Real-time transient stability prediction using neural tree networks

The application of artificial intelligence techniques to solve diagnosis and decision problems in online power system operation and control is an area of growing interest. Binary decision trees have been successfully used to solve problems such as static security assessment and transient stability analysis. The neural tree network (NTN) has recently been suggested as an alternative to decision trees. In this paper, NTNs are used to classify a transient swing as either stable or unstable on the basis of post-fault real-time phasor measurements using a 10-cycle window. Different fault locations and durations are simulated to cover a wide range of loading conditions. The performance of the NTN as a predictor is compared to that of multilayer perceptrons (MLP) classifier. Techniques are proposed to further improve the NTN performance