Bipartite Consensus of Integrator Multi‐agent Systems with Measurement Noise

Cui‐Qin Ma,Zhengyan Qin,Yun‐Bo Zhao

DOI: https://doi.org/10.1049/iet-cta.2017.0334

IF: 2.67

2017-01-01

IET Control Theory and Applications

Abstract:IET Control Theory & ApplicationsVolume 11, Issue 18 p. 3313-3320 Research ArticleFree Access Bipartite consensus of integrator multi-agent systems with measurement noise Cui-Qin Ma, Cui-Qin Ma School of Mathematical Sciences, Qufu Normal University, No. 57, Jingxuan West Road, Qufu, People's Republic of ChinaSearch for more papers by this authorZheng-Yan Qin, Zheng-Yan Qin School of Mathematical Sciences, Qufu Normal University, No. 57, Jingxuan West Road, Qufu, People's Republic of ChinaSearch for more papers by this authorYun-Bo Zhao, Corresponding Author Yun-Bo Zhao ybzhao@zjut.edu.cn College of Information Engineering, Zhejiang University of Technology, No. 288, Liuhe Road, Hangzhou, People's Republic of ChinaSearch for more papers by this author Cui-Qin Ma, Cui-Qin Ma School of Mathematical Sciences, Qufu Normal University, No. 57, Jingxuan West Road, Qufu, People's Republic of ChinaSearch for more papers by this authorZheng-Yan Qin, Zheng-Yan Qin School of Mathematical Sciences, Qufu Normal University, No. 57, Jingxuan West Road, Qufu, People's Republic of ChinaSearch for more papers by this authorYun-Bo Zhao, Corresponding Author Yun-Bo Zhao ybzhao@zjut.edu.cn College of Information Engineering, Zhejiang University of Technology, No. 288, Liuhe Road, Hangzhou, People's Republic of ChinaSearch for more papers by this author First published: 24 October 2017 https://doi.org/10.1049/iet-cta.2017.0334Citations: 18AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract The bipartite consensus problem for integrator multi-agent systems over signed fixed digraphs is investigated in the presence of measurement noise. A time-varying consensus gain is introduced and then a stochastic type protocol is proposed, whose performance is analysed using the state transition matrix of the closed-loop system. Necessary and sufficient conditions for ensuring a mean square bipartite consensus protocol are obtained in the presence of noise. Furthermore, in the absence of noise it is shown that these conditions are also necessary and sufficient for ensuring the bipartite consensus except for the quadratic integrability of the consensus gain. It is found that the signed digraph being structurally balanced and having a spanning tree are the weakest assumptions on connectivity for achieving bipartite consensus regardless of the measurement noise. In particular, if the signed digraph is structurally unbalanced, then under some mild conditions, the states of the closed-loop system converge to zero in mean square, regardless of the initial states. 1 Introduction The research on consensus control of multi-agent systems (MASs) has been active over the past two decades [1–7]. In MASs the agents may compete or cooperate with each other, and consensus is reached by their cooperative interactions. In particular, consensus may be such that the agents have the same value but with different signs, referred to as bipartite consensus [8]. Note that bipartite consensus is different from the well-known group consensus [3], in that the latter may have multiple consensus groups and each group may have different, independent consensus values. Examples of bipartite consensus can be found in diverse fields including sociology [9], physics [10] and so on. It is known that bipartite consensus usually relies on three key factors, i.e. the agent dynamic structure, the consensus protocols and the communication topology among agents. Existing studies often keep one or two such factors fixed and then investigate the role of the rest factors. For example, under given agent dynamic structure and bipartite consensus protocol, the role of the communication topology has been studied in [8, 11–14] and so on. In the pioneering work [8], it is shown that for first-order integrator MASs with linear Laplacian-like bipartite consensus protocols and strongly connected signed digraph, bipartite consensus is achieved if and only if the signed digraph is structurally balanced. Then, the connectivity condition in [8] is relaxed in [11] to having a spanning tree. The fixed communication topology in [8] is extended in [12] to time-varying signed digraphs. Moreover, bipartite consensus for high-order MASs with unknown disturbances is investigated [15–17]. In [15], an adaptive consensus protocol is designed