A Robust and Fast Operational Risk Assessment Method for Composite Power Systems With High Wind Power Penetration

Abstract

The strong stochasticity and volatility of wind power generation pose new challenges to the operational reliability of power systems. Traditional reliability assessment often struggles to balance economic efficiency and robustness under high-penetration wind power integration, yielding either optimistic or overly conservative results. We develop a data-driven distributionally robust optimization (DRO) model for operational risk assessment, which computes state-conditioned reliability indices while hedging probability misspecification and distribution shifts. To meet real-time needs, we propose a CNN-DRO method that predicts the worst-case probability distribution (WPD) and thereby collapses the tri-level “min-max-min” DRO model to a single-level OPF-consistent optimization problem solved per operating state. This preserves physical feasibility and enables minute-level updates. A hybrid CNN Classifier-Regressor design first screens secure states (no load shedding needed) and then predicts the WPD for insecure states, so only the latter invokes the single-level optimization solve. The CNN inference is negligible relative to the solver. Numerical simulations are conducted on the RTS-79 and IEEE 118-bus systems under varying wind power penetration levels, with comparisons against stochastic optimization (SO) and robust optimization (RO) methods. The results validate the superior robustness and economic benefits of the proposed DRO model in operational risk assessment for high-penetration wind power systems. Additionally, experiments demonstrate that the proposed method has much higher computational speed and accuracy when compared with the classical column and constraint generation (CCG) algorithm.