Extended horizon adaptive block predictive controller with an efficient prediction system

• US 5,424,942 A
• Filed: 08/10/1993
• Issued: 06/13/1995
• Est. Priority Date: 08/10/1993
• Status: Expired due to Fees

• Current Assignee: Orbital Research, Inc.
• Original Assignee: Orbital Research, Inc.
• Inventors: Dong, Yang, Schmidt, Robert N., Khoury, James M., Chizeck, Howard J.
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): Garland, Steven R.
• Application Number:
  US08/104,882
• Time in Patent Office: 672 Days
• Field of Search: 364/148-160, 364/164, 364/165
• US Class Current: 700/44
• CPC Class Codes: G05B 13/026 using a predictor

Abstract

An extended horizon adaptive block predictive controller includes a block control system and an adaptive block prediction system. The subject adaptive block prediction system includes means and procedures for accomplishing a block of k output predictions, consisting of output predictions at each future time step up to the prediction horizon k, for dynamic systems with N outputs, M inputs and model parameter order n. The prediction system is organized for performing the calculation of a k-step ahead predictor in a highly computationally and memory efficient manner. The subject prediction system, in particular, avoids computation of a large number of the predictor parameters associated with the prior art Diophantine equation method, so that memory requirements and computational complexity are significantly reduced. The subject block control system in conjunction with an adaptive block prediction system includes means for accomplishing predictive controls of dynamic systems. It subsumes most, if not all, existing extended horizon predictive controllers. The subject block control system is based on the use of a special computational method. A hardware implementation of the subject extended horizon predictive control system includes a computing means, memory means, and dynamic system input and output signal processing means. The subject invention is primarily devised for real time control of dynamic systems.

Claims

1. A control method including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive :

integers, the control method comprising;

receiving variables including at least one of a set of dynamic system parameters and a set of system outputs y(t) from said multivariable dynamic system;

generating a block of k filtered output z(t) predictions of said multivariable dynamic system by;

selectively filtering the set of system outputs v(t);

constructing a block state structure from the variables of said multivariable dynamic system;

using the block state structure, generating at least one of a block of k output y(t) predictions when the filtering of said set of system outputs y(t) is not performed and a block of k filtered output ψ

(t) predictions when the selective filtering of said set of system outputs y(t) is performed, each of dimension N consisting of output y(t) predictions and filtered output ψ

(t) predictions respectively at each future time step up to a prediction horizon k; and

,

using i) the-block of k output predictions when the selective filtering of said set of system outputs y(t) is not performed and ii) the block of k filtered output ψ

(t) predictions when the selective filtering of said set of system outputs y(t) is performed, to generate said block of k filtered output z(t) predictions;

combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

,

outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.

2. The control method according to claim 1, further comprising the step of removing output measurement noise by sequentially filtering the system input and output signals.

3. The control method according to claim 1, further comprising the step of minimizing an objective function by sequentially filtering the system input and output signals.

4. The control method according to claim 1 wherein the step of generating at least one of a block of k output y(t) predictions and a block of k filtered output ψ

(t) predictions using the block state structure comprises the step of representing the block of k output y(t) predictions and the block of k filtered output ψ

(t) predictions using a block state of dimension N by n, which is independent of the prediction horizon k.

5. The control method according to claim 4 wherein the block state constructing step includes summation of a partial input state and a partial output state of said multivariable dynamic system.

6. The control method according to claim 5, further comprising obtaining a partial output state utilizing N² anti-diagonal symmetric matrices each of dimension n by n, in conjunction with the multivariable dynamic system output data.

7. The control method according to claim 5, further comprising obtaining the partial input state utilizing N by M anti-diagonal symmetric matrices each of dimension n by n, in conjunction with the multivariable dynamic system input data.

8. The control method according to claim 4 wherein the step of generating at least one of a block of k output y(t) predictions and a block of k filtered output ψ

(t) predictions using the block state structure includes the use of dynamic shift Toeplitz matrices.

9. The control method according to claim 1, wherein the block of k output y(t) predictions or k filtered output ψ

(t) predictions is obtained as the sum of product of the block state structure with a block state mapping matrix extracted from the inverse of the lowest order coefficient matrix of an auto-regressive polynomial matrix of a block input output representation of the multivariable dynamic system and product of a block of future k inputs, each of dimension M, with an impulse response matrix.

10. The control method according to claim 9 wherein a step of obtaining a predictor output mapping matrix comprises the use of dynamic shift Toeplitz matrices.

11. The control method according to claim 1, wherein:

said at least one of a block of k output y(t) predictions and k filtered output ψ

(t) predictions is obtained as the product of the block state structure and a block state mapping matrix obtained from the first k terms of the inverse of an auto-regressive polynomial matrix of an input/output representation of the multivariable dynamic system.