for second-order MASs. In [16, 17], for high-order MASs, a leader, whose information is only available to a part of agents, is introduced to intervene a group of agents and a neural network-based adaptive protocol is proposed. Since a Lyapunov function candidate is usually hard to be constructed for high-order MASs, a common Lyapunov function method is employed in the convergence analysis [16, 17]. On the other hand, measurement noise is usually inevitable in practice and has been studied widely in the context of conventional consensus control [18–24]. To name a few, a stochastic approximation-type protocol with a decreasing step size is proposed for discrete-time models in [18]; the continuous-time integrator models are considered in [19], where necessary and sufficient conditions for ensuring mean square average consensus are given. Subsequently, the protocols developed in [19] are extended in [20] to time-varying topologies, in [21] to double-integrator MASs and in [22] to MASs described by a continuous-time linear time-invariant model with single input. Moreover, multiplicative noises are considered in [23, 24] and so on. In [23], the noise intensity is assumed to be proportional to the absolute value of the relative states of an agent and its neighbours, and sufficient conditions to achieve mean square consensus and strong consensus are given by properly selecting constant consensus gains, respectively. In [24], the intensity function in [23] is further extended to a vector function. However, measurement noise has not been well studied for bipartite consensus to date, thus motivating our this study. Two reasons make conventional stochastic analysis tools [18–20] unsuitable for bipartite consensus, justifying the necessity of our work. Firstly, in bipartite consensus, cooperation and competition coexist and interactions among agents are generally represented by signed graphs with positive/negative weights, which are different from conventional consensus. This means that the row sum of Laplacian may not be zero and Laplacian of a signed graph can be positive definite, different from standard graph theory for conventional consensus. Secondly, the error in conventional consensus is merely the difference, while for bipartite consensus it might be the sum of agent state and the final consensus value, making the state matrix of the error dynamics and Laplacian coupled, and therefore causing another challenge. In this work, we design distributed protocols to reach bipartite consensus in the presence of measurement noise. Inspired by [18, 19], a time-varying consensus gain is introduced, leading to a time-varying stochastic system as the closed-loop system. It is worth pointing out that in [25], stochastic Lyapunov function is constructed to analyse the closed-loop system and only sufficient conditions are given to ensure mean square weak bipartite consensus under undirected signed graphs. Here, directed signed graphs are considered. By using the theory of state transition matrix and Cauchy criterion, the dynamic characteristics of the closed-loop system are fully described. In the case with noise, necessary and sufficient conditions are obtained to describe the proposed protocol under which a mean square bipartite consensus is achieved. It is proven that the states of all the agents converge to a common random vector except for the sign. The mathematical expectation of this random vector is determined by the initial states and communication topology among agents. In the case without noise, some of the above conditions are also necessary and sufficient for ensuring a bipartite consensus. It turns out that the conditions for the signed digraph to be structurally balanced and having a spanning tree are not only sufficient but also necessary for achieving bipartite consensus. In particular, if the signed digraph is structurally unbalanced, the states of the closed-loop system will converge to zero in mean square under some mild conditions, regardless of the initial states. The remaining of the paper is organised as follows. In Section 2, we first review some basic concepts and properties associated with signed graphs, and then formulate the problem of interest, with the introduction of some necessary lemmas. The main results are then presented in Section 3. A numerical example is given in Section 4, and Section 5 concludes the paper. Notations: The following notations are used throughout this paper. is the real part of . is an N -dimensional column vector with all ones. is the zero matrix with appropriate dimension. and are the trace and rank of A, respectively. For a given matrix or vector X, is the transpose of X, its Euclidean norm, its Frobenius norm, the 1-norm and the norm, respectively. For a random vector x, is its mathematical expectation and its variance. is the Kronecker product. 