An optimal model using single-dimensional CAE-IRNN based SPOA for cyclone track prediction

Abstract

Tropical cyclones have a devastating impact on both human life and the natural environment, and accurate prediction of their tracks is critical for minimizing their negative effects. Traditional approaches to track tropical cyclones have several limitations, including low prediction accuracy and efficiency. With the accumulation of vast amounts of investigating and meteorological data, deep learning techniques have become increasingly popular for addressing the temporal and spatial features of tropical cyclones. This article proposes a three-step approach for predicting the tracks of tropical cyclones. The first step involves preprocessing various meteorological factors and cyclone tracks. The second step involves feature selection to minimize input variables and eliminate redundant data, enhancing efficiency. In the prediction process, the chosen meteorological factors and cyclone tracks are fed into a single-dimensional convolution neural network autoencoder, along with the independent recurrent neural network-based Student Psychology Optimization Algorithm and Harris Hawk optimization algorithm. The proposed method, called single-dimensional CAE-IRNN based SPH2, is designed to predict tropical cyclones accurately and efficiently. The technique is then compared with other existing methods to determine its efficiency. The comparison performance is implemented to evaluate the effectiveness of the proposed approach. Overall, this article highlights the importance of deep learning approaches for predicting tropical cyclones' tracks accurately and efficiently. The proposed three-step approach and the single-dimensional CAE-IRNN based SPH2 method contribute to enhancing the accuracy and efficiency of tropical cyclone prediction, which can help reduce their negative impact on human life and the natural environment.