Sunteți pe pagina 1din 6

256

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

Adjoint Polynomial Formulas for Nonlinear State-Space Realization


Juri Belikov, lle Kotta, Member, IEEE, and Maris Tnso

AbstractThis paper focuses on computational aspects of the realization of nonlinear multi-input multi-output systems. Instead of the algorithmic solutions, provided in earlier works, the explicit formulas are presented, which enable to compute the differentials of the state coordinates directly from the polynomial description of the nonlinear system. The solution is based on the concept of adjoint polynomials and requires a minimal amount of computations. The formulas are implemented in computer algebra system Mathematica and made available online via web Mathematica tools. Index TermsAdjoint polynomials, input-output model, non-commutative polynomials, nonlinear control system, state-space realization.

I. INTRODUCTION The state-space realization problem from nonlinear inputoutput (i/o) equations is a fundamental problem in control theory, which has been solved via different methods, that provide both the necessary and sufcient realizability conditions as well as the different algorithms to nd the state coordinates, see [1][5]. The motivation of our study is computational effectiveness; therefore, we focus on nding the explicit polynomial formulas for computing the differentials of the state coordinates. Our solution was originally inspired by the polynomial approach used in [6] for linear systems. Polynomial systems theory for linear time-invariant (LTI) systems is well-established and provides efcient analysis and design tools [7][9]. Part of this theory has been generalized for linear time-varying (LTV) and linear parameter-varying (LPV) systems, relying, instead of the commutative polynomial ring, on the non-commutative skew polynomial ring; see for example [10][12], respectively, and the references therein. In case of the nonlinear systems very few results have been obtained within the polynomial approach [13][18]. To build up the polynomial framework one has rst to linearize the nonlinear equations globally and represent the innitesimal system representation in terms of non-commutative polynomials. Polynomials act on differentials of inputs and outputs as operators and allow to represent the algorithms in a more compact and transparent way. If compared to LTV and LPV cases there are two important differences. First, the restrictions are imposed on the polynomial coefcients by the nonlinear system equations. Second, at the last step the solution in terms of differential one-forms has to be integrated to get back to the level of equations. This results in the integrability restrictions. Our results may be understood as extension of those in [18] for the multi-input multi-output (MIMO) case. The results in [18] are based on the well-known simple relations between the transfer function and the observer canonical form representations for LTI systems. The paper
Manuscript received July 17, 2012; revised March 08, 2013 and May 23, 2013; accepted June 14, 2013. Date of publication June 21, 2013; date of current version December 19, 2013. The work of J. Belikov, . Kotta, and M. Tnso was supported by the European Union through the European Regional Development Fund and the target funding project SF0140018s08 of Estonian Science Foundation. The work of J. Belikov and M. Tnso was also supported by the Estonian Science Foundation under Grant 8787 Recommended by Associate Editor P. Tabuada. The authors are with the Institute of Cybernetics, Tallinn University of Technology, Tallinn, 12618, Estonia (e-mail: jbelikov@cc.ioc.ee; kotta@cc.ioc.ee; maris@cc.ioc.ee). Digital Object Identier 10.1109/TAC.2013.2270868

