We consider consensus of a class of third-order continuous-time multiagent systems with time delay in undirected networks. By using matrix analysis and a frequency domain approach, a necessary and sufficient condition for consensus is established. A simulation result is also given to illustrate the main theoretical result.
1. Introduction
In recent years, more and more people are interested in multiagent systems due to their extensive applications in real world, such as cooperative control of unmanned aerial vehicle, autonomous underwater submarine, flocking, and network control in information. Among the corresponding theories for multiagent systems, consensus problem is a critical issue which has attracted interdisciplinary researchers from the fields of control, mathematics, biology, physics, computer, robot, communication and artificial intelligence, and so forth.
Most work on consensus focuses on protocols taking the form of first-order dynamics [1–5]. Extensions of consensus protocols to second-order dynamics are reported in [6–11]. Just as is shown in [9], the convergence of second-order consensus algorithms relies on not only information exchange topology but also the coupling strength between the information states in contrast to first-order consensus algorithms. There is also some recent work focusing on consensus protocols with both first-order and second-order dynamics [12, 13].
It is well known that the dynamic equation given by second-order consensus protocols contains the derivatives of position and velocity. In practice, the dynamics of agents may involve the derivative of accelerated speed. The derivative of accelerated speed is usually called Jerk in engineering. Jerk is a physical quantity to describe the changing rate of acceleration. A Jerk is often required in engineering, especially in transportation design and materials, and so forth. Therefore, it is necessary and significant to design a class of third-order consensus protocols by using the information of position, velocity, and accelerated speed of agents, which is the main purpose of this paper.
Recently, some people have studied some higher-order consensus protocols [14, 15]. In [14], the authors presented some necessary and sufficient conditions for consensus of high-order multiagent systems. Consensus of multiagent systems with linear higher-order agents was considered in [15], where the authors answered whether the group converges to consensus and what consensus value it eventually reaches. However, necessary and sufficient conditions for consensus analysis of higher-order protocols in [14, 15] are given based on the distribution of eigenvalues of a complex matrix, which do not clearly illustrate the relationship among the eigenvalues of the involved Laplacian matrix, coupling strengths, and time delay when the system achieves consensus. On the other hand, higher-order consensus protocols with time delay receive less attention in the literature. Therefore, in this paper we will introduce a third-order consensus protocol with time delay and provide a convergence analysis for the given third-order consensus protocol in undirected networks. A necessary and sufficient condition for consensus is established, which clearly shows the relationship among the nonzero eigenvalues of the Laplacian matrix, coupling strengths, and time delay when the system achieves consensus asymptotically.
An outline of this paper is as follows. In Section 2, some preliminaries about graph and a class of third-order protocol are introduced. The main result of this paper is proposed in Section 3. In Section 4, a simulation result is given to illustrate the main result. Section 5 concludes the whole paper.
Notations. The following notations will be used throughout this paper. · denotes Euclidean norm. j is the imaginary unit. ⊗ denotes the Kronecker product. Rn and RN×N denote the real space of n-dimensional column vectors and N×N-dimensional matrices, respectively. IN(ON)∈RN×N is an N×N-dimensional identity (zero) matrix and 1N=(1,…,1)T(0N=(0,…,0)T) denotes the N-dimensional column vector with entries 1(0).
2. Preliminaries2.1. Graph Theory
Let G=(ν,ε,A) be an N-order weighted undirected graph with set of nodes ν={ϑ1,ϑ2,…,ϑN}, set of edges ε={(ϑi,ϑj):ϑi,ϑj∈ν}⊆ν×ν, and weighted adjacent matrix A=(aij)N×N. An edge of G is denoted by eij=(ϑi,ϑj). In the undirected graph, eij∈ε⇔eji∈ε. The adjacency matrix is defined as aii=0 and aij=aji≥0 for i≠j and i,j∈{1,2,…,N}, where aij>0⇔eij∈ε. A path is a sequence of edges of the form (ϑi1,ϑi2),(ϑi2,ϑi3),…, where ϑij∈ν. If there is a path for arbitrary two nodes in the graph, we call that the graph is connected.
