sparse representations for packetized predictive networked control
DESCRIPTION
M. Nagahara, D. E. Quevedo Sparse Representations for Packetized Predictive Networked Control, IFAC 18th World Congress, pp. 84-89, Aug., 2011.TRANSCRIPT
Sparse Representations for Packetized Predictive Networked Control
Masaaki Nagahara (Kyoto Univ.)Daniel E. Quevedo (The Univ. of Newcastle)
Networked Control in Sparse Land
โข In networked control, one has to transmit control signals through unreliable networks.
โข Packetized Predictive Control (PPC) can make the system robust against packet dropouts.
โข Sparse Representation can effectively compress signals without much distortion.
โข This work is
PPC + Sparse Representation
Table of Contents
โข How does Packetized Predictive Control work?โ PPC in networked control systems with packet
dropoutโข How can one obtain sparse vectors?
โ -norm regularizationโ Fast iterative-shrinkage algorithm
โข PPC + Sparse Representationโ Is the feedback system stable? YES.
โข Examples
Table of Contents
โข How does Packetized Predictive Control work?โ PPC in networked control systems with packet
dropoutโข How can one obtain sparse vectors?
โ -norm regularizationโ Fast iterative-shrinkage algorithm
โข PPC + Sparse Representationโ Is the feedback system stable? YES.
โข Examples
Packetized Predictive Control
โข Compute a tentative control sequence for a finite horizon of future time instants.
โข Transmit the sequence as a packet to a buffer.โข If a packet is dropped out, use the control
stored in the buffer[Bemporad(1998), Casavola et al.(2006), Tang-Silva(2009), Quevedo(2007,2011)]
Controller Buffer Plant๐ฅ (๐) ๐ (๐) ๐ข(๐) ๐ฅ (๐)
Packetized Predictive Control
Controller Buffer Plant๐ฅ (0) ๐ (0) ๐ฅ (0)
๐ข0(0)๐ข1(0)
๐ข2(0)๐ข3(0)
๐ (0 )=[๐ข0 (0 ) ,๐ข1 (0 ) ,๐ข2 (0 ) ,๐ข3 (0 ) ]๐
๐ฝ (๐ )=โ๐ฅ (3|0 )โ๐2+โ๐=0
3
โ๐ฅ (๐|0 )โ๐2+๐โ๐โ๐
2
โ๐ฃโ๐2 โ๐ฃ๐ ๐๐ฃ
minimizing the cost function:
Packetized Predictive Control
Controller Buffer Plant
Assumption: The first packet is successfully transmitted to the buffer.
๐ฅ (0) ๐ (0) ๐ฅ (0)
๐ข0(0)๐ข1(0)
๐ข2(0)๐ข3(0)
Packetized Predictive Control
Controller Buffer Plant
Assumption: The first packet is successfully transmitted to the buffer. the 4 values are stored in the buffer.
๐ข0(0)๐ข1(0)
๐ข2(0)๐ข3(0)
๐ฅ (0) ๐ (0) ๐ฅ (0)
Packetized Predictive Control
Controller Buffer Plant
Assumption: The first packet is successfully transmitted to the buffer. the 4 values are stored in the buffer.
