Event-based consensus of multi-agent systems with general

合集下载

基于语义和多agent的电子政务协同工作模型研究

基于语义和多agent的电子政务协同工作模型研究
第2 第 1 5卷 1期
20 0 8年 1 1月
计 算 机 应 用 研 究
Ap l a in Re e r h o mp t  ̄ p i t s a c fCo u e c o
Vo . 5 No 1 12 . 1 NO . 2 0 V 08
基于语义和 多 a e t g n 的电子政务协 同工作模型研究
对 该领域 的共 同理解 。近几 年 , 体 已经成 为信 息检索 、 本 语 义网、 人工智能等研 究领 域 的热 门课 题 。本 体能 够解决协 同工作 过程 中语义层次上知识 的重用和共享 , 保证协 同信息在 交流过程中的语义一致性 。本 体论 的一 个重要应用是在 语义
琚春华 , 章朝华
( 浙江工商大学 计算机与信息工程学院, 杭州 30 1) 108
摘 要 :从 电子政务 的发展 趋势 出发 , 语 义网技 术 的基 础上 , 用 X / D 描 述协 同过程 中的语 义信 息 ; 在 采 MLR F 提
出了任务一 本体相 关联 的思想 , 用 R SC 采 M— A算法 , 态生成全局 本体 , F 动 形成部 门本体一 全局 本体 的双本 体层 ,
J h n h a HANG Ch oh a U C u —u ,Z a —u
( oeeo o p  ̄ c ne&I om t nE gnei Clg C m u r i c l f Se n r ai n i r g, f o e n
G nsag U i rt, agh u30 1 ,C ia ogh n nv sy H nzo 10 8 hn ) ei
中图分 类号 :T 33 P0
文献标 志码 : A
文章编 号 :10 -65 20 )13 1—3 0 139 (08 1—280

事件触发下随机非确定线性多智能体的指数同步

事件触发下随机非确定线性多智能体的指数同步

事件触发下随机非确定线性多智能体的指数同步邱丽;过榴晓【摘要】研究随机非确定线性多智能体系统在有向拓扑连接下的指数同步问题,为减少不必要的网络带宽资源的浪费,提出一种基于事件触发控制的协议.根据组合测量对系统中的所有节点设计相应的事件触发函数,使得节点之间的控制信号更新仅在事件触发时刻进行.基于Lyapunov稳定性理论和M矩阵理论,得到了多智能体系统指数同步结论,并给出了同步的收敛速度.同时,理论排除了事件触发控制过程中的芝诺(Zeno)现象.数值仿真结果进一步验证了理论分析的有效性.【期刊名称】《计算机工程与应用》【年(卷),期】2018(054)017【总页数】6页(P141-145,163)【关键词】事件触发控制;随机非确定;线性多智能体系统;指数同步;Zeno现象【作者】邱丽;过榴晓【作者单位】江南大学理学院,江苏无锡 214122;江南大学理学院,江苏无锡214122【正文语种】中文【中图分类】TP2731 引言多智能体系统是由多个能够相互作用、共同协作的个体组成的系统,其中每个个体具有自组织和通讯的能力,各个智能体能够通过彼此之间的信息交换来实现对整个系统的协调控制。

近年来,由于控制理论和应用的发展,多智能体系统已成为控制领域中一个重要的研究对象,其中多智能体系统的同步问题已取得不少成果[1-8]。

如:整体同步[1],局部同步[2],聚类同步[4],指数同步[5-8]等。

指数同步因其在收敛速度方面的优势,成为学者们研究的热点问题之一。

在许多实际的多智能体系统中,智能体自身的能量和通信信道的带宽是有限的,为减少不必要的网络带宽资源的浪费,因此需要设计合适的通信控制方案,节省资源。

周期采样控制方法[9-11]是在等距离的离散时刻点上进行状态采样和信息通讯,有利于节约资源,但如果两个连续采样数据之间相差很小,继续周期采样控制,则明显浪费资源。

与周期采样控制相比,事件触发控制则执行较少的信息通讯,即当事先设定的触发条件不成立,控制器执行更新[12-13]。

Consensus seeking in multiagent systems under dynamically changing interaction topologies

Consensus seeking in multiagent systems under dynamically changing interaction topologies
IEEE TRANSБайду номын сангаасCTIONS ON AUTOMATIC CONTROL, VOL. 50, NO. 5, MAY 2005
655
[6] F. H. Clarke, Y. S. Ledyaev, E. D. Sontag, and A. I. Subbotin, “Asymptotic controllability implies feedback stabilization,” IEEE Trans Autom. Control, vol. 42, no. 10, pp. 1394–1407, Oct. 1997. [7] R. Goebel, “Convex optimal control problems with smooth Hamiltonians,” SIAM J. Control Optim., to be published. [8] , “Duality and uniqueness of convex solutions to stationary Hamilton-Jacobi equations,” Trans. Amer. Math. Soc., vol. 357, pp. 2187–2203, 2005. [9] , “Hamiltonian dynamical systems for convex problems of optimal control: implications for the value function,” in Proc. 41st IEEE Conf. Decision and Control, Las Vegas, NV, 2002, pp. 728–732. , “Stationary Hamilton-Jacobi equations for convex control prob[10] lems—uniqueness and duality of solutions,” in Optimal Control, Stabilization, and Nonsmooth Analysis, ser. Lecture Notes in Control and Information Sciences. Heidelberg, Germany: Springer-Verlag, 2004. [11] R. Goebel and M. Subbotin, “Continuous time constrained linear quadratic regulator—convex duality approach,” in Proc. 24th Amer. Control Conf., 2005, to be published. [12] M. L. Gota and L. Montrucchio, “On Lipschitz continuity of policy functions in continuous-time optimal growth models,” Econom. Theory, vol. 14, pp. 479–488, 1999. [13] T. Hu and Z. Lin, Control Systems With Actuator Saturation. Boston, MA: Birkhäuser, 2001. [14] R. T. Rockafellar, Convex Analysis. Princeton, NJ: Princeton Univ. Press, 1970. , “Saddle points of Hamiltonian systems in convex problems of La[15] grange,” J. Optim. Theory Appl., vol. 12, no. 4, 1973. , Conjugate Duality and Optimization. Philadelphia, PA: SIAM, [16] 1974. [17] R. T. Rockafellar and R. J.-B. Wets, Variational Analysis. New York: Springer-Verlag, 1998. [18] A. Saberi, Z. Lin, and A. R. Teel, “Control of linear systems with saturating actuators,” IEEE Trans. Autom. Control, vol. 41, no. 3, pp. 368–378, Mar. 1996. [19] E. D. Sontag and H. J. Sussman, “Nonlinear output feedback design for linear systems with saturating controls,” in Proc. 29 IEEE Conf. Decision Control, Honolulu, HI, 1990, pp. 3414–3416. [20] E. D. Sontag, H. J. Sussmann, and Y. D. Yang, “A general result on the stabilization of linear systems using bounded controls,” IEEE Trans. Autom. Control, vol. 39, no. 12, pp. 2411–2424, Dec. 1994. [21] A. A. Stoorvogel, A. Saberi, and G. Shi, “Properties of recoverable region and semi-global stabilization in recoverable region for linear systems subject to constraints,” Autmomatica, vol. 40, no. 9, pp. 1481–1494, 2004.

基于辩论的多agent商务谈判产生机制研究

基于辩论的多agent商务谈判产生机制研究
Ke r y wo ds: a g me t to b s d n g tai n;ag r u n ain— a e e oito rume tg neai n; ag n e r to rume tsr tg n tae y; ag m e tprt c l r u n oo o
t ng n r i e h ns a e nc n i — s v g n ru e t l s i t n h e h ns c d d am l — jc v i e ea o m c a i b sd o o fc r o i dag m n— a ic i .T em c a i i l e u i be t e o tn m lte ln a c sfao m nu to i
基于辩论的谈判 是为 了解决 传统 的两种谈 判机 制 ( 于 基 博弈 的谈判和基于启发式的谈判 ) 的主要 问题而提 出的 , 由于 它允许 aet gn 在谈判 过程 中通过辩 论来 交换 和交 流额 外 的信
息 , 而 能够 对 其 他 aet 目标 、 识 、 为 等 施 加 动 态 的影 从 gn 的 知 行
d i 1 3 6 / .sn 1 01 3 9 . 01 . 2 0 3 o :0. 9 9 iis . 0 — 6 5 2 1 0 . 5
Ge e ain me h ns i lia e tta e ag me tto - a e e oito n r t c a im n mu t—g n r d ru nain- s d n g tain o梯 云
(. 1 北京 工业 大学 经济 与管理 学院 ,北 京 10 2 2 首都 师 范大学 教 育技 术 系,北京 10 3 ) 0 14; . 007 要 :针 对辩论 的产 生是 基 于辩 论的谈 判 中 aet gn 决策机 制 的核 心 问题 , 出 了一 个基 于 冲 突解决 和辩 论 分 提

异构非线性多智能体系统的一致性

异构非线性多智能体系统的一致性

Advances in Applied Mathematics 应用数学进展, 2023, 12(9), 3872-3885 Published Online September 2023 in Hans. https:///journal/aam https:///10.12677/aam.2023.129381异构非线性多智能体系统的一致性谢浩浩,李超越,贺 鑫长安大学理学院,陕西 西安收稿日期:2023年8月9日;录用日期:2023年9月3日;发布日期:2023年9月8日摘要针对一阶智能体和二阶智能体组成的异构多智能体系统,在无向通讯拓扑下研究了具有输入饱和与非输入饱和的异构非线性多智能体系统的一致性问题。

首先,分别提出了基于牵制控制和事件触发控制的一致性控制协议,其次,通过对每个智能体设计事件触发条件,当满足事件触发条件时,智能体才向周围的邻居传递自身的状态信息和更新控制器,且每个智能体只在自己的触发时刻进行传递和更新。

