Application of nonlinear principal component analysis technique to nuclear power plants

• Traditionally, manual calibration of sensors is required and performed during each refueling outage. If the traditional time-directed calibration is replaced by an online monitoring technique, the maintenance cost will be significantly reduced since only the abnormal sensors identified in on-line monitoring need to be re-calibrated or replaced off-line. The Nonlinear Principal Component Analysis (NLPCA), such as Auto-Associative Neural Network (AANN) and Auto-Associative Kernel Principal Component Analysis (AAKPCA), can describe the nonlinear correlation between sensors such as power, temperature, pressure and flowrate. In this paper, AANN and AAKPCA model are tested by simulated redundant data and Tennessee-Eastman process data. The results show that both of them have a high ability of prediction and a low sensitivity. Therefore, they are can be used in on-line monitoring.

Ämnesord

NATURVETENSKAP  -- Fysik -- Annan fysik (hsv//swe)

NATURAL SCIENCES  -- Physical Sciences -- Other Physics Topics (hsv//eng)

Nyckelord

Auto-associative kernel principal component analysis

Auto-associative neural network

Feature extraction

Nonlinear correlation

Online monitoring

Publikations- och innehållstyp

ref (ämneskategori)

kon (ämneskategori)