๐ข0(0)๐ข1(0)
๐ข2(0)๐ข3(0)
๐ข (0 )=๐ข0 (0)๐ฅ (0) ๐ (0) ๐ฅ (0)
Packetized Predictive Control
Controller Buffer Plant
๐ข0(1)๐ข1(1)
๐ข2(1)๐ข3(1)
๐ฅ (1) ๐ (1) ๐ฅ (1)
๐ (1 )=[๐ข0 (1 ) ,๐ข1 (1 ) ,๐ข2 (1 ) ,๐ข3 (1 ) ]๐
๐ฝ (๐ )=โ๐ฅ (3|1 )โ๐2+โ๐=0
3
โ๐ฅ (๐|1 )โ๐2+๐โ๐โ๐
2
Packetized Predictive Control
Controller Buffer Plant
๐ข0(1)๐ข1(1)
๐ข2(1)๐ข3(1)
๐ฅ (1) ๐ (1) ๐ฅ (1)
Packetized Predictive Control
Controller Buffer Plant
Packet-dropout occurs!๐ข0(1)๐ข1(1)
๐ข2(1)๐ข3(1)
๐ฅ (1) ๐ (1) ๐ฅ (1)
Packetized Predictive Control
Controller Buffer Plant
Use in the bufferas the control ๐ข0(0)
๐ข1(0)๐ข2(0)
๐ข3(0)
๐ข (1 )=๐ข1(0)๐ฅ (1) ๐ (1) ๐ฅ (1)
Design of control packetsโข At each step, we solve the following optimization
for the packet :
โข The solution is given by linear transformation of the state :
๐ฝ (๐ )=โ๐ฅ (๐|๐ )โ๐2+โ๐=0
๐
โ๐ฅ (๐|๐ )โ๐2+๐โ๐โ2
2
ยฟโ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ2
2+โ๐ฅ (๐)โ๐2
Table of Contents
โข How does Packetized Predictive Control work?โ PPC in networked control systems with packet
dropoutโข How can one obtain sparse vectors?
โ -norm regularizationโ Fast iterative-shrinkage algorithm
โข PPC + Sparse Representationโ Is the feedback system stable? YES.
โข Examples
Sparsity-Promoting Optimization
โข Energy-limiting optimization (-norm regularization):
โข Sparsity-promoting optimization (-norm regularization, optimization):
๐โ (๐ )=min๐
โ๐บ๐โ๐ป๐ฅ (๐ )โ22+๐โ๐โ2
2
๐โ (๐ )=min๐
โ๐บ๐โ๐ป๐ฅ (๐ )โ22+๐โ๐โ1
โ
Sparsity-Promoting Optimization
โข -norm regularization produces a dense vector like
โข -norm regularization (or optimization) produces a sparse vector like
โข Sparse vectors can be compressed more effectively than a dense vector.โ c.f. JPEG image compression
๐โ=[โ2.6 ,โ0.1 ,โ1.8 ,0.1 ,โ0.6 ]๐
๐โ=[โ2.6 ,0.09 ,โ2.2 ,0 ,0 ]๐
Why does promote sparsity?
โข By using the Lagrange dual, we obtain
for some .
{Uโ๐ 2 :โ๐โ1=const }
0
๐โ (๐ )=argmin๐
โ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ1
โ
ยฟargmin๐
โ๐โ1โs . t .โ๐บ๐โ๐ป๐ฅ (๐ )โ2
2โค๐
{Uโ๐ 2 :โ๐บ๐โ๐ป๐ฅโ22โค๐ }
Feasible set
ball
Why does promote sparsity?
โข By using the Lagrange dual, we obtain
for some .
๐โ (๐ )=argmin๐
โ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ1
โ
ยฟargmin๐
โ๐โ1โs . t .โ๐บ๐โ๐ป๐ฅ (๐ )โ2
2โค๐
{Uโ๐ 2 :โ๐โ1=const }
{Uโ๐ 2 :โ๐บ๐โ๐ป๐ฅโ22โค๐ }๐โ
0Sparse!
Feasible set
ball
Comparison with energy-limiting optimization
{Uโ๐ 2 :โ๐โ2=const }
{Uโ๐ 2 :โ๐บ๐โ๐ป๐ฅโ22โค๐ }๐โ
0Not sparse
ยฟargmin๐
โ๐โ22s . t .โ๐บ๐โ๐ป๐ฅ (๐ )โ2
2โค๐
๐โ (๐ )=argmin๐
โ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ2
2
Feasible set
ball
Iterative-Shrinkage Algorithm
โข The solution of
can be effectively obtained via a fast algorithm.