然后利用图论、Lyapunov 稳定性理论和LaSalle 不变集理论,证明了在满足某些条件下,该系统不仅达到了期望的一致性状态,而且减少了控制器的更新次数,有效地节省了通讯资源。

最后,通过数值模拟验证了理论的正确性。

关键词异构多智能体系统,牵制控制,事件触发控制,一致性,饱和输入,非线性Consensus of Heterogeneous Nonlinear Multi-Agent SystemsHaohao Xie, Chaoyue Li, Xin HeSchool of Sciences, Chang’an University, Xi’an ShaanxiReceived: Aug. 9th , 2023; accepted: Sep. 3rd , 2023; published: Sep. 8th, 2023AbstractThe consensus problem of heterogeneous nonlinear multi-agent systems with and without input saturation is investigated under the undirected communication topology for heterogeneous mul-ti-agent systems composed of first-order agents and second-order agents. First, consensus control protocols based on pinning control and event-triggered control are proposed respectively, and second, by designing event-triggered conditions for each agent, the agent transmits its own state information and updates its controller to its surrounding neighbors only when the event-triggered谢浩浩等conditions are satisfied, and each agent transmits and updates only at its own triggering moments. Then using graph theory, Lyapunov stability theory and LaSalle invariance principle, it is proved that the systems not only achieve the desired consensus state, but also reduce the number of con-troller updates and effectively save the communication resources under the fulfillment of certain conditions. Finally, the correctness of the theory is verified by numerical simulation. KeywordsHeterogeneous Multi-Agent Systems, Pinning Control, Event-Triggered Control, Consensus, Saturated Inputs, NonlinearThis work is licensed under the Creative Commons Attribution International License (CC BY 4.0)./licenses/by/4.0/1. 引言近年来,多智能体系统的一致性问题引起了学者们的广泛关注,并且在传感器网络[1]、编队控制[2]、群居昆虫的集群[3]、机器人[4]等具有广泛的实际应用价值。

