Receding horizon based adaptive control having means for minimizing operating costs

• US 5,301,101 A
• Filed: 04/14/1993
• Issued: 04/05/1994
• Est. Priority Date: 06/21/1990
• Status: Expired due to Term

• Current Assignee: Honeywell Incorporated (Honeywell International Inc.)
• Original Assignee: Honeywell Incorporated (Honeywell International Inc.)
• Inventors: MacArthur, J. Ward, Woessner, Michael A., Foslien, Wendy K., Wahlstedt, David A.
• Primary Examiner(s): Smith, Jerry
• Assistant Examiner(s): TOUSI, CAMERON
• Application Number:
  US08/047,768
• Time in Patent Office: 356 Days
• Field of Search: 364/156, 364/151, 364/157
• US Class Current: 700/36
• CPC Class Codes: G05B 13/026 using a predictor

Abstract

Disclosed is an adaptive and predictive controller having self tuning capabilities. The receding horizon window and sample interval are determined by the tunning mechanism automatically, based on the relationship between the controlled and manipulated variables. The interval and window may be set by the user as well. Constraints on the manipulated and controlled variables at the end of the horizon window are accommodated automatically. In addition to providing servo regulatory control the invention teaches simultaneous minimization of operating costs.

Claims

1. A controller for controlling manipulated variable inputs to a plant to achieve a desired controlled variable output trajectory which is sensitive to the outputs from said plant wherein said controller receives user input and provides output signals to effect control moves wherein said controller establishes a model indentification based on the manipulated variable inputs and the controlled variable outputs from the plant and comprises:

a tuner for establishing a control sample rate with reference to how long the controlled variable settling time is and on how many controlled and manipulated variable samples are used to identify the model, and for establishing the first m increments of a horizon window which has a total of m+1 increments and

a control law calculator means having a recursive model based predictor to predict the nominal control variable output trajectory that would occur without changes to the manipulated variables, said predictor deriving its model from the indentified model, and having a model based corrector, also based on the same identified model, to specify the changes required of the manipulated variable outputs to satisfy the following criteria;

a) the errors between the controlled variables and their respective setpoints become zero after m control moves,

b) the errors between the controlled variables and their respective setpoints are zero for 1 additional control moves, and

c) the change in the manipulated variables during the last n control moves of the m+1 horizon take on prescribed beta values which are determined by said control law calculator to be those values necessary to smooth the change in manipulated variables after m control moves and reduce the steady state controlled variable errors, and "means for controlling the controlled variable output based on said specified manipulated variable changes."

2. A controller as set forth in claim 1 wherein said controller further comprises:

optimizer calculator means for providing manipulated variable sequences that results in optimal cost performance without compromising servo-regulatory performance having;

access means for using the control law calculator predictor such that the predictor can be used to predict the effect on the plant as would be measured by the outputs of any hypothetical change made in any manipulated variable,

calculator means for determining the optimal set of manipulated variable changes during the horizon window based on the predicted effects of the changes over the horizon window.

3. A controller as set forth in claim 2 wherein said optimizer means establishes an optimal cost value set for the free manipulated variable at the current time increment by iteratively finding the costs for a range of values of the free manipulated variables and selecting the lowest cost subset using the feedforward capability of the control law.

4. A controller as set forth in claim 3 wherein the iterative cost finding function is repeated for each increment within the horizon window and the optimization process is repeated at each control step.

5. A controller as set forth in claim 1 wherein the tuner further comprises a heuristic supervisor means which receives plant output information and manipulated variable information and compares this information on an ongoing basis to detect system instability, and when detected, effects modification of the horizon window to remove the instability.

6. A controller as set forth in claim 5 wherein the determination of instability will provide for one of three alternative actions on the part of the tuner:

a) change the horizon window,

b) change the identified model,

c) initiate a retune.

7. A controller as set forth in claim 1 where the horizon window established by the tuner is based on the inflection point as well as the line-out of the plant output.

8. A controller as set forth in claim 1 wherein the user input may include specification of the horizon window by specifying any subset or all of the following:

one or both of the total lengths of m and l (either by specifying a number of set time length increment units or direct time measurements, such as seconds), the set point, the manipulated variable values in the manual control mode, the control mode (automatic, manual, tune), and a cost function.

9. A multivariable-input multivariable-output controller for controlling a process having at least one receding horizon type controller (hereinafter "RHC") as set forth in claim 1, and at least one other controller, both providing control move output for at least two of said multivariable output variables.

10. A controller as set forth in claim 9 wherein said other controller is also a receding horizon type controller (RHC), and wherein said two variables may be classified as one being primary and the other being secondary and wherein said RHC'"'"'s may be classified as primary optimum and secondary constrained controllers and wherein said primary optimum RHC produces its output at a first iteration assuming the primary variable must make a control move since the secondary variable is optimized and wherein said secondary constrained RHC produces its output assuming a constrained or no control move situation for said primary variable and produces output based on achieving setpoint using the secondary variable.

11. A controller as set forth in claim 10 wherein said primary optimum and said secondary constrained RHC'"'"'s outputs are both fed to a constraint checker which determines when said secondary variable has exceeded one or more of the constraints from the following list of constraints (a) the secondary variable suggested exceeds the range of available control moves for the actuator or (b) moving the secondary variable controlled actuator exceeds the user defined cost function for that actuator.

12. A controller as set forth in claim 11 wherein, in the event the constraint checker determines the secondary variable has exceeded these constraints from said list of constraints which are applicable, said constraint checker signals to said secondary constrained RHC this fact.

13. A multivariable-input multivariable-output controller for controlling a process having a multiplicity of receding horizon type controllers (hereinafter referred to as a "RHC") as set forth in claim 1, each providing control move output for at least one of said multivariable output variables, and each said provided output being checked in a constraint handling unit which provides an indicator signal when a constraint on any control move output or combination thereof is exceeded.

14. A controller as set forth in claim 13 wherein one of said multivariables may be classified as being primary to at least one variable and wherein said RHC'"'"'s may be classified as controlling the primary or one of said secondary variables and wherein said primary RHC produces its output at a first iteration assuming the primary variable must make a control move since any of secondary variables are optimized and wherein said secondary RHC'"'"'s produce their output assuming constrained or no control move situation for said primary variable and produce output based on achieving setpoint using its secondary variable.

15. A controller as set forth in claim 10 wherein when said constraint handler determines said primary variable has exceeded one or more of its constraints, it instructs by sending a signal to said primary RHC to reissues its primary variable output as constrained and also signals at least one of said secondary RHC'"'"'s to recompute, assuming the primary variable is constrained and its variable is to be used to effectuate the reference.

16. A process for regulating a system with a controller which provides inputs to a plant and receives outputs from the plant indicating changes in the plant comprising the steps of:

a) establishing by the controller a horizon window comprised of sampling increments of time such that the sampling interval will be long enough to provide useful control move change periods and short enough to substantially reduce the appearance of unwanted oscillations of the controlled variable trajectory.

a) predicting the nominal controlled variable output trajectory,

c) predicting the controlled output trajectory based on the control moves required to reach a user-defined setpoint trajectory at a given time of m increments within the horizon window,

d) constraining the controlled output trajectory to remain at the user defined output state for l increments beyond said m increments, and constraining the control moves to equal n user-specified betas for the last n increments of the l part of the horizon window.