Typhoon Track Prediction Based on Dual Attention LSTM Model

The accurate forecasting of typhoon paths and trajectories is of paramount importance to institutions engaged in risk assessment and disaster mitigation. The emergence of artificial intelligence has triggered the widespread implementation of deep neural network technologies in meteorology. This paper presents an innovative typhoon trajectory prediction model that incorporates Dual Attention LSTM mechanism. This model employs a mixed-mode approach to extract all relevant features from typhoon trajectory data, aiming to enhance the accuracy of typhoon trajectory predictions. To achieve this objective, the optimal typhoon track dataset from the Shanghai Typhoon Research Institute of China Meteorological Administration has been employed, where a comparative analysis of the Dual Attention LSTM model with LSTM, LSTM-3FC, TCN, Seq2Seq(LSTM), and Seq2Seq-lstm-attention models has been conducted. The results of our study imply that our proposed Dual Attention LSTM model shows higher accuracy in predicting typhoon trajectories than other models presented in this study. This model builds upon the foundation of the LSTM typhoon prediction model.