随机脉冲控制下领导灢跟随的多智能体系统一致性研究

随机脉冲控制下领导灢跟随的多智能体系统一致性研究

第38卷第6期2023年12月安 徽 工 程 大 学 学 报J o u r n a l o fA n h u i P o l y t e c h n i cU n i v e r s i t y V o l .38N o .6D e c .2023文章编号:1672-2477(2023)05-0072-05收稿日期:2022-09-28基金项目:国家自然科学基金资助项目(61873294);安徽省杰出青年科学基金资助项目(1908085J 04)作者简介:韩曼利(1997-),女,安徽宿州人,硕士研究生㊂通信作者:吴小太(1982-),男,安徽枞阳人,教授,博士㊂随机脉冲控制下领导-跟随的多智能体系统一致性研究韩曼利,吴小太*(安徽工程大学数理与金融学院,安徽芜湖 241000)摘要:针对一类非线性领导-跟随多智能体系统,研究随机脉冲控制下多智能体系统的一致性问题㊂由于脉冲控制在实际系统中,常常会受到各种随机因素的影响,并导致脉冲控制强度发生改变㊂因此,本文通过引入随机变量,设计了具有随机脉冲强度的分布式脉冲控制协议㊂随后,利用随机分析的方法给出了在随机脉冲控制下领导-跟随多智能体系统达成一致的充分条件㊂最后,作为特例还给出了拒绝服务攻击下领导-跟随多智能体系统的一致性准则㊂关 键 词:领导-跟随多智能体系统;一致性;随机脉冲中图分类号:T P 13 文献标志码:A在过去的几十年中,由于多智能体系统在生物学㊁工程学㊁人工智能等诸多领域中的广泛应用,多智能体协同控制逐渐成为控制科学界的研究热点[1-3]㊂一致性作为多智能体协同控制的基础问题,其任务是设计一个控制策略,使所有节点就某些共同的目标达成一致,从而完成一些大规模集群任务[4]㊂时至今日,有关多智能体系统的一致性研究取得了丰硕的研究成果,例如:基于采样数据[5]㊁输出[6],以及领导-跟随[7-9]的多智能体系统一致性等㊂脉冲控制是一种典型的非连续控制,具有简单㊁灵活等优良性能[10]㊂同时,相较于连续控制,脉冲控制可以极大地减少控制成本和网络负载,因而被广泛地应用于多智能体的协同控制中[7,10,11-12]㊂例如:文献[12]研究了线性多智能体系统的一致性问题,并提出了两种分布式脉冲控制协议㊂在此基础上,文献[7]针对非线性多智能体系统,提出了一种改进的分布式脉冲控制协议,并给出了相应的一致性准则㊂值得注意的是,在现有多智能体脉冲控制的研究中,所考虑的脉冲大多是确定性的,即脉冲强度和密度均被假设为确定的[13]㊂然而,在多智能体的实际控制中脉冲控制效果往往会受到大量随机因素的影响[14]㊂因此,需要引入随机模型来刻画脉冲控制的随机性㊂近年来,关于随机脉冲控制的相关研究取得了一系列重要研究成果[13-15]㊂但针对随机脉冲控制下的多智能体系统一致性问题的研究相对较少,仍有不少有意义的问题值得进一步研究㊂此外,由于单个智能体之间通过共同的网络进行信息交互,导致多智能体系统非常容易受到拒绝服务(D o S )攻击[8]㊂D o S 攻击的本质是阻止传感器和控制器的数据访问其目的地,也可以被视作一种特殊的随机脉冲扰动㊂因此,研究随机D o S 攻击下多智能体的一致性无疑是一个有意义的问题㊂基于上述讨论,本文研究了随机脉冲控制下的非线性领导-跟随多智能体系统的一致性问题㊂首先,通过引入随机参数刻画随机脉冲的强度,设计了分布式随机脉冲控制协议㊂其次,借助概率分析和L y a -pu n o v 函数稳定性分析的方法,给出了随机脉冲控制下领导-跟随多智能体的一致性准则㊂本文的贡献包含以下两个方面:①建立了一类基于随机脉冲控制的非线性领导-跟随多智能体模型,利用随机分析的方法给出了多智能体一致性的充分条件㊂相较于文献[7],本文在系统模型与研究方法上均具有一定的创新性;②给出了脉冲控制受到随机D o S 攻击时,非线性领导-跟随多智能体系统的一致性准则㊂本文中,ℝn 表示n 维实数集,I n 表示n 维的单位矩阵㊂令x T㊁‖x ‖分别表示x 的转置和欧几里德模,⊗表示K r o n e c k e r 积㊂定义λm a x a 和λm i na 分别为A 的最大和最小特征值㊂定义‖A ‖=λm a x (A T A ),μ2a =λm a x (A +A T )/2㊂N 个智能体的信息交互用图G ={V ,E ,A }表示,其中V ={1,2, ,N }表示节点集,E ⊆V ×V 表示边集,A =[a i j ]N ×N 表示邻接矩阵,当且仅当(j ,i )∈E 时,a i j >0,否则a i j =0㊂此外,假设a i i =0,i =1,2, ,N ㊂图G 的L a p l a c i a n 矩阵L =[l i j ]N ×N 被定义为:l i j =-a i j ,i ≠j ,∑N j =1a i j ,i =j {㊂1 预备知识本文考虑有N 个跟随者的多智能体系统,其跟随者的动力学方程为:x ㊃i (t )=A x i (t )+B g (x i (t ))+u i (t ),i =1,2, ,N ,(1)式中,x (t )∈ℝn 表示第i 个节点的状态,A 和B 为常数矩阵,g (㊃)为非线性函数,u i (t )∈ℝn表示控制输入㊂领导者的动力学方程为:s ㊃(t )=A s (t )+B g (s (t )),(2)式中s (t)∈ℝn 为领导者的状态㊂考虑如下控制协议:u i (t )=c γk ∑¥k =1[∑Nj =1-li j x j (t )+d i (s (t )-x i (t ))]δ(t -t k ),(3)这里c 表示耦合强度;γk 是一个随机变量,用于表征脉冲控制过程中的随机波动;d i >0表示在t 时刻领导者与第i 个节点之间存在直接联系;δ(㊃)表示狄拉克函数;{t k ,k ∈ℕ+}表示脉冲瞬间序列㊂定义误差状态e i (t )=x i (t )-s (t )㊂根据式(1)~(3),可以得到以下误差系统:e ㊃i (t )=A e i (t )+B g (e i (t ),s (t )),t ≠t k ,Δe i (t k )=c γk [∑N j =1-li j x j (t -k)+d i (s (t -k)-x i (t -k)],t =t k {,(4)式中,Δe i (t k )=e i (t k )-e i (t -k),e i (t k )=e i (t +k )=l i m h →0+e i (t k +h ),e i (t -k )=l i m h →0-e i (t k +h ),g (e i (t ),s (t ))=g (x i (t ))-g (s (t )),且在t =t k 时刻,e (t )是右连续的㊂令e (t )=[e T 1(t ),e T 2(t ),...,e T N (t )]T,误差系统(4)可改写为:e ㊃i (t )=(I N ⊗A )e (t )+(I N ⊗B )G (e i (t ),s (t )),t ≠t k ,Δe i (t k )=c γk [(L +D )⊗I N ]e (t -k ),t =t k {,(5)这里G (e i (t ),s (t ))=[G (e 2(t ),s (t ))T ,G (e 3(t ),s (t ))T , ,G (e i (t ),s (t ))T],D =d i a g {d 1,d 2,,d N }㊂接下来,在给出本文的主要结论之前,先给出一些必要的定义和假设㊂假设1[8] 非线性函数g (x )满足如下L i ps c h i t z 条件:‖g (a )-g (b )‖≤ρ‖a -b ‖,式中,a ,b ∈ℝn ,且ρ>0㊂假设2[14] 假定一组相互独立的随机变量γ{}l 为可能的脉冲强度,其中l ={1,2, ,v }且满足E γl =γ-l >0㊂定义1[8] 设N h (t ,s )为时间间隔(s ,t ]内的第h 种脉冲的出现次数,如果存在τa h >0,N 0h ≥0,有下列不等式成立:t -s τa h -N 0h ≤N (t ,s )≤t -s τa h+N 0h ,其中,τa h 和N 0h分别被称为第h 种脉冲的平均脉冲间隔和弹性系数㊂定义2[14] 对任意的x i (t 0)和s (t 0),如果存在常数M >0和λ>0,使得E ∑Ni =1‖x i (t )-s (t )‖2≤E ∑Ni =1‖x i (t 0)-s (t 0)‖2M e -λ(t -t 0),i =1,2,㊃37㊃第6期韩曼利,等:随机脉冲控制下领导-跟随的多智能体系统一致性研究则称多智能体系统(1)达成均方全局指数一致㊂2 主要结果在本节中,我们研究了基于随机脉冲控制的领导-跟随多智能体的一致性问题,这里将考虑脉冲强度随机而脉冲发生时间是确定的情况㊂定理1 若假设1和2成立,且存在一个正定矩阵P 和常数α>0,使得下列条件成立:P A +A TP +2ρ-P <αP ,(6)θ-k =λm a x (P )λm i n (P )E ‖θk ‖2<1,(7)τ<-l n θ~α,(8)其中,ρ-=η+λm a x (P )ρ2ηλm i n (P )‖B ‖2,θk =I N -c γk (L +D )⊗I N ,θ~=m a x {θ-k }㊂则在控制协议(3)的作用下,系统(4)能达成均方全局指数一致㊂证明 构建如下L y a pu n o v 函数:V (e (t ))=∑Ni =1e Ti (t )P e i (t ),(9)对V (e (t))求导,可得:V ㊃(e (t ))=∑Ni =1[e Ti (t )(A T P +P A )e i (t )+g T (e i (t ),s (t ))B T P e i (t )+e Ti P B g (e i (t ),s (t ))]㊂(10)根据假设2和Y o u n g 不等式,可知:2e Ti(t )P B g (e i (t ),s (t ))≤2(η+λm a x (P )ρ2ηλm i n (P )‖B ‖2)e T i (t )P e i (t ),(11)结合式(6)㊁(10)和(11),可以得到:V ㊃(e (t ))=∑Ni =1[e Ti (t )(A T P +P A +2ρ-P )e i (t )]<αV (e (t )),(12)那么,对于t ∈[t k ,t k +1),V (e (t ))<e α(t -t k )V (t k )㊂(13)另一方面,当t =t k 时,我们可以得出:V (e (t k ))=e T (t k )(I N ⊗P )e (t k )=e T (t -k )θT k (I N ⊗P )θk e (t -k ),(14)其中,θk =I N -c γk (L +D )⊗I N ㊂对式(14)两边同时取期望,可得:E V (e (t k ))=E [e T (t -k )θT k (I N ⊗P )θk e (t -k )]≤E [λm a x (P )‖θk ‖e T (t -k )θk e (t -k )]㊂(15)令F k =σ{γ1,γ2, ,γk }为γ1,γ2, ,γk 产生的σ域,可以得到:E [λm a x (P )‖θk ‖e T (t -k )θk e (t -k )]=λm a x (P )E [E [‖θk ‖2e T (t -k )e (t -k )|F k ]]=λm a x (P )E [e T (t -k )e (t -k )E [‖θk ‖2|F k ]]=λm a x (P )E ‖θk ‖2E [e T (t -k )e (t -k )]≤λm a x (P )λm i n(P )E ‖θk ‖2E V (e (t -k ))㊂(16)根据条件(7),可得:E V (e (t k ))≤θ-k E V (e (t -k )),(17)显然,对于t ∈[t 0,t 1),㊃47㊃安 徽 工 程 大 学 学 报第38卷E V (e (t 1))≤θ-1E V (e (t -1))≤θ-1e α(t 1-t 0)E V (e (t 0))㊂(18)同理对于t ∈[t 1,t 2),可得:E V (e (t ))≤e α(t -t 1)E V (e (t 1))≤θ-1e α(t 2-t 0)E V (e (t 0))㊂(19)根据定义1,可推出:E V (e (t ))≤E V (e (t 0))e ∑N (t ,t 0)l =1l n θ-l +α(t -t 0)≤E V (e (t 0))θ~t -t 0τ-N 0eα(t -t 0)≤E V (e (t 0))θ~-N 0e (l n θ~τ+α)(t -t 0),(20)其中θ~=m a x k ∈ℕ{θ-k }㊂根据条件(8),有E (∑Ni =1λm i n (P )‖e i (t )‖2)≤E (∑Ni =1λm a x (P )‖e i (t 0)‖2)θ~-N 0e (l n θ~τ+α)(t -t 0)㊂(21)令M =λm a x (P )λm i n (P )θ~-N 0,λ=-l n θ~τ-α,可得:E ∑N i =1‖e i (t )‖2≤E ∑Ni =1‖e i (t 0)‖2M e -λ(t -t 0)㊂(22)根据定义2可知,系统(4)在均方意义下达成全局指数一致㊂在文献[7-9]中,针对领导-跟随多智能体系统的一致性问题,一系列分布式脉冲控制协议被设计㊂与上述结果相比,定理1主要有以下两点创新:①提出了一类随机控制协议,定理1考虑了具有随机脉冲强度的分布式控制协议,其可将文献[7]中的控制协议视为特殊情况㊂②在随机意义下提出了一类新的分析方法:定理1借助概率分析法和L y a p u n o v 函数分析法,在平均脉冲间隔的假设下,给出了领导-跟随多智能体系统达成一致的充分条件㊂接下来,假设系统(4)受到随机D o S 攻击㊂引入满足如下伯努利分布的随机序列:P r o b {γk =1}=E γk =γ-,P r o b {γk =0}=1-E γk =1-γ-,(23)那么我们可以得到以下随机D o S 攻击下的领导-跟随一致性准则㊂推论1 令假设1和假设2以及条件(6)成立,若存在常数χ使得下列条件成立:χ=λm a x (P )λm i n (P )‖χ-‖2<1,(24)τ<-l n χα,(25)其中,χ-=I N -c γ-(L +D )⊗I N ㊂那么,在随机D o S 攻击下,误差脉冲系统是指数稳定的㊂证明:此证明类似于定理1的证明,故略去㊂3 结论本文研究了非线性领导-跟随多智能体系统的一致性问题,设计了脉冲强度随机的分布式控制协议,利用概率分析的方法,给出了达成均方指数一致的充分条件㊂同时本文还考虑了脉冲控制受到D o S 攻击的情况,给出了D o S 攻击下领导-跟多智能体系统的一致性准则㊂参考文献:[1] Z HO U B ,X U C ,D U A N G.D i s t r i b u t e da n dt r u n c a t e dr e d u c e d -o r d e ro b s e r v e rb a s e do u t p u t f e e d b a c kc o n s e n s u so f m u l t i -a g e n t s y s t e m s [J ].I E E ET r a n s a c t i o n s o nA u t o m a t i cC o n t r o l ,2014,59(8):2264-2270.