Robust steady-state target calculation for model predictive control
First Claim
1. A method of controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, G represents system gain parameters, and where J is a function of G, said method comprising the steps of;
(a) modeling said system according to an equation Δ
y={tilde over (G)}Δ
u, where {tilde over (G)} is a nominal estimate of G and has a known uncertainty description and where {tilde over (J)} is said objective function J applied to {tilde over (G)};
(b) computing steady-state targets for said system-manipulated variables u including extremizing said objective function J cast as a quadratic program, such that all of said system-controlled variables will remain feasible at steady-state for all possible values of the system parameters within said known uncertainty description; and
(c) using the computed steady-state targets for system-manipulated variables u to control the system.
4 Assignments
0 Petitions
Accused Products
Abstract
A method and apparatus for steady-state target calculation that explicitly accounts for model uncertainty is disclosed. In accordance with one aspect of the invention, when model uncertainty is incorporated, the linear program associated with the steady-state target calculation can be recast as a highly structured nonlinear program. In accordance with another aspect of the invention, primal-dual interior point methods can be applied to take advantage of the resulting special structure. For a system having characteristic gain parameters G having a known uncertainty description, the present invention provides a method and apparatus for selecting steady-state targets for said system-manipulated variables such that all system-controlled variables will remain feasible at steady-state for all possible values of the parameters G within the known uncertainty description. A nominal estimate {tilde over (G)} of the system parameters G is made, and in accordance with another aspect of the invention, the steady-state targets are selected such that when {tilde over (G)}=G, the system is driven to an operational steady-state in which the objective function is extremized.
-
Citations
6 Claims
-
1. A method of controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, G represents system gain parameters, and where J is a function of G, said method comprising the steps of;(a) modeling said system according to an equation Δ
y={tilde over (G)}Δ
u, where {tilde over (G)} is a nominal estimate of G and has a known uncertainty description and where {tilde over (J)} is said objective function J applied to {tilde over (G)};
(b) computing steady-state targets for said system-manipulated variables u including extremizing said objective function J cast as a quadratic program, such that all of said system-controlled variables will remain feasible at steady-state for all possible values of the system parameters within said known uncertainty description; and
(c) using the computed steady-state targets for system-manipulated variables u to control the system.
- y=GΔ
-
2. A method of controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, and G represents system gain parameters, said method comprising the steps of;(a) modeling said system according to an equation Δ
y={tilde over (G)}Δ
u, where {tilde over (G)} is a nominal estimate of G and has a known uncertainty description; and
(b) extremizing said objective function J by solving the quadratic program
wherec represents weights on said input variables u in the objective function J, d represents such weights on said output variables y in the objective function J, d represents such weights on said output variables y in the objective function, ε
=└
ε
1 . . . ε
n,┘
T≧
0∈
ny is a slack variable that allows for violation in constraints on output variables y,Au∈
4nuxnu and Ay∈
2nyxny are matrices corresponding to desired ranges of values of said system manipulated and system-controlled variables,bu∈
4nu and by∈
2ny are vectors corresponding to desired ranges of values of said system manipulated and system-controlled variables,U represents said uncertainty description, Q0∈
Rnuxnu is a positive-definite symmetric (or possibly zero) matrix that allows for quadratic tradeoffs in system manipulated variables, andQ1∈
Rnyxny is a positive-definite symmetric (or possibly zero) matrix that allows for quadratic tradeoffs in slack variables; and
(c) using a solution obtained in (b), controlling said system.
- y=GΔ
-
3. An apparatus for controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, and G represents system gain parameters having a known uncertainty description, and where J is a function of G, said apparatus comprising;(a) an estimator for deriving a nominal estimate {tilde over (G)} of said system gain parameters G, where {tilde over (G)} has a known uncertainty description and where {tilde over (J)} is said objective function J applied to {tilde over (G)}; and
(b) a computing circuit, responsive to at least one past value of said system-controlled output variables and of said system-manipulated input variables to compute steady-state targets for said system-manipulated variables u including extremizing said objective function J cast as a quadratic program, such that all of said system-controlled variables will remain feasible at steady-state for all possible values of the system parameters within said known uncertainty description, the steady-state targets computed by the computing being used to control the system.
- y=GΔ
-
4. An apparatus controlling a system in accordance with an objective function J, wherein the operation of said system is describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, and G represents system gain parameters having a known uncertainty description, said apparatus comprising;(a) estimator circuitry for deriving a nominal estimate {tilde over (G)} of system gain parameters G; and
(b) computation circuitry, responsive to at least one past value of said system-controlled output variables and of said system-manipulated input variables, and responsive to said nominal estimate {tilde over (G)}, for extremizing said objective function J by solving the equation
wherec represents weights on said input variables u in the objective function J, d represents such weights on said output variables y in the objective function, ε
=└
ε
1 . . . ε
n,┘
T≧
0∈
ny is a slack variable that allows for violation in constraints on output variables y,Au∈
4nuxnu and Ay∈
2n,xny are matrices corresponding to desired ranges of values of said system manipulated and system-controlled variables,bu∈
4nu and by∈
2ny are vectors corresponding to desired ranges of values of said system manipulated and system-controlled variables,U represents said uncertainty description, Q0∈
Rnuxnu is a positive-definite symmetric (or possibly zero) matrix that allows for quadratic tradeoffs in system manipulated variables, andQ1∈
Rnyxny is a positive-definite symmetric (or possibly zero) matrix that allows for quadratic tradeoffs in slack variables such that said computation circuitry provides target values for system manipulated input variables for controlling said system.
- y=GΔ
-
5. An apparatus for controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
- y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, and G represents system gain parameters having a known uncertainty description, and where J is a function of G, said apparatus comprising;a) estimator means for deriving a nominal estimate {tilde over (G)} of said system gain parameters G, where {tilde over (J)} is said objective function applied to {tilde over (G)}; and
b) computing means, responsive to at least one past value of said system-controlled output variables and of said system-manipulated input variables to compute steady-state targets for said system-manipulated input variables u including extremizing said objective function J cast as a quadratic program, such that all of said system-controlled variables will remain feasible at steady-state for all possible values of the system parameters within said known uncertainty description, the computed steady-state targets being used to control the system.
- y=GΔ
-
6. A computer program product comprising:
-
a computer usable medium for controlling a system in accordance with an objective function J, the operation of said system being describable by an equation Δ
y=GΔ
u, where y represents one or more system-controlled output variables, u represents one or more system-manipulated input variables, G represents system gain parameters, and where J is a function of G; and
a set of computer program instructions embodied on the computer usable medium, including instructions to;
model said system according to an equation Δ
y={tilde over (G)}Δ
u, where {tilde over (G)} is a nominal estimate of G and has a known uncertainty description and where {tilde over (J)} is said objective function J applied to {tilde over (G)}; and
compute steady-state targets for said system-manipulated variables u including extremizing said objective function J cast as a quadratic program, such that all of said system-controlled variables will remain feasible at steady-state for all possible values of the system parameters within said known uncertainty description; and
use the computed steady-state targets to control said system.
-
Specification