A two level hierarchical classifier for online transient stability assessment

The authors present a novel automatic transient stability classification approach, based on a two-stage hierarchical classifier obtained through the combination of two different pattern recognition structures. The first-level classifier uses low-cost computational variables evaluated at the fault occurrence moment. The classification structure used in this first level is a linear one and is obtained through the combination of a Fisher discriminant transformation and a Bayesian probabilistic approach for the independent term determination. The second-level classifier uses a new set of transient variables, directly related to kinetic and potential energies at the fault clearance moment. The classification structure in this second level is based on a weighted K nearest-neighbor rule using a Euclidian metric defined in an optimum discriminant plane, where all data have been previously projected. Results obtained in the CIGRE test system are presented, showing that this method provides an efficient tool for online transient stability assessment of electric power systems