robust multi-objective control for linear systems elements
TRANSCRIPT
Robust Multi-Objective Control for Linear Systems
Elements of theory and ROMULOC toolbox
OLOCEP project
Denis Arzelier & Didier Henrion & Jean Lasserre & Dimitri Peaucelle
LAAS-CNRS, Toulouse, FRANCE
CNRS-RAS cooperation
Outline
I - Uncertain LTI systems and performances
➞ Control objectives: stability, transient response, perturbation rejection...
➞ Structured parametric uncertainties: extremal values, bounded sets...
II - LMIs and convex polynomial-time optimization
➞ Semi-Definite Programming and LMIs
➞ SDP solvers and parsers
III - Conservative LMI results
➞ Methods: Lyapunov, S-procedure, Finsler lemma, Topologic Separation...
➞ The ROMULOC toolbox
Seminar at IPME 1 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ia - Control objectives
➞ Stability & D-stability
sx(t) = Ax(t) + Buu(t)
y(t) = Cyx(t) + Dyuu(t):
x ∈ Cn u ∈ Cqu
y ∈ Cpy
Design an LTI control u = Ky s.t. poles belong to a region of the complex plane
(or for given K analyse pole location):
DR = { s ∈ C : r11 + sr12 + s∗r12 + ss∗r22 ≤ 0 } , R = (rij)
Such regions are half-planes and discs
R =
0 1
1 0
R =
−1 0
0 1
R =
−2α cosψ cosψ − i sinψ
cosψ + i sinψ 0
ψ
α
Seminar at IPME 2 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ia - Control objectives
➞ H∞ & H2 & Impulse-to-peak performance
sx(t) = Ax(t) + Bww(t) + Buu(t)
z(t) = Czx(t) + Dzww(t) + Dzuu(t)
y(t) = Cyx(t) + Dyww(t) + Dyuu(t)
:w ∈ Cqw
z ∈ Cpz
H∞ performance : ‖z‖ ≤ γ∞‖w‖ ; bounded-real lemma
H2 performance : ‖z‖ ≤ γ2 for w g.w.n. ; max |z| ≤ γ2‖w‖
Impulse-to-peak performance : max |z| ≤ γi2p for w = δ.
Design an LTI control u = Ky s.t. given specification γ is fulfilled
or that minimizes γ
(or for given K analyse performance level)
Seminar at IPME 3 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ib - Robust Multi-Objective Control
∆ : errors in modeling, operating conditions, mass-production...
∆ : parametric uncertainty, assumed constant, belongs to a set ∆∆.
sx(t) = A(∆)x(t) + Bw(∆)w(t) + Bu(∆)u(t)
z(t) = Cz(∆)x(t) + Dzw(∆)w(t) + Dzu(∆)u(t)
y(t) = Cy(∆)x(t) + Dyw(∆)w(t) + Dyu(∆)u(t)
Find controllerK that fulfills robust specifications Πi defined for models Σi(∆i) with ∆i ∈ ∆∆i.
K
Σ2 F2(∆ )2
K
1(∆ )1
K
1Σ (0)1FΣ
Seminar at IPME 4 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ic - Structured uncertainties
➞ Polytopic models zw
K
u y< γ
[2]Σ
Σ[1]Σ[N]
Σ(∆)
✪ Affine polytopic models : convex hull of N vertices
A(∆) =∑ζiA
[i] , Bw(∆) =∑ζiB
[i]w . . . : ζi ≥ 0 ,
∑ζi = 1
Seminar at IPME 5 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ic - Structured uncertainties
➞ Polytopic models zw
K
u y< γ
[2]Σ
Σ[1]Σ[N]
Σ(∆)
✪ Affine polytopic models : convex hull of N vertices
A(∆) =∑ζiA
[i] , Bw(∆) =∑ζiB
[i]w . . . : ζi ≥ 0 ,
∑ζi = 1
➥ Parallelotopic models with NP axes
A(∆) = A[0] +∑ξiA
[i] , Bw(∆) = B[0]w +
∑ξiB
[i]w . . . : |ξi| ≤ 1
➾ polytope with N = 2NP vertices
➥ Interval models with NI non equal coefficients
A[1] � A(∆) � A[2] : a[1]ij ≤ aij(∆) ≤ a
[2]ij
➾ parallelotope with axes in the euclidian basis of matrices
➾ polytope with N = 2NI vertices
Seminar at IPME 6 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ic - Structured uncertainties
➞ LFT modelszΣ
∆w z
K
yu< γ
w
∆ ∆
sx(t) = Ax(t) + B∆w∆(t) + Bww(t) + Buu(t)
z∆(t) = C∆x(t) + D∆∆w∆(t) + D∆ww(t) + D∆uu(t)
z(t) = Czx(t) + Dz∆w∆(t) + Dzww(t) + Dzuu(t)
y(t) = Cyx(t) + Dy∆w∆(t) + Dyww(t) + Dyuu(t)
:w∆ ∈ Cq∆
z∆ ∈ Cp∆
Linear - Fractional Transformation:
A(∆) = A+B∆∆(I−D∆∆∆)−1C∆ , Bw(∆) = Bw+B∆∆(I−D∆∆∆)−1D∆w . . .