[18] showed that the above relations extend to the innitesimal representation of the nonlinear system when the transfer function is replaced by the adjoint transfer function. However, extension of the results in [18] to the MIMO case is not direct, since even in the LTI case no such simple relations exist as in the multi-input single-output (MISO) case, see [19]. Moreover, the proofs in [18] are based on transfer function formalism and are completely different from those presented in this paper. The polynomial method for nonlinear realization was also studied in [17], where left division of non-commutative polynomials was employed to nd the differentials of the state coordinates. However, in the continuous-time case, the method of adjoint polynomials is more efcient from the computational point of view and leads to faster computer programs. For discrete-time systems the polynomial realization method, based on division, has been studied in [20], but the adjoint polynomial method was not addressed due to the fact that the left division operation can be replaced by the very efcient cut-and-shift operator. In the continuous-time case such simplication is not possible. Finally, let us mention that while we focused on realization from the set of i/o equations, there is a rich literature on realization from i/o (response) maps (operators). This includes the early papers [21][23] as well as the recent ones addressing the systems with certain algebraic structure as polynomial or Nash systems [24], and rational systems, see for example [25] and the references therein. The paper is organized as follows. In Section II, we give a brief exposition of the algebraic framework based on differential one-forms [1] as well as polynomial system description. Section III recalls the realization problem and after that the main result is presented. In Section IV, an example is given and Section V concludes the paper. II. PRELIMINARIES A. I/O Equations Throughout the paper, it is assumed that the indices and . Consider a MIMO nonlinear system, described by a set of higher order differential equations, relating the inputs , the outputs , and a nite number of their time derivatives, i.e., , , and , (1) In (1) are supposed to be real analytic functions. Notations and are used below for system (1). Moreover, we assume that the indices in (1) satisfy the relations

(2) The conditions (2) mean that the (1) are assumed to be in the so-called doubly reduced form, suggested as a suitable starting point for realization in [26]. Since, according to [26, Prop. 2.6], an arbitrary set of i/o equations can be, at least locally, transformed into this form under mild rank conditions, the assumptions (2) are not restrictive. B. Linear Algebraic Framework Below we give a brief exposition of the linear algebraic approach [1], focusing on generic system properties that hold on open and dense subsets of suitable domains of denition, provided they hold at some point of such domains. Therefore, the functions and forms are typically but on open and dense regions in dened not on the whole ,

0018-9286 2013 IEEE

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

257

and the stated relations among them hold only on open subregions. In practice, these matters have to be taken care of. The generic approach motivates the choice of analytic functions in system description (1). Let denote the ring of analytic functions in a nite number of variables from the set and introduce the , dened by (1). In particular, operator

meaning that as a dependent variable has to be replaced by from (1). For the remaining elements of the time-derivative is dened , , in a standard manner, i.e., , , where and , and by then for

, where , by the . The notion of two-form is rule generalized to the -form and wedge product is dened for arbitrary -forms. is an exact one-form, if for some . We say that is said to be closed. Every exact oneA one-form for which form is closed, but the converse holds only locally, see [1]. A subspace is said to be closed or integrable, if it has a basis which consists only of closed one-forms. Integrability of the subspace of one-forms can be checked by the Frobenius theorem. , where Theorem 1 ([1]): Let are linearly independent over . The subspace is for all . integrable iff C. Polynomial Framework induces a non-commutative (skew) The differential eld . The elements of ring of the left differential polynomials can be uniquely written in the form , where is a formal variable, . Note that iff at least one of the is nonzero. If , then the positive functions , integer is called the degree of the polynomial and is denoted by . If , then polynomial is called monic. The addition of polynomials is dened in the standard way. For the multiplication is dened by the commutation rule (3) and in general, for one has , where is binomial coefcient [27]. Denition 2 ([28]): The adjoint of the skew polynomial ring is dened as the skew polynomial ring , where the multiplication is dened by the commutation rule for . , then the adjoint polyIf is dened by the formula nomial , where the products must be computed in , to yield1 . It is easy to see that (4) may be interpreted as A left differential polynomial , when we relate polynomial variable with an operator by (5) The denition (5) yields that . For we dene , where , for . see that D. Polynomial System Description By differentiating (1) we obtain the innitesimal system description and for , and it is easy to

Note that the (1) with restrictions (2) allow to dene explicitly the , which remarkably simplies the time-derivative operator in computations. In case when the i/o equations are in the implicit form , one has to replace the expressions , as well as all their time derivatives, by zero, whenever they appear in the computations. is a differential ring, being an integral domain. The pair , or . Let be That is, for multiplicative subset of . Consider the set of fractions of over , . Elements of are meromorphic functions in denoted as , where , . Since is integral domain, the set the form proves to be the eld of fractions. The operator can be extended . For we dene so that , , . Thus, the pair is a differential eld. Consider next the innite set of symbols , and denote by the over , namely vector eld spanned by the elements of . The differential operator brings the element from to :

