Model error bounds for identification of stochastic models for control design

• US 6,278,898 B1
• Filed: 06/30/1999
• Issued: 08/21/2001
• Est. Priority Date: 04/07/1999
• Status: Expired due to Term

• Current Assignee: Tokyo Electron Limited
• Original Assignee: Voyan Technology Incorporated (HP Inc.)
• Inventors: Shah, Sunil C.
• Primary Examiner(s): Gordon, Paul P.
• Application Number:
  US09/345,640
• Time in Patent Office: 783 Days
• Field of Search: 700/51, 700/97, 700/29, 700/30, 700/31, 700/44, 702/2
• US Class Current: 700/29
• CPC Class Codes: G05B 17/02 electric

Abstract

A method for computing model error bounds for system identification of stochastic systems is disclosed. The model error bounds take the form of additive frequency-weighted singular value bounds such that they are directly used in H_∞ and μ-synthesis robust control design methods. The largest singular value of the additive uncertainty bound is determined by performing a high number of simulations for the model uncertainty. Simulated values of the uncertainty are computed for a large data population, such that each candidate entry of simulated value lies on the 3-sigma ellipsoids defined by the covariance functions. For each simulated value of uncertainty, the maximum singular values are then determined. In order to determine the scalar uncertainty function needed for robust control design, the maximum over the population of the maximum singular values of uncertainty simulated values is then computed.

Claims

1. A computer-implemented method for computing model error bounds of stochastic models for control design, said method comprising:

computing a high order model estimate, wherein said model represents an auto-regressive with exogenous inputs model; and

computing an estimate of the model error bounds, wherein said estimated model error bounds are represented by a singular value, and wherein said computing an estimate of said model error bounds comprises computing the largest singular value of said model error bounds.

2. The method of claim 1, wherein said the estimated model error bounds are represented by an additive, frequency-weighted singular value.

3. A computer-implemented method for computing singular values of model error bounds for a stochastic control system, said method comprising:

computing a plurality of simulated values of said model error for a data population;

determining a plurality of maximum singular values of said plurality of simulated values of said model error; and

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

4. The method of claim 3, wherein said scalar uncertainty function determines the largest singular value of said model error bounds.

5. The method of claim 4, wherein said largest singular value of said model error bounds is a frequency-weighted singular value bound.

6. The method of claim 3, wherein said model error bounds are additive model error bounds.

7. The method of claim 3, wherein each of said plurality of simulated values of said model error lies on a curve defined by a computed covariance of multivariable transfer functions, wherein said multivariable transfer functions describe said stochastic control system.

8. The method of claim 7, wherein said model error lies on 3-sigma ellipsoids curve.

9. A method for computing model error bounds of stochastic models for control design, said method comprising:

computing a high order model estimate, wherein said model represents an auto-regressive with exogenous inputs model; and

computing an estimate of said model error bounds, wherein said estimated model error bounds are represented by a singular value, and wherein said computing an estimate of said model error bounds comprises;

computing a plurality of simulated values of said model error for a data population;

determining a plurality of maximum singular values of said plurality of simulated values of said model error; and

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

10. The method of claim 9, wherein said scalar uncertainty function determines the largest singular value of said model error bounds.

11. The method of claim 10, wherein said largest singular value of said model error bounds is a frequency-weighted singular value bound.

12. The method of claim 9, wherein said model error bounds are additive model error bounds.

13. The method of claim 9, wherein each of said plurality of simulated values of said model error lies on a curve defined by a computed covariance of multivariable transfer functions, wherein said multivariable transfer functions describe said stochastic control system.

14. The method of claim 13, wherein said model error lies on 3-sigma ellipsoids curve.

15. A computer readable medium having stored thereon instructions for computing model error bounds of stochastic models for control design, wherein said instructions, when executed, cause the computer to:

compute a high order model estimate, wherein the model represents an auto-regressive with exogenous inputs model;

compute an estimate of the model error bounds, wherein the estimated model error bounds are represented by a singular value, and wherein said compute an estimate of the model error bounds comprises;

computing a plurality of simulated values of said model error for a data population;

determining a plurality of maximum singular values of said plurality of simulated values of said model error; and

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

16. The method of claim 15, wherein said scalar uncertainty function determines the largest singular value of said model error bounds.

17. The method of claim 16, wherein said largest singular value of said model error bounds is a frequency-weighted singular value bound.