Any model rational in δi parameters ➾ LFT (not unique) with diagonal ∆ =diag(δ1, δ1, ..., δ2, ...).
Seminar at IPME 7 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ic - Structured uncertainties
➞ Uncertainty sets
✪ {X,Y, Z}−dissipative matrices{∆ ∈ Cqw×pz : X + Y∆ + ∆∗Y ∗ + ∆∗Z∆ ≤ O , X ≤ O , Z ≥ O
}➥ Norm-bounded uncertainties : ‖∆‖ ≤ ρ1 ➾ {−ρ2I,O, I}−dissipative
➥ Positive real uncertainties : ∆ + ∆∗ ≥ O (eg. s) ➾ {O,−I,O}−dissipative
Seminar at IPME 8 March 21st, 2005, St. Petersburg
I - Uncertain LTI systems and performances
Ic - Structured uncertainties
➞ Uncertainty sets
✪ {X,Y, Z}−dissipative matrices{∆ ∈ Cqw×pz : X + Y∆ + ∆∗Y ∗ + ∆∗Z∆ ≤ O , X ≤ O , Z ≥ O
}➥ Norm-bounded uncertainties : ‖∆‖ ≤ ρ1 ➾ {−ρ2I,O, I}−dissipative
➥ Positive real uncertainties : ∆ + ∆∗ ≥ O (eg. s) ➾ {O,−I,O}−dissipative
✪ Polytopic uncertainties
polytope N vertices
{∆ =
∑ζi∆[i] : ζi ≥ 0 ,
∑ζi = 1
}➥ Parallelotopic uncertainties
NP axes ➾ polytope N = 2NP
{∆ = ∆[0] +
∑ξi∆[i] : |ξi| ≤ 1
}➥ Interval uncertainties
NI coef. 6= ➾ polytope N = 2NI
{∆[1] � ∆ � ∆[2] : δ
[1]ij ≤ δij ≤ δ
[2]ij
}
Seminar at IPME 9 March 21st, 2005, St. Petersburg
Outline
I - Uncertain LTI systems and performances
➞ Control objectives: stability, transient response, perturbation rejection...
➞ Structured parametric uncertainties: extremal values, bounded sets...
zw
K
u y< γ
[2]Σ
Σ[1]Σ[N]
Σ(∆)zΣ
∆w z
K
yu< γ
w
∆ ∆
II - LMIs and convex polynomial-time optimization
➞ Semi-Definite Programming and LMIs
➞ SDP solvers and parsers
III - Conservative LMI results
➞ Methods: Lyapunov, S-procedure, Finsler lemma, Topologic Separation...
➞ The ROMULOC toolbox
Seminar at IPME 10 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIa - Semi-Definite Programming and LMIs
✪ Extension of LP to semi-definite matrices
min cx : Ax = b , xi ≥ 0 (LP ) | mat(x) ≥ O (SDP )
➥ Convexity, duality, polynomial-time algorithms (O(n6.5 log(1/ε))).
max bT y : AT y − cT = z , mat(z) ≥ O
✪ 1st developments and 1st results : LMI formalism & Control Theory
min∑
giyi : F0 +∑
Fiyi ≥ O
➥ The H∞ norm computation example for G(s) ∼ (A,B,C,D) :
‖G(s)‖2∞ = min γ : P > O ,
ATP + PA+ CTz Cz BwP + CT
z Dzw
PBTw +DT
zwCz −γI +DTzwDzw
≤ O
Seminar at IPME 11 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIb - SDP solvers and parsers
➞ LMI Control Toolbox ➾ Control Toolbox
1st solver, dedicated to LMIs issued from Control Theory, Matlab, owner.
➞ SDP solvers: SP, SeDuMi, SDPT3, CSDP, DSDP, SDPA...
Active field, mathematical programing, C/C++, free.
➞ Parsers: tklmitool, sdpsol, SeDuMiInterface, YALMIP
Convert LMIs to SDP solver format, Matlab (Scilab), free.
