一种基于原型引导的流匹配的多模态时序预测方法
By employing a prototype-guided flow matching method and utilizing a multi-head self-attention mechanism and flow matching network, the problems of multimodal data fusion and deterministic prediction are solved, achieving both accuracy and probabilistic prediction in weather forecasts. This method is applicable to weather forecasts in data-scarce regions and for new forecast elements.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA NORMAL UNIV
- Filing Date
- 2025-09-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing meteorological forecasting methods are insufficient in terms of multimodal data fusion and deterministic forecasting capabilities, making it difficult to achieve accurate and probabilistic forecasts in data-scarce regions or with new forecast elements.
A prototype-guided flow matching method is adopted, which extracts the temporal representation of multimodal information through a multi-head self-attention mechanism and generates future temporal prediction results, while calculating the probability distribution of the prediction results.
It improves the accuracy and probabilistic prediction capabilities of weather forecasts, enabling rapid adaptation to the forecasting needs of new regions and new forecast elements under zero-sample or low-sample conditions.
Smart Images
Figure CN121301951B_ABST