A Prediction Interval Based Cascading Failure Prediction Model for Power Systems

Power system blackouts result in massive supply interruptions leading to significant financial and societal losses. Since the majority of blackouts begin as a cascading failure (CF), early detection of this event can help stop the propagation of a single incident into a large-scale blackout. In this paper, a realtime load-based model for CF prediction is proposed. The developed method feeds phasor measurement units (PMU) data into a prediction interval (PI) neural network (NN) model reinforced with a data-fusion-based self-correction algorithm. The main contribution of the paper is that the model provides load shedding locations and prediction intervals regarding the expected blackout size so that the operator, or the automatic controller, can better react to the CF situation. The simulation results indicate that the proposed method is fast and accurate in predicting the size of the resulting blackout or load shedding following a CF.