[2] L U Y ,Z HA N G G ,S U NZ ,e t a l .A d a p t i v e c o o pe r a t i v ef o r m a t i o n c o n t r o l o f a u t o n o m o u s s u r f a c ev e s s e l sw i t hu n c e r t a i n d y n a m i c s a n de x t e r n a l d i s t u r b a n c e s [J ].O c e a nE ng i n e e r i n g ,2018,167:36-44.[3] T A N G Y ,Z HA N GD ,S H IP ,e t a l .E v e n t -b a s e d f o r m a t i o nc o n t r o l f o r n o n l i n e a rm u l t i a g e n t s y s t e m s u nde rD o S a t t a c k s ㊃57㊃第6期韩曼利,等:随机脉冲控制下领导-跟随的多智能体系统一致性研究㊃67㊃安 徽 工 程 大 学 学 报第38卷[J].I E E ET r a n s a c t i o n s o nA u t o m a t i cC o n t r o l,2021,66(1):452-459.[4] X I A OF,WA N GL.A s y n c h r o n o u s c o n s e n s u s i n c o n t i n u o u s-t i m em u l t i-a g e n t s y s t e m s w i t h s w i t c h i n g t o p o l o g y a n d t i m e-v a r y i n g d e l a y s[J].I E E ET r a n s a c t i o n s o nA u t o m a t i cC o n t r o l,2008,53(8):1804-1816.[5] Z HA N G W,T A N GY,HU A N GT,e t a l.S a m p l e d-d a t a c o n s e n s u s o f l i n e a rm u l t i-a g e n t s y s t e m sw i t h p a c k e t l o s s e s[J].I E E ET r a n s a c t i o n s o nN e u r a lN e t w o r k s a n dL e a r n i n g S y s t e m s,2017,28(11):2516-2527.[6] N G U Y E N D H.M i n i m u m-r a n k d y n a m i c o u t p u t c o n s e n s u s d e s i g n f o r h e t e r o g e n e o u s n o n l i n e a rm u l t i-a g e n t s y s t e m s[J].I E E ET r a n s a c t i o n s o nC o n t r o l o fN e t w o r kS y s t e m s,2018,5(1):105-115.[7] H E W,C H E N G,HA N QL,e t a l.N e t w o r k-b a s e d l e a d e r-f o l l o w i n g c o n s e n s u s o f n o n l i n e a rm u l t i-a g e n t s y s t e m s v i a d i s-t r i b u t e d i m p u l s i v e c o n t r o l[J].I n f o r m a t i o nS c i e n c e s,2017,380(20):145-158.[8] WA N GY,H E W.I m p u l s i v e c o n s e n s u s o f l e a d e r-f o l l o w i n g n o n l i n e a rm u l t i-a g e n t s y s t e m s u n d e rD o S a t t a c k s[C]//I E C-O N2019-45t h A n n u a lC o n f e r e n c eo ft h eI E E EI n d u s t r i a lE l e c t r o n i c sS o c i e t y.L i s b o n,P o r t u g a l:I E C O N,2019: 6274-6279.[9] L I Z,HU AC,L IK,e t a l.E v e n t-t r i g g e r e d c o n t r o l f o r h i g h-o r d e r u n c e r t a i nn o n l i n e a rm u l t i a g e n t s y s t e m s s u b j e c t t od e-n i a l-o f-s e r v i c e a t t a c k s[J].I E E ET r a n s a c t i o n s o nS y s t e m s,M a n,a n dC y b e r n e t i c s:S y s t e m s,2022,52(10):6129-6138.[10]T A N G Y,G A O H,Z HA N G W,e t a l.L e a d e r-f o l l o w i n g c o n s e n s u so f a c l a s so f s t o c h a s t i cd e l a y e dm u l t i-a g e n t s y s t e m sw i t h p a r t i a lm i x e d i m p u l s e s[J].A u t o m a t i c a,2015,53:346-354.[11]Z HA N G W,T A N GY,WUX,e t a l.S y n c h r o n i z a t i o n o f n o n l i n e a r d y n a m i c a l n e t w o r k sw i t hh e t e r o g e n e o u s i m p u l s e s[J].I E E ET r a n s a c t i o n s o nC i r c u i t s a n dS y s t e m s I:R e g u l a rP a p e r s,2014,61(4):1220-1228.[12]G U A NZ,L I UZ,F E N G G,e t a l.I m p u l s i v e c o n s e n s u s a l g o r i t h m s f o r s e c o n d-o r d e rm u l t i-a g e n t n e t w o r k sw i t h s a m p l e di n f o r m a t i o n[J].A u t o m a t i c a,2012,48(7):1397-1404.[13]T E E L A R,S U B B A R AMA N A,S F E R L A Z Z A A.S t a b i l i t y a n a l y s i s f o r s t o c h a s t i ch y b r i ds y s t e m s:as u r v e y[J].A u t o-m a t i c a,2014,50:2435-2456.[14]T A N G Y,WU X,S H IP,e t a l.I n p u t-t o-s t a t e s t a b i l i t y f o rn o n l i n e a r s y s t e m sw i t hs t o c h a s t i c i m p u l s e s[J].A u t o m a t i c a,2020,113:108766.[15]A N T U N E S D,H E S P A N HA J,S I L V E S T R E C.S t a b i l i t y o fn e t w o r k e dc o n t r o l s y s t e m s w i t ha s y n c h r o n o u sr e n e w a ll i n k s:a n i m p u l s i v e s y s t e m s a p p r o a c h[J].A u t o m a t i c a,2013,49:402-413.M e a nS q u a r eC o n s e n s u s f o rL e a d e r-f o l l o w i n g M u l t i-a g e n t S y s t e m sV i a S t o c h a s t i c I m p u l s i v eC o n t r o lH A N M a n l i,WU X i a o t a i*(S c h o o l o fM a t h e m a t i c s,P h y s i c s a n dF i n a n c e,A n h u i P o l y t e c h n i cU n i v e r s i t y,W u h u241000,C h i n a)A b s t r a c t:T h i s p a p e r i n v e s t i g a t e s t h e p r o b l e mo f c o n s e n s u s f o r a c l a s s o f n o n l i n e a r l e a d e r-f o l l o w i n g m u l t i-a g e n t s y s t e m s v i a s t o c h a s t i c i m p u l s i v e c o n t r o l.F i r s t,ad i s t r i b u t e d i m p u l s i v ec o n t r o l p r o t o c o lw i t hr a n-d o mi m p u l s i v e s t r e n g t h i s d e s i g n e db y i n t r o d u c i n g a s e t o f r a n d o mv a r i a b l e s.T h e n,s u f f i c i e n t c o n d i t i o n s f o r t h e l e a d e r-f o l l o w i n g m u l t i-a g e n t s y s t e mt or e a c hc o n s e n s u sv i as t o c h a s t i c i m p u l s i v ec o n t r o l a r eo b-t a i n e db y u s i n g a p r o b a b i l i s t i c a n a l y s i s a p p r o a c h.F i n a l l y,t h e c o n s e n s u s c r i t e r i o n i s p r o v i d e da s a s p e c i a l c a s e f o r t h e l e a d e r-f o l l o w i n g m u l t i-a g e n t s y s t e m s u n d e r d e n i a l-o f-s e r v i c e a t t a c k s.K e y w o r d s:l e a d e r-f o l l o w i n g m u l t i-a g e n t s y s t e m s;c o n s e n s u s;s t o c h a s t i c i m p u l s e s。

