Planejamento de geração de energia complementar térmica associada a energias renováveis utilizando inteligência artificial

Abstract

The Brazilian Electrical System has a diversified electric power generation matrix, nevertheless it is mainly composed by hydrothermal generation. In this sense, the operational planning of this system can be detailed as a large-scale optimization problem, where is necessary to use resources in a rational way, by operations dynamic, stochastic, interconnected and non-linear. The electric energy generation is susceptible to climatic variations, since the precipitations reduction causes a decrease in the hydroelectric plants reservoirs and consequently a falling in the electric energy production. The use of wind energy has been growing in recent years as an alternative to solve an eminent energy crisis. However, this power source requires adequate planning in order for the electric system operate in a safe a reliable way, due to its intermittent behavior and low predictability. In order to overcome the limitations of the energy sources mentioned above, it is necessary to guarantee the power service by reliable energy sources, like thermal generation, which is considered as a source of reliable energy because it does not suffer external influences. Among the thermal sources that compose the Brazilian Electric power generation matrix, Natural Gas has become the main fuel due to it being less aggressive to the environment compared to the others fossil fuels and by the proven national supply, which characterizes it as a reference for expansion in short time. Thus, this study proposes a shortterm modeling and simulation structure to predict the electric power production capacity for the southern subsystem generation park, analyzing the generation forecasting and emphasizing the complementarity of energy imposed on thermal generation, taking into account operation historical series. For the electric power generation forecasting modeling, a Multilayer Perceptron Artificial Neural Networks (MLP ANNs) structure was employed, due to its ability to learning by complex non-linear relationships between input and output variables from a data. In addition, to generate multicenary (critical, ideal and optimistic), the Monte Carlo Method (MCM) was used. The prediction results obtained by MLP ANN for the rates the MAE and RMSE respectively 3.22% and 4.01% to hydropower generation, and the 5.36% and 6.31% to wind generation. In addition, with results of MLP ANN and MCM combination proved that it is possible to quantify the energy availability of the south subsystem generation parks through in the adverse conditions, emphasizing the importance of the prediction model to improve the planning and operation of an electric system.