Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toolbox for Developing Physics Informed Neural Networks for Power Systems Components

Ioannis Karampinis, Petros Ellinas, Ignasi Ventura Nadal, Rahul Nellikkath, Spyros Chatzivasileiadis

TL;DR

Power systems with increasing renewable penetration exhibit low inertia and complex non-linear dynamics that challenge traditional solvers. The authors present an open-source, Python-based toolbox to develop physics-informed neural network (PINN) models of power-system components, enabling fast and scalable simulations that can be integrated with conventional simulators. The framework automates ODE setup, dataset generation (including Latin Hypercube Sampling) and collocation-based physics training, and is demonstrated on a 9th-order synchronous-machine model with an Automatic Voltage Regulator and Governor, achieving approximately $2.3\times 10^{-3}$ MAE with substantial inference-time gains. This work advances reproducible PINN-based component libraries for power-system dynamics and points to future extensions such as adaptive sampling, broader component coverage, and integration into commercial tools.

Abstract

This paper puts forward the vision of creating a library of neural-network-based models for power system simulations. Traditional numerical solvers struggle with the growing complexity of modern power systems, necessitating faster and more scalable alternatives. Physics-Informed Neural Networks (PINNs) offer promise to solve fast the ordinary differential equations (ODEs) governing power system dynamics. This is vital for the reliability, cost optimization, and real-time decision-making in the electricity grid. Despite their potential, standardized frameworks to train PINNs remain scarce. This poses a barrier for the broader adoption and reproducibility of PINNs; it also does not allow the streamlined creation of a PINN-based model library. This paper addresses these gaps. It introduces a Python-based toolbox for developing PINNs tailored to power system components, available on GitHub https://github. com/radiakos/PowerPINN. Using this framework, we capture the dynamic characteristics of a 9th-order system, which is probably the most complex power system component trained with a PINN to date, demonstrating the toolbox capabilities, limitations, and potential improvements. The toolbox is open and free to use by anyone interested in creating PINN-based models for power system components.

Toolbox for Developing Physics Informed Neural Networks for Power Systems Components

TL;DR

Power systems with increasing renewable penetration exhibit low inertia and complex non-linear dynamics that challenge traditional solvers. The authors present an open-source, Python-based toolbox to develop physics-informed neural network (PINN) models of power-system components, enabling fast and scalable simulations that can be integrated with conventional simulators. The framework automates ODE setup, dataset generation (including Latin Hypercube Sampling) and collocation-based physics training, and is demonstrated on a 9th-order synchronous-machine model with an Automatic Voltage Regulator and Governor, achieving approximately MAE with substantial inference-time gains. This work advances reproducible PINN-based component libraries for power-system dynamics and points to future extensions such as adaptive sampling, broader component coverage, and integration into commercial tools.

Abstract

This paper puts forward the vision of creating a library of neural-network-based models for power system simulations. Traditional numerical solvers struggle with the growing complexity of modern power systems, necessitating faster and more scalable alternatives. Physics-Informed Neural Networks (PINNs) offer promise to solve fast the ordinary differential equations (ODEs) governing power system dynamics. This is vital for the reliability, cost optimization, and real-time decision-making in the electricity grid. Despite their potential, standardized frameworks to train PINNs remain scarce. This poses a barrier for the broader adoption and reproducibility of PINNs; it also does not allow the streamlined creation of a PINN-based model library. This paper addresses these gaps. It introduces a Python-based toolbox for developing PINNs tailored to power system components, available on GitHub https://github. com/radiakos/PowerPINN. Using this framework, we capture the dynamic characteristics of a 9th-order system, which is probably the most complex power system component trained with a PINN to date, demonstrating the toolbox capabilities, limitations, and potential improvements. The toolbox is open and free to use by anyone interested in creating PINN-based models for power system components.

Paper Structure

This paper contains 21 sections, 12 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Physics-Informed Neural Networks
  3. Classical Ordinary differential equations solvers
  4. Neural Networks as function approximators
  5. Training Neural Networks with data
  6. Mathematical foundation of Physics Informed Machine Learning
  7. Training PINNs - Challenges
  8. Setting up the ODEs
  9. Generating the Dataset
  10. Generating the trajectories
  11. Preprocessing the Dataset
  12. The architecture of Neural Network
  13. Training challenges of the Neural Networks
  14. Performance assessment
  15. Experimental evaluation - Demonstration
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Architecture of the proposed PINN that approximates the state $\mathbf{x}(t)$ at time $t = t_0 + h$ based on the initial state $\mathbf{x}_0$ and the time t.
  • Figure 2: Metric scores of a PINN approximating the states of a 9th-order SM. The error metrics are benchmarked against the ODE solutions.
  • Figure 3: Results for SM variables from the ODE solver and the trained PINN approximation for 3 different sets of initial conditions