18. The method of claim 15, wherein said model error bounds are additive model error bounds.

19. The method of claim 15, wherein each of said plurality of simulated values of said model error lies on a curve defined by a computed covariance of multivariable transfer functions, wherein said multivariable transfer functions describe said stochastic control system.

20. The method of claim 19, wherein said model error lies on 3-sigma ellipsoids curve.

21. A computer-implemented method for computing model error bounds of a stochastic model for control design, said method comprising:

determining a closed-loop representation of said stochastic model;

performing a closed-loop simulation of said stochastic model to generate input and output data for said stochastic model; and

computing an estimate of the model error bounds, wherein said estimated model error bounds are represented by a singular value.

22. The method of claim 21 further comprising:

computing a high order model estimate using said input and output data, wherein said model represents an auto-regressive with exogenous inputs model, and wherein said computing an estimate of said model error bounds comprises computing the largest singular value of said model error bounds.

23. The method of claim 22, wherein said the estimated model error bounds are represented by an additive, frequency-weighted singular value.

26. The method of claim 21, wherein said computing an estimate of the model error bounds comprises:

computing a plurality of simulated values of said model error for said input and output data;

determining a plurality of maximum singular values of said plurality of simulated values of said model error; and

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

24. The method of 23 further comprises refining an identification trajectory corresponding to said stochastic model using said estimated model error bounds.

25. The method of claim 24, wherein said refining said identification trajectory comprises increasing a spectral energy in a plurality of selected frequency regions.

27. A computer-implemented method for computing model error bounds of a stochastic model for control design, said method comprising:

determining a closed-loop representation of said stochastic model;

performing an identification trajectory refinement corresponding to said stochastic model;

performing a closed-loop simulation of said stochastic model using said refined identification trajectory to generate input and output data for said stochastic model; and

computing an estimate of the model error bounds, wherein said estimated model error bounds are represented by a singular value.

28. The method of claim 27 further comprising:

computing a high order model estimate using said input and output data, wherein said model represents an auto-regressive with exogenous inputs model, and wherein said computing an estimate of said model error bounds comprises computing the largest singular value of said model error bounds.

29. The method of claim 28, wherein said the estimated model error bounds are represented by an additive, frequency-weighted singular value.

30. The method of claim 29, wherein said refining said identification trajectory comprises increasing a spectral energy in a plurality of selected frequency regions.

31. The method of claim 28, wherein said computing an estimate of the model error bounds comprises:

computing a plurality of simulated values of said model error for said input and output data;

determining a plurality of maximum singular values of said plurality of simulated values of said model error; and

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

32. A computer-implemented method comprising:

computing a model error bound for a stochastic model of a control system, wherein said computing a model error bounds comprises computing a singular value of said model error bound; and

performing an identification trajectory refinement of said stochastic model using said model error bound.

33. The method of claim 32 further comprising performing a closed-loop simulation of said stochastic model to determine input and output data of said stochastic model.

34. The method of claim 32 further comprising:

computing a high order model estimate, wherein the model represents an auto-regressive with exogenous inputs model; and

computing an estimate of the model error bounds, wherein the estimated model error bounds are represented by a singular value, and wherein said compute an estimate of the model error bounds comprises;

computing a plurality of simulated values of said model error for a data population;

determining a plurality of maximum singular values of said plurality of simulated values of said model error;

determining a scalar uncertainty function such that said model error is bounded by said scalar uncertainty function, wherein said determining a scalar uncertainty function comprises computing, for said data population, a maximum value of said plurality of maximum singular values.

35. The method of claim 34, wherein said scalar uncertainty function determines the largest singular value of said model error bounds.

36. The method of claim 34, wherein said largest singular value of said model error bounds is a frequency-weighted singular value bound.

37. The method of claim 34, wherein each of said plurality of simulated values of said model error lies on a curve defined by a computed covariance of multivariable transfer functions, wherein said multivariable transfer functions describe said stochastic control system.

38. The method of claim 37, wherein said model error lies on 3-sigma ellipsoids curve.

39. The method of claim 34, wherein performing said identification trajectory refinement comprises an iterative trajectory refinement operation based on said estimated model error bound.

40. The method of claim 32, wherein said performing said identification trajectory refinement comprises increasing a spectral energy in a plurality of selected frequency regions.

41. The method of claim 32, wherein said model error bounds are additive model error bounds.