A Comparative Study of Deep Learning Methods for Short-Term Solar Radiation Forecasting

Energy and climate challenges have led to significant growth in solar power generation, and because of this, smart grids are increasingly utilizing solar power. The solar power is intermittent because solar energy is mainly dependent on radiation and other atmospheric factors. To ensure the reliable economic operation of micro grids and smart grids, precise forecasting of solar radiation is crucial. Deep learning techniques are proposed in this study to model the solar radiation production. Four artificial intelligence and deep learning-based forecasting models, the fully connected artificial neural network (ANN), the convolutional neural network (CNN), the long short-term memory network (LSTM), and the bidirectional neural network (Bi-LSTM) were examined for this study. To compare the prediction accuracy of all models, three performance evaluation metrics RSME, MAE, and R2 are used. The results obtained indicate that all four methods produce reasonable estimates of solar radiation generation. As a result of RMSE, MSE, and R2, the Bi-LSTM forecasting model offers the best estimate of forecasting accuracy.