Flow-based Polynomial Chaos Expansion for Uncertainty Quantification in Power System Dynamic Simulation

A team of researchers has published a novel AI-powered framework called 'Flow-based Polynomial Chaos Expansion (PCE)' to tackle a critical challenge in modern power systems: accurately simulating grid dynamics amid the uncertainty introduced by renewable energy. Traditional PCE methods, while efficient for uncertainty quantification (UQ), rely on oversimplified assumptions about input data, often treating it as independent and Gaussian. In reality, inputs like wind and solar power are highly correlated and follow complex, non-Gaussian distributions. This mismatch leads to inaccurate simulations and unreliable grid planning.

The new framework bridges this gap by integrating normalizing flows—a type of deep generative model—into the PCE process. First, the flows learn an invertible map that transforms messy, real-world input data into a simple, well-behaved distribution. This learned transformation is then baked directly into the construction of the PCE surrogate model, ensuring the uncertainty propagation is faithful to the actual data. The paper also introduces a 'Map Smoothness Index (MSI)' to quantify the quality of this transformation, showing that smoother maps yield more accurate surrogates. The method was validated on large-scale benchmarks including the IEEE 14-bus system and a model of the Great Britain transmission network, demonstrating superior performance over conventional approaches under various uncertainty scenarios.