[Paper Review] Networked Synthetic Dynamic PMU Data Generation: A Generative Adversarial Network Approach

Xiangtian Zheng, Bin Wang+2 more|arXiv (Cornell University)|Aug 22, 2019

Computational Physics and Python Applications4 citations

TL;DR

This paper proposes a generative adversarial network (GAN)-based method to directly synthesize multiple, networked PMU data streams that reflect realistic electromechanical dynamics and obey Kirchhoff’s laws. The approach generates synchronized, multi-bus PMU data from historical measurements without relying on power system simulations, demonstrating high statistical and dynamic fidelity on a 39-bus and a 200-bus system.

ABSTRACT

This paper introduces a machine learning-based approach to synthetically create multiple phasor measurement unit (PMU) data streams at different buses in a power system. In contrast to the existing literature of creating synthetic power grid network and then using simulation software to output synthetic PMU data, we propose a generative adversarial network (GAN) based approach to synthesize multiple PMU measurement streams directly from historical data. The proposed method can simultaneously create multiple PMU measurement streams that reflect practically meaningful electromechanical dynamics which observe the Kirchhoff's laws. We further validate the synthetic data via the statistical resemblance and the modal analysis. The efficacy of this new approach is demonstrated by numerical studies on a 39-bus system and a 200-bus system.

Motivation & Objective

• To generate synthetic, multi-bus PMU data streams that reflect realistic electromechanical dynamics in power systems.
• To overcome limitations of existing methods that rely on power system simulations after synthetic network creation.
• To ensure generated data respects fundamental power system laws, such as Kirchhoff’s current and voltage laws.
• To validate the synthetic data using statistical resemblance and modal analysis techniques.
• To demonstrate the method’s scalability and fidelity on large systems, including a 200-bus test case.

Proposed method

• A generative adversarial network (GAN) is trained directly on historical PMU measurements to learn the underlying distribution of multi-bus phasor data.
• The generator network produces synthetic PMU data streams across multiple buses simultaneously, preserving inter-bus dependencies.
• The discriminator network distinguishes between real historical PMU data and generated synthetic data, enforcing statistical realism.
• The training process enforces physical consistency by ensuring the generated data adheres to Kirchhoff’s laws through architectural or loss function constraints.
• The method bypasses the need for power system network modeling and dynamic simulation, directly synthesizing synchronized phasor measurements.
• The approach is evaluated using statistical similarity metrics and modal analysis to verify dynamic fidelity.

Experimental results

Research questions

• RQ1 Can a GAN-based approach generate multiple, synchronized PMU data streams that reflect realistic electromechanical dynamics in power systems?
• RQ2 To what extent does the synthetic data preserve the statistical properties of real historical PMU measurements?
• RQ3 How well do the generated data streams satisfy Kirchhoff’s laws in a multi-bus network context?
• RQ4 Can the proposed method scale to large power systems, such as a 200-bus system, while maintaining data quality?
• RQ5 How does the synthetic data perform in modal analysis compared to real PMU data?

Key findings

• The proposed GAN-based method successfully generates multiple, synchronized PMU data streams that reflect realistic electromechanical dynamics.
• The synthetic data exhibit strong statistical resemblance to real historical PMU data, as validated through comparative statistical analysis.
• Modal analysis confirms that the synthetic data preserve key oscillatory modes present in real PMU measurements.
• The generated data streams satisfy Kirchhoff’s laws, ensuring physical plausibility across the networked system.
• The method demonstrates scalability and robustness on both a 39-bus and a 200-bus test system.
• The approach eliminates the need for separate power system modeling and simulation, enabling direct data generation from historical measurements.