The Laplacian matrix of the graph G is defined as L=[lij]∈RN×N, where (1)lii=-∑j=1,j≠iNlij,lij=-aij,i≠j.Obviously, the Laplacian matrix L is a positive semidefinite matrix and satisfies lii≥0, lij≤0 for i≠j and ∑j=1Nlij=0 for i=1,…,N.
The following lemma is well known.
Lemma 1 (see [2]).
If the undirected graph G is connected, its Laplacian L has a simple zero eigenvalue with the associated eigenvector 1N and all the other eigenvalues are positive real numbers.
2.2. Third-Order Consensus Protocol
Suppose the dynamics of each node is described by the following third-order system:(2)x˙it=vit,v˙it=zit,z˙it=uit,i=1,2,…,N,where t≥0, xi(t)∈Rn, vi(t)∈Rn, and zi(t)∈Rn represent the position, velocity, and accelerated speed of node ϑi, respectively, and ui(t)∈Rn is the control input (or protocol) to be designed.
In this paper we consider the following third-order protocol:(3)uit=k1∑j=1Naijxjt-τ-xit-τ+k2∑j=1Naijvjt-τ-vit-τ+k3∑j=1Naijzjt-τ-zit-τ,where i=1,2,…,N, k1>0, k2>0, and k3>0 are appropriate coupling strengths.
Definition 2.
Say system (2) with protocol (3) achieves consensus asymptotically, if limt→∞xi(t)-xj(t)=0, limt→∞vi(t)-vj(t)=0, and limt→∞zi(t)-zj(t)=0 for any i,j=1,2,…,N and any initial condition x(t)=ϕ(t), where ϕ is a continuous function defined on [-τ,0].
We first have the following lemma for system (2) with protocol (3).
Lemma 3.
If system (2) with protocol (3) achieves consensus asymptotically, one has(4)limt→∞xit-1N∑j=1Nxj0+vj0t+12zj0t2=0,limt→∞vit-1N∑j=1Nvj0+zj0t=0,limt→∞zit-1N∑j=1Nzj0=0.
Proof.
(i) Since the graph is undirected, by (2) and (3), we have ∑j=1Nz˙j(t)=0. Integrating it on both sides from 0 to t, we have(5)∑j=1Nzjt=∑j=1Nzj0.Denote eij(t)=zj(t)-zi(t)(i,j=1,…,N,j≠i). Thus, (5) will be rewritten into the following form: (6)Nzit+∑j=1,j≠iNeijt=∑j=1Nzj0.Namely,(7)zit-1N∑j=1Nzj0=-1N∑j=1,j≠iNeijt.Note that the system achieves consensus; we have limt→∞eijt=0 for i,j=1,…,N,j≠i. Therefore, we get from (7) that (8)limt→∞zit-1N∑j=1Nzj0=0.
(ii) By (2) and (5), we have (9)∑j=1Nv˙jt=∑j=1Nzjt=∑j=1Nzj0.Integrating the above equality from 0 to t yields(10)∑j=1Nvjt=∑j=1Nvj0+zj0t.Similar to the discussion in (i), we can easily get the second equality of (4).
(iii) By (2) and (10), we have (11)∑j=1Nx˙jt=∑j=1Nvjt=∑j=1Nvj0+zj0t.By integrating it from 0 to t, we get (12)∑j=1Nxjt=∑j=1Nxj0+vj0t+12zj0t2.We can similarly derive the first equality of (4) as above. The proof of Lemma 3 is complete.
3. Main Result
In the sequel, set (13)ξt=x1Tt,v1Tt,z1Tt,…,xNTt,vNTt,zNTtT.System (2) with protocol (3) can be rewritten into(14)ξ˙t=IN⊗Aξt-L⊗Bξt-τ,where (15)A=010001000⊗In,B=000000k1k2k3⊗In.
By Lemma 3, let (16)δ¯t=ξt-1N⊗αTt,βTt,γTT,where (17)αt=1N∑j=1Nxj0+vj0t+zj0t22,βt=1N∑J=1Nvj0+zj0t,γ=1N∑j=1Nzj0.A straightforward computation yields that system (14) is equivalent to(18)δ¯˙=IN⊗Aδ¯t-L⊗Bδ¯t-τ.