An arbitrary element of , called a one-form, is not necessarily differential of a function and has a form

where and only a nite number of coefcients is dened by nonzero. The operator

are

Note that operators

and

commute, i.e., for

. Starting from the space it is possible to build up the structures used in exterior differential calculus. We refer to [1] for details, whereas here we just recall some basic notions. Dene the set , where denotes the wedge product with the stanand for dard properties . Introduce the space of two-forms. The oper, called exterior derivative operator, is dened for ator

1The coefcients , can be alternatively computed using left , we divide by ; their left remainder is adjoint division. Starting from polynomial coefcient and their left quotient is the input for the next step, . Rewriting the recursive formulas as a single expression i.e., gives .

258

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

which, by (5), may be rewritten in terms of left polynomials

(6) Using the notations

Theorem 4: The nonlinear system, described by the set of i/o equations (1) under constraints (2), has an observable state-space realization , dened by (10), is integrable. The state coordiiff the subspace . nates can be obtained by integrating the exact basis vectors of In this paper, we focus on searching the explicit formulas for nding that require the minimal amount of computhe basis vectors of tations and are based on the concept of adjoint polynomials. B. Polynomial Formulas As the rst step, the adjoint operator is applied to the elements of and in the polynomial system description (8). The resulting matrices and , and their elements are of the are denoted respectively by form

for

, if if

and , (7) , , one can represent , or (8)

(11) By denition, the adjoint polynomials (11) are, in general, of degree , though some of their coefcients equal zero because of the conditions yields for , if . (2). In particular, yields for In the similar manner, the condition . The th rows of the matrices and are denoted by and , respectively. Next, introduce the set of one-forms (12) where . Note that are dened by the coefcients being in the th row of . The main result, Theorem 6 of , in terms below, represents the basis elements of of the one-forms (13) . The lemma below will be used later in the proof where of Theorem 6. Lemma 5: The one-forms (13) may be alternatively computed by recursive formula (14) Proof: Taking in (13) yields . To prove the recursive formula we rst increase the index by 1 in (13), yielding for , and then compute the derivative of as . To complete the to both sides of the last equality proof one has to add the term for and take into account that by (13), . The next theorem establishes the relation between the subspaces , and the one-forms (13). Though the realizability to be integrable, the forcondition of Theorem 4 requires only are given. Note that mulas for computation of all is not integrable if just one of the previous subspaces in the sequence is not integrable [30]. This fact allows sometimes to decide very quickly non-realizability of the i/o equation. can be calTheorem 6: For the i/o model (1), (2) the subspaces culated as (15) for . Proof: The proof is by mathematical induction.

where are left polynomials from (6) in the form using matrix notation

-dimensional matrices, respectively, and , . Equation (8) describes the globally linearized system, corresponding to the (1). Hereinafter the notations and are used for row vectors of and , respectively. III. REALIZATION: POLYNOMIAL FORMULAS A. Problem Statement , Denote Denition 3: The state-space description . (9) is said to be realization of the set of the i/o equations (1) if (1) and (9) . have the same solution sets2 The system (1) is said to be realizable if it admits a realization in the sense of Denition 3. A sequence of subspaces of is dened by

where and are whose elements are

and

(10) System (9) is said to be single-experiment observable if the observability matrix has full rank

Specifying Theorem 1 from [30], given in a unied manner for continuous- and discrete-time cases, yields the solution of the realization problem for system (1).
3 agrees with the denition in [2], but not with the one in [29], based on the differential closures of ideals, corresponding to the i/o and state equations. The choice was determined by our goal to nd simple formulas for computation of state coordinates and not to focus on the realization theory itself.
2Denition

