Further input-to-state stability subtleties for discrete-time systems
7
0
0
Hele tekst
(2) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. Further Input-to-State Stability Subtleties for Discrete-time Systems M. Lazar W.P.M.H. Heemels A.R. Teel . Department of Electrical Engineering, Eindhoven University of Technology, The Netherlands (e-mail: m.lazar@tue.nl) Department of Mechanical Engineering, Eindhoven University of Technology, The Netherlands (e-mail: m.heemels@tue.nl) Department of Electrical and Computer Engineering, University of California, Santa Barbara (e-mail: teel@ece.ucsb.edu). Abstract: This paper considers input-to-state stability (ISS) analysis of discrete-time systems using continuous Lyapunov functions. The contributions are as follows. Firstly, the existence of a continuous Lyapunov function is related to inherent input-to-state stability on compact sets with respect to both inner and outer perturbations. If the Lyapunov function is -continuous, this result applies to unbounded sets as well. Secondly, continuous control Lyapunov functions are employed to construct input-to-state stabilizing control laws for discrete-time systems subject to bounded perturbations. The goal is to design a receding horizon control scheme that allows the optimization of the ISS gain along a closed-loop trajectory. Keywords: Discrete-time, Stability, Input-to-state stability, Lyapunov methods, Predictive control. 1. INTRODUCTION This paper focuses on the design of robust stabilizing control laws in general and the design of robust model predictive control (MPC) laws in particular. A main motivation for this research is that nominally stabilizing (MPC) controllers might have no robustness properties with respect to disturbances. This aspect was for the first time shown in (Grimm et al., 2004), where it was indicated that asymptotically stable MPC closedloop systems may have zero robustness in the presence of arbitrary small perturbations. This undesired phenomenon was revealed in (Grimm et al., 2004) by showing that an asymptotically stable MPC closed-loop system is not robustly asymptotically stable for arbitrary small perturbations. More recently, in (Lazar et al., 2009) the same phenomenon was exposed for globally asymptotically stable (GAS) discrete-time systems in terms of a lack of input-to-state stability (ISS) (Jiang and Wang, 2001) to arbitrarily small inputs. The conclusion drawn in (Lazar et al., 2009) is that GAS discrete-time systems which admit a discontinuous Lyapunov function are not necessarily inherently ISS, not even locally. As such, this observation issued a valid warning for nominally stabilizing MPC schemes, as in the case of nonlinear or hybrid systems the MPC candidate Lyapunov function is typically a discontinuous function. To deal with the phenomenon of non-robustness, it would be useful to establish sufficient conditions under which nominally stable systems are inherently ISS. A conjecture that is frequently employed in the MPC literature is that the existence of a continuous Lyapunov function is sufficient for inherent ISS. The first contribution of this paper is to provide a formal statement of this conjecture along with a complete proof. To this end we will introduce a property called -continuity, which generalizes H¨older continuity on compact sets, and a property called -continuity, which generalizes global H¨older continuity. It is proven that continuity on a compact set is equivalent. with -continuity and that a stronger type of global uniform continuity is equivalent with -continuity. These results enable us to establish that every discrete-time system that admits a continuous Lyapunov function is inherently ISS on a robustly positively invariant compact set, with respect to both inner and outer perturbations. The inclusion of inner perturbations (e.g., measurement noise or estimation error) is particularly relevant for MPC, as most of the ISS results in this framework are limited to outer perturbations (e.g., additive disturbances). A previous article that considered nominal robustness of MPC in terms of both inner and outer perturbations is (Messina et al., 2005), where it was established that existence of a continuous Lyapunov function is equivalent with robust GAS (RGAS) and semiglobal practical asymptotic stability (SPAS). Also, therein it was established that RGAS and SPAS are equivalent with attenuated ISS and integral ISS, respectively. As most robust stability results in MPC make use of the ISS framework, see, e.g., (Limon et al., 2006; Magni et al., 2006; Lazar et al., 2008; Lazar and Heemels, 2009), and integral ISS does not necessarily imply ISS (Angeli et al., 2000), in this work we focus on establishing inherent ISS with respect to both inner and outer perturbations. In this context it is worth to mention the article (Roset et al., 2008), where a connection was established between ISS to outer perturbations and ISS to inner perturbations for general constrained discrete-time systems. The second contribution of the paper deals with the design of stabilizing MPC schemes that explicitly use a pre-defined continuous control Lyapunov function (CLF). The results established in the first part of the paper are used to show that inherent ISS is guaranteed for the resulting closed-loop system. Moreover, the recently introduced notion of optimized ISS (Lazar and Heemels, 2008) is employed to improve the disturbance rejection properties of the controller..
