Tropical Cyclone Intensity Prediction using Bayesian Machine Learning with Marine Predators Algorithm on Satellite Cloud Imagery

作     者：Ragab, Mahmoud 

作者机构：King Abdulaziz Univ Fac Comp & Informat Technol Informat Technol Dept Jeddah 21589 Saudi Arabia Al Azhar Univ Fac Sci Dept Math Cairo 11884 Egypt 

出 版 物：《AIN SHAMS ENGINEERING JOURNAL》 (Ain Shams Eng. J.)

年 卷 期：2025年第16卷第3期

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 

基　　金：Deanship of Scientific Research (DSR)   King Abdulaziz University  Jeddah  Saudi Arabia [D: 016-611-1434] 

主　　题：Tropical Cyclones Machine Learning Marine Predators Algorithm Bayesian Belief Network CapsNet 

摘      要：Due to its wide range of associated hazards, tropical cyclones (TC) become the costliest natural disaster worldwide. A correct diagnosis model for the TC intensity can save property and lives. Unfortunately, intensity forecasting of TC has been a bottleneck and has made it difficult to forecast weather. Several existing approaches and techniques make a diagnosis of TC wind speed through the satellite data at the specified time with varying success. Deep learning (DL)-based intensity forecasting has recently held great promise in surpassing conventional approaches. DL-based techniques have been developed in geosciences to replace traditional methods. However, weather forecasting is uncertain due to the Earth system s nonlinearity, complexity, and chaotic effects. Thus, this manuscript develops a new Bayesian Machine Learning with Marine Predators Algorithm for TC Intensity Prediction (BMLMPA-TCIP) approach. The major goal of the BMLMPA-TCIP model is to estimate the level of the TCs on satellite cloud images. To accomplish this, the BMLMPA-TCIP technique utilizes the Gaussian filtering (GF) approach to eradicate the noise in the cloud images. Additionally, the extraction of useful feature vectors is performed by using the capsule network (CapsNet) technique. Moreover, the MPA method accomplishes the hyperparameter tuning of the CapsNet method. Lastly, the BMLMPA-TCIP technique utilizes the Bayesian Belief Network (BBN) method to predict TC intensity. To authorize the performance of the BMLMPATCIP approach, a wide variety of experiments are performed under the TC image dataset. The experimental validation of the BMLMPA-TCIP approach illustrates a superior RMSE value of 5.89 over existing techniques.