Online Implementation of Instrument Surveillance and Calibration Verification Using Autoassociative Neural Networks

An autoassociative artificial neural network (AANN) instrument channel monitoring technique has been developed for sensor and associated instrument channel online calibration verification. Several AANN models, each modeling a group of interrelated signals, are utilized to provide plant-wide real-time estimation of true process values. This method utilizes a genetic algorithm search approach supplemented with standard linear correlations to empirically select appropriate groups or combinations of available variables to comprise the inputs and outputs of each of several AANNs. Robust training is realized by incorporating additional random noise in the data presented to the AANN models during training and retuning phases. The Sequential Probability Ratio Test (SPRT), a statistical decision technique, evaluates the residual obtained between measurement and network prediction, and determines when drift has occurred. The system described is an online implementation of a batch-driven (i.e., offline) AANN model-based calibration monitoring system previously developed and evaluated with data from the Crystal River Unit 3, provided by Florida Power Corporation.