Response-Surface-Based Bayesian Inference for Power System Dynamic Parameter Estimation

This paper develops a new response-surface-based Bayesian inference approach for power system dynamic parameter estimation of a decentralized generator using phasor-measurement-unit measurement. The response surface for the decentralized generator model is formulated through a polynomial-chaos-based surrogate. This surrogate allows us to efficiently evaluate the time-consuming dynamic solver at parameter values through a polynomial-based reduced-order representation. In addition, a polynomial-chaos-based analysis of variance is performed to screen out model parameters while ensuring system observability. In dealing with sampling the non-Gaussian posterior distribution for the parameters, the Metropolis-Hastings sampler is adopted. The simulations conducted in the New England system under different system events show that the proposed method can achieve a speedup factor of two orders or magnitude compared with the traditional method while providing full probabilistic distribution of model parameters and achieving the same level of accuracy.