Multicontingency steady state security evaluation using fuzzy clustering techniques

This paper provides a description of a new approach for steady state security evaluation, using fuzzy nearest prototype classifiers. The basic method has an offline training phase, used to design the fast classifiers for online purposes, allowing more than the two traditional security classes. A battery of these fuzzy classifiers, valid for a specific configuration of the network, is adopted to induce a global evaluation for all relevant single contingencies. An important feature of this approach is that it selects automatically the most appropriate number of security clusters for each selected contingency. Natural language labeling is also used to produce standardized sentences about the security level of the system, improving in this way the communication process between the system and the operator. The paper is completed by an example on a realistic model of the Hellenic interconnected power system, where seven contingencies were simulated.