DECISION TREES AND TRANSIENT STABILITY OF ELECTRIC-POWER SYSTEMS

An inductive inference method for the automatic building of decision trees is investigated. Among its various tasks, the splitting and the stop splitting criteria successively applied to the nodes of a grown tree, are found to play a crucial role on its overall shape and performances. The application of this general method to transient stability is systematically explored. Parameters related to the stop splitting criterion, to the learning set and to the tree classes are thus considered, and their influence on the tree features is scrutinized. Evaluation criteria appropriate to assess accuracy are also compared. Various tradeoffs are further examined, such as complexity vs number of classes, or misclassification rate vs type of misclassification errors. Possible uses of the trees are also envisaged. Computational issues relating to the building and the use of trees are finally discussed.