Machine Learning Application for Renewable Energy Forecasting

Abstract

Renewable energy is a clean source known as green energy. Its benefits are enough established. However, its effective use and increasing its share have become a major challenge for system operators. Due to its direct dependence on environmental and meteorological factors, there are often uncertainties and unexpected consequences for integrated energy system planning. Thus, the prediction of the production of renewable sources is a very relevant issue. This paper considers the application of ensemble machine learning models for renewable energy forecasting. As input data for the machine learning modem, historical data on power generation was used for the 2019–2021 period of renewable energy including meteorological data from the power plants operating in the central power system of Mongolia. The ensemble machine learning model allows us to determine the non-linear and non-stationary dependence of the time series and can be implemented in the task of forecasting the daily generation schedule. The proposed model creates a day-ahead forecast of the hourly generation curve of the photo-voltaic power plants under consideration with a normalized absolute percentage error of 6.5 – 8.4%, and for wind farms, 12.3-13.3%. Increasing the accuracy of renewable energy forecasting can positively affect the operation and planning of the central power system of Mongolia.