(3) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. 2. PRELIMINARIES Let , , and denote the field of real numbers, the set of non-negative reals, the set of integer numbers and the set of non-negative integers, respectively. For every
(4) and we define
(5) #
(6) # and similarly ,
(7) and
(8) #
(9) #
(10) . For a sequence
(11) *$ with *$
(12) , $
(13) , let
(14) *$ $
(15) and let
(16)
(17) *$ . For a set , we denote by the interior, by the boundary and by the closure of . For two arbitrary sets and , let
(18) + , +
(19)
(20) ,
(21) denote their Minkowski sum and let
(22) +
(23) + denote their Pontryagin difference. A polyhedron (or a polyhedral set) in is a set obtained as the intersection of a finite number of open and/or closed half-spaces. A polytope is a closed and bounded polyhedron. Let denote an arbitrary &norm. For a matrix
(24) let
(25) . denote its corresponding induced matrix norm. A real valued scalar function with for all
(26) is called a positive function. Let
(27) . A function
(28) belongs to class if it is continuous, strictly increasing and
(29) . A function belongs to class if
(30) and (
(31) . A function belongs to class if for each fixed #
(32) ,
(33) #
(34) and for each fixed (
(35) , (
(36) is decreasing and (
(37) #
(38) . Fact 1. Let
(39) ,
(40) ,
(41) and let.
(42) . Then
(43)
(44) with
(45) is a -function and is a -function.. +, where the supremum is an attainable maximum by continuity of the norm and , respectively, and Fact 7. Definition 8. A function is called -continuous on (or shortly, KC()) if there exists
(46) , for some
(47) , such that
(48) and + , . + , for all +
(49) ,
(50) . Definition 9. A function is called -continuous on (or alternatively, globally -continuous (GKC)) if there exists such that
(51) and + , + , for all +
(52) ,
(53) . Notice that the set of HC (GHC) functions is a subset of KC (GKC) functions. Lemma 10. Let be a compact set. A function is UC() if and only if it is -continuous on . The proof of Lemma 10 is given in Appendix A. Corollary 11. Let be a compact set. A function is continuous on if and only if it is -continuous on . Proof. The claim follows from Lemma 10 in combination with Fact 4 and Fact 5. Lemma 12. A function is unbounded globally uniformly continuous if and only if it is -continuous. The proof of Lemma 12 is given in Appendix B. Corollary 13. Every -continuous function is globally uniformly continuous. Proof. The claim follows from Lemma 12 and the fact that every unbounded GUC function is a GUC function. . 3. UNIFORM CONTINUITY ON COMPACT SETS. 4. INHERENT INPUT-TO-STATE STABILITY. Definition 2. A function is called uniformly continuous on (or shortly, UC()) if there exists a positive function such that for all
(54) and all +
(55) ,
(56)
(57) with + , it is holds that + , . If is UC( ), then called globally uniformly continuous (GUC). If is GUC and moreover,
(58) , is called unbounded GUC. Definition 3. A function is called H¨older continuous on (or shortly, HC()) if there exist
(59) ,
(60) such that + , + , for all +
(61) ,
(62) . If is HC( ), then is called globally H¨older continuous (GHC). If
(63) , then is called Lipschitz continuous. Fact 4. A function that is UC() is continuous on , for . Fact 5. Heine-Cantor Theorem. Let be a compact set and let denote a continuous function on . Then is UC(). Fact 6. Let
(64) be arbitrary compact sets and let denote a continuous function on . Also, let
(65) + +
(66) . Then and are compact sets. Fact 7. Let be a compact set and let denote a continuous function on . Then attains its minimum and maximum on . . Consider the discrete-time nominal system (1) +#