๐โ (๐ )=argmin๐
โ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ1
โ
๐ ๐+1=๐2๐ /๐ ( 1๐ ๐บ๐ (๐ป๐ฅ (๐)โ๐บ๐ ๐ )+๐ ๐) , ๐=0,1,2,โฆ
[Beck-Teboulle, SIAM J. Imag. Sci., 2009][Zibulevsky-Elad, IEEE SP Mag., 2010]
Iterative-Shrinkage Algorithm
โข The solution of
can be effectively obtained via a fast algorithm.๐ ๐+1=๐2๐ /๐ ( 1๐ ๐บ๐ (๐ป๐ฅ (๐)โ๐บ๐ ๐ )+๐ ๐) , ๐=0,1,2,โฆ
๐2๐/ ๐ (๐ข)
๐ข2๐ /๐
โ2๐ /๐ ๐>๐max (๐บ๐๐บ)
๐โ (๐ )=argmin๐
โ๐บ๐ โ๐ป๐ฅ (๐ )โ22+๐โ๐โ1
โ
[Beck-Teboulle, SIAM J. Imag. Sci., 2009][Zibulevsky-Elad, IEEE SP Mag., 2010]
Table of Contents
โข How does Packetized Predictive Control work?โ PPC in networked control systems with packet
dropoutโข How can one obtain sparse vectors?
โ -norm regularizationโ Fast iterative-shrinkage algorithm
โข PPC + Sparse Representationโ Is the feedback system stable? YES.
โข Examples
Stability Analysisโข The controlled plant: โข The control packet:
โข If then โข This implies that asymptotic stability will not be
achieved if is unstable even if there is no packet-dropout.
{๐ฅโโ 2:โ๐บ๐ ๐ป๐ฅโโโค2๐}
0๐ฅ (๐)
Practical Stabilityโข Assumption: The number of consecutive packet-
dropouts is always less than the prediction horizon (the size of the buffer)
[Theorem]Let and choose to satisfy
where Then for we have (practical stability)
where are constants.
๐ฝ (๐ )=โ๐ฅ (๐|๐ )โ๐2+โ๐=0
๐
โ๐ฅ (๐|๐ )โ๐2+๐โ๐โ1
โ
Terminal condition
Table of Contents
โข How does Packetized Predictive Control work?โ PPC in networked control systems with packet
dropoutโข How can one obtain sparse vectors?
โ -norm regularizationโ Fast iterative-shrinkage algorithm
โข PPC + Sparse Representationโ Is the feedback system stable? YES.
โข Examples
Examples
โข Controlled plant
the elements in and are generated by random sampling from . has 3 unstable eigenvalues.
โข The horizon length is .โข Two designs:
โ Sparsity-promoting design ()โ Energy-limiting design (regularization)
Transmitted Control Packets
Histogram of Quantized Transmitted Values
Proposed
Conventional
-norm of the state
Sparsity-promoting (proposed)
Energy-limiting (Conventional)
-norm of the state
Sparsity-promoting (proposed)
Energy-limiting (Conventional)
Proposed method leads to more effective compression than the conventional method without much distortion.
Parameter vs sparsity and performance
๐ฝ (๐ )=โ๐ฅ (๐|๐ )โ๐2+โ๐=0
๐
โ๐ฅ (๐|๐ )โ๐2+๐โ๐โ1
โ
Sparsityโ5โโ๐โ0=5โ1100
โ๐=1
100
โ๐ (๐ )โ0
Performanceโโ๐ฅโ2
Conclusionโข Sparsity-promoting optimization () for packetized
predictive control.โข Sparse representation of packets leads to efficient
compression of transmitted signals.โข The feedback system can be practically stable.โข Examples show the effectiveness of our method.โข Future work may include
โ Bit-rate analysis of optimized controlโ Robustness against disturbances in the plant
Conclusionโข Sparsity-promoting optimization () for packetized
predictive control.โข Sparse representation of packets leads to efficient
compression of transmitted signals.โข The feedback system can be practically stable.โข Examples show the effectiveness of our method.โข Future work may include
โ Bit-rate analysis of optimized controlโ Robustness against disturbances in the plant
Grazie!