High Information Content Database Generation for Data Mining based Power System Operational Planning Studies

Database generation for training is a critical aspect of the performance of data mining based power system reliability studies. Traditionally, Monte Carlo sampling of operational parameters are done to form various basecases and contingency analysis is performed to obtain the training database. This paper proposes an efficient sampling strategy that maximizes information content while minimizing computing requirements to form a training database for decision tree based operational planning studies. A Monte-Carlo variance-reduction method, namely importance sampling, is used to construct the proposed sampling approach. The method developed is tested on the Brittany area of RTE's system for a voltage stability assessment study, and decision rules are shown to have improved accuracy.