A sea typhoon profile generating method based on a downward cast typhoon sounding data and deep learning
By using a method based on downdrop typhoon sounding data and deep learning, an implicit profile network is used to generate typhoon wind profiles at sea. This solves the problems of sparsity and complexity in the generation of typhoon profiles in existing technologies, and achieves high-precision typhoon profile reconstruction and completion, which is applicable to a variety of engineering application scenarios.
Patent Information
- Application Number
- CN202610865968.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to generate continuous wind profiles of typhoons covering different radii, quadrants, intensities, and altitudes under conditions of scarce observations. Furthermore, traditional methods are insufficient in characterizing complex typhoon structures, do not make full use of discrete radiosonde data, and have weak profile continuity and generalization capabilities.
We employ a method based on downdrop typhoon sounding data and deep learning. Through an implicit profile network architecture, combined with conditional input features and normalized height, we generate radial and tangential wind components. We then use a deep learning model to generate high-precision offshore typhoon wind profiles, including conditional branching, height branching, shared trunk, dual output heads, and differential anchoring layers, and train the model using multiple loss functions.
It achieves high-precision generation of vertical profiles of typhoons at sea under sparse observation conditions, solving the problems of insufficient characterization of complex typhoon structures and inadequate utilization of discrete radiosonde data. It has strong continuous representation capabilities and physical consistency, and is suitable for engineering applications in multiple scenarios.
Smart Images
Figure CN122413987A_ABST