Study of Forecasting Methods’ Impact in Wholesale Electricity Market Participation

Abstract

Renewable energy sources have transformed the electricity market, allowing virtual power players or aggregators to participate and benefit from selling surplus energy. However, meeting demand and ensuring energy production stability can be challenging due to the intermittent nature of renewable sources. Accurate forecasting of energy consumption, generation, and electricity prices is critical to address these issues. Moreover, the selection of the best algorithm for forecasting is usually based on the predictions’ accuracy, neglecting other factors such as the impact of errors on the real context. This paper presents a study on the economic risk of price forecasting errors on the electricity market’s trading. For this, a simulation model is proposed to analyze the deviations between actual and predicted prices and how these deviations can affect trading in the electricity market, where the main purpose is to maximize profit, depending on whether the player is buying or selling electricity. The economic risk analysis and the predictions scores are used to improve the forecasts, using an approach based on reinforcement learning to evaluating and selecting which models demonstrated better performance in past transactions. The study involved simulating an aggregator’s transactions in the Iberian electricity market for two consecutive days in October 2021. Real data from the market operator between 2020 and 2021 and seven forecasting models were used. The findings showed that errors have a significant impact on profit. Including the economic impact analysis and score evaluation of forecasting methods to determine which method can offer better results has proven to be a viable approach.

The present work funds from FCT Portuguese Foundation for Science and Technology. Brígida Teixeira is supported by FCT with Ph.D. Grant 2020.08174.BD. This work has received funding from FEDER Funds through COMPETE program and from National Funds through (FCT) under the project PRECISE (PTDC/EEI-EEE/6277/2020). The authors acknowledge the work facilities and equipment provided by the GECAD research center (UIDB/00760/2020 and UIDP/00760/2020) to the project team.