Friendly Deep Learning Model for Estimating Daily Global Solar Radiation

Technologies have transformed our life: work, social life, economy and communicates, making with it a slew of new conveniences, efficiency, and connection. However, all of these wonderful advantages have resulted in an increase in energy usage. As a result, various attempts are being undertaken to make technologies more energy-efficient and to power these technical infrastructures using renewable energies such as : wind ,water and solar power. To effectively use solar energy and foresee its future, large investments in green energy research and development are required. In line with this goal, it is required to evaluate the potential of solar energy using a variety of models when forecasting and estimating it. We estimate global sun radiation on a horizontal surface in this article. These models' inputs are meteorological parameters and data. LSTM (Long and Short Term Memory Neural Networks),CNN (Convolutional Neural Networks), RNN (Recurrent Neural Networks) and MLP (Multi-Layer Perceptron) are the Deep Learning methods used in this work, and the impact of the parameter optimizer was investigated. The dataset for this study were obtained from the Helio-Clim1 site and span a ten-year period (1996 to 2005). Among the performance requirements for solar radiation forecasting models are forecast accuracy, model resilience, predictor uncertainty, and predictor dependability. Finally, in terms of error values, the CNN model outperforms the others (MAE=2,00, MSE=7,66, RMSE = 2,67, and R2 = 0,86).