Development of a pattern recognition approach to underfrequency relaying

The main objective of the practice of load shedding is to shed an appropriate amount of load in the event of a major disturbance in order to reestablish system frequency in response to a major disturbance which causes power system cascading and formation of isolated areas (islands) with the imbalance between generation and load. The design of an underfrequency protective program to realize this objective is described. The possibilities and drawbacks of some load-shedding programs are discussed. The use of artificial intelligence in a proactive, rather than the common reactive, load-shedding scheme is considered. To classify the state of the power system certain features must be extracted from the system data. Once these features are extracted, they can then be classified into an appropriate group (e.g. system acceptable, system declining, system recovering, system unacceptable). To select the most favorable load shedding, including recognition of capabilities and methods of using different types of relays, one needs to understand how the system frequency respond to discrete load-shedding operations. Therefore, mid- to long-term modelling of the power system must be conducted to extract the features necessary for classifying the state of the power system and for gaining an understanding of how the system frequency will respond to the load shedding. Potential features that are being evaluated for use in the pattern-recognition model are discussed