Operating experiences with an on-line, computer based nuclear plant surveillance and anomaly detection system based on pattern recognition and artificial intelligence

Abstract

The control room of a nuclear power plant can represent a hostile work environment for all but a highly trained operating team. Whilst disciplined training and long professional experience will guarantee assurance of safety and reliability in nuclear plant operation, any surveillance system which has the ability to minimise human error and provide additional safeguards is a desireable asset. This paper proposes a scheme whereby some key parameters of a nuclear power plant, precisely known through detailed calculation and accurate measurement are stored as a data base in an on-line computer. Through the systematic statistical analysis of key stochastic variates a comparison is made by the on-line system with the data base at regular intervals. These time intervals may be as short as seconds during periods of reactor transients such as at start-up or shut down. Alternatively, during stead state operation, the parameters are calculated and displayed at intervals of an hour or greater. An anomaly, or an indication of unusual operational behaviour is indicated both numerically and graphically by the computer if it detects a variance greater than a few percent from the mean value of the reference data base.