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

259

First, we show that formula (15) holds for (15), yields

. Taking

in (16)

and compute by (14). Note that if Suppose that , then due to condition in (2) the coefcients in (12). Analogously, the condition yields that . since polynomials are monic by Observe as well that . In general, can be (7). Thus, expressed as

and therefore the coefcients of adjoint polynomial equal to those of , in Thethe original polynomial. Then the one-forms orem 6 can be alternatively expressed via the so-called cut-and-shift operator, borrowed from [31] and dened as . Namely, , , means the repeated application of cut-and-shift where and one operator to the elements of row vectors. Finally, if for , then the state coordinates equal chooses to those given in [31], see p. 75 and the state-space realization (9) is in the observer canonical form. Remark 8: The formula (13) cannot be directly extended to nd the . However, for the i/o equations (1) can be found subspace of the matrices and . using the greatest common left divisor . The elements Then span , see [27] for the single-input of the vector single-output case. IV. EXAMPLE The realization algorithm, based on the exact formulas (13), is implemented in computer algebra system Mathematica as a part of a package called NLControl [32], [33], devoted to nonlinear control problems and largely based on the linear algebraic approach of differential one-forms. The main functions of the package, including realization, are made accessible on NLControl website [34] via webMathematica tools, so that one can solve the control problems without installing Mathematica into a local computer. We present below an example; more examples can be found on the website [34]. Example: Consider the system

where tions from

and are some func. That is is represented as a linear combination of , , being the basis vectors of , as in (10). dened by (10). Thus, (16) agrees with and we prove it to Assume next that formula (15) holds for . The proof is based on the denition of the subspaces be valid for . We have to prove that (17)

calculated according to formula (15), satises denition (10). First, we will show that the one-forms are in . It is obvious, since we have assumed formula (15) to hold for . Second, we have to prove that the derivatives of the basis one-forms of (17) . By Lemma 5, can be represented as a difference of belong to two terms (18) To deal with the rst term of (18), we consider separately two cases. In , by (12), (13) and relation , one has case

(19) that can be described by polynomial matrices

and

The property (4) allows to move the coefcients with respect to

from right to left

It is easy to check that for (19) , , , and thus the system order . According to Corollary 7, our task is to nd the one-forms (14). For that we rst need to compute the adjoint polynomials of the entries of and :

Thus, due to (8). In case , the rst term due to induction assumption. As for the second term of (18), by denition (12) the one-forms are the linear combinations of , since and are functions from . Therefore, for . Finally, we observe that the derivatives of the rest of the basis one-forms in (17) are , which are also . Thus, we have proved that , computed according to (15), in agrees with denition (10). Note that even if is completely integrable, the arbitrary basis , , are not necessarily exact. vectors, in particular Corollary 7: For the realizable i/o model (1) the differentials of the can be calculated as independent instate coordinates , . tegrable linear combinations of the one-forms Note that in the MISO case the results of Theorem 6 yield those in (1) are linear, the ring given in [18]. Moreover, if the functions is commutative, since in the commutation rule (3)

where and

According to (12), the one-forms being in the th row of of

are dened by the coefcients :

260

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

By (14) we may compute

called minimal if it is both accessible and observable or, if its state dimension is minimal. Though in the linear case these two denitions are equivalent, it was shown in [29] that the minimal realization is not necessarily accessible. This topic is further studied in [37] to show that the equivalence of denitions may be restored for nonlinear systems when one is looking for minimal realization in a larger class of time-varying systems.

