Emergency state stability control of power systems through intelligent islanding

The availability of reliable electric power is achieved via modern interconnected power systems. This reliability is obtained concomitantly with the economic gains realized from long distance power transactions. Modern power systems are also increasingly operating at higher levels of stress than was originally foreseen by system planners. Such heavily stressed power systems are particularly vulnerable to cascading disturbances spreading rapidly throughout the system via the interconnections. The focus of this dissertation is an emergency stability control tool to detect a sequence of low probability events that have the potential of causing uncontrolled system separation due to global synchronous instability. System specific decision trees are developed, to predict a severe loss of system security in a disturbed power system. The decisions trees are empirically built from the simulation data for the particular system. The offline training and online performance aspects of these decision trees are examined in detail. A fuzzy classifier combination technique is further proposed, wherein multiple decision trees coalesce to form a robust online transient stability assessment tool. Finally, an adaptive islanding strategy is developed on the basis of the decision tree system, which relies on strategically placed R-Rdot relays to initiate controlled separation of the system, when armed by the decision trees. To demonstrate the tool, a large multi-machine power system is employed. Simulations with the decision tree based islanding warning system show that it is robust in the sense that all contingencies for a particular operating point, seen and unseen during training, were classified correctly. The Sugeno integral based fuzzy aggregation leads to better interpretation of multiple decisions from independent decision trees. A key finding from this research is that for emergency security assessment, the robustness of the decision tree may significantly degrade due to changing pre-event system operating conditions. It is concluded that the decision tree system trained for a power system be associated with it particular pre-fault operating conditions, for robust online transient security assessment.