(67) +#
(68) #
(69)
(70) and its perturbed counterpart (2) +#
(71) +#
(72) #
(73) #
(74) #
(75)
(76) where + is the state trajectory, is an unknown inner perturbation trajectory, is an unknown outer perturbation trajectory and , are nonlinear maps with +
(77)
(78)
(79) + for all +
(80) and
(81) . The reason for distinguishing between inner and outer perturbations will be made clear later in the section. For ease of notation we will use +, and , respectively, to also denote a vector in . Let , and denote subsets of that contain the origin in their interior. Definition 14. A set with
(82) is called a positively invariant (PI) set for system (1) if for all +
(83) it holds that +
(84) . A set with
(85) is called a robustly positively invariant (RPI) set for system (2) and
(86) , or shortly, RPI(
(87) ), if for all +
(88) it holds that +
(89)
(90)
(91) for all
(92)
(93) . Definition 15. We call system (1) asymptotically stable in , or shortly AS(), if there exists a -function such that for each +
(94) it holds that +# +
(95) #, #
(96) . We call system (1) GAS if it is AS( ). Definition 16. We call system (2) input-to-state stable in for inputs in and , or shortly ISS(,, ), if there exist a function and -functions , such that, for each +
(97) ,. be a Let be a compact set and let continuous function on . Let
(98) + and
(99).
(100) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. all
(101) $ with $
(102) for all $
(103) and all
(104) $ with $
(105) for all $
(106) , it holds that the corresponding state trajectory of (2) satisfies +# +
(107) #
(108)
(109) for all #
(110)
(111) . The system (2) is globally ISS if it is ISS( , , ). Throughout this article we will employ the following sufficient conditions for analyzing ISS. Theorem 17. Let
(112)
(113)
(114) ,
(115)
(116) , with
(117) . Let be a function with
(118) and consider the following inequalities: (3a) + + +
(119) +
(120)
(121) + + (3b) (i) If is a RPI(, ) set for system (2) and inequalities (3) hold for all +
(122) ,
(123) and all
(124) , then system (2) is ISS(,, ). If inequalities (3) hold for all +
(125)
(126)
(127) , then system (2) is globally ISS. (ii) If is a PI set for system (1) and inequalities (3) hold for all +
(128) (+
(129) ),
(130)
(131) and
(132)
(133) , then system (1) is AS() (GAS). Definition 18. A function that satisfies the hypothesis of Theorem 17-! for some sets
(134) is called an ISS Lyapunov function on for system (2), or shortly, an ISS(,, ) Lyapunov function. An ISS( , , ) Lyapunov function is called a global ISS Lyapunov function. Definition 19. A function that satisfies the hypothesis of Theorem 17-!! is called a Lyapunov function on for system (2). A Lyapunov function on is called a global Lyapunov function. The interested reader is referred to (Jiang and Wang, 2001; Lazar et al., 2008) for a proof of Theorem 17. Notice that in contrast to the continuous-time case, in discrete-time the above sufficient conditions for ISS (GAS) only require the continuity of the system dynamics and the (ISS) Lyapunov function at +
(135) , as indicated in (Lazar et al., 2006, 2008). However, in what follows we will focus on continuous Lyapunov functions. The interested reader is referred to (Lazar et al., 2009) for ISS subtleties for discrete-time systems regarding discontinuous Lyapunov functions. The next two theorems consider the case when the perturbed system (2) satisfies +
(136)
(137)
(138) + for all +
(139)
(140)
