Importance Sampling Based Intelligent Test Set Generation for Validating Operating Rules Used in Power System Operational Planning

Decision tree based machine learning methods find a great deal of application in power system reliability assessment studies, wherein essential knowledge in the form of operating rules or guidelines are produced that help operators maneuver the system away from insecurity. Independent test sets are generally used to validate these rules, with the motivation of estimating their classification accuracy and error rates, apart from checking their performance against some interesting situations. This paper proposes an importance sampling based method to generate intelligent test set for validating operating rules. The method is applied for testing decision tree rules derived against voltage collapse problems in western regions of the French power system, and is seen to produce test sets at lesser computation that estimates the rule´s classification errors with good accuracy. For a given computation, it also provides richer information on critical operating conditions for which the rule is vulnerable, which helps in further improving the rules.