A linearized drift-diffusion equation solving method based on physical information neural network

By solving the linearized drift physics equations using a Physical Information Neural Network (PINN), the problems of low computational efficiency in traditional methods and data dependence in machine learning models are solved, thus achieving efficient and reliable tokamak plasma physics calculations.

CN122412733APending Publication Date: 2026-07-17ANHUI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ANHUI UNIV
Filing Date
2026-04-17
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Traditional numerical methods are computationally inefficient in solving the linearized drift physics equations in tokamak plasma physics, making it difficult to meet the requirements of real-time control and nonlinear integrated simulation. Furthermore, machine learning surrogate models rely on large amounts of data and lack physical constraints, resulting in unreliable results.

Method used

A Physical Information Neural Network (PINN) is used to quickly solve the linearized drift physical equations. The complex coefficients are processed by Fourier transform and decoupling. The total loss function is constructed and trained using MLP and Adam optimizer to ensure that the model output satisfies the physical equations and boundary conditions.

Benefits of technology

Achieving high physical fidelity and rapid computation under unlabeled data conditions significantly improves solution efficiency and result consistency, and is suitable for rapid computation in plasma physics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122412733A_ABST
    Figure CN122412733A_ABST
Patent Text Reader

Abstract

The application provides a linearized drift kinetic equation solving method based on a physical information neural network, and relates to the controllable nuclear fusion technical field.The method comprises the following steps: after obtaining plasma experimental data, linearized drift kinetic equations are constructed by calculating total collision frequency, drift frequency and anti-bounce frequency; based on the experimental data, each coefficient in the equation is calculated, the coefficients are taken as input features, and the coefficients are arranged into a data set for model training; a physical information neural network (PINN) model fusing the physical constraint of the linearized drift kinetic equation is constructed, and self-consistent solving of the equation is realized by minimizing a total loss function composed of physical equation residual loss and boundary condition loss.The application directly drives the physical information neural network learning through the physical equation, breaks through the dependence of the traditional data-driven method on labeled data while ensuring the solving accuracy, and ensures the physical consistency of the prediction result due to the physical constraint nature of the solution.
Need to check novelty before this filing date? Find Prior Art