(141) , which exposes the difference between inner and outer perturbations. The next result relates existence of a continuous Lyapunov function to inherent ISS for system (1). Theorem 20. Let , and be compact subsets of with the origin in their interior. Suppose that is an RPI(
(142) ) set for system (2). Furthermore, suppose that system (1) admits a continuous Lyapunov function on . Then, system (2) is ISS(,, ). Proof. The hypothesis implies that there exists a continuous function that satisfies (3a) for all +
(143) . Thus, it satisfies (3a) for all +
(144) as well. Next, we prove that satisfies (3b) for all +
(145)
(146)
(147) . Let +
(148) + . As is a Lyapunov function on for system (1), by Definition 19 it follows that +
(149) + + . +
(150) . (4). Since is a RPI(
(151) ) set for system (2), from Corollary 11, Fact 6, the reverse triangle inequality and using +
(152) + we also have that there exist
(153)
(154) such that (5a) + +
(155) + +
(156) (5b) + +
(157) (5c) for all +
(158)
(159)
(160) . Then, using the fact that
(161) for any
(162)
(163) and adding (5b) and (4) yield + + + for all +
(164) and all
(165) . Adding and subtracting + and + in the above inequality and using (5a) and (5c), respectively, along with the fact that
(166) for any
(167)
(168) , yield + + +. .
(169). . for all +
(170)
(171)
(172) . Letting
(173)
(174) and
(175)
(176) yields that satisfies (3b) for all +
(177)
(178)
(179) . Hence, the claim follows from Theorem 17-!. A global correspondent of Theorem 20 is stated next. Theorem 21. Suppose that system (1) admits a -continuous global Lyapunov function that satisfies (3b) for all +
(180) ,
(181)
(182) and
(183)
(184) with a -continuous
(185) . Then, system (2) is globally ISS. Proof. The claim follows via the reasoning used in the proof of Theorem 20, in combination with Definition 9. Consider next the discrete-time nominal system with a control input (6) +#
(186) +#
(187) )+#
(188) #
(189)
(190) and its perturbed counterpart +#
(191) +#
(192) )+# # #
(193) #
(194)
(195) (7) where ) is a state-feedback control law and is a nonlinear map with
(196)
(197) . For ease of notation we will use ) to also denote a vector in . Let be a subset of with
(198) . Definition 22. Let and be compact sets. A map is called uniformly -continuous on if there exists
(199) , for some
(200) , such that
(201) and +
(202) ) ,
(203) ) + , for all )
(204) and all +
(205) ,
(206) . A map is called uniformly -continuous if there exists
(207) such that +
(208) ) ,
(209) ) + , for all )
(210) and all +
(211) ,
(212) . Definition 23. Let and . Suppose that ) is a known map with )
(213) . A set with
(214) is called a PI set for system (6) if for all +
(215) it holds that +
(216) )+
(217) . A set with
(218) is called a RPI(
(219) ) set for system (7) if for all +
(220) it holds that +
(221) )+
(222) for all
(223)
(224) . Theorem 24. Let , , and be compact sets with the origin in their interior and let ) be a known map with )
(225) . Suppose that is a RPI(
(226) ) set for system (7). Furthermore, suppose that system (6) admits a continuous Lyapunov function on and the map is uniformly -continuous on . Then, the system (7) is ISS(,, )..
(227) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. Proof. Let +
(228) + . Observe that for any +
(229) and
(230) , +
(231) and thus, +
(232) ) +
(233) , since system (6) admits a Lyapunov function on . This further implies implies that there exists a continuous function such that +
(234) +
(235) (8) +
(236) ) + + From Corollary 11 and Definition 22 it follows that there exist.