ODE–热方程级联系统的事件触发控制

ODE–热方程级联系统的事件触发控制

第40卷第8期2023年8月控制理论与应用Control Theory&ApplicationsV ol.40No.8Aug.2023 ODE–热方程级联系统的事件触发控制杨辉1,宗西举1,2†,郑泽阳1,徐秀妮3(1.济南大学自动化与电气工程学院,山东济南250022;2.济南大学信息科学与工程学院,山东济南250022;3.陇东学院电气工程学院,甘肃庆阳745000)摘要:本文针对常微分方程(ODE)耦合偏微分方程(PDE)建模的分布式参数多智能体系统进行研究,针对一致性同步问题,提出了事件触发的网络化ODE–热方程级联系统多智能体一致性边界交互协议.本文考虑的热方程左边界为Neumann边界条件,并且与ODE系统耦合,右边界为绝热边界条件.假设网络化多智能体系统的连接方式为全联通有向拓扑图,给出ODE–热方程级联系统的多智能体的一致性控制协议.另外针对现有数字式控制器,设计了事件触发的一致性控制协议,并利用李雅普诺夫函数验证了在事件触发条件下ODE–热方程级联系统的稳定性.最后给出了由5个ODE–热方程级联的多智能体系统的仿真结果,验证了事件触发控制器的有效性.关键词:ODE–热方程级联系统;有向拓扑图;多智能体系统;李雅普诺夫函数;事件触发控制器引用格式:杨辉,宗西举,郑泽阳,等.ODE–热方程级联系统的事件触发控制.控制理论与应用,2023,40(8): 1349–1356DOI:10.7641/CTA.2022.20285Event-triggered control of ode-heat equation cascade systemYANG Hui1,ZONG Xi-ju1,2†,ZHENG Ze-yang1,XU Xiu-ni3(1.School of Electrical Engineering,University of Jinan,Jinan Shandong250022,China;2.School of Information Science and Engineering,University of Jinan,Jinan Shandong250022,China;3.School of Electrical Engineering,Longdong University,Qingyang Gansu745000,China)Abstract:In this paper,a distributed parameter multi-agent system modeled by the ordinary differential equation(ODE) coupled with the partial differential equation(PDE)is studied,and an event-triggered networked ODE-heat equation cas-cade system with multi-agent consensus boundary interaction protocol are proposed.The left boundary of the heat equation considered in this paper is the Neumann boundary condition,which is coupled with the ODE system,and the right bound-ary is the adiabatic boundary condition.It is assumed that the connection mode of the networked multi-agent system is fully connected and directed the topology diagram,the multi-agent consensus control protocol of the ODE-heat equation cascade system is given.In addition,for the existing digital controller,event-triggered-based consensus control protocol is designed,and the Lyapunov function is used.The stability of the ODE-heat equation cascade system is verified under the event-triggered condition.Finally,the simulation results of the multi-agent system composed offive ODE-heat equation cascades are given,which verifies the effectiveness of the event-triggered controller.Key words:ODE-heat equation cascade system;directed topology;multi-agent system;Lyapunov function;event-triggered controllerCitation:YANG Hui,ZONG Xiju,ZHENG Zeyang,et al.Event-triggered control of ode-heat equation cascade system. Control Theory&Applications,2023,40(8):1349–13561引言随着传感器技术、人工智能技术和分布式网络的迅速崛起,多智能体系统(multi-agent systems,MAS)成为控制科学和人工智能等领域的研究热点,引起了广大学者的关注[1].智能体之间的信息交互通过网络拓扑结构进行,相对于单个个体(智能体)而言,多智能体系统具有更强的控制能力和协调能力,从而能够解决单个个体无法实现的功能,因此被广泛应用于电网经济调度优化、无人机编队联合侦查与搜索、卫星集群通信、智能交通与物流等领域.多智能体系统的协调控制研究包括多个方面,如编队控制、同步稳定、蜂拥控制等,而其中最基本的问题是多智能体系统的一收稿日期:2022−04−17;录用日期:2022−11−18.†通信作者.E-mail:cse**************.cn;Tel.:+86531-89736515.本文责任编委:郭宝珠.山东省高等学校青年创新科技计划项目(2019KJN029),国家自然科学基金项目(12026215)资助.Supported by the Youth Innovation and Technology Program of Shandong Province(2019KJN029)and the National Natural Science Foundation of China(12026215).1350控制理论与应用第40卷致性问题[2–6].一致性问题的研究成为各类企业提高生产率的关键,蓝宝石生产工艺便是其中之一.工业蓝宝石是氧化铝的晶体,是国民经济、国防工业和科学技术发展必不可少的基础材料和重要的战略物资[7].在蓝宝石结晶初期,融融状态的蓝宝石引晶点处温度检测和控制对后续整个结晶过程起到至关重要的作用,由于工业蓝宝石的熔点温度为2050◦C,所以目前不存在有效的自动温度检测手段与检测设备来获得如此高温,工程中现行的检测方法是利用一根细长的、外部包裹了透明合金材料的晶体深入加热炉内,并置细晶体棒的一端于熔融状态的蓝宝石中心上方2∼3cm,通过人眼不定期地观察晶体棒端点来近似估计熔融状态蓝宝石中心的温度.很多蓝宝石企业,为了提高成产效率,引入了多智能体系统.在工业生产中,将多台加热炉联接成网络,通过控制其中部分加热,利用加热炉之间的信息传递,可以实现自动化与网络化同步控制,既能节约成本又能提高生产效率,这就是本文要研究的网络化系统的状态一致性控制问题.目前,一方面由于实际工程中许多系统模型均不能质点化,并且一、二阶常微分方程(ordinary differen-tial equation,ODE)建模的多智能体系统相对完善,但基于偏微分方程(partial different equation,PDE)建模的多智能体系统的研究没有得到同样程度的重视,所以在这方面存在很大的挖掘空间[8–10].一般来说,利用ODE或PDE对一些系统建模都不能完美的接近实际的模型,但ODE与PDE相互耦合的级联系统却能完美的逼近实际系统[11–13].其中,由于大部分实际工程的需要,如水利、钻井、电加热炉等常见工业生产模型,其分布参数控制系统往往需要在边界耦合集中参数控制系统.并且大部分都是通过对边界进行控制.正如文献[14]研究了麦克斯韦方程与热方程的耦合;文献[15]用一类ODE与KDV(korteweg-de vries equa-tion)方程边界级联的形式,来简化存在机械振动的潜水波模型;文献[16]利用反映–扩散–对流方程对冶金固定床内部颗粒发生化学反应的动力学行为进行建模.所以本文选择ODE–热方程级联的方式对蓝宝石加热炉进行建模.另一方面,由于实际传感器测量速度的限制,网络带宽限制以及系统处理速度等各类因素的限制,显然连续时间控制器很难满足控制要求,而离散信号是很容易获取的,并且连续时间控制器的离散和采样已成为有限维多智能体系统的研究热点[5,9].虽然出现了许多优秀的控制方案,但触发频率较高,仍然可能会造成计算资源和能量的不必要损耗.为了解决上述问题,受到文献[10,17–18]等文章的启发,针对ODE–热方程建模的加热炉多智能体系统,为降低控制器的要求,本文提出了事件触发控制方案,并利用Lyapunov 函数验证在事件触发条件下该系统的稳定性.本文所考虑的热方程左边界为Neumann边界条件,与ODE系统耦合,右边界为绝热边界条件,采用这种方式对系统进行建模,更加符合实际工程,并且所有智能体通过全联通有向拓扑进行通信.设计了ODE–热方程级联系统的同步控制器与事件触发的同步控制器,与之比较,事件触发控制值只在触发时刻进行更新,很大程度上,节约了计算资源,降低了控制器的负载.通过Lyapunov稳定性分析,证明了误差系统可以收敛到一致的值,选取5个智能体进行仿真验证,证明了事件触发控制器的有效性.2问题描述2.1代数图论为了便于研究多智能体系统,研究学者引入了图论知识G=(V,E,A),用于分析所有智能体之间的信息交换,其中节点集V={1,2,···,N},代表系统中的每个智能体;边集E=V×V,表示智能体之间的通信连接.N i={j∈V:(i,j)∈V},代表智能体i 的邻点集.邻接矩阵A=[a ij]∈R n×n表示节点之间的连接关系,其中a ij=1表示节点j能够获取节点i 的信息,否则a ij=0.度矩阵D用来描述与之连接的所有节点的连接度.定义节点V i的入度为deginV i= n∑j=1a ji,出度为degoutV i=n∑j=1a ij,那么在图G的度矩阵D=[D ij]中,D ij=0,∀i=j且D ii=deg out V i[8,18–19].2.2模型描述在竞争激烈的今天,每个细节都能决定企业的成功与否.其中制药工程、晶体结晶、发酵工艺等都对温度有严格的要求.正如蓝宝石工艺,在加热的过程中,会形成固体与液体交接的情况.蓝宝石的温度分布应该保证固体与液体交接点的变化满足工程实际需要,即固体与液体交接点的变化速度不能过快或者过慢,否则会导致蓝宝石结晶过程中晶体因受热不均而发生炸裂,或者与加热炉粘连,从而影响晶体质量.然而蓝宝石加热炉是通过给外部的钨棒网通电进行加热的,所以最终目的是通过改变钨棒网的电压来实现温度控制,从而使固液交接点处的温度分布均匀,其加热炉如图1所示.图1中:1表示CCD摄像机;2表示目镜;3表示旋转挡板驱动;4表示旋转挡板把手;5表示挡板;6表示隔热层;7表示孔洞;8表示提拉杆;9表示坩埚壁;10表示籽晶;11表示钨加热器;12表示炉壁;13表示熔融状蓝宝石.利用各向同性性质,归一化处理后,可以将熔融状态的蓝宝石温度分布用抽象的一维反应扩散方程近第8期杨辉等:ODE–热方程级联系统的事件触发控制1351似描述,因此系统模型可表示为˙z i(t)=u i(t),i=1,2,···,N,ωit(x,t)=ωixx(x,t),ωix(0,t)=0,ωix(1,t)=cz i(t),z i(0)=z i,ωi(x,0)=ωi(x),(1)其中:N为智能体的个数,i表示第i个智能体,z(t)表示电源控制器内部的状态变量和加热炉内钨棒网的温度,ω(x,t)为蓝宝石溶液在x位置t时刻的温度分布,其状态空间为H:=L2(0,1),x表示空间坐标.通过对电源z(t)的控制改变蓝宝石温度分布ω(x,t)以保证固体与液体交接点x的变化满足工程实际需要.钨棒网与溶液边界直接耦合,由于位置x的对称性,交接点两侧的温度近似相同,所以交接点不存在热交换,因此可以将另一侧边界假设为绝热条件.图1蓝宝石加热炉Fig.1Sapphire heating furnace根据图论知识,针对具有N个智能体的多智能体系统,第i个智能体的状态为ωi(x,t),若系统中每个智能体的状态都达到一致,需要满足limt→∞∥ωi(x,t)−ωj(x,t)∥H=0,i=j.本文目的是通过控制电源,最终使所有状态能够自发的达成一致,也就是说,液体热量最终能够在各个位置相同.接下来将设计通信交互协议,并证明该通信协议能够使系统稳定,并给出了事件触发控制方案,从而能够节约成本.假设1假设系统(1)的连接方式由全联通有向拓扑图G=(V,E)来描述.图G的拉普拉斯矩阵L可以表示为L=D−A,或常用L(A)=L(G)来表示图G的拉普拉斯矩阵.定义1设X是线性赋范空间,X上的单参数强连续有界线性算子族T(t)称为算子半群,简称C0半群[20].对任何t>0,T(t)都是线性有界算子,且满足T(0)=I,T(t+s)=T(t)T(s),limt→0∥T(t)x−x∥=0,∀x∈X.定义2在线性赋范空间X上满足∥T(t)∥ 1,∀t 0的C0半群称为压缩C0半群.定义3线性赋范空间X上一线性算子A称为耗散的,如果对任意x∈D(A),∃x∗∈D′(A),使Re⟨A x,x∗⟩ 0,若A还满足R(λ−A)=X,∀λ>0,则称A为耗散的[21].定理1设A是线性赋范空间X中的一稠定算子,则以下条件等价:若A是耗散的,且∃λ0>0使得R(λ0−A)=X,则A生成X上的压缩C0半群.若A生成X上的压缩C0半群,则A是耗散的,且R(λ−A)=X,∀λ>0,此外Re⟨A x,x∗⟩ 0,∀x∈D(A),x∗∈D′(A). 2.3一致性控制协议设计由于级联系统存在耦合的边界条件,传统的ODE 或PDE的多智能体一致性控制已不能满足需求,若继续使用传统的一阶多智能体控制器的设计形式,稳定性证明时则会产生不稳定的耦合项,因此本文根据级联系统特性重新设计一致性同步协议并予以证明.对于型如系统(1)的多智能体ODE–热方程级联系统,设计控制器u(t)=−c Lω(1,t)−kIz(t),(2)其中:c>0,k>0均是常数,L是拉普拉斯矩阵,I表示单位矩阵,ω(x,t)和z(t)均为N维向量,将控制器(2)代入系统(1),可以得到级联系统如下:˙z(t)=−c Lω(1,t)−kIz(t),ωt(x,t)=ωxx(x,t),ωx(0,t)=0,ωx(1,t)=cz(t),z(0)=z0,ω(x,0)=ω0(x).(3)引理1(文献[8]引理3.1)对于任意的z(t)∈L2loc(0,∞),ω0(x)∈H2(0,1),系统(3)存在唯一解ω(x,t),使ω(x,t)∈C1(0,∞;L2(0,1))∩C(0,∞; H2(0,1)).此外,通过H2(0,1)⊂C[0,1],那么ω(1,t)对于时间t>0是连续的、有界的.考虑状态空间H=(L2(0,1))N×R N.(4)1352控制理论与应用第40卷定义内积如下:⟨ϕ1,ϕ2⟩H =1ωT1(x,t )L ω2(x,t )d x +12z T1(t )z 2(t ),(5)其中∀ϕi =(ωTi (x,t ),z i (t )),i =1,2.定义算子A :D (A )→H :A (ω(x,t ),z (t ))=(ωxx (x,t ),−c L ω(1,t )−kz (t )),D (A )={(ω,z )∈(H 2(0,1))N ×R N |ωx (0)=0,ωx (1)=cz }.(6)将方程(3)写为发展方程形式如下:˙ϕ(t )=A ϕ(t ).(7)定理2算子A 在空间H 上生成压缩C 0半群.在式(6)中,计算可以得出Re ⟨A ϕ,ϕ⟩=−kz T(t )z (t )−1ωTx (x,t )ωx (x,t )d x 0.(8)因此得出算子A 在空间H 上是能量耗散的,由于初值条件给定,其具有唯一解,因此可以得出A −1存在.根据Sobolev 嵌入定理得出A −1在H 中是紧的.利用定理1(Lumer-Phillips)可以得出,算子A 在空间H 上生成压缩C 0半群.本文的目标是使系统同步,分析可知同步的一个充分条件是使边界无能量流动,同时使得所有状态趋近于0.又由于边界绝热条件的充分条件是使ODE 部分收敛到0,因此初步分析ODE 部分特征值需要为负,而热方程部分系统的特征值所在空间仅有一个零特征值,其余为负.根据以上思路设计的控制器(2).接下来证明在控制器(2)作用下系统的稳定性,定义Lyapunov 函数如下:V =12 10ωT (x,t )L ω(x,t )d x +12z T (t )z (t ).(9)对Lyapunov 函数求导得˙V = 10ωT (x,t )L ωt (x,t )d x +z T (t )z t (t )=10ωT (x,t )L ωxx (x,t )d x +z T(t )(−c L ω(1,t )−kz (t ))=ωT (x,t )L ωx (1,t )− 1ωT x (x,t )L ωx (x,t )d x +z T(t )(−c L ω(1,t )−kz (t ))=cωT (1,t )L z (t )− 1ωT x (x,t )L ωx (x,t )d x −cω(1,t )L z T (t )−kz T(t )z (t )=− 1ωTx (x,t )L ωx (x,t )d x −kz T (t )z (t ) 0.(10)通过LaSalle 不变集原理,可以找出有且仅有一个稳定解收敛,即系统状态最终可以达成同步∥(ω(·,t )),z (t ))∥H →0,t →∞.(11)以上是连续系统ODE–热方程级联系统的控制器及闭环系统的证明.下文将设计事件触发控制器的控制规则,并证明该系统依然保持收敛性能.3事件触发控制器设计上文提到的控制器是时间连续形式的,但是连续控制需要在时间尺度上,不间断的采集并获取信息,同时需要实时不间断的对信息进行处理.