MODEL-PREDICTIVE PH CONTROL USING REAL-TIME NARX APPROACH

A new comprehensive real-time identification/control methodology based on the concept of nonlinear autoregressive exogenous input (NARX) models and adaptive, nonlinear, model-predictive control (ANMPC) is applied to a pH neutralization process. The existing NARX model theory has been extended by incorporating measured disturbances. NARX models have shown superior predictive characteristics in comparison to linear models. The proposed real-time methodology uses a pointer vector being created during an initial identification and model structure selection procedure. Using this pointer vector, which allocates the chosen elements from the pool of all possible linear and nonlinear combinations, one needs no explicit information about the model structure for the closed-loop control. The nonlinear programming problem encountered in ANMPC is solved by a gradient-based modified Marquardt and finite difference methods. The design procedure and explicit algorithms are discussed for the multiinput/multioutput case. A pH wastewater neutralization process used illustrates and verifies the procedure by computer simulations and real-time laboratory-scale experiments.