TL;DR
This paper introduces a novel measurement-based method to estimate generator participation factors (PFs) in power systems, crucial for understanding oscillation dynamics. By transforming measurement data into a new space where symmetry is optimized, the approach accurately estimates PFs even without a complete system model or perfectly symmetric initial conditions, a significant improvement over existing methods.
What Problem Does This Paper Address?
Power system oscillations, especially low-frequency inter-area ones, can threaten grid stability. Participation Factors (PFs) are vital for identifying which generators contribute most to these oscillations, enabling targeted mitigation strategies. Traditionally, PFs are calculated using model-based approaches that rely on the system's linearized model and its eigenvectors. However, obtaining accurate and complete linearized models for modern, complex power grids, especially those with a high penetration of inverter-based resources (IBRs), is challenging. Existing measurement-based methods often require specific conditions (like perfectly symmetric initial states) or rely on black-box models with carefully designed perturbations, limiting their practical applicability. This paper aims to bridge this gap by providing a robust measurement-based PF estimation method that works even with incomplete system information and non-ideal measurement conditions.
Methods
The core of the proposed method is a four-step process that transforms the problem into a more amenable space:
- Optimal Initial State Selection: The first step involves identifying a subset of measurement data (initial states) that exhibits the most symmetric distribution around the system's equilibrium point. This is achieved by optimizing for symmetric pairs of initial states, even if they don't perfectly form a parallelotope in the original state space.
- Coordinate Transformation: A linear transformation matrix, denoted by H, is determined. This transformation maps the original measurement space (x-space) to a new space (z-space). The goal of this transformation is to create a z-space where the selected initial states exhibit an "ideally symmetric condition." This condition is crucial for the subsequent steps.
- PF Computation in z-space: In this transformed z-space, the Extended Participation Factors (EPFs) are computed. Because the z-space is designed to satisfy the symmetric condition, these EPFs are equivalent to the true PFs in that space.
- Inverse Transformation: Finally, the computed PFs from the z-space are transformed back to the original x-space using the inverse of the transformation matrix H. These transformed PFs are termed Measurement-based Participation Factors (MPFs).
This approach effectively relaxes the strict symmetry requirement of traditional measurement-based methods by finding an optimal transformation that recovers symmetry in a new domain.
Datasets
The paper evaluates the proposed method using two primary datasets:
- Two-Area System: A modified Kundur's two-area system is used. Measurements are generated from simulations under various disturbances, including cases with full observability (all generators monitored) and partial observability (some generators not monitored). The study also considers scenarios with inverter-based resources (IBRs) replacing some generators.
- NPCC 48-Machine System: A larger, more complex 140-bus, 48-machine system is used to test the method's scalability. Two types of datasets are generated:
- Fault-based: Simulations of 50 three-phase faults are used to create a measurement dataset.
- Random Initial States: 50 scenarios are generated starting from randomly selected initial states within the state space.
These datasets are derived from simulations rather than real-world field measurements, but they are designed to represent realistic power system dynamics and disturbances.
Key Insights and Ideas
- Symmetry is Key, But Not Always Present: The paper highlights that while ideal symmetry in initial measurement states is required for direct PF estimation from measurements, real-world data rarely meets this criterion.
- Transforming to a "Symmetric" Space: The core innovation is the idea of finding a linear transformation that maps the original, potentially asymmetric, measurement space into a new space where symmetry is optimized. This allows for accurate PF estimation in the transformed space.
- EPFs as a Bridge: Extended Participation Factors (EPFs) serve as a crucial intermediate step. They are equivalent to true PFs under symmetric conditions, and by transforming the data to satisfy this condition, EPFs in the z-space become reliable estimates.
- Model-Free Applicability: The method is designed to be applicable even when a detailed system model is unavailable, making it highly valuable for real-time monitoring and analysis.
- Impact of Observability: The study demonstrates that the accuracy of MPFs is significantly influenced by the number of monitored generators (PMU coverage). Higher observability generally leads to more accurate estimations.
- Coherence and Error: The paper touches upon the relationship between generator coherence (how closely their responses track each other) and estimation error, noting that high coherence can lead to ill-conditioned sampling matrices and amplified errors.
Key Results
- Two-Area System:
- With full observability, the MPFs closely matched model-based PFs for the inter-area mode, with small errors attributed to imperfect symmetry in initial states and Prony analysis.
- Under partial observability, errors increased as more generators lacked PMU coverage, but the method still provided reasonable rankings.
- The presence of IBRs led to faster mode frequencies, and the MPFs generally aligned with PFs, though with some discrepancies for generators with less coherence.
- NPCC 48-Machine System:
- Fault-based dataset: MPFs for generators 78 and 26 were close to true values, but others showed inaccuracies due to strong coherence among generators, leading to ill-conditioned sampling matrices.
- Random initial states dataset: Utilizing data from as few as 20 simulations could identify the generator with the largest PF. With 40 simulations, the ranking of PFs was correct, and with all 50 scenarios, the MPFs closely matched the true PFs, demonstrating the method's effectiveness on a large-scale system.
Limitations and Future Work
- Dataset Generation: The current study relies on simulated data. Future work could involve validating the approach with real-world power system measurements.
- Prony Analysis Errors: Errors introduced by Prony's analysis in the z-space can affect the accuracy of the MPFs.
- Coherence Issues: For systems with highly coherent generators, the sampling matrix can become ill-conditioned, leading to significant estimation errors. The paper suggests that understanding and mitigating this coherence is an area for future research.
- Computational Complexity: While the method is designed to be practical, the computational cost of finding optimal transformations and performing analysis on large systems might be a consideration.
- Nonlinear Dynamics: The current approach focuses on linearized system models. Extending it to handle strongly nonlinear oscillations would be a valuable future direction.
Summary updated