Short-Term Solar Irradiance Forecasting from Future Sky Images Generation

Hoang Chuong Nguyen, Miaomiao Liu

Abstract: Solar irradiance prediction is critical for the integration of the solar power to the existing power system. A recent trend in the literature is to adopt deep learning-based methods to predict future solar irradiance from sky images. While these models have achieved significant improvements, they are only capable of making predictions for a fixed forecasting horizon due to their non-recurrent nature. To this end, we propose a deep learning network that is capable of predicting solar irradiance in an autoregressive manner, which allows predictions across a long time horizon. Particularly, we reduce the problem to first generating future sky images which are then used to predict future solar irradiance. We evaluate our models on TSI880 and ASI16 datasets, and show that our model achieves superior performance compared to previous works for 4-h ahead-of-time predictions. Furthermore, we also demonstrate that the solar irradiance forecast of our model is not limited to only 4 h, but can be extended for even longer horizon.