Typhoon Intensity Prediction Based On Deep Learning

A typhoon is a devastating and complex atmospheric motion system.Currently,typhoon intensity prediction is in the stage of exploration and research in the world.The demand for automation degree of the typhoon monitoring system has been increased,for changing weather enterprise and services modernisation.Over 40 years of development,forecasters are gradually applying artificial intelligence to typhoon intensity prediction.Convolutional Neural Networks(CNN)based on deep learning have become a subject undergoing intense study in recent years.This technology of objects classification has become mature gradually.However,there were some defects in the actual operation—such as high subjectivity,low accuracy,and weak promptness.Particularly in applying to typhoon intensity prediction,this technology needs to be enhanced.According to the results of various studies,this thesis analysed and compared the characteristics of Convolutional Neural Networks(CNN),Long Short-Term Memory(LSTM)and e Xtreme Gradient Boosting(XGBoost)models in typhoon intensity prediction,as well as the strengths and weaknesses in the application.Thus,it was proposed that the optimal method for typhoon intensity prediction with the XGBoost model.Using the meteorological data from 2009 to 2019 transmitted by FY-2 Satellite for experiments,relevant parameters for the optimisation model have been determined.Several comparative analysis experiments were designed and conducted among the optimisation model,the original model,CNN model and LSTM model to verify the feasibility and superiority of the optimisation model.Other than the training data sets,some typical typhoon cases were chosen for predictive experiments.The experimental results showed that applying the optimisation model in typhoon intensity prediction was valid,with higher accuracy and better stability in the short-term typhoon intensity prediction.The proposed application of the xgboost optimisation model in typhoon intensity prediction has extended the methods for typhoon intensity prediction and provided a new thought for applying and developing deep learning related techniques in the meteorological field.