Seminar at IPME 12 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIc - SDP-LMI issues and prospectives
✪ Any SDP representable problem is ”solved” (numerical problems due to size and structure)
➥ Find ”SDP-ables” problems
(linear systems, performances, robustness, LPV, saturations, delays, singular systems...)
➥ Equivalent SDP formulations ➾ distinguish which are numerically efficient
➥ New SDP solvers: faster, precise, robust (need for benchmark examples)
Seminar at IPME 13 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIc - SDP-LMI issues and prospectives
✪ Any SDP representable problem is ”solved” (numerical problems due to size and structure)
➥ Find ”SDP-ables” problems
(linear systems, performances, robustness, LPV, saturations, delays, singular systems...)
➥ Equivalent SDP formulations ➾ distinguish which are numerically efficient
➥ New SDP solvers: faster, precise, robust (need for benchmark examples)
✪ Any ”SDP-able” problem has a dual interpretation
➥ New theoretical results, new proofs (Lyapunov functions = Lagrange multipliers)
➥ SDP formulas numerically stable (KYP-lemma)
Seminar at IPME 14 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIc - SDP-LMI issues and prospectives
✪ Any SDP representable problem is ”solved” (numerical problems due to size and structure)
➥ Find ”SDP-ables” problems
(linear systems, performances, robustness, LPV, saturations, delays, singular systems...)
➥ Equivalent SDP formulations ➾ distinguish which are numerically efficient
➥ New SDP solvers: faster, precise, robust (need for benchmark examples)
✪ Any ”SDP-able” problem has a dual interpretation
➥ New theoretical results, new proofs (Lyapunov functions = Lagrange multipliers)
➥ SDP formulas numerically stable (KYP-lemma)
✪ Non ”SDP-able” : Robustesse & Multi-objective & Relaxation of NP-hard problems
➥ Optimistic / Pessimistic (conservative) results
➥ Reduce the gap (upper/lower bounds) while handling numerical complexity growth.
Seminar at IPME 15 March 21st, 2005, St. Petersburg
II - LMIs and convex polynomial-time optimization
IIc - SDP-LMI issues and prospectives
✪ Any SDP representable problem is ”solved” (numerical problems due to size and structure)
➥ Find ”SDP-ables” problems
(linear systems, performances, robustness, LPV, saturations, delays, singular systems...)
➥ Equivalent SDP formulations ➾ distinguish which are numerically efficient
➥ New SDP solvers: faster, precise, robust (need for benchmark examples)
✪ Any ”SDP-able” problem has a dual interpretation
➥ New theoretical results, new proofs (Lyapunov functions = Lagrange multipliers)
➥ SDP formulas numerically stable (KYP-lemma)
✪ Non ”SDP-able” : Robustesse & Multi-objective & Relaxation of NP-hard problems
➥ Optimistic / Pessimistic (conservative) results
➥ Reduce the gap (upper/lower bounds) while handling numerical complexity growth.
✪ Develop software for ”industrial” application / adapted to the application field
➾ ROMULOC toolbox
Seminar at IPME 16 March 21st, 2005, St. Petersburg
Outline
I - Uncertain LTI systems and performances
➞ Control objectives: stability, transient response, perturbation rejection...
➞ Structured parametric uncertainties: extremal values, bounded sets...
zw
K
u y< γ
[2]Σ
Σ[1]Σ[N]
Σ(∆)zΣ
∆w z
K
yu< γ
w
∆ ∆
II - LMIs and convex polynomial-time optimization
➞ Semi-Definite Programming and LMIs
➞ SDP solvers and parsers ➾ YALMIP and all solvers
III - Conservative LMI results
➞ Methods: Lyapunov, S-procedure, Finsler lemma, Topologic Separation...
➞ The ROMULOC toolbox
Seminar at IPME 17 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIa - Nominal performance analysis V (x) = xTPx Lyapunov function (P > O)
✪ Stability ATP + PA < O | ATPA− P < O
✪ D-Stability[
I A∗] r11P r12P
r∗12P r22P
I
A
< O
✪H∞ norm
ATP + PA+ CTz Cz PBw + CT
z Dzw
BTwP +DT
zwCz −γ2I +DTzwDzw
< O
✪H2 normATP + PA+ CT
z Cz < O
trace(BTwPBw) < γ2
✪ Impulsion-to-peakATP + PA < O BT
wPBw < γ2I
CTz Cz < P DT
zwDzw < γ2I
Seminar at IPME 18 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIb - Robust performance analysis V (x,∆) parameter-dependent Lyapunov function.
✪ Nominal analysis (LMI)→ Robust analysis (NP-hard)
∃ P : LΣ(P ) < O → ∀∆ ∈ ∆∆ , ∃ P (∆) : LΣ(∆)(P (∆)) < O
Test over sample values in ∆∆ gives optimistic results.
