Metodologia de Previsão de Cenários do Preço da Curva Forward de Energia Elétrica nos Horizontes de Médio e Longo Prazo

Abstract

In the electricity market, the issue of contract price negotiation between generators, traders, and buyers is of particular importance, as an accurate contract modeling approach leads to increased financial returns and enhanced business sustainability for the various participating agents. In the Brazilian case, electricity prices are published weekly by the Chamber of Electric Energy Commercialization (CCEE) through the execution of computational models such as NEWAVE, DECOMP, and DESSEM. The primary output of these models is the Preço de Liquidação das Diferenças (PLD), which serves as the official price reference in the electricity market and is used to settle surpluses and deficits of market agents in both the regulated and free contracting environments. However, the variables considered in the PLD calculation—primarily aimed at reflecting the system’s operational costs across short-, medium-, and long-term horizons—do not accurately capture market expectations, nor do they incorporate the risk premium intrinsic to the energy trading process. This misalignment is more pronounced in medium- and long-term horizons, where prices are less influenced by operational variables and more sensitive to factors modeling the business-related risks. Given this decoupling in medium- and long-term horizons, it becomes essential to establish methodologies that generate price curve scenario projections that account not only for the intrinsic aspects of grid operation but also for the risks associated with energy trading. This would enable the optimization of contract modeling among market agents. In this context, a more precise methodological approach was sought to incorporate the dynamic nature of the energy market while ensuring a robust and user-experience-oriented modeling process, allowing for reliable electricity price projections over medium- and long-term horizons (3 to 10 years). As a methodological proposal, the statistical technique of Dynamic Bayesian Networks (DBN) was adopted for scenario forecasting, combined with Genetic Algorithms to automatically optimize model performance by defining the topologies used by the DBN. To determine the most relevant variables for topology creation, a detailed study was conducted to identify which factors most significantly impact electricity price formation. Finally, the results obtained from the proposed methodology were compared with those generated by traditional techniques, such as Linear Regression, Support Vector Regression, and Extreme Gradient Boosting, using performance evaluation metrics to assess their effectiveness.