A PROTOTYPE NEURAL-NETWORK TO PERFORM EARLY WARNING IN NUCLEAR-POWER-PLANT

The paper presents some results of research work in the field of artif icial neural networks (ANN) applied to nuclear safety. It shows how a priori knowledge in the form of qualitative physical reasoning can pro vide a powerful basis for designing a set of ANN-based detection subsy stems. In particular, it explains how each ANN is in charge of modelli ng a physical relationship between a set of state variables (thermal b alance, mass balance, etc.) by trying to predict one particular variab le from other ones; then, the residual signal, defined by the differen ce between the predicted value and the real one is used to decide whet her abnormalities are present. As far as the decision logic is concern ed, the paper describes how robustness can be improved by adequate fil ters on the residuals. The proposed approach is then validated on data coming from a fullscope simulator of one of the Belgian nuclear power units: the neural-based detection system is trained on ''normal'' sce narios and is able, after learning, to detect reliably and rapidly mos t of the incidental situations chosen as tests.