Harnessing Weather-Informed Data Analytics and Machine Learning for Performance Optimisation in VSAT Communication Networks

Singapore’s tropical and humid climate is a major factor to consider when evaluating the efficiency of satellite communication systems. Under heavy downpour, the Radio Frequency (RF) waves that satellites use undergo heavy attenuation and degrade the signal, an effect known as rain attenuation. This decreases the signal to noise ratio (SNR) until the receiver is no longer able to detect the RF wave, the link is lost and thus leading to unexpected service outages during extreme weather events. Current fade mitigation techniques are overly conservative or insufficiently robust, increasing electrical consumption and operating cost. This investigation utilises open-source weather data to create an Extreme Gradient Boosting (XGBoost) model that will be used to predict the intensity of rain attenuation on the Kacific-1 satellite link in the Ka band (26.5-40.0 GHz) and subsequently assess link reliability and evaluate the potential for proactive fade mitigation in Singapore’s tropical climate. Machine learning models (Random Forests, Extreme Gradient Boosting, Support Vector Machines, Long Short-Term Memory and Generalised Linear Models) are used to identify the algorithm that best fit the nature of weather data. Extreme Gradient Boosting was the best performing model and was further optimised to increase its accuracy. The model had a root mean squared error of 1.100 mm.