Feature selection and rule extraction for the estimation of power system transient stability

This paper proposes a novel scheme to abstract transient stability assessment (TSA) rules directly from training samples. The comparison and selection of proper input features for the transient stability state estimation is carried out through a genetic algorithm (GA) based feature selection algorithm. Significant events are then identified by pattern discovery algorithm based on statistic analysis applying the residual analysis and recursive partitioning which reflect the sample distribution patterns in the input feature space. Stability estimation tree as well as production rules can be easily obtained based on patterns. Applications of the proposed scheme in the New England 39-bus power system are introduced in detail, including the kernel feature set selected, patterns discovered as well as the decision tree and rules abstracted for the stability assessment. Vast stability assessment tests prove the validity and good performance of the proposed TSA scheme.