REFERENCES
(20) For the one-forms (20) the conditions of Theorem 1 are fullled; thus, the set of one-forms is integrable and the system (19) is realizable. However, only in (20) is exact one-form, therefore we have to , , and by the independent integrable linear replace combinations of one-forms (20), to obtain the differentials of the state as follows: coordinates. We may dene , , , , . Integrating , yields the state . Differentiating the state coordinates gives . To obtain the state equations, one has to substitute and into from the i/o equations (19), solve for and substitute them into . This yields the state equations
[1] G. Conte, C. H. Moog, and A. M. Perdon, Algebraic Methods for Nonlinear Control Systems. London, U.K.: Springer-Verlag, 2007. [2] A. J. van der Schaft, On realization of nonlinear systems described by higher-order differential equations, Math. Syst. Theory, vol. 19, no. 1, pp. 239275, 1987. [3] E. Delaleau and W. Respondek, Lowering the orders of derivatives of controls in generalized state space systems, J. Math. Syst., Estim. , Control, vol. 5, no. 3, pp. 127, 1995. [4] P. E. Crouch and F. Lamnabhi-Lagarrigue, State space realizations of nonlinear systems dened by input-output differential equations, in Analysis and Optimization of Systems, ser. LNCIS. London, U.K.: Springer-Verlag, 1988, vol. 111, pp. 138149. [5] . Kotta and T. Mullari, Equivalence of realizability conditions for nonlinear control systems, Proc. Estonian Acad. Sci., vol. 55, no. 1, pp. 2442, 2006. [6] P. Rapisarda and J. C. Willems, State maps for linear systems, SIAM J. Control Optimiz., vol. 35, no. 3, pp. 10531091, 1997. [7] H. Blomberg and R. Ylinen, Algebraic Theory for Multivariable Linear Systems. London, U.K.: Academic, 1983. [8] H. H. Rosenbrock, State-space and Multivariable Theory. New York, NY, USA: Wiley Interscience Division, 1970. [9] W. A. Wolowich, Linear Multivariable Syst.. New York, NY, USA: Springer, 1974. [10] A. Ilchmann and V. Mehrmann, A behavioral approach to time-varying linear systems. Part 2: Descriptor systems, SIAM J. Control Optimiz., vol. 44, no. 5, pp. 17481765, 2005. [11] H. Bourls and B. Marinescu, Linear Time-varying Systems: Algebraic-analytic Approach, ser. LNCIS. New York, NY, USA: Springer-Verlag, 2011, vol. 410. [12] R. Tth, H. S. Abbas, and H. Werner, On the state-space realization of LPV input-output models: Practical approaches, IEEE Trans. Control Syst. Technol., vol. 20, no. 1, pp. 139153, Jan. 2012. [13] Y. Zheng, J. C. Willems, and C. Zhang, A polynomial approach to nonlinear system controllability, IEEE Trans. Autom. Control, vol. 46, no. 11, pp. 17821788, Nov. 2001. [14] R. Ylinen, Application of polynomial systems theory to nonlinear systems, in Proc. 16th IFAC World Congr., Prague, Czech Republic, 2005, pp. 564564. [15] M. Hals, An algebraic framework generalizing the concept of transfer functions to nonlinear systems, Automatica, vol. 44, no. 5, pp. 11811190, 2007. [16] M. Hals, . Kotta, and C. H. Moog, Transfer function approach to the model matching problem of nonlinear systems, in Proc. 17th IFAC World Congr., Seoul, Korea, 2008, pp. 15,19715,202. [17] J. Belikov, . Kotta, and M. Tnso, Minimal realization of nonlinear MIMO equations in state-space form: Polynomial approach, in Proc. IEEE Conf. Decision Control Eur. Control Conf., Orlando, FL, USA, 2011, pp. 77357740. [18] M. Hals and . Kotta, A transfer function approach to the realisation problem of nonlinear control systems, Int. J. Control, vol. 3, no. 85, pp. 320331, 2012. [19] T. Glad and L. Ljung, Control Theory: Multivariable and Nonlinear Methods. New York, NY, USA: Taylor & Francis, 2000. [20] . Kotta and M. Tnso, Realization of discrete-time nonlinear input-output equations: Polynomial approach, Automatica, vol. 48, pp. 247458, 2012. [21] B. Jakubczyk, Existence and uniqueness of realizations of nonlinear systems, SIAM J. Control Optimiz., vol. 18, no. 4, pp. 455471, 1980. [22] H. Sussman, Existence and uniqueness of minimal realizations of nonlinear systems, Math. Syst. Theory, vol. 10, pp. 263284, 1977.