(237)
(238) such that: +
(239) ) + +
(240) ) + +
(241) ) + +
(242) ) + for all +
(243)
(244) . Adding and subtracting +
(245) ) + in (8) and using the above inequality along with the fact that
(246) for any
(247)
(248) , yield + +
(249) ) + + Then, the claim follows via the reasoning used in the proof of Theorem 20, by considering a perturbed system (2) with +
(250)
(251)
(252) +
(253) )+ for all +
(254)
(255)
(256) . Theorem 25. Let ) be a known map with )
(257) . Suppose that system (6) admits a -continuous global Lyapunov function that satisfies (3b) for all +
(258) ,
(259)
(260) and
(261)
(262) with a -continuous
(263) . Furthermore, suppose that is a uniformly continuous map. Then, system (7) is globally ISS. Proof. The claim follows via the reasoning used in the proof of Theorem 24 in combination with the reasoning employed in the proof of Theorem 21 and Definition 22. Remark 26. The local inherent ISS results of Theorem 20 and Theorem 24 are obtained using a similar reasoning as the one employed in (Messina et al., 2005) to establish local inherent robustness (RGAS and SPAS) from a global continuous Lyapunov function. The ISS results established in the abovementioned theorems require the existence of a suitable RPI set, which is an additional requirement compared to (Messina et al., 2005). Similarly to the results in (Messina et al., 2005), the result of Theorem 20, which applies to systems without a control input, does not require continuity of the system dynamics and the result of Theorem 24, which applies to systems with a control input, does not require continuity of the statefeedback control law. The same holds for the ISS results of Theorem 21 and Theorem 25, respectively, which do not a have a correspondent in (Messina et al., 2005). Notice that if
(264) , then the continuity assumptions on can be removed in Theorem 24 and Theorem 25. . Moreover, as it was recently pointed out in (Lazar and Heemels, 2008), besides guaranteeing inherent ISS, it would be desirable to optimize the ISS gain of the closed-loop system, i.e., the gain of the functions
(265)
(266) . In what follows we briefly recall some of the results in (Lazar and Heemels, 2008), which consider outer perturbations only, and provide some new insights and extensions. For clarity of exposition we will treat the case of outer perturbations separately from the case of inner perturbations, as there is a crucial difference between the two cases. Let us begin with the case of outer perturbations. To optimize disturbance attenuation for the closed-loop system, at each time instant #
(267) and for a given +#
(268) , it would be desirable to simultaneously compute a control action )+#
(269) that satisfies !. +#
(270) )+# +# +#
(271)
(272) (9). for (
(273) #( ,
(274) , #
(275) and (ii) minimize #. Next, we recall a solution to this problem that was given in (Lazar and Heemels, 2008). Let be a polytope and let , %
(276)
(277) , be the vertices of . Next, consider a finite set of simplices
(278)
(279) with each simplex equal to the convex hull of the origin and a subset of the vertices of , and such that
(280) ,
(281) for !
(282) ",
(283) for all !. More precisely,
(284)
(285)
(286)
(287) and
(288)
(289)
(290)
(291) (i.e., %
(292)
(293) %
(294)
(295) ) with
(296)
(297) linearly independent. For each simplex we define the matrix
(298)
(299) , which is invertible. Let #, #
(300) , be optimization variables associated with each vertex . In what follows, when the time dependency is irrelevant, it will be omitted for brevity of presentation. Consider the following set of inequalities depending on ) and : +
(301) )+ + +
(302) (10a) +
(303) )+ + +
(304) %
(305)
(306) (10b) Theorem 28. (Lazar and Heemels, 2008). Let be a convex function. Suppose that for some
(307) and +
(308) there exist )+
(309) and #
(310) such that (10a) and (10b) hold. Then (9) holds for the same )+, with (
(311) ( and. 5. OPTIMIZED INPUT-TO-STATE STABILITY This section illustrates how a continuous control Lyapunov function (CLF) (Kellett and Teel, 2004) can be employed to design a stabilizing MPC scheme that is inherently input-tostate stabilizing. As such, let us formally define a CLF for system (6). Definition 27. Let be a candidate Lyapunov function on , i.e., a function that satisfies (3a) for all +
(312) . is called a control Lyapunov function on for system (6) if there exists a map ) with )
(313) such that is a PI set for system (6) and +
(314) )+ + +
(315) +
(316) A control Lyapunov function on is called a global control Lyapunov function. .