2 Preliminaries and problem formulation 2.1 Preliminaries on signed graph A signed digraph can be represented by where is the set of nodes, and is the set of edges. An edge of is denoted by an ordered pair (j, i), where j is the parent vertex of i, and i is the child vertex of j. denotes the neighbour set of agent i. is the weighted adjacency matrix. . means cooperation and means competition between agent i and j. Throughout this paper, we always assume that and , . is Laplacian of , where . Let , and a signed digraph is weight balanced if . A digraph is a directed tree if each of its vertices except the root has exactly one parent. A spanning tree of a digraph is a directed tree that contains all the vertices of the digraph. A graph is a subgraph of if , and all non-zero elements of belong to . is an input-isolated subgraph if no edge comes from to , i.e. . A signed digraph is structurally balanced if there exist two subsets , , , such that , , . It is structurally unbalanced otherwise. Laplacian of a signed digraph is closely related to the concept of structural balance, as stated below. Lemma 1 [13, 26].Laplacian of a signed digraph has at least one zero eigenvalue and all of the non-zero eigenvalues are in the open right half-plane, if is structurally balanced. Furthermore, has exactly one zero eigenvalue if and only if has a spanning tree. Lemma 2.Let be a signed digraph with Laplacian . Then, all the eigenvalues of are in the open right half-plane if and only if is neither structurally balanced nor contains an input-isolated structurally balanced (ISB) subgraph. Proof.Consider system . It is clear that , is equivalent to that all the eigenvalues of are in the open right half-plane.According to Theorem 1 in [12], is equivalent to the statement that is neither structurally balanced nor contains an ISB subgraph. This completes the proof. □ 2.2 Problem formulation Suppose that the agents have the following dynamics: (1)where and are the state and control input for the i th agent, respectively. Assume that the i th agent receives information from its neighbours with measurement noise where is n -dimensional independent standard white noise. The main objective here is to design a distributed protocol under measurement noise in order that for any initial value the states of agents in (1) converge to values with equal modulus (they may have different signs though). To that end, a time-varying consensus gain is introduced for each agent, as follows: (2) where is a piecewise continuous function. We know that for conventional consensus a similar protocol to (2) may be designed by dropping , which guarantees that each agent changes its state such that it gets close to its parent. Here, due to the existence of competition in our problem setting and the use of the term , the protocol in (2) guarantees that each agent changes its state in such a way that it gets close to its cooperative neighbours but far away from its compete neighbours, and finally reaching bipartite consensus. Let . Given the protocol in (2), the state of the integrator agents in (1) evolves as follows: (3)where is an block diagonal matrix with , , , and are standard Brownian motions. The closed-loop system in (3) is a time-varying stochastic differential equation. To characterise the asymptotic behaviour of agents in (3), we first introduce the definition of bipartite consensus for stochastic system. Definition 1.A distributed protocol is a mean square bipartite consensus protocol if for any given , there exist and random vector such that where , is determined by the initial state and the communication topology among agents. Definition 1 shows that in the non-trivial case (), the states of the agents converge to or in mean square. Definition 2.A distributed protocol is said to be mean square stabilising if it renders system (1) to satisfy Finally, we list the following assumptions which may be used in the lemmas and theorems that follows: . . is structurally balanced. has a spanning tree. Remark 1. (i) and are often used in stochastic approximation theory [18]. A possible choice of satisfying both and was given in [19]: , where , , , , . (ii) From Theorem 1 in [12] it is known that and are the weakest