Page 1
VARIABLE ROBUSTNESS
CONTROL:
PRINCIPLES and ALGORITHMS
Marco C. Campi
Simone Garatti
Page 2
thanks to :
Algo Care’
Simone GarattiGiuseppe Calafiore
Maria Prandini
Page 3
PART I: Principles
Page 4
Uncertainty
controller synthesis
noise compensation
prediction
optimization
program
Optimization
Page 5
U-OP:
Uncertain Optimization Program
Page 6
U-OP:
not well-defined
Uncertain Optimization Program
Page 9
[J.C. Doyle, 1978], [G. Zames, 1981]
Uncertainty
Page 10
Probabilistic uncertainty
Page 11
Probabilistic uncertainty
Page 12
Probabilistic uncertainty
Page 13
Probabilistic uncertainty
Page 14
Probabilistic uncertainty
Page 15
Probabilistic uncertainty
R.F. Stengel, L.R. Ray, B.R. Barmish, C.M. Lagoa …
R. Tempo, E.W. Bai, F. Dabbene, P.P. Khargonekar, A. Tikku, …
Page 16
Probabilistic uncertainty
Page 17
[A. Charnes, W.W. Cooper, and G.H. Symonds, 1958]
Probabilistic uncertainty
chance-constrained approach:
Page 18
[A. Charnes, W.W. Cooper, and G.H. Symonds, 1958]
Probabilistic uncertainty
chance-constrained approach:
almost neglected by the systems
and control community:
(i) tradition;
(ii) lack of algorithms.
Page 19
[A. Charnes, W.W. Cooper, and G.H. Symonds, 1958]
Probabilistic uncertainty
chance-constrained approach:
almost neglected by the systems
and control community:
(i) tradition;
(ii) lack of algorithms.
GOALS: 1. excite interest in the chance-constrained approach
2. provide algorithmic tools
Page 20
a look at optimization in the space
Page 21
performance cloud
Page 24
chance-constrained approach
Page 25
chance-constrained approach
very hard to solve!
Page 26
VRC – Variable Robustness Control
Page 27
performance - violation plot
Page 28
performance - violation plot
Page 29
icicle geometry [C.M. Lagoa & B.R. Barmish, 2002]
Page 30
icicle geometry [C.M. Lagoa & B.R. Barmish, 2002]
… let the problem speak
Page 31
PART II: Algorithms
(convex case)
Page 32
The “scenario” paradigm
[G. Calafiore & M. Campi, 2005, 2006]
Page 33
SPN = scenario program
The “scenario” paradigm
SPN is a standard finite convex optimization problem
[G. Calafiore & M. Campi, 2005, 2006]
Page 34
Fundamental
question: how robust is ?
Page 35
Example: feedforward noise compensation
Page 36
Example: feedforward noise compensation
ARMAX
System
Page 37
Example: feedforward noise compensation
CompensatorARMAX
System
Page 38
Example: feedforward noise compensation
CompensatorARMAX
System
Objective: reduce the effect of noise
Page 39
Example: feedforward noise compensation
CompensatorARMAX
System
ARMAX System:
Compensator:
Goal:
Page 40
Example: feedforward noise compensation
CompensatorARMAX
SystemCompensator:
ARMAX System:
Page 41
Example: feedforward noise compensation
CompensatorARMAX
SystemCompensator:
Easy:
ARMAX System:
Page 42
Example: feedforward noise compensation
CompensatorARMAX
System
Page 43
Example: feedforward noise compensation
system parameters unknown:
Page 44
Example: feedforward noise compensation
system parameters unknown:
PERTURBED
SystemNominal
Compensator
Page 45
Example: feedforward noise compensation
sample:
solve:
scenario approach:
Page 46
Fundamental
question: how robust is ?
Page 47
Fundamental
question: how robust is ?
that is: how guaranteed is against all
Page 48
Fundamental
question: how robust is ?
that is: how guaranteed is against all
from the “visible” to the “invisible”
Page 54
Comments
generalization need for structure
Good news: the structure we need
is only convexity
Page 55
… more comments
N often tractable by standard solvers
N easy to compute
N independent of Pr
permits to address problems otherwise intractable
Ex: feedforward noise compensation
Page 56
Example: feedforward noise compensation
Page 57
Example: feedforward noise compensation
Page 58
Example: feedforward noise compensation
Page 59
Example: feedforward noise compensation
sample:
solve:
Page 60
Example: feedforward noise compensation
sample:
solve:
Page 61
Example: feedforward noise compensation
Output variance below 5.8 for all plants but a
small fraction ( = 0.5%)
Page 62
Example: feedforward noise compensation
performance profile
Output variance below 5.8 for all plants but a
small fraction ( = 0.5%)
Page 63
Variable Robustness Control
Page 64
Variable Robustness Control
Page 65
Variable Robustness Control
Page 66
Variable Robustness Control
Page 67
Variable Robustness Control
Page 68
Variable Robustness Control
Page 69
Variable Robustness Control
Page 73
Comments
the result does not depend on the
algorithm for eliminating k constraints
Page 74
Comments
the result does not depend on the
algorithm for eliminating k constraints
… do it greedy
Page 75
Comments
the result does not depend on the
algorithm for eliminating k constraints
… do it greedy
value can be inspected
violation probability is guaranteed
by the theorem
Page 76
performance - violation plot
Page 77
Example: feedforward noise compensation
Page 78
Example: feedforward noise compensation
sample:
solve:
Page 79
Example: feedforward noise compensation
sample:
solve:
Page 80
Example: feedforward noise compensation
Page 81
Example: feedforward noise compensation
Page 82
performance profile
Example: feedforward noise compensation
Page 83
performance profile
Example: feedforward noise compensation
Page 84
performance profile
Example: feedforward noise compensation
Page 85
performance profile
Example: feedforward noise compensation
Page 86
performance profile
Example: feedforward noise compensation
Page 87
performance profile
Example: feedforward noise compensation
Page 88
performance profile
Example: feedforward noise compensation
Page 89
performance profile
Example: feedforward noise compensation
Page 90
performance profile
Example: feedforward noise compensation
Page 91
Example: feedforward noise compensation
CompensatorARMAX
System
Page 92
Example: feedforward noise compensation
PERTURBED
SystemCompensator
Page 93
Conclusions
The VRC approach is a very general tool to trade
robustness for performance
Page 94
Conclusions
It is based on a solid and deep theory, but its practical
use is very simple
The VRC approach is a very general tool to trade
robustness for performance
Page 95
Conclusions
It is based on a solid and deep theory, but its practical
use is very simple
Applications in:
- prediction
- robust control
- engineering
- finance
The VRC approach is a very general tool to trade
robustness for performance
Page 96
REFERENCES
M.C. Campi and S. Garatti.
Variable Robustness Control: Principles and Algorithms.
Proceedings MTNS, 2010.
M.C. Campi and S. Garatti.
The Exact Feasibility of Randomized Solutions of Uncertain Convex Programs.
SIAM J. on Optimization, 19, no.3: 1211-1230, 2008.
G. Calafiore and M.C. Campi.
Uncertain Convex Programs: randomized Solutions and Confidence Levels.
Mathematical Programming, 102: 25-46, 2005.
G. Calafiore and M.C. Campi.
The Scenario Approach to Robust Control Design.
IEEE Trans. on Automatic Control, AC-51: 742-753, 2006.