Download - emgr - Empirical Gramian Framework
![Page 1: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/1.jpg)
emgr - Empirical Gramian Framework
Christian Himpe ([email protected])Mario Ohlberger ([email protected])
WWU MünsterInstitute for Computational and Applied Mathematics
RB summer school 20.-23.08.2013
![Page 2: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/2.jpg)
Overview
1 Control Systems2 Gramian-Based Model Reduction3 Empirical Gramians4 Neural Networks
![Page 3: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/3.jpg)
Control Systems(in a hurry)
![Page 4: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/4.jpg)
Control Systems
General control system:
x = f (x , u, θ)
y = g(x , u, θ)
with:Input u ∈ Rm
State x ∈ Rn
Output y ∈ Ro
Parameters θ ∈ Rp
![Page 5: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/5.jpg)
Control Systems
Linear control system:
x = Ax + Buy = Cx + Du
with:Input u ∈ Rm
State x ∈ Rn
Output y ∈ Ro
Parameters θ ∈ Rp
![Page 6: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/6.jpg)
Control Systems
Linear control system:
x = Ax + Buy = Cx
with:Input u ∈ Rm
State x ∈ Rn
Output y ∈ Ro
Parameters θ ∈ Rp
![Page 7: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/7.jpg)
Control Systems
Nonlinear control system:
x = A tanh(x) + Buy = Cx
with:Input u ∈ Rm
State x ∈ Rn
Output y ∈ Ro
Parameters θ ∈ Rp
![Page 8: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/8.jpg)
Control Systems
Parametrized linear control system:
x = Aθx + Buy = Cx
with:Input u ∈ Rm
State x ∈ Rn
Output y ∈ Ro
Parameters θ ∈ Rp
![Page 9: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/9.jpg)
Control Theory Vocabulary
Linear System: Σ = (A,B,C )
Adjoint System: ΣT = (AT ,CT ,BT )
Markov parameter: m(k) = CAkBImpulse response: h(t) = CeAtBInput-to-state map: x(t) = eAtBu(t)
State-to-output map: y(t) = CeAtx0
Transfer function: G (s) = C (s1− A)−1BSystem gain: S = tr(CA−1B) = G (0)
![Page 10: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/10.jpg)
Model Order Reduction
General Control System:
x = f (x , u, θ)
y = g(x , u, θ)
with:dim(x) >> 1dim(θ) >> 1
dim(x) < dim(x)
dim(θ) < dim(θ)
y ≈ y
![Page 11: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/11.jpg)
Model Order Reduction
State Reduced System:
˙x = f (x , u, θ)
y = g(x , u, θ)
with:dim(x) >> 1
dim(θ) >> 1
dim(x) < dim(x)
dim(θ) < dim(θ)
y ≈ y
![Page 12: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/12.jpg)
Model Order Reduction
Parameter Reduced System:
˙x = f (x , u, θ)
y = g(x , u, θ)
with:
dim(x) >> 1
dim(θ) >> 1
dim(x) < dim(x)
dim(θ) < dim(θ)
y ≈ y
![Page 13: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/13.jpg)
Model Order Reduction
Combined Reduced System:
˙x =˜f (x , u, θ)
y = ˜g(x , u, θ)
with:dim(x) >> 1dim(θ) >> 1dim(x) < dim(x)
dim(θ) < dim(θ)
y ≈ y
![Page 14: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/14.jpg)
Gramian-Based Model Reduction(and System Identification)
![Page 15: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/15.jpg)
Hankel Operator
For a linear control system Σ = (A,B,C ).The Hankel operator H(t) maps past inputs to future outputsusing the impulse response
H(t) =
∫ 0
−∞h(t − τ)u(τ)dτ =
∫ 0
−∞CeA(t−τ)Bu(τ)dτ
We are interested in the Singular Values of H(t)!
![Page 16: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/16.jpg)
Controllability Gramian (WC )
Controllability quantifies how well a state can be driven by input orcontrol.
The controllability gramian WC characterizes the controllability of alinear system and is computed as the smallest semi-positive definitesolution of the Lyapunov equation:
AWC + WCAT = −BBT .
For asymptotic stable systems WC is also given by:
WC =
∫ ∞0
eAtBBT eAT tdt.
