Comparação entre as redes LVQ e MLP na previsão da atenuação provocada pela chuva

Abstract

Predicting the rate of rain and attenuation caused by it plays an essential role in the areas of communications. In other words, predicting attenuation due to rainfall is not a simple task to be solved mathematically, because the variables that cause the phenomenon are often nonlinear and have complicated behavior. This project is expected to predict rain attenuation in Brazil using the Artificial Neural Network (RNA). Ten-year rainfall data were collected from measurements performed in different geographical locations. The local covered all five states, representing all geographic areas of Brazil. RNA was trained to predict rain attenuation in these locations using annual rainfall data from 2012 to 2021. Two types of network architecture were used: Multilayer Perceptron Network (MLP) and Vector Quantization by Learning Network (LVQ), both with the same objective. The results of rain attenuation provided by RNA were compared with the results provided by the International Telecommunications Union (ITU) model, which is a well-established model. In terms of performance, they show that the predicted attenuation of RNA is in line with the forecast of the ITU model. On the other hand, the resulting RNA training is a useful tool for communication engineers to predict the attenuation of the rain of subsequent years and proactively solve the problem of signal attenuation inherent to the operation of the satellite path above 10 GHz.