(317).
(318) . . (11).
(319)
(320) , !
(321)
(322)
(323) . where
(324)
(325)
(326) and let be a Let for some function that satisfies .
(327)
(328) ; for example,
(329) Let
(330) , let denote an arbitrary cost function that is zero at zero and define
(331) ) . and Problem 29. Let
(332) , , be given. At time #
(333) let the state +# be known and minimize the cost over #
(334) #, subject to.
(335) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. #
(336)
(337) #
(338)
(339) +#
(340) #
(341)
(342) (12a) +#
(343) # +# +#
(344) (12b) +#
(345) # +# +# #
(346) %
(347)
(348) (12c) Set (13) )+#
(349) feas # . Theorem 31. Let
(350) , let be a convex function and let
(351) + be an affine map of + for all ). If for +
(352) and + +
(353) and
(354)
(355) there exist ) such that (15a) and (15b) hold, then (14) holds for the same ) +, with (
(356) ( and
(357) (16)
(358) . In the above problem feas # denotes a control law that selects the first element of an arbitrary feasible sequence of inputs #
(359) for all #
(360) . Moreover, because , and (12) likewise, is a function of +#, feas # is a function of +# as well. Theorem 30. Let
(361)
(362)
(363) , a continuous and convex be given. CLF , a cost and a cost Suppose that Problem 29 is feasible for all +
(364) and the map is uniformly -continuous on . Then the closed-loop system (7)-(13) is ISS(, , ).. The proof of the above theorem, which is similar, mutatis mutandis, to the proof of Theorem 28 is omitted due to space limitations.. Proof. For any #
(365) and +#
(366) it holds that +#
(367) +#
(368) )+# #
(369) for all #
(370) by (12a). Hence, Problem 29 is recursively feasible for all +
(371) . Then, the result follows from (12b) and Theorem 24 with
(372) . It is important to observe that Problem 29, although it inherently guarantees a constant ISS gain, it provides freedom to optimize the ISS gain of the closed-loop system, by minimizing As such, in the variables #
(373)
(374) # via the cost . reality the gain # of the function can be much smaller for # # , for some #
(375) , depending on the actual state trajectory. Next, consider the case of inner perturbations. The goal is now to simultaneously compute a control action ) +#
(376) at time #
(377) that satisfies ! +#
(378) )+# +# +#
(379)
(380) (14) for (
(381) #( ,
(382) , #
(383) and (ii) minimize #. As done for outer perturbations, let be a polytope and let , *
(384)
(385)
(386) , be the vertices of . Next, consider a finite set of simplices
(387)
(388) with each simplex equal to the convex hull of a subset of the vertices of and the origin, and such that
(389) ,
(390) for !
(391) ",
(392) for all !. More precisely,
(393)
(394)
(395)
(396) and
(397)
(398)
(399)
(400) (i.e., *
(401)
(402) *
(403)
(404) ) with
(405)
(406) linearly independent. For each simplex we define the matrix
(407)
(408) , which is invertible. Let
(409) be variables associated with each vertex . Next, suppose that + is known. Notice that the assumption that + is known is only used here to show how one can transform (14) into a finite dimensional problem. The dependence on + will be removed later, leading to a main stability result and an MPC algorithm that only use the perturbed state + for feedback, see Problem 32. Consider the following set of constraints: (15a) +
(410) ) + + +
(411) + + +
(412) +
(413) ) (15b) *
(414)
(415) . . .
(416)
(417) and is the where corresponding induced matrix norm.. Based on the result of Theorem 31 we are now able to formulate a finite dimensional optimization problem that results in closedloop ISS with respect to inner perturbation # and moreover, in optimization of the closed-loop trajectory-dependent ISS gain. This will be achieved only based on knowledge of the perturbed state + # and the set .
(418)
(419)
(420) and let be defined similarly as done Let for outer perturbations. Define next: +#
(421) + (17) min
(422)
(423)
(424) . and max +#
(425). .
