Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On Physics-Informed Neural Network Control for Power Electronics

Peifeng Hui, Chenggang Cui, Pengfeng Lin, Amer M. Y. M. Ghias, Xitong Niu, Chuanlin Zhang

Abstract

Considering the growing necessity for precise modeling of power electronics amidst operational and environmental uncertainties, this paper introduces an innovative methodology that ingeniously combines model-driven and data-driven approaches to enhance the stability of power electronics interacting with grid-forming microgrids. By employing the physics-informed neural network (PINN) as a foundation, this strategy merges robust data-fitting capabilities with fundamental physical principles, thereby constructing an accurate system model. By this means, it significantly enhances the ability to understand and replicate the dynamics of power electronics systems under complex working conditions. Moreover, by incorporating advanced learning-based control methods, the proposed method is enabled to make precise predictions and implement the satisfactory control laws even under serious uncertain conditions. Experimental validation demonstrates the effectiveness and robustness of the proposed approach, highlighting its substantial potential in addressing prevalent uncertainties in controlling modern power electronics systems.

On Physics-Informed Neural Network Control for Power Electronics

Abstract

Considering the growing necessity for precise modeling of power electronics amidst operational and environmental uncertainties, this paper introduces an innovative methodology that ingeniously combines model-driven and data-driven approaches to enhance the stability of power electronics interacting with grid-forming microgrids. By employing the physics-informed neural network (PINN) as a foundation, this strategy merges robust data-fitting capabilities with fundamental physical principles, thereby constructing an accurate system model. By this means, it significantly enhances the ability to understand and replicate the dynamics of power electronics systems under complex working conditions. Moreover, by incorporating advanced learning-based control methods, the proposed method is enabled to make precise predictions and implement the satisfactory control laws even under serious uncertain conditions. Experimental validation demonstrates the effectiveness and robustness of the proposed approach, highlighting its substantial potential in addressing prevalent uncertainties in controlling modern power electronics systems.
Paper Structure (28 sections, 21 equations, 8 figures, 3 tables)

This paper contains 28 sections, 21 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. PRELIMINARIES
  3. Power Electronic Systems with Uncertainties
  4. Physics-Informed Machine Learning
  5. Constructing Hybrid-Physics-Data Models
  6. Hybrid Physics and Data-driven Loss Function
  7. PINN Controller Design
  8. Framework Design
  9. Data-driven Module
  10. Physics Module
  11. Control Module
  12. Loss Function Module
  13. Physics Module Design
  14. Control Module Design
  15. Data-driven Module Design
  16. ...and 13 more sections

Figures (8)

  • Figure 1: The process of fitting a black-box converter model using PINN
  • Figure 2: Structure of the proposed hybrid intelligent control framework and its training process.
  • Figure 3: Validation of the proposed control method through adaptability testing and comparison with alternative methods.
  • Figure 4: Experimental setup.
  • Figure 5: The experiment implementation process of the proposed method.
  • ...and 3 more figures