부산대학교 도서관

Typhoons are exceptionally destructive tropical cyclones that possess the capacity to inflict significant harm upon society. For institutions engaged in risk assessment and disaster mitigation, accurately forecasting typhoon paths and trajectories is of paramount importance. In this study, we propose a Typhoon Track Image Prediction model based on a Comprehensive Long Short-Term Memory (LSTM) Attention Model in the field of Deep Learning. This model significantly improves the accuracy of typhoon track prediction. To assess its performance, we utilize typhoon data from the China Typhoon Network spanning from 1949 to 2022. The Comprehensive LSTM Attention Model is compared against other popular models such as Temporal Convolutional Networks (TCN), traditional LSTM, and Seq2Seq models for forecasting the typhoon track over a 12-hour period. The results demonstrate the superiority of our proposed model, showcasing its effectiveness in terms of superior accuracy in typhoon track prediction.