Neural NLMPC schemes for the control of the activated sludge process

Abstract Nonlinear model predictive controllers based on neural networks are implemented in this paper to regulate the activated-sludge process. The simulation protocol BSM1 is used to apply the predictive controller schemes and study the closed loop process behavior in different situations. Also input-output data are gathered from the benchmark for the neural networks training. Control results under dry-weather perturbations are satisfactory when a combined NLMPC - Classic PI control system is tested. This scheme has shown the best performance when compared to a centralized NLMPC scheme, a decoupled NLMPC scheme or the classic PI system developed on the BSM1. An economic analysis indicates that the neural predictive algorithm for nitrates control improves the effluent quality while also decreasing the pumping energy consumption by optimizing the internal recirculation flowrate variations.