Adoption of Deep Neural Network Model for the Prediction of Rain Attenuation

Abstract

Weather, rain, and atmospheric forces cause attenuation in communication satellites and devices, which affects the quality of calls and internet access. Communication signals can seriously be affected by rainfall and various research has entailed developing adequate rainfall prediction models that can accurately predict rain attenuation. Predicting rainfall attenuation from measured data is a challenging process and several rainfall attenuation models exist, but their estimates are not close to actual measured rainfall attenuation data. This paper predicts rain attenuation using a Deep Neural Network (DNN) and modelling process to propose a probabilistic model that gives better results. The proposed model was built using information from the International Telecommunications Union Radiocommunication for propagation statistics and prediction techniques needed for the construction of Earth-space telecommunication systems. Rainfall data spanning a period of seven years (January 2014 to December 2020) in Oyo State were collected from Nigerian Meteorological Agency (NIMET). Results showed that the proposed model was able to predict well even with new data (2021) and this was used to show how well the DNN model can perform using rainfall features and elimination complex mathematical models to describe the mappings between rainfall attenuation and rainfall data. This is also evident from the performance graph which shows an RMSE of 0.21 and a Loss of 0.021 at the 100 epochs. The results also reveal that the proposed model results observed against predicted, performed relatively well. From the performance metrics, it shows a value of 0.7172 for Pearson’s linear correlation, showing a strong similarity of predicted versus observed. The mode efficiency index nt on the other hand, shows a value of 0.1744 and gives near acceptable results. In conclusion, the research showed that the Deep Neural Network model was able to predict well compared to the ITU-R model.