A comparison of nonlinear control techniques for continuous stirred tank reactors

Abstract Globally linearizing control, a differential geometry-based technique (continuous), and nonlinear predictive control, an optimization-based approach (discrete), are compared for temperature control of a classical exothermic CSTR. The two strategies can be tuned to have identical performance for setpoint changes or measured disturbances when there are no bounds on the manipulated variable. As the sample time is decreased, the two approaches also yield identical performance for unmeasured disturbances or uncertain models. However, NLPC performs better in the presence of constraints on the manipulated variable. An open-loop observer for the unmeasured state variable (composition) has been used. The system studied is minimum-phase, allowing a filtered deadbeat control law for the nonlinear predictive control strategy.