Comparison of Decision Tree Attribute Selection Methods for Static Voltage Stability Margin Assessment

Abstract

Decision tree (DT) as an effective data mining method has been widely used in voltage stability assessment. The selection of decision tree’s input attributes is critical because input attributes affect the accuracy and efficiency of the decision tree. This paper compares two attribute selection methods: participation factor method and Relief-F algorithm. Participation factor method is based on modal analysis of Jacobi matrix, while Relief-F algorithm is a mathematical approach that does not require power system knowledge. Two DTs with the same number of input attributes identified by participation factor analysis and Relief-F algorithm respectively are constructed for comparison in term of accuracy and efficiency. A case study on a practical power system indicates that two methods identify similar attributes and the accuracy of two DTs are close.