Prediction of electricity generation based on solar irradiance: A case study

Abstract

In the face of the need for greater sustainability, there has been an increasing pursuit of the use of renewable energy sources such as wind, solar, and biomass. Due to Brazil’s climatic conditions, the participation of wind and solar energy sources has rapidly grown in the energy matrix. This work addresses the prediction of energy generation in a solar complex. To do so, machine learning techniques are proposed to determine the energy production to be generated throughout a day, according to the solar irradiance measured by solar stations. Real data from a solar complex located in the north of Minas Gerais were used as a case study. Various methods and parameters were evaluated through the development of models using Random Forest (RF) and Artificial Neural Networks (ANNs). In total, six types of ANNs and three types of RFs were evaluated using solar irradiance data as input. In the tests conducted, we evaluated how the frequency of the input data influences the prediction quality. Comparative tests of the results obtained using data every 15, 30, and 60 minutes indicate that there is no difference in the use of the different frequencies evaluated. Regarding the different methods and parameters evaluated, the results obtained after 30 rounds of each method demonstrate that the implemented methods can be used for solar energy generation prediction, with special emphasis on the ANN with relu activation function and 10 neurons which shows the best metrics.