available in the series to make the
forecasts with the smallest possible error.