由于目前广泛应用的数字式执行器执行测量、运算或输出等操作均有最小时钟周期限制,因此数字式执行器无法做到无限小时刻采样并对信息进行处理.为解决数字式控制难以处理连续性控制的缺点,所以本文提出事件触发的ODE–热方程级联系统一致性控制协议,给出具体的触发条件,在达成触发条件时改变控制输入的值,而其余时刻则保持之前状态不变.这样不仅仅能够解决连续性控制的缺点,同时还可以节约控制器的计算资源.下面介绍事件触发控制器的设计,以及触发条件的确认.根据第2.3节得出的一致性同步协议可以假设ODE–热方程系统事件触发控制器如下:{u (t )=−c L ω(1,t )−kz (t ),t =t k ,u (t )=u (t k ),(12)其中:u (t )为ODE–热方程级联系统事件触发控制器,t ∈[t k ,t k +1),∀k =0,1,2,···且t 0为初始时刻,触发间隔时间点t k 满足t k =arg min t>t k −1{∥e (1,t )∥=σ1∥z (t )∥|c |∥L∥,∥ϵ(t )∥=σ2∥z (t )∥k},(13)其中:argmin 表示∥e (1,t )∥和∥ϵ(t )∥达到最小值时t 的取值,∀σ1,σ2∈R +满足条件∥e (1,t )∥ σ1∥z (t )∥|c |∥L∥,∥ε(t )∥ σ2∥z (t )∥k ,k −σ1−σ2<0.(14)为便于证明在事件触发控制器作用下系统仍保持稳定,首先引入误差系统{e (x,t )=ω(x,t k )−ω(x,t ),ϵ(t )=z (t i )−z (t ),(15)将误差状态(15)代入式(12)中,可以得到控制器在触发时刻的误差表示形式u (t )=−c L e (1,t )−c L ω(1,t )−kϵ(t )−kz (t ).(16)第8期杨辉等:ODE–热方程级联系统的事件触发控制1353为了证明控制器(16)能够使系统(1)的每个子系统保持稳定,并能够使各个子系统最终状态达成一致,利用Lyapunov 稳定性判据对事件触发系统进行稳定性判别.设计Lyapunov 函数如下:V (t )=12(z T(t )z (t )+ 10ωT (x,t )L ω(x,t )d x ).(17)对上式左右两边关于时间变量求导,可得˙V(t )=d d t 12(z T(t )z (t )+ 10ωT (x,t )L ω(x,t )d x )=z T (t )u (t )+ 10ωTt (x,t )L ω(x,t )d x =z T (t )u (t )+ 10ωT xx (x,t )L ω(x,t )d x =z T (t )u (t )+ωT x (x,t )L ω(x,t )|10−10ωT x (x,t )L ωx (x,t )d x =z T(t )u (t )+ωTx (1,t )L ω(1,t )− 1ωT x (x,t )L ωx (x,t )d x.(18)结合耦合边界条件cz (t )=ωx (1,t ),可以得出˙V(t )=−cz T (t )L [e (1,t )+ω(1,t )]−kz T (t )[ϵ(t )+z (t )]+ωT x (1,t )L ω(1,t )−1ωTx (x,t )L ωx (x,t )d x =−ωTx (1,t )L e (1,t )−kz T (t )ϵ(t )−kz t(t )z (t )− 1ω1x ωT x (x,t )L ωx (x,t )d x∥ωx (1,t )∥∥L∥∥e (1,t )∥+k ∥z (t )∥∥ϵ(t )∥−k ∥z (t )∥2−1ωT x (x,t )L ωx (x,t )d x.(19)设∃σ1,σ2>0,满足以下不等式组条件:∥e (1,t )∥ σ1∥z (t )∥|c |∥L∥,∥ε(t )∥ σ2∥z (t )∥k ,k −σ1−σ2<0.(20)事实上,对于∀σ1,σ2∈R +,总是至少存在着一组解满足不等式组(20),这是很容易得出的,因为前两个不等式组在出发时刻∥e (1,t )∥和∥ϵ(t )∥都等于0.所以σ1,σ2>0即可满足要求.并且由于设定参数k >0,所以必然存在σ1+σ2<k 满足第3个不等式.将假设的不等式组代入式(19),可以得出˙V(t ) (k −σ1−σ2)∥z (t )∥∥ϵ(t )∥−k ∥z (t )∥2− 1ωT x (x,t )L ωx (x,t )d x.(21)根据Lyapunov 稳定性判据可以得出,该系统是Ly-apunov 意义下稳定的,同时通过计算其稳定点所构成的不变集,容易得出最终有且仅有一个共同收敛的点,因此该ODE–热方程级联系统事件触发控制器最终能够使得系统(1)中的状态达成一致.根据不等式组(20)的条件,对于k =1,2,···,以及σ1,σ2满足式(20)条件,设触发时刻为t k =arg min t>t k −1{∥e (1,t )∥=σ1∥z (t )∥|c |∥L∥,∥ϵ(t )∥=σ2∥z (t )∥k}.(22)通过上述证明,可以得出控制器u (t )在式(12)的作用下能够使系统(1)最终状态达成一致.由于PDE 部分边界条件经由ODE 中的控制器引入了负实部极点,因此系统具有一定的鲁棒性,即存在一定的扰动的情况下,系统仍然可以按照预期的目标达成一致,下面针对上述事件触发一致性协议设计计算机仿真程序来进行验证.4数值仿真验证为验证ODE–热方程级联系统事件触发一致性控制协议的实际有效性,本文采用数字计算机平台进行仿真模拟验证.这里仿真离散格式中,对于空间采用中心差分法进行离散,时间采用前向差分方法进行离散化,ODE 部分采用前向差分离散化.选取仿真步长d t =0.001s,空间步长d x =0.05m,仿真总时长T =15s,PDE 选取长度为1m.设系统参数c =5,k =10/3.假设系统仿真的初值条件如下:ω(x,0)= 4+2sin(2πx +1)+0.7(x −0.5)−2+1.4sin(3πx +2)−0.4(x −0.7)9−6sin(1.5πx +3)−0.9(x −0.3)9−1.4sin(2πx +2)+0.9(x −0.4)4−2sin(1.5πx +1)+0.7(x −0.2),z (0)=[412−1−4]T ,考虑多智能体个数为5,且为全连通有向拓扑,假设权重均为1,其连接方式如图2.通过上述假设的参数进行仿真,能够得出仿真所用控制系统如下:˙z (t )=−5 1−100001−100001−100001−1−10001 ω(1,t )−103z (t ),ωt (x,t )=ωxx (x,t ),ωx (0,t )=0,ωx (1,t )=5z (t ).(23)1354控制理论与应用第40卷因此根据本文所给出的计算方法,设计控制器u(t)=˙z(t)=−51−100001−100001−100001−1−10001ω(1,t)−103z(t),u(t)=u(t k),t∈(t k,t k+1),t k=arg mint>t k−1{∥e(1,t)∥=σ1∥z(t)∥|c|∥L∥,∥ϵ(t)∥=σ2∥z(t)∥k}.(24)其中选取σ1=0.43,σ2=11.1.仿真结果如图3–10所示.图2多智能体连接拓扑图Fig.2Multi-agent connectiontopology图3多智能体系统ODE部分状态图Fig.3Partial ODE state diagram of multi-agentsystem图4多智能体系统PDE部分状态1Fig.4Multi-agent system PDE partial State1图5多智能体系统PDE部分状态2Fig.5Multi-agent system PDE partial State2图6多智能体系统PDE部分状态3Fig.6Multi-agent system PDE partial State3图7多智能体系统PDE部分状态4Fig.7Multi-agent system PDE partial State4图3描述了多智能体系统(24)中的ODE部分随时间趋向于0,其物理含义是为了最终使PDE系统能够稳定在一致温度下,热系统边界随着时间增加,其热流动逐渐停止.图4–8五个多智能体系统的PDE部分在边界耦合ODE系统的控制下,最终能够收敛到一致状态.图9–10表示了多智能体共同决策得出的触发条件,其中触发时刻用“*”着重标记.根据仿真结果可以看出,通过利用本文提出的基于事件触发的ODE–热方程级联系统多智能体一致性控制协议,能够使系统在大约12s左右通过较少的触发次数达到同步,表第8期杨辉等:ODE–热方程级联系统的事件触发控制1355明事件触发控制器具有优越的性能,能在很短的时间内让蓝宝石加热炉达到一致的工作状态.图8多智能体系统PDE部分状态5Fig.8Multi-agent system PDE partial State5图9多智能体系统事件触发条件1Fig.9Multi-agent system event trigger Condition1图10多智能体系统事件触发条件2Fig.10Multi-agent system event trigger Condition25结论与展望本文研究了基于事件触发的ODE–热方程级联系统多智能体一致性控制协议的设计,通过控制与PDE部分边界耦合的ODE系统的输入,使PDE部分的热系统最终能够稳定,快速的达成一致,进一步给出了事件触发控制的条件,通过仿真验证可以看出该同步协议具有一定的鲁棒性.相较于之前已有的成果来说,本文的创新在于采用不同的(Neumann)边界条件,控制输入和被控对象分别处于不同的系统,同步采用边界反馈,并证明了事件触发需要满足的条件及其稳定性,但因本文系统存在耦合项,Zeno现象的避免未能给出严格的证明,这将在后续的工作中进行.参考文献:[1]TIAN Yuan.Consensus control for multi-agent systems with impul-sive effects.Chongqing:Southwest University,2020.(田袁.脉冲作用下的多智能体系统一致性控制.重庆:西南大学, 2020.)[2]JI Lianghao,LIAO Xiaofeng.Consensus analysis of multi-agent sys-tem with multiple time delays.Journal of Physics,2012,61(15):8–16.(纪良浩,廖晓峰.具有不同延时的多智能体系统一致性分析.物理学报,2012,61(15):8–16.)[3]FU Q,YU P F,XU G Z,et al.Containment control for partial differ-ential multi-agent systems.Physica A:Statistical Mechanics and its Applications,2019,529:121549.[4]ZHOU Tuo,LIU Quanli,WANG Dong,et al.Leader-following con-sensus for linear multi-agent systems based on integral-type event-triggered strategy.Control and Decision,2022,37(5):1258–1266.(周托,刘全利,王东,等.积分事件触发策略下的线性多智能体系统领导跟随一致性.控制与决策,2022,37(5):1258–1266.)[5]OLFATI-SABER R,MURRAY R.Consensus problems in networksof agents with switching topology and time-delays.IEEE Transac-tions on Automatic Control,2004,49(9):15201533.[6]XIE Jun,CHEN Kaixuan,YUE Dong.Distributed economic dispatchbased on consensus algorithm of multi agent system for power sys-tem.Electric Power Automation Equipment,2016,36(2):112–117.(谢俊,陈凯旋,岳东.基于多智能体系统一致性算法的电力系统分布式经济调度策略.电力自动化设备,2016,36(2):112–117.)[7]LI L C,FENG J S.Current situation and development trend of sap-phire crystal growth technology in China.Journal of Synthetic Crys-tals,2012,41(S1):221–226.[8]MENG X Y,MENG Z Y,CHEN T W,et al.Pulse width modulationfor multi-agent systems.Automatica,2013,49(7):2125–2132.[9]XI J X,WANG C,YANG X J,et al.Limited-budget output consen-sus for descriptor multiagent systems with energy constraints.IEEE Transactions on Cybernetics,2020,50(11):4585–4598.[10]OLFATI-SABER R,ALEX F,MURRAY R M.Consensus and coop-eration in networked multi-agent systems.Proceedings of the IEEE, 2007,95(1):215–233.[11]CHRISTOFIDES P D.Nonlinear and robust control of PDE systems:Methods and applications to transport-reaction processes.Applied Mechanics Reviews,2002,55(2):29–30.[12]WANG J,KRSTIC M.Delay-compensated control of sandwichedode-pde-ode hyperbolic systems for oil drilling and disaster relief.Automatica,2020,120:109131.[13]JIN F F,GUO B Z.Performance boundary output tracking for one-dimensional heat equation with boundary unmatched disturbance.Automatica,2018,96:1–10.[14]ALAMT T M,PAQUET L.The coupled heat maxwell equations withtemperature-dependent permittivity.Journal of Mathematical Analy-sis and Applications,2022,505(1):125472.1356控制理论与应用第40卷[15]AY ADI H.Exponential stabilization of cascade ode-linearized KDVsystem by boundary dirichlet actuation.European Journal of Control, 2018,43:33–38.[16]XU Zenghe,ZHAI Yuchun,JI Zhiling.An analytical solution for cou-pled convection-diffusion and chemical reaction infixed-beds.The Chinese Journal of Nonferrous Metals,2004,14(11):1918–1925.(徐曾和,翟玉春,纪智玲.固定床中对流扩散与反应耦合问题的一个解析解.中国有色金属学报,2004,14(11):1918–1925.)[17]DIMAROGONAS D V,FRAZZOLI E,JOHANSSON K H.Dis-tributed event-triggered control for multi-agent systems.IEEE Trans-actions on Automatic Control,2012,57(5):1291–1297.[18]LIU T F,JIANG Z P.Distributed control of multi-agent systems withpulse-width-modulated controllers.Automatica,2020,119:109020.[19]LU Renwei.Consistency of nonlinear second-order multiagent withdirected topology and delay.Yangzhou:Yangzhou University,2012.(鲁仁伟.具有有向拓扑和时滞的非线性二阶多智能体的一致性.扬州:扬州大学,2012.)[20]SI Qin,LUO Cheng.Global attractors of operator semigroups ontopological space.Acta Analysis Functionalis Applicata,2017,19(4): 412–418.(斯琴,罗成.拓扑空间上算子半群的全局吸引子.应用泛函分析学报,2017,19(4):412–418.)[21]LIU Rui,ZHAO Huaxin,MA Qiangqiang.Generalized-semigroupsand dissipative operator.Journal of Shenyang Normal University (Natural Science Edition),2011,29(3):362–364.(刘瑞,赵华新,马强强.广义C0半群与耗散算子.沈阳师范大学学报(自然科学版),2011,29(3):362–364.)作者简介:杨辉硕士研究生,目前研究方向为分布参数控制系统理论与应用,E-mail:1104660379@;宗西举博士,教授,英国奥斯特大学兼职博士生导师,目前研究方向为分布参数控制系统理论、复杂系统控制理论、控制理论在电力系统中的应用,E-mail:cse**************.cn;郑泽阳硕士研究生,目前研究方向为分布参数控制系统理论与应用,E-mail:*****************;徐秀妮讲师,目前研究方向为分布参数控制系统理论与应用, E-mail:***************.。

