Thunderstorm gale prediction method, system and device based on centralized sampling and multi-branch fusion

Abstract

The application discloses a thunderstorm gale prediction method and system based on centralized sampling and multi-branch fusion, and equipment, relates to the meteorological prediction technical field, and includes: adopting centralized sampling to intercept the radar echo sequence of a prediction area; performing multi-branch feature extraction on the radar echo sequence to obtain double-dimension splicing features, and performing branch feature interaction on the double-dimension splicing features to obtain global fusion features; extracting physical constraint features of the prediction area, and constructing physical-global fusion features based on the physical constraint features and the global fusion features; performing coordinate regression and logistic regression calculation based on the physical-global fusion features to obtain the thunderstorm gale occurrence probability of the thunderstorm gale drop point in a short term. The application is used to solve the problem that the thunderstorm gale drop area and the thunderstorm gale occurrence probability in a short term cannot be accurately predicted.