RECURRENT NEURAL CONTROL OF WASTEWATER TREATMENT BIOPROCESS VIA MARQUARDT LEARNING

The aim of this paper is to propose a new Recurrent Neural Network (RTNN) topology and a dynamic recursive Levenberg-Marquardt algorithm of its learning capable to estimate the states and parameters of a highly nonlinear wastewater treatment bioprocess. The proposed RTNN identifier is implemented in a direct adaptive control scheme incorporating feedback/feedforward recurrent neural controllers and a noise rejecting filter. The proposed control scheme is applied for continuous wastewater treatment bioprocess plant model, taken from the literature, where a good convergence, noise filtering and a low Mean Squared Error of reference tracking is achieved.