assumptions on communication topology for ensuring a bipartite consensus in the absence of noise. Remark 2. (i) In the absence of noise, i.e. , let , the protocol in (2) is exactly the same bipartite consensus protocol as in [8, 11–13, 26]. However, in the presence of noise, for constant the resulting evolution of agent states leads to fluctuations. To deal with measurement noise, the time-varying consensus gain is thus introduced. (ii) If all communication weights , in , then becomes a standard digraph in conventional consensus. Under this situation, (2) is the same protocol as in [19, 20]. Furthermore, only the states of agent i and its neighbours are used and hence (2) is a distributed protocol. Remark 3. (i) In Definition 1, the requirement on is necessary, since otherwise agents in (1) will not have to communicate with each other, regardless of the initial value nor the communication content among agents. In such a case one needs only to design protocols based on states of themselves (for example, taking , then one has ). (ii) Definition 1 here is not equivalent to that in [25]. The latter requires the convergence of the state differences/sums between/of different agents, regardless of the convergence of the states themselves. In the absence of measurement noise, the states of the agents are all deterministic processes. Then Definition 1 is equivalent to bipartite consensus in [8, 12, 13, 26]. 2.3 Useful lemmas To derive the main results, some useful lemmas are presented first. Lemma 3.The state transition matrix of the following system is , (4)where , and is a Jordan block with on its diagonal. In addition, if and . Proof.By definition By direct calculation, one has (5)Therefore, and , which implies that is the state transition matrix of (4).In addition, if and , then The above together with (5) leads to . □ Let be the state transition matrix of the closed-loop system in (3). Clearly, by the definition of matrix exponential function, . If Laplacian has exactly one zero eigenvalue and all of the non-zero eigenvalues are in the open right half-plane, then there exists an invertible matrix T such that (6)where is an Jordan block with on its diagonal, . Obviously, are eigenvalues of and , . This together with Lemma 3 gives the state transition matrix , where are defined as in Lemma 3. In addition, if , then (7) Thus there exists such that (8) If the protocol in (2) is a mean square bipartite consensus protocol, the following lemma will provide a useful insight into the state transition matrix of the closed-loop system in (3). It plays an essential role in the asymptotic behaviour analysis. Lemma 4.If the protocol in (2) is a mean square bipartite consensus protocol, then there exist and such that .If the protocol in (2) is mean square stabilising, i.e. , then . Proof.If the protocol in (2) is a mean square bipartite consensus protocol, then according to Definition 1, for any given , there exist and random vector such that . From (3), one immediately obtains . Without loss of generality, one may assume that converges to in mean square. Thus, where . Noting that is arbitrary. Hence, for and , there exist (), and , such that and . Therefore, one derives that (9)Then we must have . If otherwise, take as an example. In this case, besides , (9) contains at least one n -dimensional component which is . So . This contradicts the statement that is determined by and the communication topology among agents. Other cases can be similarly eliminated.Combining with the above equations one has (10)where is a linear combination of , and . Let . Then each column has entries with equal absolute value. Analogously, any linear combination has n -dimensional components with equal modulus.If , then by taking , the statement of Lemma 4 is true. If at least one column of , say, is non-zero, then (, say, ). Note that for , , , we have . Therefore, if for some , all n -dimensional components of have equal modulus if and only of , . If , let , and then for any j and . Noticing that , we have .If the protocol in (2) is mean square stabilising, then repeating the above arguments, we have . □ The following lemma is instrumental for the main result in the following section. Lemma 5.If assumptions hold, then for any given , there exists random vector such that of the closed-loop system in (3) converges to in mean square, i.e. . Proof.Since and hold, Lemma 1 implies that has exactly one zero eigenvalue and all of the non-zero eigenvalues are in the open right half-plane. This together with assumption