An enhanced prediction model for the on-line monitoring of the sensors using the Gaussian process regression

Abstract

The auto-associative kernel regression (AAKR) and Gaussian process regression (GPR) have been used for estimating the condition of the sensors in the on-line monitoring system of the nuclear power plants. The estimations of the condition could be biased by the data of an unhealthy sensor, even though GPR generates its predictive uncertainty as a part of the predictions which AAKR may not provide. An effective modification to GPR, which enables early detection of the unhealthy sensor based on the prediction uncertainty and the residuals of estimations, is proposed to eliminate the influences of the biases. The proposed method which is named as an enhanced GPR (EGPR) shows a better performance in estimating the states of the sensors than that of AAKR and GPR with the test data from the flow system.