![Page 17: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/17.jpg)
Observability Gramian (WO)
Observability quantifies how well a change in state is reflected bythe outputs.
The observability gramian WO characterizes the observability of alinear system and is computed as the smallest semi-positive definitesolution of the Lyapunov equation:
ATWO + WOA = −CTC .
For asymptotic stable systems WO is also given by:
WO =
∫ ∞0
eAT tCTCeAtdt.
![Page 18: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/18.jpg)
Balanced Truncation1 (BT)
Hankel Singular Values:
σi =√λ(WCWO)
Balancing Transformation:
∃U,V : VWCV T = UTWOU =
σ1
σ2. . .
σn
thus:
x = Ax + Buy = Cx
1[Moore’81]
![Page 19: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/19.jpg)
Balanced Truncation1 (BT)
Hankel Singular Values:
σi =√λ(WCWO)
Balancing Transformation:
∃U,V : VWCV T = UTWOU =
σ1
σ2. . .
σn
thus:
x = VAUx + VBuy = CUx
1[Moore’81]
![Page 20: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/20.jpg)
Balanced Truncation1 (BT)
Hankel Singular Values:
σi =√λ(WCWO)
Balancing Transformation:
∃U,V : VWCV T = UTWOU =
σ1
σ2. . .
σn
thus:
x = Ax + Bu
y = C x1[Moore’81]
![Page 21: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/21.jpg)
Balanced Truncation1 (BT)
Hankel Singular Values:
σi =√λ(WCWO)
Balancing Transformation:
∃U,V : VWCV T = UTWOU =
σ1
σ2. . .
σn
thus:
x =
(A11 A12
A21 A22
)(x1x2
)+
(B1
B2
)u
y =(C1 C2
)(x1x2
)1[Moore’81]
![Page 22: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/22.jpg)
Balanced Truncation1 (BT)
Hankel Singular Values:
σi =√λ(WCWO)
Balancing Transformation:
∃U,V : VWCV T = UTWOU =
σ1
σ2. . .
σn
thus:
˙x = A11x1 + B1u
y = C1x1
1[Moore’81]
![Page 23: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/23.jpg)
Cross Gramian (WX )
The cross gramian characterizes both, the controllability and theobservability of a linear system.
Given the system has the same number of inputs and outputs, thecross gramian is computed as the smallest solution of the Sylvesterequation:
AWX + WXA = −BC
For asymptotic stable systems WX is also given by:
WX =
∫ ∞0
eAtBCeAtdt
![Page 24: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/24.jpg)
Direct Truncation (Approximate Balanced Reduction2)
For SISO or symmetric MIMO (has symmetric transfer function)systems (A,B,C ) there exists a J such that:
J = JT , AJ = JAT , BT = CJ
then:
WOWC = W 2X ⇒ σi = |λ(WX )| ⇒WX = UDV ≈ U
σ1
σ2. . .
σn
V
thus:
x = Ax + Buy = Cx
2[Sorensen’01], [Sorensen’02]
![Page 25: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/25.jpg)
Direct Truncation (Approximate Balanced Reduction2)
For SISO or symmetric MIMO (has symmetric transfer function)systems (A,B,C ) there exists a J such that:
J = JT , AJ = JAT , BT = CJ
then:
WOWC = W 2X ⇒ σi = |λ(WX )| ⇒WX = UDV ≈ U
σ1
σ2. . .
σn
V
thus:
x = VAUx + VBuy = CUx
2[Sorensen’01], [Sorensen’02]
![Page 26: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/26.jpg)
Direct Truncation (Approximate Balanced Reduction2)
For SISO or symmetric MIMO (has symmetric transfer function)systems (A,B,C ) there exists a J such that:
J = JT , AJ = JAT , BT = CJ
then:
WOWC = W 2X ⇒ σi = |λ(WX )| ⇒WX = UDV ≈ U
σ1
σ2. . .
σn
V
thus:
x = Ax + Bu
y = C x
2[Sorensen’01], [Sorensen’02]
![Page 27: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/27.jpg)
Direct Truncation (Approximate Balanced Reduction2)
For SISO or symmetric MIMO (has symmetric transfer function)systems (A,B,C ) there exists a J such that:
J = JT , AJ = JAT , BT = CJ
then:
WOWC = W 2X ⇒ σi = |λ(WX )| ⇒WX = UDV ≈ U
σ1
σ2. . .