gives (7) and (8). By Itô formula, the solution of the closed-loop system in (3) can be expressed as , leading to the convergence in mean square of to a random vector by Cauchy's criterion [27].By assumption , for , there exists such that . By (7) and (8), for the above , there exists such that for , , .Let . Then Therefore, . Obviously, . Noticing that and , one obtains that . Similarly, . Thus, .Due to Cauchy's criterion and arbitrariness of , there exists such that converges to in mean square. Therefore, converges in mean square to . By (7), one obtains □ 3 Main results 3.1 Mean square bipartite consensus Theorem 1.For the system in (1), the protocol in (2) is a mean square bipartite consensus protocol if and only if assumptions hold. In what follows, we prove Theorem 1 step by step. Proof.We prove the sufficiency in the three steps. S.I. Construct the consensus grouping: By assumption , vertex set of can be divided into two subsets , , where and , and furthermore for and for . Since vertices can be renumbered as needed, we may assume and . Taking for and for , by the definition of Laplacian , one has . S.II. Prove : From Lemma 5 we know that converges to in mean square, i.e. . Without loss of generality one may assume that converges to in mean square, i.e. . Our goal is to show that converges in mean square to , i.e. , . Let , where , . It suffices to show that . To this end, we analyse the evolution of the error dynamics, and use the similar non-singular matrix S as introduced in [14], i.e. Let , and hence, , where . Since by (3), one has (11)It is straightforward that and have the same eigenvalues except one zero eigenvalue. Therefore, there exists an invertible matrix Q such that , where are defined as in (6). Notice that the state transition matrix of (11) is , where , are defined as in Lemma 3. Therefore , i.e. for , such that for any , , and hence such that , . By Itô formula, Due to the fact that , one has This, together with , , yields that for any , Since is arbitrary, . Thus . S.III. Determine the statistic characteristics of : It is evident that . Then . Suppose the first column of T is and the first row of is . Then . By (6), one has and . Thus and , i.e. and are the right and left eigenvectors of corresponding to eigenvalue 0, respectively. Since has only one zero eigenvalue, the eigenspace corresponding to the zero eigenvalue is 1-dimensional (1D). From S.I we know that . Therefore, () and . Immediately, one obtains . Clearly, is dependent on the communication topology and hence is determined by the initial state and the communication topology among agents. By (7), it is obvious that is uniformly bounded. Then, for any given , there exists such that (12)Let . Then there exists such that for all , . Combining this with (12), where . According to this and , it can be obtained that . Therefore, . By Definition 1, protocol (2) is a mean square bipartite consensus protocol. Necessity: We prove the necessity in the four steps, as follows. N.I. Prove the necessity of , i.e. : If does not hold, then there exists a constant such that . Therefore, . Since is invertible, we obtain that . However, from Lemma 4 we know that , and hence . This is a contradiction. Thus, holds. N.II. Prove that Laplacian has and only has one zero eigenvalue : First, we prove that 0 is an eigenvalue of . If 0 is not an eigenvalue of , then is a Hurwitz matrix. From and Lemma 3 we know that . This together with Lemma 4 yields that . Obviously, has nothing to do with the initial state , and thus is independent from . This contradicts the statement that is determined by and the communication topology among agents. Therefore, 0 is an eigenvalue of . Suppose is a Jordan block corresponding to eigenvalue 0. Then it is 1D. Suppose by contradiction that is -dimensional and . By (5) and assumption , we get that does not exist and hence does not exist. This contradicts Lemma 4. Thus is 1D. Second, we prove that 0 is an eigenvalue of algebraic multiplicity 1. Assume the multiplicity of 0 is m. If , then we take as an example. In this case, due to the fact that every Jordan block corresponding to eigenvalue 0 is 1D, one has Thus which implies that . However, from Lemma 4 we know that . This is a contradiction. Other cases can be proved similarly. Therefore, has exactly one zero eigenvalue. N.III. Prove the necessity of assumptions : From the above analysis, we know that has exactly one zero eigenvalue. Combining this with obtains (7). This together with Lemma 4 gives (13) Note that is the first