This paper presents a real-time application of a data-mining method with the purpose of indirect contingency screening, by rapid recognition of hazardous, reoccurring power-system operating conditions. The method, suitable for real-time applications, is demonstrated using the Slovenian power-system model for first-swing stability issues. The presented demonstration is segmented into a few parts. First, a database containing a set of pre-fault operating conditions (represented by a measurement matrix) and critical clearing times corresponding to several contingencies is constructed. Measurements matrix from the database is decomposed using the principal component analysis method and represented in a coordinate system defined by the principal components as dense clusters of points. Consequentially, monitoring real-time steady-state conditions in terms of seeking similarity to existing database clusters is facilitated (global screening). This is established by identifying the shortest Euclidean distance metric in the mentioned coordinate system. Finally, a second-stage (local) screening is applied in order to adaptively account for the fluctuating sensitivity of contingencies to variations in pre-fault operating conditions. In this manner, an accurate indication of each contingency impact is provided rapidly as long as a similar operating state exists within the database. Otherwise, the case is thoroughly investigated and added to the database as a new entry. This approach is applicable to a wide spectrum of dynamic problems, considering that problem-relevant sets of input data are available.
Read full abstract