FAST REACTOR CORE MONITORING BY ANALYSIS OF TEMPERATURE NOISE
Following a summary of the principles of pattern recognition, the report describes the development of discrimination methods for use in situations where access to all possible states of the system is not possible. The method for handling new data points accepts the possibility of not being able to classify a point, either because it lies between classes or because it lies a long way outside existing classes. When a sufficient number of rejected points have been collected, they are re-analysed to determine if they constitute one or more new classes. The appearance of a new class signals a change in the system and as such, can be used as a diagnostic indicator. The complete algorithm, covering the initial learning phase and the analysis of rejected points and their merging into existing classes or acceptance as new classes, has been applied to experimental data. The algorithm detected new states and the classifications obtained were satisfactorily homogeneous. Finally the evolution of the characteristics of the system following a change of state was studied. An animated display was created to reveal more fully the evolution of the pattern recognition process. Some results obtained of an artificial Gaussian data set are presented.
Bibliographic Reference: EUR 9364 EN (1984) MF, 99 P., BFR 240, BLOW-UP COPY BFR 495, EUROFFICE, LUXEMBOURG, POB 1003
Availability: Can be ordered online
Record Number: 1989123010800 / Last updated on: 1987-01-01
Available languages: en