column of T. Then by (6), . From (13), it follows that , where . Thus, . By the definition of Laplacian , we obtain that for any j, . Noticing that and , , we get . Thus, . This implies that or all non-zero terms have the same sign with . Let and . Then and . Obviously, if , then by , we have and . By definition, is structurally balanced, or equivalently, holds. From N.II., and Lemma 1, we know that has a spanning tree, i.e. holds. N.IV. Prove the necessity of assumption : By contradiction, we assume . Note that is the first row of and . Then by (3), one has , i.e., . By Definition 1, converges in mean square to with . Then, as , converges in mean square to a random vector with . Thus, . On the other hand, This leads to a contradiction. Thus, holds. We thus complete the proof. On the premise of structural balance, a bipartite consensus problem can be converted into a conventional consensus problem on a standard graph with the help of a gauge transformation [8, 26, 28]. This is also true in the case with noise [25]. Hence, if is weight balanced, then under assumptions bipartite average consensus can be derived from average consensus results in [19]. Corollary 1.If is weight balanced and assumptions hold, then , where Remark 4.The key tool in [19] is symmetrised graph, which heavily relies on the balance of . However, in Theorem 1 we only require that has a spanning tree, but not on its weight balance. This means that the analysis tool in [19] does not apply directly to the sufficiency proof of Theorem 1.In the necessity proof of Theorem 1, structural balance does not exist as a prerequist but as a condition to be proved. Therefore, gauge transformation is invalid and thus the results in [19–22] are not directly applicable to Theorem 1. Moreover, convergence analysis of [19] heavily relies on the tool of balanced graph. So, the results in [19] cannot be applied here neither. From Theorem 1 one sees that structural balance is necessary for ensuring the mean square bipartite consensus in the presence of noise. One may be further interested to know the evolution of the closed-loop system in (3) for structurally unbalanced . For that purpose, we give the following assumptions. is structurally unbalanced. does not contain an ISB subgraph. An input-isolated subgraph of means that agents of the subgraph cannot receive information from other agents of . If such a subgraph is structurally balanced, then its agents may achieve bipartite consensus independent from the remaining agents of . Theorem 2.The protocol in (2) is mean square stabilising under assumptions , , and , i.e. for any given , .On the other hand, assumptions , and must hold if the protocol in (2) is mean square stabilising. Proof.Since and hold, Lemma 2 implies that all the eigenvalues of are in the open right half plane. For the closed-loop system in (3), adopting arguments similar to the proof after (11) in the sufficiency proof of Theorem 1, we know from and that for any given , , i.e. (2) is mean square stabilising.On the other hand, if protocol (2) is mean square stabilising, then repeating the similar procedure as in N.I of Theorem 1, we obtain that holds. According to N.II and Lemma 4, we obtain that 0 is not an eigenvalue of . Therefore, all the eigenvalues of are in the open right half-plane. Combining Lemma 2, assumptions and hold. □ Theorem 2 indicates that, regardless of the initial state, the agent state under protocol (2) converges to zero in mean square under mild conditions, for structurally unbalanced . 3.2 Bipartite consensus in the absence of measurement noise In this section, we show that under , , the protocol in (2) is also a bipartite consensus protocol in the absence of measurement noise. Suppose , the protocol in (2) is reduced to (14) Notice that for , protocol (14) is equivalent to the protocol in [8, 11–13, 26]. Applying (14) to (1) leads to the closed-loop system . Theorem 3.For the system in (1), the protocol in (14) is a bipartite consensus protocol if and only if , and hold. Proof.Using the similar arguments as in Theorem 1, one immediately obtains the above theorem. □ Remark 5.The role of can be seen from Theorems 1 and 3. Indeed, is the only requirement for to ensure a bipartite consensus in the case without measurement noise. However, itself is not sufficient in the case with noise, since in such a case the stochastic process is required to be square integrable, which can only be true in the presence of . Remark 6.If we take , then protocol (14) becomes the protocol in [8, 11–13, 26]. In this situation, assumption naturally holds, and therefore Theorem 3 degenerates to Theorem 1 in [12], i.e. bipartite consensus is achieved if and only if is structurally balanced and has a spanning tree. Compared with [8], where is required to be strongly connected, here having a spanning tree is less stringent. Corollary 2.For the system in (1), the protocol in (14) is stabilizing if and only if assumptions , and hold. Remark 7.When , Corollary 2 is consistent with Theorem 1 in [12]. In [29], for the system to be stabilizing, is required to be strongly connected. Clearly, it is a special case of . 