Gypsum crystallization inhibitor screening method and system fusing first principle and depth map network

Abstract

This invention discloses a method and system for screening gypsum crystallizers that integrates first-principles calculations and deep graph networks. The method includes: first, obtaining a benchmark dataset containing the adsorption structure and adsorption energy of crystallizer molecules on gypsum crystal surfaces through first-principles calculations; second, converting the structural information in the dataset into gypsum crystal surface diagrams and crystallizer molecule diagrams; then, training a dual-tower graph neural network model containing gypsum crystal surface towers and crystallizer molecule towers to obtain a screening model that can accurately predict adsorption energy; finally, using this model to perform high-throughput virtual screening on a large-scale candidate molecule library and outputting high-priority candidate crystallizers. This invention, by constructing a dedicated dual-tower neural network architecture and a cross-tower interaction mechanism, replaces time-consuming first-principles calculations with a high-precision surrogate model, achieving second-level, high-precision screening of gypsum crystallizers, and solving the technical problems of low screening efficiency, high cost, and long cycle in existing technologies.

Images