Attention mechanism fusion neural network for typhoon path prediction

Timely and accurately predicting typhoon movement paths is essential to prevent typhoon disasters and reduce property losses. However, the existing typhoon path prediction models have weak spatiotemporal correlation feature extraction capabilities and insufficient consideration of the importance of different dimensional features, resulting in low prediction accuracy. In this paper, we propose an attention mechanism-based fusion neural network prediction model for typhoon paths to address this issue. This model first uses a Temporal Convolutional Network (TCN) with residual units (R-TCN) to extract time-dependent features from typhoon trajectory data. Then, we propose a fusion self-attention-based ResNeXt-50 model (AT-ResNeXt-50) to obtain position attention between atmospheric data image channels, thus having better spatiotemporal feature extraction capabilities. Finally, we use attention aggregation to automatically weigh and fuse the two data features, achieving high-precision typhoon path prediction based on multidimensional feature data. The experimental results show that our model has higher accuracy and is significantly superior to the several existing data-driven typhoon path prediction methods.