4 Numerical example Consider a signed digraph with , as shown in Fig. 1, where four types of communications are considered and blue lines represent competitive interactions among agents. Fig. 1Open in figure viewerPowerPoint Four types of communications of the signed digraph In (1), (2) and (3) of Fig. 1, either does not have a spanning tree or is structurally unbalanced, while in (4), has a spanning tree and is also structurally balanced. If we take and , respectively, then seven state trajectories can be found in Figs. 2 and 3. Fig. 2Open in figure viewerPowerPoint State trajectories of seven agents with Fig. 3Open in figure viewerPowerPoint State trajectories of seven agents with One can observe that only when satisfies (4) in Fig. 1, can bipartite consensus be achieved. This is consistent with Theorem 3. Comparing with (4) in Figs. 2 and 3, one can see that the time of reaching bipartite consensus is about 60 s and 50 s respectively. This means that the faster approaches , the faster the considered system reaches bipartite consensus. Let satisfy (4) in Fig. 1. In the presence of measurement noise, if we still take , then the states of the seven agents fluctuate randomly with time, as shown in Fig. 4. Fig. 4Open in figure viewerPowerPoint Fluctuations of seven agents with Clearly, as time goes on, the fluctuation will not die off. We know from Theorem 1 that and are necessary and sufficient conditions on for achieving bipartite consensus in the presence of noise. Hence let . To demonstrate the convergence performance more clearly, a signed digraph of 100 nodes is considered. The stochastic protocol in (2) is used in Fig. 5, which shows the bipartite consensus in the presence of measurement noise. Fig. 5Open in figure viewerPowerPoint Bipartite consensus of 100 agents in the case with noise In particular, if the signed digraph with is structurally unbalanced and does not contain an ISB subgraph, such as in Fig. 6, then we apply the protocol in (2) to the system in (1) with satisfying and . Fig. 6Open in figure viewerPowerPoint Signed digraph with structural unbalance and no ISB subgraph In this case, the conditions of Theorem 2 hold. The initial state is given by . The states of the closed-loop system are described in Fig. 7. Fig. 7Open in figure viewerPowerPoint Seven state trajectories under signed digraph with structural unbalance and no ISB subgraph It can be seen that mean square stabilising is achieved asymptotically. 5 Concluding remarks Bipartite consensus for first-order integrator MASs is investigated in the context of measurement noise, where existing protocols often lead to fluctuations. A new stochastic type protocol with time-varying consensus gain is proposed, and necessary and sufficient conditions for ensuring mean square bipartite consensus are given. Moreover, under the assumption of structural unbalance, the states of the agents are proved to converge to zero in mean square, regardless of the initial states. The results in this work are only for first-order integrator MASs under fixed topologies, and hence time-varying topology becomes interesting in our future work. 6 Acknowledgments The work was supported in part by the National Natural Science Foundation of China under grants 61104136, 61673350, the Natural Science Foundation of Shandong Province under grants ZR2010FQ002, ZR2016FQ09 and the Qufu Normal University Domestic Visit Program for Enhancing Capabilities of Young Scholars. 