By Lemma 1, we can denote the nonzero eigenvalues of L by λ2,λ3,…,λN and set 0<λ2≤λ3≤⋯≤λN. Then there exists an orthogonal matrix W such that (19)Λ¯=WTLW=diag0,λ2,…,λN,WT⊗I3δ¯t=0nT,0nT,0nT,δTtT,where δ(t) is a column vector of appropriate dimension. Premultiplying both sides of (18) by WT⊗I3, we have(20)δ˙t=IN-1⊗Aδt-Λ⊗Bδt-τ,where Λ=diag{λ2,…,λN}.
Theorem 4.
System (2) with protocol (3) achieves consensus asymptotically if and only if(21)λik2ωcosωτ+k3ω2sinωτ-k1sinωτω3<1,where i=2,3,…,N; ω satisfies the following equation:(22)k2ωtanωτ=k3ω2-k1.
Proof.
Based on the above analysis, consensus of system (14) is equivalent to the asymptotic stability of system (20). On the other hand, system (20) is asymptotically stable if and only all its eigenvalues lie on the open left half plane (LHP). Thus, it only requires showing that all the eigenvalues of system (20) lie on the open LHP if and only if (21) and (22) hold. By using Laplace transform on both sides of (20), we have δ(s)=Gτ-1(s)δ(0), where (23)Gτs=sI3nN-1-IN-1⊗A+Λ⊗Be-τs.Note that (24)IN-1⊗A-Λ⊗Beτs=diagA-λ2eτsB,…,A-λneτsB.By a direct calculation, we get the determinant of Gτ(s):(25)detGτs=∏i=2NdetsI3n-A+λie-τsB=∏i=2NdetsI3-Mi⊗In=∏i=2Ns3+λik3e-τss2+λik2e-τss+λik1e-τsn,where (26)Mi=010001-λik1-λik2-λik3,i=2,3,…,N.By (25), all the zeros of Gτ(s) lie on the open LHP if and only if all the roots of equations(27)s3+λik3e-τss2+λik2e-τss+λik1e-τs=0,i=2,…,N,are all located on the open LHP. Since (27) is equivalent to(28)1+λik3e-τss2+λik2e-τss+λik1e-τss3=0,for i=2,…,N, all the roots of (27) lie on the open LHP if and only if all the roots of (28) are located on the open LHP. By the Nyquist stability criterion, all the roots of (28) lie on the open LHP if and only if the Nyquist curve(29)z=λik3e-τjωjω2+λik2e-τjωjω+λik1e-τjωjω3does not enclose the point (-1,0j), where j is an imaginary number unit and i=2,…,N. Using Euler’s formula for (29), we have (30)z=-λik2ωcosωτ+k3ω2sinωτ-k1sinωτω3+-k3ω2cosωτ+k2ωsinω+k1cosωτλiω3j.Hence, all the Nyquist curves (29) do not enclose the point (-1,0j) if and only if (21) holds, where ω is the solution of equation (31)-k3ω2cosωτ+k2ωsinω+k1cosωτλi=0.Namely, (22) holds. The proof of Theorem 4 is complete.
For the particular case when τ=0, Theorem 4 reduces to Corollary 5 in [16].
Corollary 5.
System (2) with protocol (3) and τ=0 achieves consensus asymptotically if and only if the coupling strengths k1, k2, and k3 and the eigenvalue λ2 of the Laplacian matrix L satisfy (32)λ2>k1k2k3.
4. Simulation
In this section, we will give a simulation to illustrate the main result. The topology graph in the simulation is shown in Figure 1 which has four agents and 0-1 weights.
An undirected graph.
It is easy to see the corresponding Laplacian matrix (33)L=2-10-1-12-100-12-1-10-12.
