Robust design of GPC for processes with time delay

Robust Nonlinear Predictive Control Applied to a Solar Collector Field in a Solar Desalination Plant

“…Nevertheless, in some cases, tuning of is not trivial [19]. In this algorithm, is proposed as [20], and thus, the disturbance model is given by (14) For computing the predictions, let us consider the following diophantine equation: (15) where . Using (15) and disturbance model (14), can be found by (16) As the terms in are in the future, the best prediction of the disturbance (16) at the time is given by (17) Notice that, if varies between 0 and 1, the disturbance prediction is given by the low-pass filtered disturbance .…”

“…The predictions at , , can be computed using (20) The nonlinear optimization process uses (20) to compute the predictions in the desired range ( to ). In this case, the predictions are based on the prediction error at which is computed by the FSP.…”

“…In this case, the predictions are based on the prediction error at which is computed by the FSP. Notice that filter does not appear explicitly in (20), although it affects output prediction . Therefore, tuning of will not affect the optimization procedure and the final control structure can be written as an FSP followed by a nonlinear MPC as in Fig.…”

Robust Nonlinear Predictive Control Applied to a Solar Collector Field in a Solar Desalination Plant

“…Nevertheless, in some cases, tuning of is not trivial [19]. In this algorithm, is proposed as [20], and thus, the disturbance model is given by (14) For computing the predictions, let us consider the following diophantine equation: (15) where . Using (15) and disturbance model (14), can be found by (16) As the terms in are in the future, the best prediction of the disturbance (16) at the time is given by (17) Notice that, if varies between 0 and 1, the disturbance prediction is given by the low-pass filtered disturbance .…”

“…The predictions at , , can be computed using (20) The nonlinear optimization process uses (20) to compute the predictions in the desired range ( to ). In this case, the predictions are based on the prediction error at which is computed by the FSP.…”

“…In this case, the predictions are based on the prediction error at which is computed by the FSP. Notice that filter does not appear explicitly in (20), although it affects output prediction . Therefore, tuning of will not affect the optimization procedure and the final control structure can be written as an FSP followed by a nonlinear MPC as in Fig.…”

Model predictive control springer, Berlin, 1999, ISBN 3540762418, 280 pages

“…The chapter also presents some simulation examples using GPC and DMC and a section with the formulation problems caused by the dead-time in the multivariable case. Thus, the chapter can be used as an introduction to MIMO-MPC giving the basis for a more deeply study of the subject that must include, as in the SISO case, a robustness analysis of the controllers [17], the tuning procedures and also the effect of the predictor in the performance and robustness of the closedloop [18], which are not included in the chapter.…”

Improved generalized predictive controllers for decentralized control

“…Moreover, stronger T ltering does not always provide better robustness, as pointed out with counter-examples in ref [78]. The diculty of choosing T is overcome in ref [49] where T is used only for the predictor block (in Figure(1.2.2)) while the GPC optimization that yields C and W do not involve T . Another preltering strategy is the Youla (or Q) Parameterization [1] of the GPC; the T -method is a special case of this Q-method.…”

Robust Nonlinear Predictive Control Applied to a Solar Collector Field in a Solar Desalination Plant

“…Nevertheless, in some cases, tuning of is not trivial [19]. In this algorithm, is proposed as [20], and thus, the disturbance model is given by (14) For computing the predictions, let us consider the following diophantine equation: (15) where . Using (15) and disturbance model (14), can be found by (16) As the terms in are in the future, the best prediction of the disturbance (16) at the time is given by (17) Notice that, if varies between 0 and 1, the disturbance prediction is given by the low-pass filtered disturbance .…”

“…The predictions at , , can be computed using (20) The nonlinear optimization process uses (20) to compute the predictions in the desired range ( to ). In this case, the predictions are based on the prediction error at which is computed by the FSP.…”

“…In this case, the predictions are based on the prediction error at which is computed by the FSP. Notice that filter does not appear explicitly in (20), although it affects output prediction . Therefore, tuning of will not affect the optimization procedure and the final control structure can be written as an FSP followed by a nonlinear MPC as in Fig.…”

Model predictive control springer, Berlin, 1999, ISBN 3540762418, 280 pages

“…The chapter also presents some simulation examples using GPC and DMC and a section with the formulation problems caused by the dead-time in the multivariable case. Thus, the chapter can be used as an introduction to MIMO-MPC giving the basis for a more deeply study of the subject that must include, as in the SISO case, a robustness analysis of the controllers [17], the tuning procedures and also the effect of the predictor in the performance and robustness of the closedloop [18], which are not included in the chapter.…”

Improved generalized predictive controllers for decentralized control

“…Moreover, stronger T ltering does not always provide better robustness, as pointed out with counter-examples in ref [78]. The diculty of choosing T is overcome in ref [49] where T is used only for the predictor block (in Figure(1.2.2)) while the GPC optimization that yields C and W do not involve T . Another preltering strategy is the Youla (or Q) Parameterization [1] of the GPC; the T -method is a special case of this Q-method.…”

Robust Nonlinear Predictive Control Applied to a Solar Collector Field in a Solar Desalination Plant

“…Nevertheless, in some cases, tuning of is not trivial [19]. In this algorithm, is proposed as [20], and thus, the disturbance model is given by (14) For computing the predictions, let us consider the following diophantine equation: (15) where . Using (15) and disturbance model (14), can be found by (16) As the terms in are in the future, the best prediction of the disturbance (16) at the time is given by (17) Notice that, if varies between 0 and 1, the disturbance prediction is given by the low-pass filtered disturbance .…”

“…The predictions at , , can be computed using (20) The nonlinear optimization process uses (20) to compute the predictions in the desired range ( to ). In this case, the predictions are based on the prediction error at which is computed by the FSP.…”

“…In this case, the predictions are based on the prediction error at which is computed by the FSP. Notice that filter does not appear explicitly in (20), although it affects output prediction . Therefore, tuning of will not affect the optimization procedure and the final control structure can be written as an FSP followed by a nonlinear MPC as in Fig.…”

Model predictive control springer, Berlin, 1999, ISBN 3540762418, 280 pages

“…The chapter also presents some simulation examples using GPC and DMC and a section with the formulation problems caused by the dead-time in the multivariable case. Thus, the chapter can be used as an introduction to MIMO-MPC giving the basis for a more deeply study of the subject that must include, as in the SISO case, a robustness analysis of the controllers [17], the tuning procedures and also the effect of the predictor in the performance and robustness of the closedloop [18], which are not included in the chapter.…”

Improved generalized predictive controllers for decentralized control

“…Moreover, stronger T ltering does not always provide better robustness, as pointed out with counter-examples in ref [78]. The diculty of choosing T is overcome in ref [49] where T is used only for the predictor block (in Figure(1.2.2)) while the GPC optimization that yields C and W do not involve T . Another preltering strategy is the Youla (or Q) Parameterization [1] of the GPC; the T -method is a special case of this Q-method.…”