Artificial Intelligence Design Method for Energetic Molecules

By constructing a performance-based molecular generation model and adaptive cyclic training, the problems of scarce sample size and low diversity in energetic molecule design were solved, and efficient generation and screening of high-energy density candidates were achieved, thereby improving molecular design efficiency and screening effects.

CN117079744BActive Publication Date: 2025-09-19INST OF CHEM MATERIAL CHINA ACADEMY OF ENG PHYSICS
2 Cites -1 Cited by

Patent Information

Application Number
CN202310984813.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-08-07
Publication Date
2025-09-19
Estimated Expiration
2043-08-07

AI Technical Summary

Technical Problem

To quickly obtain effective energetic molecular candidates in chemical space, existing technologies face the problems of scarce sample size and low diversity, resulting in inefficient molecular design.

Method used

An energetic molecule artificial intelligence design method is adopted to construct a performance-inverse molecular generation model, establish a performance prediction model through learning or fitting, combine the molecular generation model and the property prediction model, conduct adaptive cyclic training, and generate and screen potential high energy density candidates.

Benefits of technology

It improves the efficiency of molecular design and the screening hit rate, reduces computing resources and costs, and generates molecular samples with excellent comprehensive properties.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117079744B_ABST
    Figure CN117079744B_ABST
Patent Text Reader

Abstract

The present invention discloses an artificial intelligence design method for energetic molecules. The method designs energetic molecular structures by combining a molecular generative model with a property prediction model. The method comprises three elements: a molecular generative model (G) for creating molecular structures, a property prediction model (D) for predicting performance, and an adaptive molecular design and screening cycle (L). G can reversely generate molecular structures according to performance intervals, and a computer can autonomously generate a specified number of molecules to form a virtual screening molecule library. D uses a molecular graph as the only input to predict the performance of the virtual screening molecules and screen out energetic molecule candidates (C). C is used to expand the molecular structure data set and retrain G to make it easier to generate molecules with excellent performance. L is composed of the application process of G-D-C-G...D-C, which can simultaneously realize generative model optimization and target energetic molecule screening. The present invention realizes the rapid discovery of potential new energetic molecules with limited resources.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Computer aided design system for predicting energetic molecule based on machine learning performance

    CN110728047A

  • Eutectic prediction method based on graph neural network and deep learning framework

    CN111882044A