Robust online dynamic security assessment using adaptive ensemble decision-tree learning

Online dynamic security assessment (DSA) is examined in a data-mining framework by taking into account the operating condition (OC) variations and possible topology changes of power systems during the operating horizon. Specifically, a robust scheme is proposed based on adaptive ensemble decision tree (DT) learning. In offline training, a boosting algorithm is employed to build a classification model as a weighted voting of multiple unpruned small-height DTs. Then, the small-height DTs are periodically updated by incorporating new training cases that account for OC variations or the possible changes of system topology; the voting weights of the small-height DTs are also updated accordingly. In online DSA, the updated classification model is used to map the real-time measurements of the present OC to security classification decisions. The proposed scheme is first illustrated on the IEEE 39-bus test system, and then applied to a regional grid of the Western Electricity Coordinating Council (WECC) system. The results of case studies, using a variety of realized OCs, illustrate the effectiveness of the proposed scheme in dealing with OC variation and system topology change.