Hybrid Solar Forecasting Method Using Satellite Visible Images and Modified Convolutional Neural Networks

This paper proposes a new hybrid method to predict global horizontal irradiance (GHI) at temporal horizons of 1, 2, 3 and 4 hours, combining the satellite visible images and meteorological information. First, the satellite visible images are preprocessed to remove the diurnal effects caused by the solar zenith angle. Then the cloud cover factors are extracted from satellite visible images by using the modified convolutional neural network (CNN). After that, the GHI forecasting model is developed which is based on the combined use of meteorological information and cloud cover factors. The sensitivity of the prediction accuracy to a variety of CNN structures with different widths, depths, and pooling methods is also explored in the paper. Meanwhile, a cloud motion forecasting method using predicted wind speeds is developed. The forecasting skills of the proposed method for different time horizons are demonstrated by comparing with several benchmark models.