12. The control method according to claim 1, wherein a computation of at least one of a block of k output predictions and a block of k filtered ψ

(t) predictions is performed recursively for increasing horizons from 1 through k.

13. The control method according to claim 12, further comprising using a dynamic shift Toeplitz formulation to recursively compute rows of a block state mapping matrix.

14. The control method according to claim 12, further comprising using system identification methods which provide estimated parameters of the multivariable dynamic system based on measurements of input and output signals.

15. The control method according to claim 1 wherein the step of using said k filtered output ψ

(t) predictions to generate said k filtered output z(t) predictions includes filtering said k filtered output ψ

(t) predictions to generate k filtered output z(t) predictions.

16. The control method according to claim 15 further comprising estimating parameters of the multivariable dynamic system based on measurements of input and output signals combined with system identification methods.

17. The control method according to claim 15 wherein each of the steps are performed in real time.

18. The control method according to claim 1 further comprising using the block of output y(t) predictions or filtered output z(t) predictions, in combination with blocks of inputs, outputs and desired outputs to determine a feedback control of a multivariable dynamic system.

19. The control method according to claim 16 wherein each of the steps are performed in real time.

20. The control method according to claim 1 wherein the steps of at least one of said steps of generating y(t) prediction and filtered output z(t) prediction is performed in real time.

21. A system controller, including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive integers comprising:

means adapted for receiving variables from an associated multivariable dynamic system, which variables include at least one of a set of dynamic system parameters and a set of system outputs

predictor means for generating output predictions of said multivariable dynamic system, the predictor means including;

selective filtering means for selectively filtering said set of system outputs y(t);

means for constructing a block state structure from the variables of said multivariable dynamic system;

generating means for generating at least one of a block of k output y(t) predictions when the filtering of said set of system outputs y(t) is not performed and a block of k filtered output ψ

(t) predictions when the selective filtering of said set of system outputs y(t) is performed in accordance with the block state structure, each of dimension N consisting of output y(t) predictions and filtered output ψ

(t) predictions respectively at each future time step up to a prediction horizon k, and

the generating means including means for generating a block of k filtered output z(t) predictions using i) the block of k output y(t) predictions when the selective filtering of said set of system outputs y(t) is not performed and ii) the block of k filtered output ψ

(t) predictions when the selective filtering of said set of system outputs y(t) is performed;

combining means for combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

output means for outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.

22. The system controller of claim 21 further comprising a filter means for filtering output measurement noise by sequentially filtering the system input and output signals.

23. The control system of claim 22 wherein the predictor means includes means for minimizing an objective function by sequentially filtering the system input and output signals.

24. A block predictive controller for control and prediction of an associated multivariable dynamic system approximated as a linear system having N dynamic output signals, M dynamic input signals, and model parameter order n where N, M, and n are finite positive integers, the controller receiving a block of dynamic system outputs and generating a block of dynamic system inputs to effect said control, the controller comprising:

a block predictor system means adapted for receiving measurements of said dynamic system outputs, the block predictor system means including means for combining said measurements with dynamic system parameters to generate a block of predicted filtered system outputs;

a block control system means for determining a block of potential system inputs based upon a block of desired output trajectory and said block of predicted filtered system outputs; and

,

a control implementation unit receiving said block of potential system inputs and generating said block of dynamic system inputs according to a predetermined control algorithm to control said multivariable dynamic system.

25. The block predictive controller according to claim 24 wherein said block predictor system means includes means for generating a block of filtered output z(t) predictions of said multivariable dynamic system as said block of predicted filtered system outputs by:

constructing a block state structure from said dynamic system parameters;

using the block state structure, generating a block of k filtered output ψ

(t) predictions of dimension consisting of filtered output ψ

(t) predictions at each future time step up to a prediction horizon k; and

,

using said block of k filtered output predictions, generating said block of k filtered output z(t) predictions.

26. An adaptive block predictive controller for control and prediction of an associated multivariable dynamic system approximated as a linear system having N dynamic output signals, M dynamic input signals, and model parameter order n where N, M, and n are finite positive integers, the controller receiving a block of dynamic system outputs and generating a block of dynamic system inputs to effect said control, the controller comprising:

identifier means for identifying parameters of said multivariable dynamic system and generating a block of identified parameters;

a block predictor system means adapted for receiving measurements of said dynamic system outputs, the block predictor system means including means for combining said measurements with a one of dynamic system parameters from said multivariable dynamic system and said block of identified parameters to generate a block of predicted filtered system outputs;

a block control system means for determining a block of potential system inputs based upon a block of desired output trajectory and said block of predicted filtered system outputs; and

,

a control implementation unit receiving said block of potential system inputs and generating said block of dynamic system inputs according to a predetermined control algorithm to control said multivariable dynamic system.

