Source item inversion method and system based on improved Transform and storage medium

By improving the source term inversion method constructed by Transformer and utilizing Gram corner field transformation and self-attention quantum network, the problems of high computational cost and insufficient accuracy of source term inversion in nuclear accidents are solved, and rapid and accurate inversion of multi-component nuclide release rates is achieved, thereby improving emergency response capabilities.

CN121766068APending Publication Date: 2026-03-31CHINA INST FOR RADIATION PROTECTION
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-12
Publication Date
2026-03-31

AI Technical Summary

Technical Problem

Existing source term inversion methods suffer from high computational costs, insufficient model accuracy, poor robustness, and limited ability to invert multi-component nuclides in nuclear accident emergency response, making it difficult to meet the needs for rapid and accurate emergency response.

Method used

A source term inversion method based on an improved Transformer is adopted. By constructing a Gram angle field transformation and a self-attention subnetwork, and combining meteorological data for feature fusion, a source term inversion model is constructed, including shallow feature extraction, a self-attention subnetwork and a feature fusion subnetwork. The Swin Transformer module is used to enhance feature representation and fusion capabilities.

Benefits of technology

It achieves high-precision and robust source term inversion, can quickly respond to nuclear accidents, effectively integrate multi-source data, capture complex time-series features, and improve the accuracy of nuclide release rate inversion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121766068A_ABST
    Figure CN121766068A_ABST
Patent Text Reader

Abstract

The invention discloses an improved Transform-based source item inversion method and system and a storage medium, and the method comprises the steps: obtaining a training data set which comprises a plurality of groups of training samples, and each group of training samples comprises source item data, gamma dose rate time sequence data and corresponding meteorological data; performing Gamma angle field conversion on the gamma dose rate time sequence data, and performing normalization processing on the meteorological data; constructing a source item inversion model, the input data of the source item inversion model being gamma dose rate time sequence data and corresponding meteorological data, and the output data being source item data obtained by inversion; processing the dose rate shallow layer features through a self-attention sub-network to obtain dose rate deep layer features; fusing the dose rate deep layer features and the meteorological shallow layer features through a feature fusion sub-network, and inputting the fused features into a second multi-layer perceptron for training to obtain a source item inversion model; and carrying out source item inversion by adopting the source item inversion model.
Need to check novelty before this filing date? Find Prior Art