Cloud Change Trend Prediction System Based On Deep Learning

Satellite-to-ground laser communication is the main method to realize high-speed backbone link between space-based resources and ground resources.It has the advantages of high speed,wide bandwidth,anti-interference and high confidentiality,which also makes laser gradually become the mainstream transmission medium for satellite-to-ground communication in the future.However,due to the application scenarios of satellite-to-ground communication,laser links need to pass through the atmosphere,and are easily affected by the absorption and scattering of clouds in the atmosphere,background light interference,atmospheric turbulence and other factors.which posing a greater challenge for the application.In order to ensure the reliability of the communication link,this paper focuses on the cloud cluster change trend prediction system based on deep learning.This paper proposes a cloud change trend prediction system(Cloud Net1.0)based on deep learning.First,the cloud dataset above the ground station is used to classify the cloud image according to the cloud amount and cloud shape standard.According to the results of the classification,a cloud prediction model based on multi-layer stacking of deep prediction networks is proposed on the basis of the time series prediction model,which introduces multiple sizes of convolution kernels into the convolution calculation.The experimental results show that When the prediction duration is 100 s and the error of 5%is allowed,the prediction accuracy rate reaches 81%.Then,in order to verify the influence of meteorological factors on the prediction of cloud change,we select four types of meteorological data that affect the generation of cloud and propose a multi-input cloud prediction model called Cloud Net2.0 whose input including meteorological data.The experimental results show that the multi-input model with meteorological data is better than the original model.Although it is not as good as Cloud Net1.0,the prediction accuracy rate can reach 79%.The reasons for the occurrence were further analyzed in view of the experimental results that did not meet the expectations.In summary,this article focuses on the problem of link reliability in satellite-toground laser communication,proposes a cloud change trend prediction system based on deep learning to avoid complex physical modeling and theoretical derivation processes.The system is able to overcome the problem of laser link interruption caused by the cloud by knowing the changing trend of the cloud in advance and scheduling the link,which is of great significance for improving the reliability of the satellite-toground high-speed backbone link.