27. The adaptive block predictive controller according to claim 26 wherein said a block predictor system means includes means for generating a block of filtered output z predictions of said multivariable dynamic system as said block of predicted filtered system outputs by:

constructing a block state structure from said dynamic system parameters and said block of identified parameters;

using the block state structure, generating a block of k filtered output ψ

(t) predictions of dimension N consisting of filtered output ψ

(t) predictions at each future time step up to a prediction horizon k; and

,

using a block of k filtered output ψ

(t) predictions, generating a block of k filtered output z(t) predictions.

28. A system output prediction method in an apparatus for predicting at least one of the filtered outputs z(t) of a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive integers, the control method comprising:

receiving variables including a set of dynamic system parameters and a set of system outputs from said multivariable dynamic system;

generating a block of k filtered output z(t) predictions of said multivariable dynamic system by;

constructing a block state structure from the variables of said multivariable dynamic system;

using the block state structure, generating a block of k filtered output ψ

(t) predictions of dimension N consisting of filtered output ψ

(t) predictions at each future time step up to a prediction horizon k; and

,

using a block of k filtered output ψ

(t) predictions, generating said block of k filtered output z(t) predictions.

29. A control method including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive integers, the control method comprising:

receiving variables including at least one of a set of dynamic system parameters and a set of system outputs y(t) from said multivariable dynamic system;

generating a block of k filtered output z(t) predictions of said multivariable dynamic system by;

constructing a block state structure from the variables of said multivariable dynamic system;

using the block state structure, generating a block of k output y(t) predictions of dimension N consisting of output y(t) predictions at each future time step up to a prediction horizon k; and

,

filtering the block of k output y(t) predictions to generate said block of k filtered output z(t) predictions;

combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

,

outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.

30. A control method including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive integers, the control method comprising:

receiving variables including at least one of a set of dynamic system parameters and a set of system outputs y(t) from said multivariable dynamic system;

generating a block of k filtered output z(t) predictions of said multivariable dynamic system by;

filtering the set of system outputs y(t);

constructing a block state structure from the variables of said multivariable dynamic system;

using the block state structure, generating a block of k filtered output ψ

(t) predictions of dimension N consisting of filtered output ψ

(t) predictions at each future time step up to a prediction horizon k; and

,

filtering the block of k filtered output ψ

(t) predictions to generate said block of k filtered output z(t) predictions;

combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

,

outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.

31. A control method including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n where N, M, and n are finite positive integers, the control method comprising:

receiving variables including at least one of a set of dynamic system parameters and a set of system outputs y(t) from said multivariable dynamic system;

generating a block of k filtered output z(t) predictions of said multivariable dynamic system by;

filtering the set of system out,puts y(t) using a first set of filtering polynomials;

constructing a block state structure from the variables of said multivariable dynamic system, the variables including the filtered set of system outputs y(t);

using the block state structure, generating a block of k filtered output ψ

(t) predictions of dimension N consisting of filtered output ψ

(t) predictions at each future time step up to a prediction horizon k; and

,

filtering the block of k filtered output ψ

(t) predictions to generate said block of k filtered output z(t) predictions;

combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

,

outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.

32. The control method according to claim 31 wherein the step of filtering the set of system outputs y(t) includes filtering the set of system outputs y(t) using an identity matrix.

33. The control method according to claim 31 wherein the step of filtering the set of system outputs y(t) includes filtering the set of system outputs y(t) using a first filtering matrix other than an identity matrix.

34. A system controller, including system output prediction in an apparatus effecting control over a multivariable dynamic system approximated as a linear system having N output signals, M input signals, and model parameter order n whence N, M, and n are finite positive integers comprising:

means adapted for receiving variables from an associated multivariable dynamic system, which variables include at least one of a set of dynamic system parameters and a set of system outputs y(t);

predictor means for generating output predictions of said multivariable dynamic system, the predictor means including;

means for constructing a block state structure from the variables of said multivariable dynamic system;

generating means for generating a block of k filtered output ψ

(t) predictions in accordance with the block state structure, the block of k filtered output ψ

(t) predictions being of dimension N and comprising filtered output ψ

(t) predictions at each future time step up to a prediction horizon k, and

the generating means including means for generating a block of k filtered output z(t) predictions using the block of k filtered output ψ

(t) predictions;

combining means for combining said block of k filtered output z(t) predictions with a desired output trajectory to generate block control signals; and

,

output means for outputting said block control signals as input signals to said multivariable dynamic system to effect said control over the system.