Using CMAC for adaptive nonlinear MPC and optimal setpoint identification of an activated sludge process

This paper proposes both an adaptive nonlinear model predictive control and a method to identify an optimal setpoint. Local discrete-time linear models, estimated from output measurements, are stored in a Cerebellar Model Arithmetic Computer (CMAC). The CMAC provides a practical way to store, access, and interpolate the models in real-time and for future-time predictions. A finite-horizon nonlinear optimization decides on a desired control signal for training a CMAC controller. In order to search for on an optimal setpoint in the case of a measured disturbance, another set of local linear models is produced that depends on only outputs and disturbances. A Lyapunov-based method ensures stability (uniformly ultimately bounded signals) in the cases of a cart-pendulum system and an activated sludge process for wastewater treatment. Simulation results show successful trajectory tracking and setpoint identification for both systems in simulation.