Real-time prediction of event-driven load shedding for frequency stability enhancement of power systems

Maintaining frequency stability is one of the three dynamic security requirements in power system operations. As an emergency control, event-driven load shedding (ELS), which is determined preventively and triggered immediately after fault occurrence, can effectively prevent frequency instability. This study proposes a methodology for real-time predicting required ELS against severe contingency events. The general idea is to train an extreme learning machine-based prediction model with a strategically prepared ELS database, and apply it on-line for real-time ELS prediction. The methodology can overcome the shortcomings of conventional deterministic approaches by its high generalisation capacity and accuracy. It can either be an individual tool or a complement to deterministic approaches for enhancing the overall reliability of the ELS strategy. Its feasibility and accuracy are verified on the New England 10-machine 39-bus system, and the simulation results show that the prediction is acceptably accurate and very fast, which is promising for practical use.