Implementation of Artificial Intelligence Methods for Solar Energy Prediction

Abstract

In order to satisfy the growing world energy demand and decreasing the emission of greenhouse gases the requirement for more green energy has led to an increased focus on research related to forecasting solar energy recently. In this study we aim to develop forecast models, based on Artificial Neural Network and Random Forrest algorithms to predict daily solar energy based on daily historical meteorological data measured between 2019 and 2021. The accuracy and the performance of each model are compared using mean squared error, mean absolute percentage error, mean absolute error, max error and R-squared for evaluation. The prediction of daily solar energy from the daily maximum, minimum and average values of the metrological variables using Artificial Neural Network and Random Forest was carried out. The results obtained indicate that both models can predict daily solar energy with good accuracy (MAPE = 13%). On the one hand, the RF model showed excellent accuracy during the training phase (MAPE = 8%, R2 = 0.97), but it failed to show same results during the testing phase (MAPE = 13%, R2 = 0.79). On the other hand, the ANN was able to maintain the same results during training and testing (MAPE = 13%, R2 = 0.81).