Enhancement of Rainfall Prediction Model using Convolutional Neural Network Optimized by Chameleon Particle Swarm Optimization for Satellite Communication Application

Abstract

Accurate rainfall prediction is critical for dependable wireless and satellite communications, especially in tropical areas where rain attenuation significantly degrades the reliability of these systems. While Convolutional Neural Networks (CNN) have proven effective in rainfall modelling due to rainfall's non-linear characteristics, achieving optimal performance requires careful tuning of CNN's hyperparameters. This research introduces a Chameleon Particle Swarm Optimization (CNN-CPSO) aimed at the automatic optimization of CNN hyperparameters with improved rainfall prediction capability. The simulation was performed using monthly rainfall data from Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) obtained from the National Aeronautics and Space Administration (NASA), 1980-2020. Nigerian Meteorological Agency (NiMet) observation data were used for model evaluation. The rainfall data was pre-processed and normalized, and then partitioned into training, validation, and testing sets. The CNN model was implemented in the deep learning paradigm, while the Chameleon Particle Swarm Optimization (CPSO) technique was used to optimize some of the CNN’s key hyperparameters, including learning rate, batch size, dropout rate, number of convolutional layers, and filters. The performance of the developed model was determined using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination ().

Compared to the conventional CNN and CNN optimized with the standard Particle Swarm Optimization (CNN-PSO), the proposed CNN-CPSO showed improved prediction performance with a lower MSE of 1.12, RMSE of 1.06, and MAE of 0.89, an  of 0.94, and a prediction accuracy of 93.30%. This compared to 87.90% for CNN-PSO and 84.60% for the conventional CNN, respectively. Therefore, the CNN-CPSO is recommended for use in weather forecasting, rain fade prediction and mitigation in satellite communications, hydrological modeling, and climate change studies.