Efficient Database Generation for Decision Tree Based Power System Security Assessment

Decision tree based planning tools provide operators with the most important system attributes that guide them in deciding as to what situation requires operator action. Key to this approach is the manner in which different operating conditions are sampled to form a database for training. This paper develops an efficient sampling strategy that maximizes database information content while minimizing computing requirements. The approach involves two stages: stage-I to find the high information content region in the multidimensional operating parameter state space and stage-II to bias the sampling towards that region using importance sampling. The proposed approach is applied for deriving operating rules against voltage stability issues on the Brittany region of the French EHV system. The results show that the decision trees produced by the proposed efficient sampling approach have significantly improved classification performance and offer economic benefits compared to conventional sampling strategies, all at greatly reduced computational requirements.