Seminar at IPME 19 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIb - Robust performance analysis V (x,∆) parameter-dependent Lyapunov function.
✪ Nominal analysis (LMI)→ Robust analysis (NP-hard)
∃ P : LΣ(P ) < O → ∀∆ ∈ ∆∆ , ∃ P (∆) : LΣ(∆)(P (∆)) < O
Test over sample values in ∆∆ gives optimistic results.
➥ Choice of P (∆) for having a finite number of decision variables :
➙ “Quadratic Stability”: P (∆) = P
➙ Polytopic PDLF: P (∆) =∑ζiP
[i]
➙ P (∆) polynomial w.r.t. ζi
➙ Quadratic-LFT PDLF: P (∆) =[
I ∆TC
]P
[I
∆C
]: ∆C = (I−∆D∆∆)−1∆C∆
➙ P (∆) polynomial w.r.t. A(∆)
Seminar at IPME 20 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIb - Robust performance analysis V (x,∆) parameter-dependent Lyapunov function.
✪ Nominal analysis (LMI)→ Robust analysis (NP-hard)
∃ P : LΣ(P ) < O → ∀∆ ∈ ∆∆ , ∃ P (∆) : LΣ(∆)(P (∆)) < O
Test over sample values in ∆∆ gives optimistic results.
➥ Choice of P (∆) for having a finite number of decision variables :
➙ “Quadratic Stability”: P (∆) = P
➙ Polytopic PDLF: P (∆) =∑ζiP
[i]
➙ P (∆) polynomial w.r.t. ∆ coefficients
➙ Quadratic-LFT PDLF: P (∆) =[
I ∆TC
]P
[I
∆C
]: ∆C = (I−∆D∆∆)−1∆C∆
➙ P (∆) polynomial w.r.t. A(∆)
➥ LMIs over infinite number of variables
∀∆ ∈ ∆∆ , ∃ P (∆) : LΣ(∆)(P (∆)) < O
⇐ ∃ P [i] : ∀∆ ∈ ∆∆ , LΣ(∆)(P (∆)) < O
Seminar at IPME 21 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIc - Conservative LMIs for polytopic models
Example : stability of x = A(∆)x with A(∆) =∑ζiA
[i] : ζi ≥ 0 ,∑ζi = 1
Seminar at IPME 22 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIc - Conservative LMIs for polytopic models
Example : stability of x = A(∆)x with A(∆) =∑ζiA
[i] : ζi ≥ 0 ,∑ζi = 1
➙ “Quadratic Stability”: P (∆) = P
V (x) < 0 ⇔ AT (∆)P + PA(∆) < O ⇔ A[i]TP + PA[i] < O
Seminar at IPME 23 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIIc - Conservative LMIs for polytopic models
Example : stability of x = A(∆)x with A(∆) =∑ζiA
[i] : ζi ≥ 0 ,∑ζi = 1
➙ “Quadratic Stability”: P (∆) = P
V (x) < 0 ⇔ AT (∆)P + PA(∆) < O ⇔ A[i]TP + PA[i] < O
➙ Polytopic PDLF: P (∆) =∑ζiP
[i] x
x
T O P (∆)
P (∆) O
x
x
< 0 :[A(∆) −I
] x
x
= 0
⇔ Finsler Lemma O P (∆)
P (∆) O
+G(∆)[A(∆) −I
]+
AT (∆)
−I
GT (∆) < O
⇐ G(∆) = G & convexity O P [i]
P [i] O
+G[A[i] −I
]+
A[i]T
−I
GT < O
Seminar at IPME 24 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId - Conservative LMIs for LFT models
Example : stability of x = Ax+B∆w∆ with w∆ = ∆z∆ = ∆C∆x+ ∆D∆∆w∆
Seminar at IPME 25 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId - Conservative LMIs for LFT models
Example : stability of x = Ax+B∆w∆ with w∆ = ∆z∆ = ∆C∆x+ ∆D∆∆w∆
➙ “Quadratic Stability”: P (∆) = P
V (x) =
x
w∆
∗ I O
A B∆
∗ O P
P O
I O
A B∆
x
w∆
< 0
:[
∆ −I] C∆ D∆∆
O I
x
w∆
= 0
⇔ Finsler Lemma
M∗A
O P
P O
MA < τM∗C
∆∗
−I
[∆ −I
]MC
Seminar at IPME 26 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId - Conservative LMIs for LFT models
Example : stability of x = Ax+B∆w∆ with w∆ = ∆z∆ = ∆C∆x+ ∆D∆∆w∆
➙ “Quadratic Stability”: P (∆) = P
V (x) =
x
w∆
∗ I O
A B∆
∗ O P
P O
I O
A B∆
x
w∆
< 0
:[
∆ −I] C∆ D∆∆
O I
x
w∆
= 0
⇔ Finsler Lemma
M∗A
O P
P O
MA < M∗CΘMC ≤ τM∗
C
∆∗
−I
[∆ −I
]MC
with[
I ∆∗]Θ
I
∆
≤ O
Seminar at IPME 27 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId - Conservative LMIs for LFT models
Example : stability of x = Ax+B∆w∆ with w∆ = ∆z∆ = ∆C∆x+ ∆D∆∆w∆
➙ “Quadratic Stability”: P (∆) = P
V (x) =
x
w∆
∗ I O
A B∆
∗ O P
P O
I O
A B∆
x
w∆
< 0
:[
∆ −I] C∆ D∆∆
O I
x
w∆
= 0
⇔ Finsler Lemma
M∗A
O P
P O
MA < M∗CΘMC ≤ τM∗
C
∆∗
−I
[∆ −I
]MC
with[
I ∆∗]Θ
I
∆
≤ O
➙ Quadratic-LFT PDLF - same methodology.