σn
V
thus:
x =
(A11 A12
A21 A22
)(x1x2
)+
(B1
B2
)u
y =(C1 C2
)(x1x2
)2[Sorensen’01], [Sorensen’02]
![Page 28: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/28.jpg)
Direct Truncation (Approximate Balanced Reduction2)
For SISO or symmetric MIMO (has symmetric transfer function)systems (A,B,C ) there exists a J such that:
J = JT , AJ = JAT , BT = CJ
then:
WOWC = W 2X ⇒ σi = |λ(WX )| ⇒WX = UDV ≈ U
σ1
σ2. . .
σn
V
thus:
˙x = A11x1 + B1u
y = C1x1
2[Sorensen’01], [Sorensen’02]
![Page 29: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/29.jpg)
Nonsymmetric Cross Gramian
Symmetricity is strong requirementFor nonsymmetric systems only
k∑i=1
σi ≥k∑
i=1
λ(WX )i ∧k∑
i=1
λ(WX )N−i+1 ≥k∑
i=1
σN−i+1
holds.
But:embed into symmetric systemorthogonal symmetricityuse approximate nonsymmetric Cross Gramian
![Page 30: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/30.jpg)
Sensitivity Analysis
Given a parametrized system, with many parameters...Which parameters are important? Which are redundant?⇒ Parameter IdentificationFor a generic model, which parameters can be excluded here?⇒ Parameter Reduction
So what is the dominant subspace of the parameter space?Wait a minute, that sounds familiar...
![Page 31: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/31.jpg)
Sensitivity Gramian3 (WS)
Treating the parameters as additional inputs of dim(θ) with steadyinput θ (if possible) gives:
x = f (x , u) +P∑
k=1
f (x , θk)
⇒WC = WC ,0 +P∑
k=1
WC ,k
Sensitivity Gramian WS :
WS ,ii = tr(WC ,i ).
The state controllability gramian WC is a byproduct!3based on [Sun’06]
![Page 32: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/32.jpg)
Identifiability Gramian4 (WI )
Augmenting the system by dim(θ) constant states with initial valuebeing θ gives:
xa =
(xθ
)=
(f (x , u, θ)
0
)xa(0) =
(x0θ
)⇒WO,a =
(WO WMW ∗
M WP
).
Identifiability Gramian WI :
WI = WP −W ∗MWO
−1WM ,
WI ≈WP .
The state observability gramian WO is a byproduct!4[Geffen’08]
![Page 33: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/33.jpg)
Combined Reduction
High-dim state space → state reductionHigh-dim parameter space → parameter reductionHigh-dim state and parameter spaces → combined reduction
![Page 34: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/34.jpg)
Joint Gramian5 (WJ)
Augmenting the system by dim(θ) constant states with initial valuebeing θ:
xa =
(xθ
)=
(f (x , u, θ)
0
)xa(0) =
(x0θ
)⇒WJ,a =
(WX WM0 0
).
Cross-Identifiability Gramian WII :
WII = −W ∗M diag(WX + W T
X )−1WM .
The state cross gramian WX is a byproduct!5[Himpe’13]
![Page 35: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/35.jpg)
Empirical Gramian-Based Model Reduction(going nonlinear)
![Page 36: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/36.jpg)
Empirical Gramian6 / Empirical Covariance Matrix7
Concept: POD MethodWC = mean
u∈U(∫∞0 x(t)x∗(t)dt)
WO = meanx0∈X
(∫∞0 ρ(y∗(t)y(t))dt)
WX = meanu∈U,x0∈X
(∫∞0 ϕ(x(t), y(t))dt)
withU some perturbation to the input uX some perturbation to the initial state x0
Empircal Covariance Matrices correspond to the EmpiricalGramians, but allow arbitrary input instead of delta-impulses.
