ANN_ITU: Predicting rain attenuation with a hybrid model for earth-space links

Abstract

Rain attenuation prediction of earth-space links is of vital significance for the application and development of satellite communication. Recently, most rain attenuation prediction methods are based on semi-empirical models or data-driven models, the former suffering from incompleteness problem, the latter faced with limited performance due to scarce data. In order to realize higher rain attenuation prediction performance, we propose a novel hybrid model ANN_ITU that combines advantages of the semi-empirical model and the artificial neural network. In ANN_ITU framework, the semi-empirical model ITU-R P.618-12 is leveraged to predict rain attenuation, and a six-layer artificial neural network is utilized to correct the rain attenuation predicted by ITU-R P.618-12, thus generating the final rain attenuation value. What’s more, we also present theories of two machine-learning based rain attenuation prediction methods, namely, random forest and support vector regression. Last but not least, we expound on processes of DBSG3 dataset filtering and data preprocessing. Experiments on DBSG3 dataset are carried out. Experimental results demonstrate that the hybrid ANN_ITU algorithm outperforms purely semi-empirical algorithms and data-driven algorithms. The evaluation indexes mean value, standard deviation, and root mean square value are 0.0355%, 19.63%, and 19.63%, respectively, which prove the effectiveness and precision of our rain attenuation prediction model ANN_ITU.