coordinates

V. CONCLUSION The paper addressed the realization problem of nonlinear MIMO systems, described by the set of i/o differential equations. The explicit formulas for deriving the differentials of the state coordinates were proposed, based on the notion of adjoint polynomials. From the computational point of view the polynomial approach of this paper, if compared to the earlier results, has a number of advantages. First, it is straightforward, meaning that there is no need to compute step-by-step subspaces in order to nd as was by denition (10) all the as in [2], or iterproposed in [1], or the sequence of distributions ative Lie brackets of the vector elds as in [3]. In other words, using the polynomial representation of the globally linearized system, one may immediately nd, by simple polynomial operations, the basis el, dening the differentials of the state ements of the subspace coordinates. One possible direction for the future research is to extend the results of this paper to nonlinear models where the system variables are not partitioned a priori into the inputs and the outputs, i.e., for the model used in the so-called behavioral approach, see, for example, [12], [35], [36] for linear systems. Regarding more intrinsic aspects of nonlinear realization theory, there is a number of questions that have not yet a complete answer. One such issue is the intrinsic denition of the state, similar to that in [6] and whether this denition at the level of trajectories can be characterized algebraically. Another possible issue is that of minimal realization, as raised in [29]. The realization is typically

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 59, NO. 1, JANUARY 2014

261

[23] E. D. Sontag, Realization theory of discrete-time nonlinear systems. Part IThe bounded case, IEEE Trans. Circuits Syst., vol. CS-26, no. 4, pp. 342356, May 1979. [24] J. Nmcov, M. Petreczky, and J. H. van Schuppen, Realization theory of Nash systems, in Proc. IEEE Conf. Decision Control, Shanghai, China, 2009, pp. 59355940. [25] J. Nmcov and J. H. van Schuppen, Realization theory for rational systems: The existence of rational realization, SIAM J. Control Optimiz., vol. 48, no. 4, pp. 28402856, 2009. [26] A. J. van der Schaft, Transformations of nonlinear systems under external equivalence, in New Trends in Nonlinear Control Theory, ser. LNCIS. Berlin, Germany: Springer, 1989, vol. 122, pp. 3343. [27] . Kotta, Z. Bartosiewicz, E. Pawluszewicz, and M. Wyrwas, Irreducibility, reduction and transfer equivalence of nonlinear input-output equations on homogeneous time scales, Systems Control Lett., vol. 58, no. 9, pp. 646651, 2009. [28] S. A. Abramov, H. Q. Le, and Z. Li, Univariate Ore polynomial rings in computer algebra, J. Math. Sci., vol. 131, no. 5, pp. 58855903, 2005. [29] J. Zhang, C. H. Moog, and X. Xia, Realization of multivariable nonlinear systems via the approaches of differential forms and differential algebra, Kybernetika, vol. 46, no. 5, pp. 799830, 2010. [30] J. Belikov, P. Kotta, . Kotta, and M. Tnso, Practical polynomial formulas in MIMO nonlinear realization problem, in Proc. Conf. Decision Control, Maui, HI, USA, 2012, pp. 12531258. [31] J. C. Willems, The behavioral approach to open and interconnected systems, IEEE Control Syst. Mag., vol. 27, no. 6, pp. 4699, Dec. 2007. [32] M. Tnso, H. Rennik, and . Kotta, WebMathematica-based tools for discrete-time nonlinear control systems, Proc. Estonian Acad. Sci., vol. 58, no. 4, pp. 224240, 2009. [33] J. Belikov, . Kotta, and M. Tnso, Symbolic polynomial tools for nonlinear control systems, in Full paper Preprint Volume: 7th Vienna Int. Conf. Math. Modelling, Vienna, Austria, 2012, pp. 16. [34] 2013, NLControl Website [Online]. Available: www.nlcontrol.ioc.ee [35] P. Rapisarda and A. J. van der Schaft, Canonical realization by factorization of constant matrices, Syst. Control Lett., vol. 61, no. 8, pp. 827833, 2012. [36] A. J. van der Schaft and P. Rapisarda, State maps from integration by parts, SIAM J. Control Optimiz., vol. 49, no. 6, pp. 24152439, 2011. [37] . Kotta, C. H. Moog, and M. Tnso, The minimal time-varying realization of a nonlinear time-invariant system, in Proc. 9th IFAC Symp. Nonlinear Control Syst. (NOLCOS), Toulouse, France, 2013, submitted for publication.

