A hybrid neural network---fuzzy logic approach to nuclear power plant transient identification

A methodology is presented that couples pretrained artificial neural networks (ANNs) to rule-based fuzzy logic systems, for the purpose of distinguishing different transients in a Nuclear Power Plant (NPP). A model referenced approach is utilized in order to provide timely concise and task specific information about the status of the system under consideration. A rule based system integrated with a set of neural networks, that typify steady-state operation as well as different transients, diagnoses the state of the system and identifies the type of transient under development. ANNs produce their response in the form of membership functions which independently represent individual transients and the steady-state. Membership functions condense functionally relevant information in order for the overall system to successfully perform transient identification, in a time span faster or at least comparable to that of the transient development. To demonstrate the proposed methodology simulated accidents corresponding to a particular category of transients are used. The results obtained demonstrate the excellent noise tolerance of the ANNs and suggest a new approach for transient identification within the framework of fuzzy logic