Power Grid Reliability Estimation via Adaptive Importance Sampling

Abstract

Renewable energy sources (RES) has become common in modern power systems, helping to address decarbonization and energy security goals. Despite being attractive, RES such as solar and have low inertia and high uncertainty, thus compromising power grid stability and increasing the risk of energy blackouts. Stochastic (chance-constrained) optimization and other state-of-theart algorithms to optimize and control power generation under uncertainty either explicitly assume the distribution of renewables, or use data-driven approximations. The latter becomes time-consuming and inaccurate, esp. when optimizing over multiple time steps.This paper considers a discrete-time chance-constraint direct current optimal power flow control problem for minimizing power generation costs subjected to power balance and security constraints. We propose an importancesampling-based data-driven approximation for the optimal automated generation control, which allows to improve accuracy and reduce data requirements compared to stateof-the-art methods. We support the proposed approach theoretically and empirically. The results demonstrate the approach superior performance in handling generation uncertainty, enhancing the stability of renewable-integrated power systems, and facilitating the transition to clean energy.

Data-Driven Stochastic AC-OPF using Gaussian Processes

“…Monte Carlo scenario-based CC-OPF [7,8], with nonlinear AC-PF equations, does not computationally scale for large systems or a large number of scenarios [9]. Consequently, advanced sampling policies (importance sampling, active sampling) have been recommended to reduce the number of samples necessary [10,11,9,12,13]. In contrast, analytical approach to CC-OPF states the chance constraints using distributional information of the uncertainty and is the focus of this work.…”