7 References 1Olfati-Saber R., and Murray R.: 'Consensus problems in networks of agents with switching topology and time-delays', IEEE Trans. Autom. Control, 2004, 49, (9), pp. 1520– 1533 2Ren W., and Beard R.M.: 'Consensus seeking in multiagent systems under dynamically changing interaction topologies', IEEE Trans. Autom. Control, 2005, 50, (5), pp. 655– 661 3Yu J.Y., and Wang L.: 'Group consensus in multi-agent systems with switching topologies and communication delays', Syst. Control Lett., 2010, 59, (6), pp. 340– 348 4Ma C., and Zhang J.: 'Necessary and sufficient conditions for consensusability of linear multi-agent systems', IEEE Trans. Autom. Control, 2010, 55, (5), pp. 1263– 1268 5Liu S. Xie L., and Zhang H.: 'Distributed consensus for multi-agent systems with delays and noises in transmission channels', Automatica, 2011, 47, (5), pp. 920– 934 6Xin Y., and Cheng Z.: 'R-consensus control for discrete-time high-order multi-agent systems', IET Control Theory Appl., 2013, 7, (17), pp. 2103– 2109 7Xin Y., and Cheng Z.: 'R-consensus control for high-order multi-agent systems with digraph', Asian J. Control, 2013, 15, (5), pp. 1524– 1530 8Altafini C.: 'Consensus problems on networks with antagonistic interactions', IEEE Trans. Autom. Control, 2013, 58, (4), pp. 935– 946 9Wasserman S., and Faust K.: ' Social network analysis: methods and applications' ( Cambridge University Press, 1994) 10Hu J. Chen S., and Chen L.: 'Adaptive control for anti-synchronization of Chua's chaotic system', Phys. Lett. A, 2005, 339, (6), pp. 455– 460 11Hu J., and Zheng W.X.: 'Emergent collective behaviors on coopetition networks', Phys. Lett. A, 2014, 378, (26-27), pp. 1787– 1796 12Proskurnikov A.V. Matveev A., and Cao M.: 'Opinion dynamics in social networks with hostile camps: consensus vs. polarization', IEEE Trans. Autom. Control, 2016, 61, (6), pp. 1524– 1536 13Meng D. Du M., and Jia Y.: 'Interval bipartite consensus of networked agents associated with signed digraphs', IEEE Trans. Autom. Control, 2016, 61, (12), pp. 3755– 3770 14Valcher M.E., and Misra P.: 'On the consensus and bipartite consensus in high-order multi-agent dynamical systems with antagonistic interactions', Syst. Control Lett., 2014, 66, (4), pp. 94– 103 15Hu J., and Zhu H.: 'Adaptive bipartite consensus on coopetition networks', Phys. D, Nonlinear Phenom., 2015, 307, pp. 14– 21 16Wu Y. Hu J., and Zhang Y., et al.: 'Interventional consensus for high-order multi-agent systems with unknown disturbances on coopetition networks', Neurocomputing, 2016, 194, pp. 126– 134 17Hu J., and Wu Y.: 'Interventional bipartite consensus on coopetition networks with unknown dynamics', J. Franklin Inst., 2017, 354, (11), pp. 4438– 4456 18Huang M., and Manton J.H.: 'Coordination and consensus of networked agents with noisy measurements: stochastic algorithms and asymptotic behavior', SIAM J. Control Optim., 2009, 48, (1), pp. 134– 161 19Li T., and Zhang J.F.: 'Mean square average-consensus under measurement noises and fixed topologies: necessary and sufficient conditions', Automatica, 2009, 45, (8), pp. 1929– 1936 20Li T., and Zhang J.F.: 'Consensus conditions of multi-agent systems with time-varying topologies and stochastic communication noises', IEEE Trans. Autom. Control, 2010, 55, (9), pp. 2043– 2057 21Cheng L. Hou Z.G., and Tan M., et al.: 'Necessary and sufficient conditions for consensus of double-integrator multi-agent systems with measurement noises', IEEE Trans. Autom. Control, 2011, 56, (8), pp. 1958– 1963 22Cheng L. Hou Z.G., and Tan M.: 'A mean square consensus protocol for linear multi-agent systems with communication noises and fixed topologies', IEEE Trans. Autom. Control, 2014, 59, (1), pp. 261– 267 23Ni Y., and Li X.: 'Consensus seeking in multi-agent systems with multiplicative measurement noises', Syst. Control Lett., 2013, 62, (5), pp. 430– 437 24Li T. Wu F., and Zhang J.F.: 'Multi-agent consensus with relative-state-dependent measurement noises', IEEE Trans. Autom. Control, 2014, 59, (9), pp. 2463– 2468 25Ma C.Q., and Qin Z.Y.: 'Bipartite consensus on networks of agents with antagonistic interactions and measurement noises', IET Control Theory Appl., 2016, 10, (17), pp. 2306– 2313 26Zhang H., and Chen J.: ' Bipartite consensus of general linear multi-agent systems'. Proc. American Control Conf., Portland, Oregon, USA, 4–6 June 2014, pp. 808– 812 27Chow Y.S., and Teicher H.: ' Probability theory: independence, interchangeability, martingales' ( Springer, 1997, 3rd edn.) 28Zhang H., and Chen J.: 'Bipartite consensus of multi-agent systems over signed graphs: state feedback and output feedback control approaches', Int. J. Robust Nonlinear Control, 2017, 27, (1), pp. 3– 14 29Zhang H., and Chen J.: ' Bipartite consensus of linear multi-agent systems over signed digraphs: an output feedback control approach'. Proc. 19th IFAC World Congress, Cape Town, South Africa, 24–29 August 2014, pp. 4681– 4686 Citing Literature Volume11, Issue18December 2017Pages 3313-3320 FiguresReferencesRelatedInformation