Temporal neural network for the identification of nuclear power pant transients

A layered spatiotemporal neutral network is proposed for the identification of nuclear power plant transients. The developed layered spatiotemporal network is inspired by the formal avalanche structure developed by S. Grossberg and offers advantages compared with the stationary pattern approach using the perceptron paradigm. Each layer in the network is trained to recognize a separate time-dependent accident scenario. Within each scenario, the temporal behavior of the revelant parameters such as pressurizer pressure, pressurizer water volume, cold and hot legs temperatures, vessel flow, and power, are considered. Numerical, vessel flow, and power, are considered where the proposed methodology is applied to two nuclear power plant anticipated transient scenarios: the Station Blackout and the Anticipated Transient without Scram transients in a pressurized water reactor.