1 / 57

GUARANTED SET COMPUTATION Ellipsoidal approach

GUARANTED SET COMPUTATION Ellipsoidal approach. S. Lesecq GIPSA-Lab, Département Automatique INPG-UJF-CNRS Suzanne.lesecq@gipsa-lab.inpg.fr. Motivation Notations and definitions Outer bounding ellipsoid Algorithm(s) Parameterised family Simple example Factorisation.

espen
Download Presentation

GUARANTED SET COMPUTATION Ellipsoidal approach

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. GUARANTED SET COMPUTATIONEllipsoidal approach S. Lesecq GIPSA-Lab, Département Automatique INPG-UJF-CNRS Suzanne.lesecq@gipsa-lab.inpg.fr

  2. Motivation Notations and definitions Outer bounding ellipsoid Algorithm(s) Parameterised family Simple example Factorisation Convergence conditions Common difficulties Choice of the “bound” Industrial context: input? Extra Inner bounding Uncertainty in the model References Outline

  3. Motivation - 1 • Discrete-time dynamic system • Observation ykmeasured output vector uk known input vector unknown: state, process perturbation, measurement noise • Identification (output linear in the parameters)

  4. Define two parallel hyperplanes S3 S1 2 2 Feasible set Compute a simpler set enclosing S S2 S dk 1 1 Motivation - 2 • Consider a SISO model dk : regressor,  : parameter vector Error bound ek

  5. 2 S S 1 Motivation - 3 • Possible simpler sets • Polytopes • Parallelotopes • Orthotopes • Ellipsoids… • Outer bounding ellipsoid Mininum size • Inner bounding ellipsoid Maximum size 2  Criterion to be optimised 1

  6. det trace det trace Motivation - 4 • Usual criteria • “Determinant criterion” (actually log(det)…) • “Trace criterion”

  7. Notations & definitions - 1 • Norm : Euclidian • Unit ball centred on the origin: • Bounded ellipsoid E(c,P) with non empty interior c: centre P = PT > 0 : shape and orientation

  8. 2 1 Notations & definitions - 2 • Particular cases • Strip = unbounded ellipsoid • Empty interior ellipsoid Centre c not unique, just satisfy: y = Cc 2 1

  9. Notations & definitions - 3 • Sum of K ellipsoids (prediction) • Intersection of K ellipsoids (correction)

  10. Motivation Notations and definitions Outer bounding ellipsoid Algorithm(s) Parameterised family Simple example Factorisation Examples : Identification Convergence conditions Common difficulties Choice of the “bound” Industrial context: input? Extra Inner bounding Uncertainty in the model References Outline

  11. 2 Sk Ek-1 1 Ek Weighted sum of Ek-1 and Sk Outer bounding ellipsoid - 1 • Recursive algorithm (Fogel and Huang, 1982) not normalised form (intersection): P = PT > 0

  12. Outer bounding ellipsoid - 2 • OBE algorithm Choice ofk, k ?

  13.  No update Outer bounding ellipsoid - 3 • “Basic OBE algorithm not optimal” (Belforte and Bona, 1985 and 1990) • Void intersection  must be detected  add intersection test (e.g. Pronzato and Walter, 1994) • Other classical situation

  14. l2 l1 Outer bounding ellipsoid - 4 • Usual criteria • In the litterature, two sets of programs (Favier and Arruda, 1996; Tran Dinh, 2005) • Set 1: minimize the geometrical size of Ek • Set 2 : insure the convergence of  2

  15. Fogel and Huang, 1982, (FH) Parametrised family (Durieu, et al., 1996) 2 families  Set Membership Set Approximation SMSA (Nayeri, et al., 1994) Same criterion (tr or det)  same ellipsoids Outer bounding ellipsoid - 5 • Set 1 (size)

  16. Dasgupta and Huang, 1987 (DH) If then Criterion: If then Lozano-Leal and Ortega, 1987 (LO) Criterion: If then Tan, et al., 1997 (TAN) Criterion : Outer bounding ellipsoid - 6 • Set 2

  17. Outer bounding – 7 (parameterised family) • Normalised problem (Durieu, et al., 2001) • MIMO models • State estimation (identification)  Sum, Intersection • Analytical results K ellipsoids • Optimisation problem addressed in details Empty interior Unbounded (strip)

  18. Outer bounding – 8 (parameterised family) • Sum of K ellipsoids (Durieu, et al., 2001) Problem: find • Theorem 4.1 The centre of the optimal ellipsoid E* for both problems is given by: Empty interior possible

  19. Outer bounding – 9 (parameterised family) Parameterized family  optimisation can be done • Theorem 4.2 • c*: independent of  • Solve for  the problem:  generally, suboptimal solution of:

  20. Outer bounding – 10 (parameterised family) • Necessary condition (Lemma 4.1) for E* to be optimal solution of • Theorem 4.4: Trace criterion  explicit solution • Theorem 4.5: recursive  nonrecursive approximating ellipsoid • Determinant criterion  no explicit solution

  21. Unbounded ellipsoid possible Outer bounding – 11 (parameterised family) • Intersection of K ellipsoids (Durieu, et al., 2001) Problem: find

