Deep Reservoir Computing aplicado na previsão do preço de energia elétrica no mercado de spot

Abstract

The Brazilian electric system is a world reference for its particularities in the form of energy generation, transmission and distribution. The energy matrix is mostly composed of hydroelectric and thermoelectric plants, having seen a sharp increase in wind and solar sources in recent years. The dispatch of hydroelectric and thermoelectric power plants is performed centrally by ONS. The softwares NEWAVE and DECOMP is used for optimization aiming the lowest total cost over the entire study horizon. In addition to the dispatch of each hydro and thermal plant, these softwares disclose the liquidation price of the differences, used by the Electric Energy Trading Chamber to value the short-term energy price. This price is also used for bilateral negotiations between market agents, either for trading or buying and selling electricity to consumers. Market players use this software to project short-term electricity prices. However, the software needs high computational processing to perform the optimization, and the dispatch calculation is very time consuming. Thus, this paper presents a computational model using Deep Reservoir Computing to project the spot market electricity price. This work demonstrates that Deep Reservoir Computing can be used as a support tool for projecting electricity prices in the short term market. The proposed model was developed in Matlab software and, for the experiments performed, obtained better results than more popular neural networks, such as Multilayer Perceptron, where the correlation was 3.08% higher in Deep Reservoir Computing in addition to the absolute mean percentage error being 8,8% lower