Tropical cyclone path forecasting method improved by attention mechanism

The invention discloses a tropical cyclone path forecasting method improved by using an attention mechanism, and relates to the technical field of marine weather forecasting, and the method comprises the steps: fusing the two-dimensional features of a tropical cyclone and the three-dimensional features of the tropical cyclone and the periphery of the tropical cyclone through a Deepframe, and constructing an end-to-end model; for the Cross component, extracting two-dimensional features of the tropical cyclone through a feature crossing method; for the Deep component, a Multi-TrajGRU method based on a convolution attention module CBAM is used for extracting three-dimensional features of the tropical cyclone and the periphery of the tropical cyclone, the convolution attention module CBAM can conduct modeling on a three-dimensional structure of the tropical cyclone, the problems that the tropical cyclone is affected by an isobaric surface and the three-dimensional structure is large in size are solved, and theref