Convergence Rate of Distributed Consensus for Second-Order Multi-agent Systems

Convergence Rate of Distributed Consensus for Second-Order Multi-agent Systems
Zhong Yu and Lihua Xie are with the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, E-mail: {yuzh0005,elhxie}@.sg.
I. I NTRODUCTION The distributed consensus problem for networked multiagent systems has attracted a lot of attentions in recent years due to its broad applications in distributed computation, sensor fusion, formation flight, network synchronization, etc. Consensus control means to design a networked interaction protocol such that all the agents reach an agreement on their states asymptotically or in finite time. In 1995, Vicsek et al.[1] proposed a model to describe agents’ motion in a plane with the same speed but different headings and Jadbabaie provided a simple consensus protocol which updates the heading of an agent by averaging the headings of its neighbors [2]. Olfati-Saber and Murray studied the problem using graph theory and provided conditions for consensus under switching topologies and constant time delays [3]. Ren and Beard then generalized their results and proposed a more relaxed condition for the network topology to reach consensus. The condition is given in terms of the existence of a spanning tree. Other studies for different models and consensus protocols can also be found, see e.g. [4] and references therein. It is worth noting that the consensus problem has primarily been studied for first-order kinematics. However, in many applications, both position and velocity information may be of interest, for instance, in multi-robot formation control. The study of the second-order consensus problem can be traced back to 2003, when Olfati-Saber and Murray[5] investigated the flocking problem. Xie and Wang[6] also proposed a protocol and discussed the consensus convergence rate. Ren and Atkins[7] proposed a protocol similar to that of [5] and

多Agent谈判中议题相关性及权重度量研究

多Agent谈判中议题相关性及权重度量研究
教授
收稿 日期 :2 1 0—7 01 2 — 4
Ema :za@m ihseu n ・ i g l n a. td. lu . c
18 6

算 机


2 1 年 9月 2 01 0日
当前的交易来判 断用户的偏好 ,未能利用历史数据 ,而且买
合并用符号 “ 0”表示 ,其规则为 :先求集合 的并集 ,
Re e r h 0 s ue l tv t n e g t e s r m e t s a c n I s sRea i iy a d W i h a u e n M i u t- e tNe o i to n M liAg n g ta i n
GAN o b n ZHU u - i XI Za - i , Ch n x , AO o q a g Gu - i n ( c o l f o ue c n e n e h oo y Hu z o gUnv ri f ce c n e h o o y Wu a 3 0 4 C i a S h o mp t S i c d c n lg , a h n ie s yo i ea dT c n lg , h 4 0 7 , h n ) oC r e a T t S n n
和事例推理法使 用户能 更灵活地 设定权 重 , 根据单个议题权重初值 的确 定函数以及议 题组的权重求解规则 ,给出分组变化 时的权 重动态 变 更算法。分析结果表 明,该算法能解决 多议 题谈判中的议题相关性 问题和议题权重 的动态度量 问题 ,使 多议 题谈 判更加科 学合 理。
关奠词 :多 A et gn 谈判 ;议题分组 ;动态组权重 ;事例推理 ; 动平均法 移
要性 。
通 过从历史数据 中获取较多的个性 化信息来实现 ; 3 关性 (相 ) 处理与权重度量是动态的 ,贯穿于整个谈判过程之中。 基于此 ,本文对谈判 中的多个议题 间的相关性 问题 以及 议 题的权重度量做了研究。