A Mean Square Consensus Protocol for Linear Multi-Agent Systems With Communication Noises and Fixed Topologies
Long Cheng, Zeng-Guang Hou, and Min Tan

AbstractThe mean square consensus of linear multi-agent systems with communication noises is studied in this note. Each agent is modeled by a continuous-time linear time-invariant dynamics and the xed communication topology is described by a digraph. The proposed consensus protocol is composed of two parts: the agents own state feedback and the relative states between agent and its neighbor agents. Due to the existence of communication noises, the relative states cannot be obtained accurately. To atis applied to tenuate the noise effect, a time-varying gain vector the inaccurate relative states. It is proved that: 1) if the communication topology has a spanning tree and every node has at least one parent node, then the proposed protocol can solve the mean square consensus problem if and only if satises and ; and all roots of the polynomial whose coefcients are the elements of vector are in the left half complex plane; 2) if the communication topology has a spanning tree and there exists one node without any parent node, then is only sufcient but not necessary; and the condition 3) if the communication topology has no spanning tree, then the proposed protocol cannot solve the mean square consensus problem. Index TermsConsensus, multi-agent systems, noise. linear time-invariant, mean square,

I. INTRODUCTION Recently, the distributed consensus of multi-agent systems (MASs) has attracted considerable attention from the control community due to its fundamental role in other coordination applications. However, most existing papers assume that MASs work in an ideal communication environment, i.e., each agent can receive the information accurately. In practice, communication channels are inevitably interfered by stochastic noises. In the literature, there is a growth of interest in designing proper consensus protocols to reduce the noise effect [1][8]. In [1][3], Huang and his collaborators rst proposed several stochastic approximation type consensus protocols for rst-order integral MASs with communication noises under xed topologies and switching topologies, respectively. In [4] and [5], Li and Zhang further investigated these stochastic approximation type protocols and presented the necessary and sufcient conditions for ensuring the mean square/almost sure average consensus. Based on these results, Liu and Xie studied the mean square consensus of rst-order integral MASs with consideration of both communication noises and communication delays [6]. Hu and Feng also provided some results on the mean square leader-following consensus of rst-order integral MASs with communication noises [7]. It is noted that the above papers all concentrate on the study of communication topologies and assume that the agent is modeled by a simple dynamics (single integrator). For a comprehensive understanding on the consensus of MASs with
Manuscript received August 06, 2011; revised February 15, 2012; accepted October 28, 2012. Date of publication June 21, 2013; date of current version December 19, 2013. This work was supported in part by the National Natural Science Foundation of China (Grants 61370032, 61004099, 61225017, 61273326, 61175076, 61273352), Beijing Natural Science Foundation (Grant 4123105), Beijing Nova Program (Grant Z121101002512066), the startup fund for the recipient of Presidential Award of Chinese Academy of Sciences, and the Early Career Development Award of SKLMCCS. Recommended by Associate Editor C. De Persis. The authors are with the State Key Laboratory of Intelligent Control and Management of Complex Systems (SKLMCCS), Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China (e-mail: chenglong@compsys.ia.ac.cn; zengguang.hou@ia.ac.cn; tan@compsys.ia.ac. cn). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TAC.2013.2270873

0018-9286 2013 IEEE

S-ar putea să vă placă și