If we consider k1=k2=k3=1 and τ=0.3, conditions (21) and (22) are satisfied. By Theorem 4, consensus of system (2) with protocol (3) is achieved asymptotically. However, when τ=0.356, conditions (21) and (22) are not satisfied. Therefore, system (2) with protocol (3) does not achieve consensus asymptotically. The states of position, velocity, and accelerated speed of agents with τ=0.3 and τ=0.356 are shown in Figures 2 and 3, respectively.
Trajectories of system (2) with k1=k2=k3=1, τ=0.3.
Trajectories of system (2) with k1=k2=k3=1, τ=0.356.
5. Conclusions
In this paper, we study the consensus problem of the third-order dynamic multiagent system with time delay in undirected graphs. A necessary and sufficient condition for consensus of the system has been established. A simulation result illustrates the effectiveness of the theoretical result. Consensus of third-order multiagent systems with time delay in directed graphs will be further studied in the future.
Competing Interests
The authors declare that they have no competing interests.
Acknowledgments
This work was supported by the Natural Science Foundation of Shandong Province under Grant no. JQ201119 and the National Natural Science Foundation of China under Grant nos. 61174217, 61374074, and 61473133.
JadbabaieA.LinJ.MorseA. S.Coordination of groups of mobile autonomous agents using nearest neighbor rules2003486988100110.1109/tac.2003.812781MR19862662-s2.0-0038548185Olfati-SaberR.MurrayR. M.Consensus problems in networks of agents with switching topology and time-delays20044991520153310.1109/tac.2004.834113MR20869162-s2.0-4644244041RenW.BeardR. W.Consensus seeking in multiagent systems under dynamically changing interaction topologies200550565566110.1109/tac.2005.846556MR21415682-s2.0-20344399896ShenJ.CaoJ.Consensus of multi-agent systems on time scales201229450751710.1093/imamci/dns006MR3002709ZBL1256.930682-s2.0-84871186689JiZ.LinH.YuH.Protocols design and uncontrollable topologies construction for multi-agent networks201560378178610.1109/tac.2014.2335971MR33184032-s2.0-84923345301RenW.BeardR. W.2008London, UKSpringYuW.ChenG.CaoM.KurthsJ.Second-Order consensus for multiagent systems with directed topologies and nonlinear dynamics201040388189110.1109/TSMCB.2009.20316242-s2.0-77952581942YuW.ChenG.CaoM.Some necessary and sufficient conditions for second-order consensus in multi-agent dynamical systems20104661089109510.1016/j.automatica.2010.03.006MR28771922-s2.0-77953129248RenW.Second-order consensus algorithm with extensions to switching topologies and reference modelsProceedings of the American Control Conference (ACC '07)July 20071431143610.1109/acc.2007.42822042-s2.0-39549110331XuG.-H.GuanZ.-H.HeD.-X.ChiM.WuY.-H.Distributed tracking control of second-order multi-agent systems with sampled data2014351104786480110.1016/j.jfranklin.2014.06.003MR32591432-s2.0-84908151770PeiH.ChenS.LaiQ.A local flocking algorithm of multi-agent dynamic systems201588112242224910.1080/00207179.2015.1039595MR3420623ZBL1334.93018ZhengY.ZhuY.WangL.Consensus of heterogeneous multi-agent systems20115161881188810.1049/iet-cta.2011.0033MR29067432-s2.0-81255193942ZhengY.WangL.Finite-time consensus of heterogeneous multi-agent systems with and without velocity measurements201261887187810.1016/j.sysconle.2012.05.009MR29563692-s2.0-84863496064RenW.MooreK.ChenY. Q.High-order consensus algorithms in cooperative vehicle systemsProceedings of the IEEE International Conference on Networking, Sensing and Control2006Fort Lauderdale, Fla, USA45746210.1109/ICNSC.2006.1673189WielandP.KimJ. S.ScheuH.AllgöwerF.On consensus in multi-agent systems with linear high-order agentsProceedings of the 17th World Congress of the International Federation of Automatic ControlJuly 2008Seoul, Republic of Korea15411546CaoY.SunY.Consensus analysis for third-order multiagent systems in directed networks20152015994082310.1155/2015/940823MR3372946