Thermal power estimation by fouling phenomena compensation using wavelet and principal component analysis

A small percentage of reactor thermal power can be overestimated because of fouling phenomena in a secondary feedwater flowmeter. This study proposes a signal processing technique for the compensation of a degraded flowmeter such a secondary feedwater flowmeter in nuclear power plants. The technique proposed is mainly focused on noise classification and step-by-step noise reduction. The noises focused are classified into the rapid distortion caused by environmental interference, the flow fluctuation according to plant state transition and the degradation by fouling phenomena qualitatively. The multi-step de-noising technique reduces each noise by three techniques step-by-step. The wavelet analysis as a low frequency pass filter to remove the rapid distortion, the linear principal component analysis (PCA) to predict a steady-state value from the fluctuation, and the non-linear PCA implemented as an autoassociative neural network (AANN) to predict an original value from the signal including fouling phenomena are developed. The main purpose of this approach is to make an AANN concentrate on compensating the degradation by fouling phenomena itself. For the demonstration the signals from a simulator and signal modeling were used so that the role and the performance of each noise removal step was represented. In addition a thermal power deviation estimator is proposed to recognize the degradation effect of each operating parameter for reactor thermal power calculation.