Predicción del precio de la energía eléctrica en Colombia mediante un enfoque de machine learning

Abstract

In this research, numerous predictive models are developed, including regression models, VAR models, ARIMA models, ARIMAX models and SARIMAX models, which were further used to estimate and predict the electricity spot price, and therefore obtaining an approximate value for the sale of a kilowatt-hour, a critical input for calculating the revenues in the valuation models of electric power generations projects in Colombia. This was accomplished using the historical records from XM’s databases, analyzing the relationship between the historical spot price for electricity in the frame of time from January 2000 to July 2023, other input variables were also considered such as hydrological contributions, hydrological discharges and hydrological reserves expressed in terms of energy, as well as the potential effects of climatological phenomena like the El Niño Southern Oscillation (ENSO) that occurs in the country. The results of the research indicate that the prices of the kilowatt-hour are affected by the rainy season and specially by the occurrence of the El Niño phenomena, during which prices increase triggering the scarcity price of the system, which can be observed in the years 2015 and 2016. Finally, as a result, all models follow the price behavior trends. The models were subjected to different time horizon tests, finding that the model to be used depends on the time horizon that the investor needs to analyze: VAR models for the short-term, SARIMAX models for the medium-term and multiple regression models for the long-term.