A Conservative Prediction Model of Power System Transient Stability

With the emerging of Wide-Area Measurement System (WAMS) and Phasor Measurement Units (PMUs), their applications enable the online diagnosis and prediction of the operating status of power systems. Unlike most existing pattern recognition methods for transient stability prediction, a conservative prediction model for power system transient stability is proposed in this paper, aiming at improving accuracy when predicting the unstable cases. The model is established as an ensemble learning model using multiple Support Vector Machines (SVMs) as sub-learning machines. The generator power angle, rotor speed and the bus voltage amplitudes within a few fundamental cycles after the clearance of the fault, are used as the input features. A proof of the conservatism of the proposed method is also presented based on the conditional probability theory. Numerical tests with the New England 39-bus test system, show that the prediction accuracy for unstable cases is 98.26%, which are at least 2.5% higher than those from six comparative models. The proposed model is more conservative to provide reliable information for potential on-line stability or emergency control applications.