A neural model for transient identification in dynamic processes with “don't know” response

Abstract This work presents an approach for neural network based transient identification which allows either dynamic identification or a “don't know” response. The approach uses two “jump” multilayer neural networks (NN) trained with the backpropagation algorithm. The “jump” network is used because it is useful to dealing with very complex patterns, which is the case of the space of the state variables during some abnormal events. The first one is responsible for the dynamic identification. This NN uses, as input, a short set (in a moving time window) of recent measurements of each variable avoiding the necessity of using starting events. The other one is used to validate the instantaneous identification (from the first net) through the validation of each variable. This net is responsible for allowing the system to provide a “don't know” response. In order to validate the method, a Nuclear Power Plant (NPP) transient identification problem comprising 15 postulated accidents, simulated for a pressurized water reactor (PWR), was proposed in the validation process it has been considered noisy data in order to evaluate the method robustness. Obtained results reveal the ability of the method in dealing with both dynamic identification of transients and correct “don't know” response. Another important point studied in this work is that the system has shown to be independent of a trigger signal which indicates the beginning of the transient, thus making it robust in relation to this limitation.