Enhanced Power System Security Assessment Through Intelligent Decision Trees

Abstract

Power system security assessment involves ascertaining the post-contingency security status based on the pre-contingency operating conditions. A system operator accomplishes this by the knowledge of critical system attributes which are closely tied to the system security limits. For instance, voltage levels, reactive power reserves, reactive power flows are some of the attributes that drive the voltage stability phenomena, and hence provide easy guidelines for the operators to monitor and maneuver the highly stressed power system to a secure state. With tremendous advancements in computational power and machine learning techniques, there is increased ability to produce security guidelines that are highly accurate and robust under a wide variety of system conditions. Particularly, the decision trees, a data mining tool, has lend itself well in extracting highly useful and succinct knowledge from a very large repository of historical information. The most vital and sensitive part of such a decision tree based security assessment is the stage of training database generation, a computationally intensive process which involves sampling many system operating conditions and performing power system contingency assessment simulations on them. The classification performance of operating guidelines under realistic testing scenarios depend heavily on the quality of the training database used to generate the decision trees. So the primary objective of this chapter is to develop an improvised database generation process that creates a satisfactory training database by sampling the most influential operating conditions from the input operating parameter state space prior to the stage of power system contingency simulation. Embedding such intelligence to the system scenario sampling process enhances the information content in the training database, while minimizing the computing requirements to generate it. This chapter will clearly explain and demonstrate the process of identifying such high information contained sampling space and the advantage of deriving security guidelines from decision trees that exclusively use such an enhanced training database.