6[Lall’99],[Lall’02]7[Hahn’02]
![Page 37: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/37.jpg)
Perturbation Sets
U = Eu × Ru × Qu
Eu = {ei ∈ Rj ; ‖ei‖ = 1; eiej 6=i = 0; i = 1, . . . ,m}Ru = {Si ∈ Rj×j ; S∗i Si = 1; i = 1, . . . , s}Qu = {ci ∈ R; ci > 0; i = 1, . . . , q}
X = Ex × Rx × Qx
Ex = {fi ∈ Rn; ‖fi‖ = 1; fi fj 6=i = 0; i = 1, . . . , n}Rx = {Ti ∈ Rn×n; T ∗i Ti = 1; i = 1, . . . , t}Qx = {di ∈ R; di > 0; i = 1, . . . , r}
![Page 38: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/38.jpg)
Empirical Controllability Gramian (WC)
For sets Eu, Ru, Qu, input u(t) and input during the steady statex , u, the empirical controllability gramian is given by:
WC =1
|Qu||Ru|
|Qu |∑h=1
|Ru |∑i=1
m∑j=1
1c2h
∫ ∞0
Ψhij(t)dt
Ψhij(t) = (xhij(t)− x)(xhij(t)− x)∗ ∈ Rn×n.
With xhij being the states for the input configurationuhij(t) = chSieju(t) + u.
![Page 39: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/39.jpg)
Empirical Observability Gramian (WO)
For sets Ex , Rx , Qx and output y during the steady state x , y , theempirical observability gramian is given by:
WO =1
|Qx ||Rx |
|Qx |∑k=1
|Rx |∑l=1
1d2kTl
∫ ∞0
Ψkl (t)dt T ∗l
Ψklab = (ykla(t)− y)∗(yklb(t)− y) ∈ R.
With ykla being the systems output for the initial stateconfiguration xkla
0 = dkSl fa + x .
![Page 40: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/40.jpg)
Empirical Cross Gramian8 (WX)
For sets Eu, Ex , Ru, Rx , Qu, Qx , input u during steady state x withoutput y , the empirical cross gramian is given by:
WX =1
|Qu||Ru|m|Qx ||Rx |
|Qu |∑h=1
|Ru |∑i=1
m∑j=1
|Qx |∑k=1
|Rx |∑l=1
1chdk
∫ ∞0
TlΨhijkl (t)T ∗l dt
Ψhijklab (t) = f ∗b T ∗k ∆xhij(t)e∗i S∗h ∆ykla(t)
∆xhij(t) = (xhij(t)− x)
∆ykla(t) = (ykla(t)− y).
Where xhij and ykla being states and output for the inputuhij(t) = chSieju(t) + u and initial state xkla
0 = dkTl fa + xrespectively.
8for SISO: [Streif’06], for MIMO: [Himpe’13]
![Page 41: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/41.jpg)
Empirical Sensitivity Gramian (WS)
The sensitivity gramian encapsulates the controllability gramian.→ Same computation with the empirical controllability gramian.
![Page 42: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/42.jpg)
Empirical Identifiability Gramian (WI)
The identifiability gramian encapsulates the observability gramian.→ Same computation with the empirical observability gramian.
![Page 43: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/43.jpg)
Empirical Joint Gramian (WJ)
The joint gramian encapsulates the cross gramian.→ Same computation with the empirical cross gramian.
![Page 44: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/44.jpg)
Research Question(ay, there is the rub)
![Page 45: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/45.jpg)
Hyperbolic Network Model
x(t) = Aθ tanh(Kθx(t)) + Bθu(t)
y(t) = Cθx(t)
State and parameter space dimensions:
dim(x) = ndim(u) = dim(y) = m
⇒ dim(θ) = n2 + 2mn + n
![Page 46: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/46.jpg)
Hyperbolic Network Model
x(t) = Aθ tanh(Kθx(t)) + Bθu(t)
y(t) = Cθx(t)
State and parameter space dimensions:
dim(x) = ndim(u) = dim(y) = m
⇒ dim(θ) = n2 + 2mn + n
![Page 47: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/47.jpg)
Bayesian Inverse Problem
h(x , θ) = y + ε
Noise: ε = N(0, σ2)
Bayes Rule: P(θ|yd ) = P(yd |θ)P(θ)P(yd )
∝ P(yd |θ)P(θ)
Prior: P(θ)
Likelihood: P(yd |θ)
Evidence: P(yd )
Posterior: P(θ|yd )
![Page 48: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/48.jpg)
Robust Reduction
If the parameters are uncertainin example, if they are a Gaussian distributionreduction needs to consider this uncertainty
Empirical Gramians to the rescue!
for WI and WJ or WC and WS
parameters are or can be included in the perturbed componentsthus parameters uncertainty is taken care of.
