(389a) Networked Model Predictive Control

This paper presents a networked approach to large-scale process control. The networked approach views a physical system as networked processes with certain intrinsic properties that characterize how a process stabilizes itself and the manifestation of cause and effect phenomenological relationships. The approach is process centric and utilizes a control system framework that includes both hierarchical and distributed control. Open control architecture is synthesized in accordance with the intrinsic process behavior that deals with disturbances, adaptability and uncertainties. The networked MPC is designed to deliver improved quality of control performance. It is quality by design approach in which a number of innovative solutions are brought together to tackle the entrenched and intractable problems of MPC as practiced in the industry presently. The networked MPC incorporates the following four key architectural elements:

1.0 Explicit stability criteria

An explicit process-centric stability criteria is augmented to a traditional MPC formulation. Application of the stability criteria determines control structure design in that it delineates process boundary in a way that supports both hierarchical and distributed control structure consistently. The stability criteria permeate both internally and externally to the control structure. The stability criteria is independent variables based and therefore can be used with linear as well non-linear process. It allows a nested and networked control systems relationship for large-scale process application with recycle streams. It predominately separates out the variables in accordance with the innate characteristics of the process stability. It clearly sets requirements for at least a two level control structure comprising a regulatory control level and an optimizing level (supervisory level). The method is based on self-stabilizing aspect of a process and its sub processes and on the premise that a stable process can be optimized and controlled in much the same way. In this regard, speed of optimization plays an important and critical role in ensuring coordination of changes in the manipulated variables in unison.

The stability criteria supports asymmetrical control actions capability that allows a process to perform a complete cycle control that includes start up, normal operation and shut down. Both acceleration and de-acceleration of process changes can be done under operator's control. Intrinsically, the stability criteria makes a MPC both modular and robust and hence the MPC incorporating it is considered to a modular robust MPC.

2.0 PV-based Models (in place of traditional SP or OP based Models)

An alternate method of modeling is devised independent of how control actuation is applied to a process. This makes the control models independent of regulatory controller tuning effects and control valve non-linearity effects. The method of modeling is termed as process value based (PV-based) models and is markedly different from the traditional set point based (SP-based) and control output based (OP-based) models. The PV-based models permits embedding of regulatory controllers within a MPC. This allows the MPC to compensates for unmeasured disturbance effects closest to the source while eliminating propagation of model mismatch error to other variables. Most importantly, the MPC continues to operate unaffected by the breakdown of models when one or more control valves saturate. It can operate with mix of fast to slow response time without loss of approximation with multiple control cycle frequency. It eliminates use of a whole class of models relating to controller output that are noisy and non-linear and requires update of one model only for any change in control valve parameter. It can be upgraded with the least amount of changes for any process design change. The PV-based models can be identified with the least amount of plant testing. With the adoption of the explicit stability criteria and the PV-based models, a modular robust model predictive controller can be built that is network enabled.

3.0 Hybrid Dynamic Control

A flexible method of dynamic MPC is devised in which steady state and dynamic state constrained optimization is performed simultaneously as part of a one-step solution. The method of dynamic MPC is flexible in constituting a hybrid controller in which a selected sub-set of variables is chosen for simultaneous steady state and dynamic state constrained optimization while the rest of the variables set is chosen for separate steady state and dynamic state optimization. The Hybrid Dynamic MPC (H-DMPC) can be constituted in real time based on certain closed loop performance criteria such as the variance of dynamic violations of the variables. A number of adaptations of the H-DMPC offer novel methods of control that include dynamic regulatory controller and dynamic feed forward constraints. These adaptations of the H-DMPC can be used for feed tank switching and product grade change.

In a networked MPC, the H-DMPC plays a vital role in responding to dynamic violations with localized and targeted control actions without unduly affecting the rest of the network. The H-DMPC performs dynamic adaptive control actions without resorting to solve a much larger and unwieldy dynamic control problem. It is an efficient and effective solution to the dimensionality curse of solving large-scale dynamic control problem.

4.0 Variance Control

In spite of the best performance of a control system, a certain amount of variance of violations is unavoidable for the reasons of measured and unmeasured disturbances. Therefore, in a networked MPC, such uncontrolled violations must be dealt with locally. This is achieved by a further enhancement to the modular robust MPC that includes what is described herein as n-sigma violations control and variance stabilization. The variance control relates to both the controlled variables and the manipulated variables. The controlled variable variance control relates to avoidance of violations whereas the manipulated variable variance control relates to variance stabilization by subduing the manipulated variables moves.

The method of networked MPC presented allows scalability of a control system in progressive manner without compromising the stability and operability of the large-scale process. With its modular robust MPC as building block, a new node can be added in a plug and play manner with no changes to configurations or tuning to the rest of the network nodes.