A randomized approximation algorithm for the minimal-norm static-output-feedback problem☆
Introduction
The problem of finding necessary and sufficient conditions, for the existence of static-output-stabilizing-feedbacks (SOF), which can be computed in reasonable (i.e. polynomial) time, has a long history (see Syrmos, Abdallah, Dorato, and Grigoradis (1997) for a survey). The problem is known to be NP-hard if structural-constraints or bounds are imposed on the entries of the controllers (see Blondel and Tsitsiklis (1997) and Nemirovskii (1993)), but unknown and suspected to be NP-hard, otherwise. Obviously, in real-life, when one searches a minimal-norm SOF, it is always done in a bounded region. Thus, minimal-gain SOF with bounded entries is obviously NP-hard problem. Pole-placement and simultaneous stabilization via SOF are also NP-hard (see Fu (2004) and Blondel and Tsitsiklis (1997), resp.). Thus, practically, one should expect that only approximation or randomized algorithms will be able to cope with these problems.
Many problems can be reduced to the constrained SOF problem. These include the reduction of the minimal-degree dynamic-feedback problem, robust or decentralized stability via static-feedback and PID controllers to this problem (see Blondel and Tsitsiklis (1997), Mesbahi (1999) and Zheng, Wang, and Lee (2002), resp.). A formulation of the reduced-order filter problem as constrained SOF problem, is considered in Borges, Calliero, Oliveira, and Peres (2011).
The solution of the SOF problem is important for systems which models structural dynamics, and naturally needs a static feedback that can be built into the structure (e.g. buildings and bridges subject to earthquakes, strong winds and unpredicted dynamic loadings—see Spencer and Sain (1997), Xu and Teng (2002), Yang, Lin, and Jabbari (2003) and Polyak, Khlebnikov, and Shcherbakov (2013), where it is shown that optimal SOF’s, although costless, may achieve similar performance as optimal dynamic feedbacks). Note that the long-term memory of dynamic feedbacks is useless in the case of unpredicted dynamic loadings.
The suggested algorithm that will be called the Ray-Shooting (RS) algorithm, does not make use of any heavy tools like Semi-Definite Programming (SDP) or Linear, or Bilinear Matrix Inequalities (LMI’s and BMI’s resp.). Note that the existence of BMI’s solutions is NP-hard problem (see Toker and Özbay (1995)). The least-rank dynamic-output-feedback problem is considered in Mesbahi (1999) using SDP. This can be solved with the RS method, by applying a Binary-Search on the rank and using its formulation as a SOF problem, with much less computational efforts and without the need of finding any feasible initial point. Also, the problem of optimal abscissa via SOF can be solved with the RS method, by applying Binary-Search on the abscissa.
In Vidyasagar and Blondel (2001) the problem of a common Lyapunov matrix, and the problems of static-stability and simultaneous static stability, are treated, using the probabilistic method (i.e. the “generate and check” strategy). The article deals with the problem of counting the minimal number of samples that guarantee a given probability threshold of success, through the use of the Chernoff bound and bounds on the Vapnik–Chervonenkis dimension (VC-dimension) of the problem, without knowing the specific distribution of successful examples and without using the structure of the given system (i.e. the specific matrices). In this article, it is shown that at least samples are needed to guarantee with confidence that the empirical probability is uniformly-close to the exact probability (with ), where is the VC-dimension of the problem, where for the SOF problem it is shown there that ( being the state-space dimension). Thus, for at least samples are needed ().
One way to overcome this “curse of dimensionality” is presented in Tempo, Calafiore, and Dabbene (2005) and Tempo and Ishii (2007). Concerning the problem of Robust Stability via SOF, it is proved in Tempo et al. (2005) that the empirical performance measure is probability-close to the exact performance with probability at least and confidence at least , if one takes samples of the controller parameters and samples of the system uncertainty (). Thus, for one needs samples of the controller parameters and samples of the system uncertainty, which results in 32,367,984 evaluations of the performance measure. The RS algorithm seems to practically overcome this obstacle and at least suggests another way to cope with this problem.
The structure of the article is as follows: in Section 2, we set notions, definitions, and give some lemmas. In Section 3, we introduce a lemma which provides the basis for the RS algorithm. We next introduce an approximation algorithm that applies the RS method again, in order to find a minimal-gain SOF. In Section 4, we consider the Alternating-Projections (AP) algorithm introduced in Yang and Orsi (2006), which solves the problem of pole-placement via static-feedback, and we revise this algorithm with some improvement. We also consider the Hide-And-Seek algorithm, introduced in Bélisle (1992), which solves the global optimization problem of continuous functions on compact domains. In Section 5, we compare the results of the RS algorithm with the algorithms: AP, Hide-And-Seek, Mixed LMI/Randomized Method (see Arzelier, Gryazina, Peaucelle, and Polyak (2010)), HIFOO (see Gumussoy, Henrion, Millstone, and Overton (2009)) and HINFSTRUCT (see Apkarian and Noll (2006)). In Section 6, we revise a proof for the convergence in probability of the RS algorithm and discuss its complexity under some reasonable assumptions. In Section 7, we conclude with some remarks concerning the comparison between the algorithms and discuss the limitations of the RS method.
Section snippets
Preliminaries
The complex, real and rational fields are denoted by , resp. We denote by the open unit disk and by the open left half-plane. For we denote by the set . For we denote by , its real and imaginary parts, resp. By we denote the set of nonnegative real numbers. For a set , we denote by the linear span of . For a square matrix , we denote by the spectrum of . For a matrix , we denote by its conjugate transpose, and by its
Description and correctness of the RS algorithm
This section contains three subsections. In the first one we provide a procedure for pole-placement via static-state-feedback, which will be used by the RS algorithm to find a starting point for the SOF feasibility problem. Any other algorithm for pole-placement via static-state-feedback can be used instead (see Pandey et al. (2014) for a survey). In the second subsection we describe the RS feasibility algorithm and in the third one we describe the RS optimization algorithm.
