Efficient Data-Driven Uncertainty Propagation in Power System Dynamics Using Low-Rank Randomized Koopman Operator Approximation

In this paper we propose an efficient model-free data-driven algorithm for propagating uncertainty through nonlinear power system dynamics. We characterize the uncertainty in terms of the moments of the probability distribution of the dynamic states. The proposed approach relies on efficiently learning a linear representation of the power system by using the Koopman operator for propagating the moments. We use a randomized singular value decomposition to accelerate the computation of the Koopman operator. The proposed approach is implemented in Julia for propagating initial condition uncertainties for multiple power systems. The results demonstrate that the speedup of the proposed approach compared with conventional approaches increases as the system size increases and reaches an acceleration of 6x for the New England test system with comparable accuracy.