(426)
(427)
(428) . +. (18). and be Problem 32. Let
(429) , , # be known given. At time #
(430) let the perturbed state + and minimize the cost over #
(431) #, subject to #
(432)
(433) #
(434)
(435) (19a) +
(436) #
(437)
(438) +
(439) +#
(440)
(441) + #
(442) (19b) +# max +#
(443) (19c) +#
(444) # min +# +#
(445) # min (19d) max +# #
(446) *
(447)
(448) Set (20) ) +#
(449) feas # In the above problem feas # denotes a control law that selects the first element of an arbitrary feasible sequence of inputs #
(450) for all #
(451) . Moreover, because , and (19) # as likewise, is a function of + #, feas # is a function of + well. Theorem 33. Let
(452)
(453)
(454) , a continuous and convex be given. CLF , a cost and a cost Suppose that Problem 32 is feasible for all +
(455) and is an affine map of + for all ). Then the closed-loop system (7)-(20) is ISS(, , ). #
(456) it holds that Proof. For any #
(457) and + # . Hence, it follows +#
(458) +#
(459)
(460) + that +#
(461) +#
(462) ) +#
(463) for all #
(464) by (19b). Hence, Problem 32 is recursively feasible for all +
(465) . Then, the result follows from (19c) and Theorem 24 with
(466) . Notice that a more restrictive condition is imposed on the system dynamics for inner perturbations, i.e., should be an affine map in + for all ), than for outer perturbations, i.e., .
(467) UKACC International Conference on CONTROL 2010, 7-10 September, 2010, Coventry University, UK. should be a uniformly -continuous map. Under the stronger condition, the proposed MPC schemes can be combined in one algorithm that yields optimized ISS(,, ). 6. CONCLUSIONS Input-to-state stability analysis of discrete-time systems using continuous Lyapunov functions was considered. Firstly, the existence of a continuous Lyapunov function was related to inherent input-to-state stability on compact sets with respect to both inner and outer perturbations. For -continuous Lyapunov functions it was shown that this result applies to unbounded sets as well. Secondly, continuous control Lyapunov functions were employed to construct input-to-state stabilizing control laws for discrete-time systems subject to bounded perturbations.. Appendix A. PROOF OF LEMMA 10 Let us begin with the only if part. As is UC(), without any loss of generality we can take to be a positive, non-decreasing function. Let
(468) . Next, let
(469)
(470)
(471)
(472)
(473)
(474)
(475) for Notice that and, if
(476) , then all
(477) . Also,
(478) . Observe that the function is non-decreasing and it extends the domain of . Next, we prove that there exists
(479) such that
(480) and for all
(481) . Define
(482) #
(483) (A.1) (
(484) . REFERENCES Angeli, A., Sontag, E.D., and Wang, Y. (2000). A characterization of integral input-to-state stability. IEEE Transactions on Automatic Control, 45(6), 1082–1097. Grimm, G., Messina, M.J., Tuna, S.E., and Teel, A.R. (2004). Examples when nonlinear model predictive control is nonrobust. Automatica, 40(10), 1729–1738. Jiang, Z.P. and Wang, Y. (2001). Input-to-state stability for discrete-time nonlinear systems. Automatica, 37, 857–869. Kellett, C.M. and Teel, A.R. (2004). Discrete-time asymptotic controllability implies smooth control-Lyapunov function. Systems & Control Letters, 52, 349–359. Lazar, M. and Heemels, W.P.M.H. (2008). Optimized inputto-state stabilization of discrete-time nonlinear systems with bounded inputs. In American Control Conference. Seattle. Lazar, M. and Heemels, W.P.M.H. (2009). Predictive control of hybrid systems: Input-to-state stability results for suboptimal solutions. Automatica, 45(1), 180–185. Lazar, M., Heemels, W.P.M.H., and Teel, A.R. (2009). Lyapunov functions, stability and input-to-state stability subtleties for discrete-time discontinuous systems. IEEE Transactions on Automatic Control, 54(10), 