Um modelo híbrido-adaptativo para previsão de demanda de energia elétrica baseada em técnicas de inteligência artificial

Abstract

Among the challenges of electrical power sector companies, there is a need to maintain the quality and continuity on electrical power supply. For that, these companies must perform studies and develop procedures and methodologies to optimize and improve services rendered by them. One of these needs is the knowledge of power demand on future states, which helps operators, engineers, and decision-making managers. In this context, a redicting model of electrical power demand, having a very short term as the projection horizon and considering the influence of climate variables, contributes and supports decisions on the company where it is used. It may also allow identifying possible risk situations while operating the system. To build the previously cited model, companies may use more than one artificial intelligence technique, and also count on probabilistic methods, thus developing a hybrid-adaptive model. This predicting model configures its own parameters, in order to improve the method, adjusting itself to different scenarios, minimizing the difference between the value of the estimated demand and the real measured value. For this, we used self-organizing maps, which allow the creation of a probabilistic model, also known as Markov Chain, based on historical data from climate variables. We also added genetic algorithm techniques to make the proposed methodology adaptive to different scenarios, allowing the definition of input parameters. The research originating this work was applied on a power supplier company in Rio Grande do Sul, Brasil, which was interested in performing a power demand prediction, considering a very short term as its projection horizon. Thus, this method is reliable and robust for decision making, and helps on planning and brings economic benefits, since it provides reliable data for decision makers.