Forecasting Solar Radiation in Diverse Climates in Iran Utilizing Deep Learning Methods

Abstract

Fossil fuel consumption not only leads to climate change and global warming but also destroys the environment with causing drought, heavy rains, and storms, created by increasing emissions of greenhouse gases. Hence, producing energy from renewable resources is more necessary than ever. Although Iran has a high potential for producing energy from renewable sources, the share of these energies in electricity production is only 1.1 percent. Therefore, solar energy can be utilized with a practical strategy as an alternative source for transitioning from fossil fuels and achieving sustainable energy production. This study analyzes deep learning models for predicting solar radiation under different meteorological conditions in Iran, emphasizing the necessity for accurate renewable energy forecasts due to rising fossil fuel use and environmental concerns.  This research evaluates the performance of Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and hybrid models with real solar radiation data. The data reveal that the CNN-GRU model attained a substantial R² value of 0.895, while the CNN-LSTM model exhibited the highest prediction accuracy with the lowest RMSE. These findings emphasize the tremendous potential of hybrid architectures to improve solar energy forecasting, thereby easing Iran's transition to renewable energy.  Moreover, identified research gaps underscore the imperative of amalgamating various climatic data sources, including satellite imagery and localized meteorological observations. Future research should concentrate on improving forecasting methods by investigating hybrid modelling methodologies and integrating machine learning algorithms to augment predictive precision in solar radiation forecasting.