Network-Cognizant Voltage Control for Wind Farms Using Risk-Aware Learning

Abstract

Recent grid codes impose more stringent low-voltage ride-through (LVRT) and post-fault voltage recovery requirements at the Point of Common Coupling (PCC). The PCC voltage response is determined by the combined effect of locally regulated bus voltages and the STATCOM response at the PCC. However, because individual buses operate according to local control objectives, the resulting voltage response is generally suboptimal from a wide-area coordination perspective. This paper proposes a coordinated framework to enhance LVRT performance by addressing the two main factors governing voltage dynamics during faults: the pre-fault operating condition (OC) and the controller response to disturbances. To obtain a suitable initial OC, a preventive control strategy based on loss minimization and tight voltage regulation across all buses is developed using a centralized convex quadratic optimization problem. To achieve locally implementable responses aligned with global system objectives, a data-driven voltage control policy is proposed through a risk-aware learning strategy relying solely on local measurements, thereby avoiding dependence on ultra-fast communication infrastructure. Simulations based on a high-fidelity dynamic model of a wind farm comprising 20 double-fed induction generators (DFIGs) and a Static Synchronous Compensator (STATCOM) connected to the IEEE 9-bus system demonstrate that the proposed approach improves system resilience and fault tolerance while satisfying stringent LVRT requirements.