Application of adaptive Kalman filtering technique for the diagnostic system of nuclear power plants

The authors propose a diagnostic system of nuclear power plants which is composed of three blocks, i.e. 1) detection and classification block, 2) disturbance estimation block and 3) storage of past observed signals. In the block-1, a set of observed signals is identified with one of the categories prescribed to present the normal and several anomalous situations in multidimentional space, where the linear discriminant functions basing maximum likelihood technique are utilized. An approximate linear dynamic model for the individual prescribed anomalous state is identified beforehand, where the disturbance and several assumed variables are utilized in a dynamic model and a observed vector is composed of several selected observed signals. The Kalman filters for all anomalous categories are obtained using corresponding dynamic models, and they are provided in the block-2. When the present state is identified to one of the prescribed anomalous situations by the block-1, a Kalman filter corresponding to the identified category is selected from the block-2, and the disturbance is estimated using the past observed signals obtained from the block-3 and future coming signals. The linear discriminate functions and the approximate linear dynamic models are derived using the data base of prescribed categories obtained from the accurate plant simulator. The database will be improved by the experience of actual plant. The effectiveness of this diagnostic system was examined by the computer experiment. The results show that classification of the present operating state and estimation of disturbance are available with reasonable reliability and reasonable computation time.