Seminar at IPME 28 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId -Conservative LMIs for LFT models
➞ LMI conservative choices of Quadratic separators Θ
✪ {X,Y, Z}−dissipative matrices ∆∆ = {∆ : X + Y∆ + ∆∗Y ∗ + ∆∗Z∆ ≤ O }
[I ∆∗
]Θ
I
∆
≤ O : ∀∆ ∈ ∆∆ ⇔ Θ ≤ τ
X Y
Y ∗ Z
, τ ≥ 0
Seminar at IPME 29 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId -Conservative LMIs for LFT models
➞ LMI conservative choices of Quadratic separators Θ
✪ {X,Y, Z}−dissipative matrices ∆∆ = {∆ : X + Y∆ + ∆∗Y ∗ + ∆∗Z∆ ≤ O }
[I ∆∗
]Θ
I
∆
≤ O : ∀∆ ∈ ∆∆ ⇔ Θ ≤ τ
X Y
Y ∗ Z
, τ ≥ 0
✪ {X,Y, Z}−dissipative repeated scalars ∆∆ = {∆ = δI : x+ yδ + δ∗y∗ + δ∗zδ ≤ 0 }
[I δ∗I
]Θ
I
δI
≤ O : ∀δ ∈ ∆∆ ⇐ Θ ≤
xQ yQ
y∗Q zQ
, Q ≥ O
Seminar at IPME 30 March 21st, 2005, St. Petersburg
III - Conservative LMI results
IIId -Conservative LMIs for LFT models
➞ LMI conservative choices of Quadratic separators Θ
✪ {X,Y, Z}−dissipative matrices ∆∆ = {∆ : X + Y∆ + ∆∗Y ∗ + ∆∗Z∆ ≤ O }
[I ∆∗
]Θ
I
∆
≤ O : ∀∆ ∈ ∆∆ ⇔ Θ ≤ τ
X Y
Y ∗ Z
, τ ≥ 0
✪ {X,Y, Z}−dissipative repeated scalars ∆∆ = {∆ = δI : x+ yδ + δ∗y∗ + δ∗zδ ≤ 0 }
[I δ∗I
]Θ
I
δI
≤ O : ∀δ ∈ ∆∆ ⇐ Θ ≤
xQ yQ
y∗Q zQ
, Q ≥ O
✪ Polytopic uncertainties ∆∆ ={
∆ =∑ζi∆[i] : ζi ≥ 0 ,
∑ζi = 1
}[
I ∆[i]∗]Θ
I
∆[i]
≤ O ,[
O I]Θ
O
I
≥ O
Seminar at IPME 31 March 21st, 2005, St. Petersburg
Outline
I - Uncertain LTI systems and performances
➞ Control objectives: stability, transient response, perturbation rejection...
➞ Structured parametric uncertainties: extremal values, bounded sets...
zw
K
u y< γ
[2]Σ
Σ[1]Σ[N]
Σ(∆)zΣ
∆w z
K
yu< γ
w
∆ ∆
II - LMIs and convex polynomial-time optimization
➞ Semi-Definite Programming and LMIs
➞ SDP solvers and parsers ➾ YALMIP and all solvers
III - Conservative LMI results
➞ Methods: Lyapunov, S-procedure, Finsler lemma, Topologic Separation...
➞ The ROMULOC toolbox
Seminar at IPME 32 March 21st, 2005, St. Petersburg