Transient Stability Prediction of Power Systems Based on Deep Belief Networks

This paper proposed a transient stability prediction method based on deep belief networks. Synchronously sampled values provided by phasor measurement units (PMUs) of the generator rotor angles collected immediately after clearing a fault are utilized as the original data. A novel data pre-processing method are used to obtain the statistical features called trajectory cluster features as inputs to a deep belief networks. The number of the input features is independent from the scale of power systems. Studies on the New England 10-machine 39-bus system indicated that more accurate and more robust predictions could be made through the proposed method. The accuracy and robustness of the method was extensively tested under different load levels, network topology and incomplete measurements, which indicated that the proposed method was effective.