2421–2425. Lazar, M., Heemels, W.P.M.H., Weiland, S., and Bemporad, A. (2006). Stabilizing model predictive control of hybrid systems. IEEE Transactions on Automatic Control, 51(11), 1813–1818. Lazar, M., Mu˜noz de la Pe˜na, D., Heemels, W.P.M.H., and Alamo, T. (2008). On input-to-state stability of min-max nonlinear model predictive control. Systems & Control Letters, 57, 39–48. Limon, D., Alamo, T., Salas, F., and Camacho, E.F. (2006). Input to state stability of min-max MPC controllers for nonlinear systems with bounded uncertainties. Automatica, 42, 797–803. Magni, L., Raimondo, D.M., and Scattolini, R. (2006). Regional input-to-state stability for nonlinear model predictive control. IEEE Transactions on Automatic Control, 51(9), 1548–1553. Messina, M.J., Tuna, S.E., and Teel, A.R. (2005). Discretetime certainty equivalence output feedback: allowing discontinuous control laws including those from model predictive control. Automatica, 41, 617–628. Roset, B.J.P., Heemels, W.P.M.H., Lazar, M., and Nijmeijer, H. (2008). On robustness of constrained discrete-time systems to state measurement errors. Automatica, 44(4), 1161–1165.. Then, define.
(485) . (
(486) (A.2) ( ( for all
(487)
(488) and all #
(489) ,
(490) . Observe that (
(491) , which implies that is continuous at zero. As
(492)
(493) for all #
(494) , (
(495) and (
(496) , it follows that is continuous on
(497) . Next, observing that (
(498)
(499)
(500) ( for all #
(501) yields that is strictly increasing. Hence, the constructed function
(502) is of class and, from (A.1) and (A.2) it follows that for all
(503) . As
(504) , by Fact 1 it holds that
(505)
(506) is a function. As + , for any +
(507) ,
(508) , we can define '
(509) + , and *
(510) '. Since is UC() it follows that for all +
(511) ,
(512) , + ,
(513) '
(514) * *. + , *
(515) + ,. Observing that
(516)
(517) is a function completes the only if part of the proof. The if part of the proof proceeds as follows. Let and take
(518)
(519) . Then, by Definition 8, for all +
(520) ,
(521) with + , it holds that + , + ,
(522)
(523) which completes the proof. Appendix B. PROOF OF LEMMA 12 The claim for the only if part follows mutatis mutandis by applying the reasoning used in the proof of Lemma 10. The difference is that
(524) and as such, it suffices to construct a function such that for all . It is straightforward to verify that the function (
(525)
(526)
(527)
(528) #
(529)
(530) ( ( and
(531) , where (
(532)
(533) . #
(534) . satisfies the desired properties. Similarly, for the if part it suffices to observe that
(535) ..
(536)
GERELATEERDE DOCUMENTEN
Uit het onderzoek van Karlè Sýkora van Wageningen UR naar de ontwikkeling van stroomdalgraslanden in de Millingerwaard blijkt onder andere dat de oppervlakte zuiver
De Dienst Ver- keerskunde heeft de SWOV daaro m verzocht in grote lijnen aan te geven hoe de problematiek van deze wegen volgens de principes van 'duurzaam veilig' aangepakt
In an exploratory study conducted in Kenya and Uganda among young people on HIV voluntary counseling and testing by Horizons Programme, participants reported a high level
This paper advances results in model selection by relaxing the task of optimally tun- ing the regularization parameter in a number of algorithms with respect to the
Er bestaat een verband tussen het percentage licht dat door een reageerbuis met bacteriën wordt doorgelaten en de zogeheten optische dichtheid... Hierin is L het percentage
The input estimate is obtained from the innovation by least-squares estimation and the state estimation problem is transformed into a standard Kalman filtering problem.. Necessary
Its main objective is to improve the classification of brain tumours through multi-agent decision support over a distributed network of local databases or Data Marts..