The Alternating-Projection and the Hide-And-Seek algorithms
In this section, we briefly mention the two randomized algorithms to be compared with the RS algorithm in Section 5. Let denote the set of all possible closed-loop matrices, i.e. the set of all for any . Let be the closed-set of -stable matrices, related to discrete-time systems, where , and let be the closed-set of -stable matrices, related to continuous-time systems, where . The Alternating-Projection taken from Yang
Some experiments
In the following experiments we used examples taken from Leibfritz (2003), Leibfritz and Lipinski (2004) and Leibfritz (2003). In our experiments the continuous-time systems AC1-AC18, BDT2, DIS4, DIS5, HE1, HE3-7, IH, JE2-3, NN1-3, NN5-7, NN9, NN12-17, PAS, ROC1-7, TF1-3, WEC1, TMD, HF2D10-11 and CSE2 were considered. In all the experiments we used , as a bound on the number of iterations, and , , where , for the Quasi-Binary-Search in Step 4 of
Convergence and complexity issues
The results of this section are based mainly on Solis and Wets (1981). We revise their algorithm and make it more adaptive and more flexible in order to fit it to the suggested algorithm—on each one of its phases. Their proof of convergence is generalized here to adaptive probabilities and to flexible generating kernels for the new sample points (i.e. generating relations instead of generating functions). Let and let be the closed sets constructed from defined in (8), by
Concluding remarks
Concerning the comparison between the RS algorithm and the above mentioned algorithms, we conclude the following:
The Hide-And-Seek algorithm, had a total failure, regarding the SOF problem. The AP algorithm performs better than the RS-PHASE-I in terms of run-time, for all the considered systems. For all the systems considered, the RS-PHASE-I algorithm outperforms the AP algorithm in the percent of success. The RS algorithm outperforms HIFOO and HINFSTRUCT w.r.t. the CPU time, in almost all the
Acknowledgments
I wish to thank my learning companion, Mr. Eliezer Fetterman, for his deep questions and for the fruitful discussions with him on the philosophical and practical meaning of static-feedbacks. I also wish to thank to anonymous reviewers for their helpful and encouraging comments and for bringing to my attention many of the references.
Yossi (Joseph) Peretz was born in Beer-Sheva, Israel, in 1963. He has received B.Sc. in Mathematics and Computer Sciences in 1988, B.Sc. in Mechanical Engineering in 1995, M.Sc. in Mathematics and Computer Sciences in 1995 and Ph.D. in Mathematics and Computer Sciences in 1999, all from the Ben-Gurion University of The Negev, Beer-Sheva, Israel. Since 1998 he serves as a senior lecturer at the Lev Academic Center, Jerusalem College for Technology (JCT), Jerusalem, Israel. His interests are in
References (48)
Univariate polynomials: nearly optimal algorithms for numerical factorization and root-finding
Elsevier Journal of Symbolic Computation
(2002)A characterization of all the static stabilizing controllers for LTI systems
Linear Algebra and its Applications
(2012)- et al.
Static output feedback: A survey
Automatica
(1997) - et al.
Monte Carlo and Las Vegas randomized algorithms for systems and control
European Journal of Control
(2007) - et al.
Probabilistic solutions to some NP-hard matrix problems
Automatica
(2001) - et al.
Generalized pole placement via static output feedback: A methodology based on projections
Automatica
(2006) - et al.
On the design of multivariable PID controllers via LMI approach
Automatica
(2002) - Apkarian, P. Comparison of HINFSTRUCT Matlab Robust Control Toolbox R2010b and HIFOO 3.0 with HANSO 2.0....
- et al.
Nonsmooth synthesis
IEEE Transactions on Automatic Control
(2006) - et al.
Generalized Eigenproblem algorithms and software for algebraic Riccati equations
Proceedings of the IEEE
(1984)
The rate of convergence in the method of alternating projections
St. Petersburg Mathematical Journal
On projection algorithms for solving convex feasibility problems
SIAM Review
On the method of cyclic projections for convex sets in Hilbert space
Convergence theorems for a class of simulated annealing algorithms on
Applied Probability
Numerical solution of algebraic Riccati equations
NP-hardness of some linear control design problems
SIAM Journal on Control and Optimization
Numerical methods for algebraic riccati equations
design with pole placement constraints: an LMI approach
IEEE Transactions on Automatic Control
Robust pole placement in LMI regions
IEEE Transactions on Automatic Control
Pole placement by static output feedback for generic linear systems
SIAM Journal on Control and Optimization
Pole placement via static output feedback is NP-hard
IEEE Transactions on Automatic Control
Cited by (0)
Yossi (Joseph) Peretz was born in Beer-Sheva, Israel, in 1963. He has received B.Sc. in Mathematics and Computer Sciences in 1988, B.Sc. in Mechanical Engineering in 1995, M.Sc. in Mathematics and Computer Sciences in 1995 and Ph.D. in Mathematics and Computer Sciences in 1999, all from the Ben-Gurion University of The Negev, Beer-Sheva, Israel. Since 1998 he serves as a senior lecturer at the Lev Academic Center, Jerusalem College for Technology (JCT), Jerusalem, Israel. His interests are in control theory and its applications, optimization, randomized algorithms, parallel algorithms, linear algebra algorithms, complexity, computability and cryptography.
- ☆
The material in this paper was not presented at any conference. This paper was recommended for publication in revised form by Associate Editor Fabrizio Dabbene under the direction of Editor Richard Middleton.
- 1
Tel.: +972 2 6751016; fax: +972 2 6751046.