![Page 49: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/49.jpg)
Implementation and Numerical Experiments(almost over, now its mainly pictures)
![Page 50: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/50.jpg)
emgr - Empirical Gramian Framework
Gramians:Empirical Controllability GramianEmpirical Observability GramianEmpirical Cross GramianEmpirical Sensitivity GramianEmpirical Identifiability GramianEmpirical Joint Gramian
Features:Uniform InterfaceCompatible with MATLAB & OCTAVEVectorized & ParallelizableOpen-Source licensed
More info at: http://gramian.de
![Page 51: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/51.jpg)
Simplified Hyperbolic Network Model
x(t) = Aθ tanh(x(t)) + Bu(t)
y(t) = Cx(t)
Assumptions:
A = AT
C = BT
P(θ) = (−1, 1)
Dimensions:
n = {16, 25, 36, 49, 64}⇒p = {256, 625, 1296, 2401, 4096}
m =o = {4, 5, 6, 7, 8}
![Page 52: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/52.jpg)
State Reduction (HNM)
20 25 30 35 40 45 50 55 600
0.5
1
1.5
2
2.5
3
3.5
States
Tim
e (
s)
Offline Time
Balanced POD
Balanced Truncation
Cross Gramian
20 25 30 35 40 45 50 55 600.0085
0.009
0.0095
0.01
0.0105
0.011
0.0115
0.012
0.0125
States
Tim
e (
s)
Online Time
Full Order
Balanced POD
Balanced Truncation
Cross Gramian
20 25 30 35 40 45 50 55 600
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
StatesE
rro
r
Relative L2−Error
Balanced POD
Balanced Truncation
Cross Gramian
![Page 53: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/53.jpg)
Parameter Reduction (HNM)
20 25 30 35 40 45 50 55 600
500
1000
1500
2000
2500
States
Tim
e (
s)
Offline Time
Full Order
Sensitivity Gramian
Identifiability Gramian
Joint Gramian
20 25 30 35 40 45 50 55 600
50
100
150
200
250
States
Tim
e (
s)
Online Time
Sensitivity Gramian
Identifiability Gramian
Joint Gramian
20 25 30 35 40 45 50 55 600
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
StatesE
rro
r
Relative L2−Error
Full Order
Sensitivity Gramian
Identifiability Gramian
Joint Gramian
![Page 54: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/54.jpg)
Combined Reduction (HNM)
20 25 30 35 40 45 50 55 600
500
1000
1500
2000
2500
States
Tim
e (
s)
Offline Time
Full Order
Sensitivity + Observability Gramian
Controllability + Identifiability Gramian
Joint Gramian
20 25 30 35 40 45 50 55 600
50
100
150
200
250
States
Tim
e (
s)
Online Time
Sensitivity + Observability Gramian
Controllability + Identifiability Gramian
Joint Gramian
20 25 30 35 40 45 50 55 600
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
StatesE
rro
r
Relative L2−Error
Full Order
Sensitivity + Observability Gramian
Controllability + Identifiability Gramian
Joint Gramian
![Page 55: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/55.jpg)
FAQ
Why the Cross-Gramian?→ No balancing required!
Why Empirical Gramians?→ Allow nonlinear systems, can be data-driven!
Why the Joint Gramian?→ Best compromise between speed and flexibility!
Why Bayesian Inversion?→ Prior distribution helps snapshot generation!
Why Octave?→ Compatible to Matlab and it is Open Source!
![Page 56: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/56.jpg)
tl;dl
Combined Reduction: reduction of states and parameters.Joint Gramian: enables combined reduction.Empirical Joint Gramian: nonlinear combined reduction.
Get the source code: http://j.mp/hnm13 .
Thanks!
![Page 57: emgr - Empirical Gramian Framework](https://reader033.vdocuments.us/reader033/viewer/2022052600/55838837d8b42a11178b4967/html5/thumbnails/57.jpg)
Shameless Plug
Use emgr
for:Model Order Reduction
State ReductionParameter ReductionCombined Reduction
System IdentificationParameter IdentificationSensitivity AnalysisDecentralized Control
with:First & Second OrderLinear & NonlinearParametrized & Parametric
Systems
Download, Documentation, Demos:http://gramian.de