A Probabilistic Data-Driven Method For Response-Based Load Shedding Against Fault-Induced Delayed Voltage Recovery in Power Systems

Aiming at timely and adaptive remedial control for the fault-induced voltage recovery (FIDVR) events in power systems, this paper develops a probabilistic data-driven method for response-based load shedding (RLS). In the proposed method, a scalable Gaussian process (SGP) model is developed to estimate the required load shedding (LS) amount and the confidence of the corresponding predictions. Based on the probabilistic information, a 2-stage LS process is designed to enhance the effectiveness and efficiency of the scheme, using the mean value for LS at the first stage and the upper-bound value for LS at the second stage. The 1(st) stage shed the amount of load with the largest likelihood, aiming to alleviate system stress with the least control cost. The 2(nd) stage serves as a safety net to ensure system stability considering the possible prediction error. Compared with the conventional RLS schemes and other state-of-the-art approaches, simulation results verify that the proposed method can effectively mitigate FIDVR with a much less load shedding amount.