Model identification by neuro-fuzzy techniques: Predicting the water level in a steam generator of a PWR

In this paper we present a neuro-fuzzy technique which allows building a predictive model of an evolving signal. The fuzzy if-then rules are inferred from the available input-output data through a training procedure. During operation, in correspondence of each incoming input pattern the corresponding output is predicted and a measure of the strength of the model rules is computed: the largest strength value can be used as an indicator for detecting, in real-time, any deviation of the process due to a component failure or sensor malfunction. Applications of the prediction approach are presented with respect to a chaotic time series of literature and to the water level in the steam generator of a PWR.