Power system online security assessment using synchronized phasor measurements and decision trees

Since modern power systems are being operated at a more stressed level due to deregulation and the difficulty of increasing transmission capacity, severe security problems can be initiated by critical contingencies that eventually lead to a major blackout. So far, detailed security analysis for N-k contingencies (k = 1, 2, 3, 4...) in a near real-time setting is still a great challenge because of the super-high computational burden. This dissertation presents a decision tree (DT) assisted scheme to rapidly and accurately evaluate various system security issues using synchronized phasor measurements, including the assessment of transient stability, voltage stability, voltage magnitude violation and thermal violation problems following severe contingencies. The main procedure includes three steps: (1) offline DT training, (2) periodic DT update and (3) online application.;From exhaustive offline security simulations, DTs can identify critical system attributes as security indicators and provide predictive models with excellent performance. Periodically updated nomograms are also defined using critical system attributes, showing operating regions with different risk levels. For online application, the corresponding phasor measurement unit (PMU) measurements are collected and compared with the offline-defined thresholds in the DT models to obtain an assessment result if a critical contingency occurs. An insecure result indicates that control actions need to be armed as quickly as possible. The proposed scheme provides an important alternative for system operators to make near real time decisions and arm remedial control actions. The proposed approach is tested on three realistic power system models in the U.S. for different security assessment objectives, including the Entergy system, the American Electric Power (AEP) system and the Salt River Project (SRP) system.;The results show that properly trained decision trees can obtain very high prediction accuracy and demonstrate good robustness under changing system conditions. Several innovative methods to improve decision tree performance are proposed and tested, including the applications of 'multiple optimal DTs' and 'corrective DTs'. In addition, the idea of tuning critical system attributes to design preventive control actions is introduced for enhancing system security.