Controllo e gestione intelligente degli impianti di depurazione

This work proposes a way to combine two alternative criteria for the real-time monitoring and analysis of the denitrification reaction in a Sequencing Batch Reactor for the treatment of urban sewage. The data, acquired by probes, are analysed by neural networks of class SOM and Elman. The outputs of the networks are mapped onto fuzzy predicates and further elaborated using rules. The results of the two modules are combined as if produced by a committee of experts to reduce the uncertainty associated to the individual responses. Eventually, the a-posteriori process state estimate is compared to the expected a-priori behaviour and, in case of strong mismatches, more rules are activated, which model and apply the diagnostic procedures to detect the type and cause of malfunctionings.