Robust steady-state target calculation for model predictive...

Details Robust steady-state target calculation for model predictive... Robust steady-state target calculation for model predictive...

1999-03-11

2002-04-30

Grant, William (Department: 2121)

Data processing: generic control systems or specific application

Generic control system, apparatus or process

Optimization or adaptive control

C700S028000, C703S002000

Reexamination Certificate

active

06381505

ABSTRACT:

FIELD OF THE INVENTION
This invention relates generally to the field of computer-based control systems and control algorithms therefor, and more particularly relates to a Model Predictive Control class of computer control algorithms.
BACKGROUND OF THE INVENTION
Model Predictive Control (“MPC”) refers to a class of computer control algorithms that compute sequences of manipulated variable adjustments (control moves) in order to optimize the future behavior of a system. Computerized control systems making use of MPC technology are particularly well-suited for controlling such complex industrial systems as manufacturing plants, chemical processing plants, petroleum refineries, and the like. Such systems have many parameters (flow rates, temperatures, pressures, and so on) which require continuous real-time or nearly real-time adjustment or tuning.
As the name suggests, MPC uses an explicit model to predict how a process will evolve in time. The prediction is used to determine “optimal” control moves that will bring the process to a desired state. The optimal control moves are the result of an online optimization which is, in general, a full-blown nonlinear program (“NLP”). In practice, though, because linear models are used, the resulting optimizations are linear and quadratic programs. While originally developed to address the needs of power plants and petroleum refineries, MPC technology is now used in a wide variety of commercial and industrial applications ranging from food processing to pulp and paper production.
In many modem industrial processing plants, MPC is implemented as part of a multi-level hierarchy. of control functions.
FIG. 1
depicts, in block diagram form, a control system
10
suitable for controlling, for example, an industrial plant or the like having a plurality of operational components which require parametric control of one or more (usually a plurality) of controlled variables to maintain a desired operational status for the plant overall. In one embodiment, for example, controller system
10
may be a chemical processing plant. Those of ordinary skill in the art would appreciate that such a plant would have various operational components, such as electronically-controlled flow valves, pressure valves, heating and cooling systems, and the like, which require continuous monitoring and parametric adjustment or control in order to ensure proper functioning of the plant as a whole.
The block diagram of
FIG. 1
illustrates the differences between a standard conventional control hierarchy and one using MPC. In
FIG. 1
, a “conventional” control hierarchy is represented within dashed line
12
, and an MPC control hierarchy is represented within dashed line
14
. As would be apparent to those of ordinary skill in the art, each of the control hierarchies
12
and
14
functions to generate control signals, represented by lines
16
in
FIG. 1
, to be applied to controlled systems or components in, for example, an industrial plant of some sort. In the exemplary embodiment of
FIG. 1
, the “controlled components” include flow controllers
18
, pressure controllers
20
, temperature controllers
22
, and level controllers
24
as examples, although it is to be understood that the class of controlled components and systems includes many other types of functional units adapted to be adjusted or controlled by means of one or more electrical control signals.
In
FIG. 1
, conventional control hierarchy
12
is shown being comprised of several discrete components. Central to control hierarchy
12
is one or more single-parameter controllers
26
, one type of which being commonly referred to as a proportional, integrating, derivative (“PID”) controller. Those of ordinary skill in the art will appreciate that various types of PID controllers are available as commercial, off-the-shelf components from, by way of example but not limitation, Honeywell, Inc., Minneapolis, Minn. Such controllers implement simple control algorithms to generate a single parameter control output signal in response to one or more inputs. Controller
12
in
FIG. 1
is shown to further comprise other logic blocks, such as lead/lag (“L/L”) block
28
, summation (“SUM”) blocks
30
, and high/low select logic block
32
. Those of ordinary skill in the art will appreciate that this additional logic associated with PIDs
26
enables controller
12
to accomplish the multi-parameter control functionality that can be done with a single MPC controller such as that designated with reference numeral
50
in FIG.
1
.
The design and operation of a conventional controller
12
such as that in
FIG. 1
is well understood by those of ordinary skill in the art, and the details of its operation will not be discussed in further detail herein. It is sufficient to state that controller
12
operates in response to various inputs reflecting the state of the system to generate one or more output control signals for adjustment of one or more operational parameters of the system.
With continued reference to
FIG. 1
, at the top of the control hierarchy, a plant-wide optimizer designated with reference numeral
34
determines optimal steady-state settings for each controlled unit in the plant. These settings may be sent to local optimizers
36
associated with each controlled unit. Such local optimizers may run more frequently or consider a more detailed unit model than is possible at the plant-wide level. Each local optimizer
36
computes control settings corresponding to optimal, economic steady-states and passes this information to a dynamic constraint control system for implementation. As indicated by dashed line
35
in
FIG. 1
, the dynamic constraint control system comprises high-low select logic block
32
, PIDs
26
, lead/lag block
28
, and summation blocks
30
. The dynamic constraint control
35
must move the plant from one constrained steady-state to another while minimizing constraint violations along the way.
Those of ordinary skill in the art will appreciate that in conventional control structures such as that designated with reference numeral
12
in
FIG. 1
, tuning and design can take considerable engineering effort. This is in part because each PID
26
is general-purpose and is typically only capable of controlling one controlled variable, and as an off-the-shelf component must be tuned for a particular application.
MPC is an optimal-control based method in the sense that it determines the optimal control moves by minimizing an objective function. The objective function depends on both the current and predicted future states of the system and the future inputs to the system. The states andthe inputs are related through an explicit process model. The theoretical framework which has developed around MPC does not depend on the particular model form and allows for many variations. The models can be linear or nonlinear, continuous-time or discrete time, state-space or input-output, deterministic or stochastic, and more. The flexibility in choosing the type of model provides a powerful advantage to MPC in accordance with the embodiments of the invention disclosed herein.
With continued reference to
FIG. 1
, within the MPC control structure, the plant-wide optimization represented by block
34
is a system-wide process that is typically conducted less frequently relative to other levels of optimization and control in the overall hierarchy. On the other hand, local optimization represented by blocks
36
in
FIG. 1
, also known as “real-time optimization” or “RTO,” may be conducted somewhat more frequently, for example every 6 hours, or perhaps every 30 minutes or so, depending on the particularities of the system being controlled.
The MPC function, represented by block
50
in
FIG. 1
, is customarily divided into a steady-state calculation and a dynamic calculation, where each of these phases of MPC may be performed every one to two minutes, for example. The dynamic MPC calculation has been studied extensively (see, e.g., S. J. Qin and T. A. Badgwell, “An Overview of Industrial Model Predictive Control

Affiliated with

Also associated with

Rating

Robust steady-state target calculation for model predictive... does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Robust steady-state target calculation for model predictive..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Robust steady-state target calculation for model predictive... will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2817015