Nonlinear control of a wastewater pH neutralisation process using adaptive NARX models

Abstract

This paper describes a comprehensive approach to the control of a wastewater pH neutralisation process which consists of a linear dynamic part with a long time delay and a severely nonlinear static part with an extremely high gain. Varying buffer capacity is also a characteristic of the process. Nonlinear predictive control (NPC) schemes are used based on adaptive nonlinear autoregressive with exogenous inputs (NARX) models. The initial structure and parameters are identified off-line with a modified orthogonal forward regression (MOFR) algorithm. In order to avoid estimator windup due to poor excitation of online data, the modified least squares incorporating exponential resetting and forgetting algorithm (EFRA) and setpoint perturbation are applied to adaptively adjust the parameters of a NARX model. Several aspects of NPC are discussed. Comparisons of nonadaptive control, adaptive control without and with setpoint perturbation, and adaptive control with the noisy measurement are carried out to demonstrate the performance of the proposed schemes.