Design of an integrated decision support system for power system security assessment

The complexity of power system security assessment requires approximate system performance (ASP) models (i.e., scalar performance indices or linearized system models) to predict the system security status. Each model is based on assumptions which may or may not hold true under actual operating conditions. Therefore, selection of which ASP models that give more accurate security prediction is difficult. This thesis presents a solution through data fusion techniques. We use as many ASP models as possible and then integrate the decisions derived from them, thereby obtaining a security assessment for the power system which exceed the assessment of each ASP model alone.; This thesis develops an integrated decision support (IDS) system for assessing the security of power systems. The IDS system integrates the decisions from the various ASP models (termed local decision makers, LDMs). The integration is performed so as to minimize the risk of making wrong global decisions under changing operating conditions.; The key to the IDS system is the fusion rule which does the integration of binary (yes/no) decisions of the LDMs. The fusion rule consists of a scalar index representing the weighted sum of the decisions made by a set of LDMs and a threshold. We provide a method for finding optimal fusion rules (i.e., optimal weighting coefficient and threshold) based on the Neyman-Pearson, Bayes Risk and Minimax criteria. The design of the IDS system for detecting thermal limit violation and static voltage collapse is illustrated on a 14 bus power system. Numerical examples verify that global security assessment is always superior (and in no case inferior) to the traditional method, which predicts through a single LDM. Moreover, a method is developed for updating the optimal fusion rule for each contingency. This method makes the IDS system useful for improving on-line contingency classification.; Along with the method developed for fusion rule design, this thesis provides a new method for evaluating the effectiveness of each LDM in predicting the insecurity of power systems. Although approximate, this method is computationally efficient, and thus be useful in on-line applications.