Satellite Internet-Based High-Precision Space-Ground Coordinated Ultra-Short-Term Photovoltaic Power Prediction

The high-precision ultra-short-term photovoltaic power prediction is the key guarantee for efficient solar energy utilization. However, various factors significantly challenge the accurate photovoltaic (PV) ultra-short-term power prediction, including the remote location of PV power plant, high communication cost and significant delay. In this work, a satellite internet-based ultra-short-term power prediction (SUPP) framework is proposed that data from space and ground. The core algorithm of SUPP is established with multilayer heterogeneous neural network consisting of convolutional neural network (CNN), gated recurrent unit (GRU) and attention mechanism to utilize the technical advantages of high bandwidth, wide coverage and low latency of satellite internet. The proposed SUPP framework can realize real-time transmission and fusion of local cloud map information collected by ground-to-air cameras and wide-area meteorological data collected by remote sensing satellites. Simulation results show that the proposed SUPP framework can provide a reliable decision basis for system-level scheduling and control by accurately predicting the trend and value of PV power generation with the efficient fusion of multivariate data.