Effective numerical simulations of synchronous generator system

Jiawei Zhang, Aiqing Zhu, Feng Ji, Chang Lin, Yifa Tang

TL;DR

This paper introduces a more robust way to simulate synchronous generators by reformulating them as port-Hamiltonian descriptor systems. By using structure-preserving Gauss methods that respect the system's underlying "Dirac structure," the researchers achieved significantly better long-term computational stability and accuracy compared to industry-standard software like PSCAD/EMTDC.

What Problem Does This Paper Address?

Simulating the electromechanical transients of power systems is notoriously difficult. Traditional "Electro-Magnetic Transient Program" (EMTP) softwares typically split the simulation into three independent modules: the electrical circuit, the generator, and the mechanical shaft.

This modularity creates a "one-time-step delay" as data is exchanged between modules, leading to a decline in numerical accuracy. Furthermore, standard models often treat network equations as algebraic constraints, meaning node voltages aren't "state variables" in the differential equations. This makes it hard to apply sophisticated numerical solvers. The authors aim to bridge this gap by creating a unified model that integrates these components into a single, mathematically consistent framework.

Methods

The authors propose two main categories of numerical integrators for a generator system based on the "first benchmark model of subsynchronous resonance":

  1. Predictor-Corrector (P-C) Methods: These are generalizations of the Adams-Bashforth-Moulton method. They use a fast explicit "predictor" step (like the Euler method) followed by an implicit "corrector" step (trapezoidal rule). This is computationally efficient but can struggle with long-term stability.
  2. Structure-Preserving Methods (Gauss Methods): This is the paper's primary innovation. The authors reformulate the generator equations into a port-Hamiltonian descriptor system. This framework explicitly accounts for energy dissipation (like friction and resistance) while preserving a geometric property called a Dirac structure. They then apply s-stage Gauss collocation methods, which are a specific class of Runge-Kutta methods known for their high order of accuracy and stability.

Datasets

The study does not use a "dataset" in the machine-learning sense. Instead, it uses the IEEE First Benchmark Model for Subsynchronous Resonance, a standard physical configuration used in power systems engineering to test simulation accuracy. The parameters (inductance, resistance, stiffness factors, etc.) are derived from established literature in the field.

Key Insights and Ideas

  • Geometric Preservation Matters: The "aha moment" is that preserving the physical structure (the Dirac structure) of the system during discretization is more important for long-term stability than simply having a high-order solver.
  • Eliminating Singularity: The authors found a way to reformulate the flux linkage equations to eliminate mathematical singularities, allowing the system to be written as index-1 differential-algebraic equations (DAEs).
  • Slow-Variable Exploitation: The P-C methods succeed by exploiting the physical reality that mechanical rotary speeds change much more slowly than electromagnetic fluxes, allowing for a hybrid explicit-implicit approach.

Key Results

  • Superiority over Industry Standards: In short-term simulations (0–10 seconds), the proposed methods converged rapidly to equilibrium, while the professional software PSCAD/EMTDC showed significantly larger fluctuations.
  • Long-Term Stability: In long-term tests (up to 1500 seconds), the "Predictor-Corrector I" method eventually blew up (diverged). However, the Structure-Preserving Gauss method remained perfectly stable.
  • Accuracy Gains: The structure-preserving method maintained an error level roughly half that of the best Predictor-Corrector method throughout the 1500-second simulation.
  • Theoretical Proof: The authors provided a rigorous proof for the global error estimates of these Gauss methods when applied to this specific generator model.

Limitations and Future Work

  • Idealized Friction: The authors noted that the friction loss coefficient ($D$) was set to zero for some simulations to represent an "ideal" situation, though they claim this doesn't markedly change the results.
  • Computational Cost: While Gauss methods are highly stable, they are implicit and require solving systems of equations at each step, which can be more computationally expensive than simpler explicit methods for massive-scale power grids.
  • Scaling: The paper focuses on a single generator benchmark. Future work could involve testing these structure-preserving methods on large-scale, multi-machine power systems to see if the stability benefits hold up as the system's dimensionality increases.

Summary updated

Abstract

Synchronous generator system is a complicated dynamical system for energy transmission, which plays an important role in modern industrial production. In this article, we propose some predictor-corrector methods and structure-preserving methods for a generator system based on the first benchmark model of subsynchronous resonance, among which the structure-preserving methods preserve a Dirac structure associated with the so-called port-Hamiltonian descriptor systems. To illustrate this, the simplified generator system in the form of index-1 differential-algebraic equations has been derived. Our analyses provide the global error estimates for a special class of structure-preserving methods called Gauss methods, which guarantee their superior performance over the PSCAD/EMTDC and the predictor-corrector methods in terms of computational stability. Numerical simulations are implemented to verify the effectiveness and advantages of our methods.
BibTeX
@article{zhang2024effective,
  author = {Jiawei Zhang and Aiqing Zhu and Feng Ji and Chang Lin and Yifa Tang},
  title = {{Effective numerical simulations of synchronous generator system}},
  journal = {Simul.},
  year = {2024},
  doi = {10.1177/00375497241231986},
}

Loading related content...

Figure