Ultra-short-term Irradiation Prediction Based on Ground-based Cloud Images and Deep Learning

Photovoltaic (PV) power generation has been widely used due to its advantages of green and clean, easy installation. However, since its output is mainly determined by irradiation, the intermittency, randomness, and instability of irradiation make PV large-scale grid-connectedness a lousy impact on the safety and economic operation of the power grid. Therefore, irradiation prediction can suppress randomness and instability, indirectly improving the quality of PV power generation. In this paper, firstly, image segmentation processing and feature extraction are carried out on the ground-based cloud images to digitize the images in low dimensions, where a new adaptive threshold image segmentation method based on RGB-Grey-OTSU is proposed to make the cloud images three-valued, separating the sun, clouds, and sky, and comparing it with the traditional threshold segmentation method. Feature extraction is carried out on the processed cloud images. In this paper, three feature values of weather, the percentage of sun, and cloud are extracted, and the strong correlation between the extracted feature values and irradiation is verified. This paper adopts and compares the two deep-learning prediction models, LSTM and GRU, and their performance of ultrashort-term irradiation prediction on different prediction time scales from 5 minutes to 1 hour.