  22. Outer bounding – 12 (parameterised family) • Theorem 5.1 • Proposition 5.1

  23. x2 H1 H2 S H3 x1 Outer bounding ellipsoid - 13 • Ellipsoid with parallel cuts algorithm (Goldfarb and Todd, 1982) • Sequential algorithm • Equivalent to (modified) OBE (Pronzato and Walter, 1994) • “Recursively optimal”  i.e. minimal volume ellipsoid containing E(ck-1,Pk-1)  Bk x2 Hk Ek Ek-1 x1

  24. uk yk Outer bounding ellipsoid - (example) • Simulated data ykmeasured output vector unknown: measurement noise (uniform distribution) SNR  50 dB • Identification

  25. 2 c 1  = * = 0.003 Outer bounding ellipsoid - (example) • Results: xtheoretic = [0.95, 0.05] c0 = 0, 02P = 106I Set 1 determinant Set 2

  26. Evolution of Det (k2P) LO = Lozano-Leal and Ortega, 1987 TAN = Tan, et al., 1997 DH = Dasgupta and Huang, 1987 FH = Fogel and Huang, 1982 Evolution of k2 1 4 2 3 3 2 4 k 1 k Outer bounding ellipsoid - (example)

  27. Outer bounding ellipsoid - (example) • Parameter a evolution Figure 2.4 : Evolution des paramètres estimés.

  28. Outer bounding ellipsoid - (example) • Parameter b evolution

  29. Outer bounding ellipsoid : summary • Two sets of algorithms (Favier and Arruda, 1996; Tran Dinh, 2005): • Minimise the geometrical size of E • Minimise 2 • Different formulations of the algorithm • Equivalence provable for some of them (OBE-EPC…) (Pronzato and Walter, 1994, Tran Dinh, 2005) • (Durieu, et al., 1996 and 2001): • Parameterised family • K ellipsoids • Convexity of criteria

  30. Motivation Notations and definitions Outer bounding ellipsoid Algorithm(s) Parameterised family Simple example Factorisation Examples : Identification Convergence conditions Common difficulties Choice of the “bound” Industrial context: input? Extra Inner bounding Uncertainty in the model References Outline

  31. M L LT = U 0 A H = Factorisation - 1 • Prerequisite • Orthogonal matrix: HT = H-1 • Numerically highly suitable • Orthogonal factorisation • Factorisation of a product of matrices

  32. Factorisation - 2 • Factorisation of a sum of matrices • Least Square problem solution • Recursive Least Square problem: also factorised…

  33. Factorisation - 3 • Parameterised family (Durieu, et al., 1996) hypothesis (not restrictive): y R Sum of 2 symmetric matrices  factorise! Intersection • Goldfarb and Todd, 1982 • LDL factorisation • Cholesky suggested

  34. LS: = XT M X Factorisation - 4 • Reformulation  optimisation problem (Lesecq and Barraud, 2002) Let then

  35.  = s2 > 0 ck, Mk: independent Factorisation - 5 • Factorised algorithm (Lesecq and Barraud, 2002) • Theoretical property   [0, 1[ : simpler demonstration

  36. = P Factorisation - 6 • Directly with P (general formulation) (Tran Dinh, 2005) ck, Mk: dependent

  37. Factorisation - summary • Absolutely necessary to ensure numerical stability • Academic example (Lesecq and Barraud, 2002) n = 8, M = hilb(8)=[hij = 1/(i+j-1)], c = ones (8,1), y = 1 temp = invhilb(9), d = temp(1:8,9)  = 0.001  not factorised = - 47.6 and factirised = 1.7 10-2 • Practical problem (identification) • Theoretical properties: easier demonstration • Parameterised family (Durieu, et al., 1996)  P and M algorithms (Lesecq and Barraud, 2002) • General formulation  P and M algorithms (Tran Dinh, 2005)

  38. Motivation Notations and definitions Outer bounding ellipsoid Algorithm(s) Parameterised family Simple example Factorisation Examples:Identification Convergence conditions Common difficulties Choice of the “bound” Industrial context: input? Extra Inner bounding Uncertainty in the model References Outline

  39. Examples - 1 • Industrial Data: • 1st example: industrial Looks like 1st order, 2 parameters aim: model identification  diagnosis • 2nd example: LIRMM robot 14 parameters 14 000 regressors! aim: model identification, large problem

  40. Examples - 2 • 1st example: recorded on a process (valve) output input

  41. No ellipsoid update Examples - 3 • 1st example • Data re-used several times • Determinant criterion •  = 0.002 • Measurement and regressor known Determinant criterion 1st circulation of data

  42. 2 c 1 Examples - 4 Ellipsoid updating • 1st example: parameters

  43. Examples - 5 • 1st example det(10) det(1) Gain properly identified Trace(10) Trace(1)

  44. Examples - 7 • 2nd example: LIRMM parallel robot N = 3500

  45. Examples – 8 • 2nd example: Recorded data (for instance) Sequential algorithm

  46. criterion Examples - 9 No empty intersection  = 6 Nm • 2nd example: parameters (60 circulations)

  47. Parallel robot, det, family of ellipsoids Factorisation interest No factorisation Thanks to N. Ramdani Work done by N. Ramdani P. Poignet LIRMM Factorisation

  48. Examples - 10 • 2nd example: Model reconsidered  split in several models  = 6 Nm 1st model

  49. Examples - 11 Center, obtained with  = 6 Nm and 60 circulations • 2nd example: Adaptation of the bound Far from hypothesis! • Analysis of data • Reject “outliers” • “heavy tail”

  50. Examples - 12 60 circulations • 2nd example: Adaptation of the bound:  “heavy tail”

More Related