文献——精选推荐

文献——精选推荐

⽂献徐胜元简介:徐胜元,男,南京理⼯⼤学⾃动化学院教授、博⼠、博⼠⽣导师。

毕业于南京理⼯⼤学控制理论与控制⼯程专业,获得博⼠学位。

研究⽅向:1、鲁棒控制与滤波2、⼴义系统3、⾮线性系统2017年SCI1.Relaxed conditions for stability of time-varying delay systems ☆TH Lee,HP Ju,S Xu 《Automatica》, 2017, 75:11-15EI1.Relaxed conditions for stability of time-varying delay systems ☆TH Lee,HP Ju,S Xu 《Automatica》, 2017, 75:11-152.Adaptive Tracking Control for Uncertain Switched Stochastic Nonlinear Pure-feedback Systems with Unknown Backlash-like HysteresisG Cui,S Xu,B Zhang,J Lu,Z Li,...《Journal of the Franklin Institute》, 20172016年SCI1..Finite-time output feedback control for a class of stochastic low-order nonlinear systemsL Liu,S Xu,YZhang《International Journal of Control》, 2016:1-162.Universal adaptive control of feedforward nonlinear systems with unknown input and state delaysX Jia,S Xu,Q Ma,Y Li,Y Chu《International Journal ofControl》, 2016, 89(11):1-193.Robust adaptive control of strict-feedback nonlinear systems with unmodeled dynamics and time-varying delaysX Shi,S Xu,Y Li,W Chen,Y Chu《International Journal of Control》, 2016:1-184.Stabilization of hybrid neutral stochastic differential delay equations by delay feedback controlW Chen,S Xu,YZou《Systems & Control Letters》, 2016, 88(1):1-135.Multi-agent zero-sum differential graphical games for disturbance rejection in distributed control ☆Q Jiao,H Modares,S Xu,FL Lewis,KG Vamvoudakis《Automatica》, 2016, 69(C):24-346.Semiactive Inerter and Its Application in Adaptive Tuned Vibration AbsorbersY Hu,MZQ Chen,S Xu,Y Liu《IEEE Transactions on Control Systems Technology》, 2016:1-77.Decentralised adaptive output feedback stabilisation for stochastic time-delay systems via LaSalle-Yoshizawa-type theoremT Jiao,S Xu,J Lu,Y Wei,Y Zou《International Journal of Control》, 2016, 89(1):69-838.Coverage control for heterogeneous mobile sensor networks on a circleC Song,L Liu,G Feng,S Xu《Automatica》, 2016, 63(3):349-358EI1.Finite-time output feedback control for a class of stochastic low-order nonlinear systemsL Liu,S Xu,YZhang《International Journal of Control》, 2016:1-162.Unified filters design for singular Markovian jump systems with time-varying delaysG Zhuang,S Xu,B Zhang,J Xia,Y Chu,...《Journal of the FranklinInstitute》, 2016, 353(15):3739-37683.Improvement on stability conditions for continuous-time T–S fuzzy systemsJ Chen,S Xu,Y Li,Z Qi,Y Chu《Journal of the Franklin Institute》, 2016, 353(10):2218-22364.Universal adaptive control of feedforward nonlinear systems with unknown input and state delaysX Jia,S Xu,Q Ma,Y Li,Y Chu《International Journal ofControl》, 2016, 89(11):1-195.H∞ Control with Transients for Singular Systems Z Feng,J Lam,S Xu,S Zhou 《Asian Journal of Control》, 2016,18(3):817-8272015年SCI1.Pinning control for cluster synchronisation of complex dynamical networks withsemi-Markovian jump topologyTH Lee,Q Ma,S Xu,HP Ju《International Journal of Control》, 2015, 88(6):1223-12352..Anti-disturbance control for nonlinear systems subject to input saturation via disturbance observer ☆Y Wei,WX Zheng,S Xu《Systems & ControlLetters》, 2015, 85:61-693.Exact tracking control of nonlinear systems with time delays and dead-zone inputZ Zhang,S Xu,B Zhang《Automatica》, 2015, 52(52):272-276EI1.Further studies on stability and stabilization conditions for discrete-time T–S systems with the order relation information of membership functionsJ Chen,S Xu,Y Li,Y Chu,Y Zou《Journal of the Franklin Institute》, 2015, 352(12):5796-5809 .2 .Stability analysis of random systems with Markovian switching and its application T Jiao,J Lu,Y Li,Y Chu,SXu《Journal of the Franklin Institute》, 2015, 353(1):200-220 3.Exact tracking control of nonlinear systems with time delays and dead-zone inputZ Zhang,S Xu,B Zhang《Automatica》, 2015, 52(52):272-2764.Event-triggered average consensus for multi-agent systems with nonlinear dynamics and switching topologyD Xie,S Xu,Y Chu,Y Zou《Journal of the Franklin Institute》, 2015, 352(3):1080-1098葛树志简介:葛树志,男,汉族,1963年9⽉20⽇⽣于⼭东省安丘县景芝的葛家彭旺村。

ResponseLetter投稿回复信

ResponseLetter投稿回复信

Response LetterPaper number: NODY-D-15-00088Paper title: Event-Triggered Control for Multi-Agent Network with Limited Digital CommunicationAuthors:Dear Editor-in-chief, Associate Editor and Anonymous Reviewers,We would like to thank you for your efforts in reviewing our manuscript and providing many helpful comments and suggestions. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments very carefully. Based on your criticisms, comments and suggestions, we have revised the manuscript accordingly. The details are explained below, where the number of the response is in correspondence with the number of the reviewers’ comments and su ggestions.Reply to the Associate EditorAccording to the AE’s and reviewers’ criticisms, comments and suggestions, we have modified the manuscript carefully. The description of a substantial revision and the detailed points to the review reports can be seen in the following responses and in the new revision. Moreover, we have also checked other derivations throughout the paper and some necessary explanations are also included.We would like to thank the reviewer’s great efforts in reading our manuscript and for your constructive comments and suggestions. Our responses to the comments and suggestions are listed as follows:1. Consensus with communication constraints is indeed a quite interesting topic in field of multi-agent systems, the following work on consensus of second-order multi-agent systems may be briefly mentioned: Int. J. Robust and Nonlinear Control, 22(2):170-182, 2012.Reply:The relevant works of communication constraints in Int. J. Robust and Nonlinear Control is really worth mentioning, and this reference has been added in new revision.2. The communication topology is assumed to be undirected, whether it is possible to do some further work on directed or switching topologies. One more remark may be added to the manuscript to state this issue.Reply:This suggestion is very nice and reasonable. The directed and switching topologies cases will be our future works, and the remark has been provided in the future works part of conclusion.We would like to thank the reviewer ’s great efforts in reading our manuscript and for your constructive comments and suggestions. Our responses to the comments and suggestions are listed as follows:1. The proof of Theorem 1 is not clear. It didn’t show what is the convergence setvery important obviously.Reply: This suggestion is very helpful, and I have rewritten the Theorem 1. I’m sure the new version is much clearer than the old one.2. There are some errors in the proof of Theorem 1. For example,(i) How to determine l in the last line of formula (16). There is no any constraint for l.(ii) The same problem appeared in the last line and previous line of formula (18).Reply: I am very sorry for my carelessness. The last expression ˆ(t )l ll k xin formula (16) and (18) should be replaced by ˆ(t )i ii k x. Now the total four mistakes in formula (16) and (18) have been corrected in revised version. To avoid the similar mistakes, I have also checked the other derivations throughout the manuscript. Again thanks for your carefulness and tolerance.3. What is the function of parameter i σ in the event triggering condition (8). Which performance does it affect? How to choose this parameter according to the demands of performance? The analysis should be given.Reply: This suggestion is very reasonable. Actually, this parameter’s main function is to adjust the performance of event triggering mechanism. Each agent’s event frequency has a great relationship with the parameter i σ. The larger i σ, the eventtimes are less and the performance is better. To obtain the best performance, we directly set 1σ=in revised version, i.e., we no longer define this parameter iexplicitly in revised version.Reply to Referees #3We would like to thank the reviewer ’s great efforts in reading our manuscript and for your constructive comments and suggestions. Our responses to the comments and suggestions are listed as follows:1. The main advantage of this work should be further strengthened in Introduction. Reply: Sincerely thanks for your helpful suggestion. I have rewritten the contribution part in Introduction, and I’m sure the new version is much clearer than the old one.2. What are the novelty in the proposed scheme in this paper?Reply: There are four main novelties in this paper. First, we designed an integrated communication framework for digital multi-agent network, in which the event-triggered strategy and dynamic encode/decode scheme play an important role in communication process. Second, a distributed triggering condition that only depends on local state information of neighbor agents is developed and the corresponding consensus analysis is provided. Third, we gave the specific communication algorithm considering dynamic encode/decode scheme under event-triggered strategy, and we also proposed a self-adaptive quantization algorithm that builds a connection between quantization level and quantization factor. Last, we proposed an improved communication strategy named one-bit quantized scheme such that the global consensus can still be achieved based on only one bit information exchange between agents at each quantized transmission.3. In this Reviewer's opinion, in (6), \hat{e}_i(t) is infeasible since there is both $t$ and $t_k$. The authors should explain this point.Reply: Actually, it is feasible. Here we give the detailed explanation. Just like the statements before the Algorithm 1, we assume each agent i has a memory that canstore its own instant state ()i x t , state estimate ˆ()i xt , and its all neighbor stateestimates ˆ(),j i xt j N ∈. Furthermore, the initial states of all agents are given as ()1(0)(0),,(0)TN x x x =⋯, the all initial event time 0i t and all state estimates ˆ(0),i 1,,i xN = are initialized to 0. Then the Algorithm 1 can be carried out step by step. According to the Algorithm 1 and Remark 1, we can know that the work time of encoder/decoder is only the event time of relevant agents, once the event is triggered, then the corresponding measurement state estimate is updated and rewritten to thememory . As a result, the actuator i can directly obtain ()i x t and ˆ()ii i k x t from its memory to compute the measurement error ˆ()i et .4. The authors should check some typos.Reply: I have checked the manuscript again, and found there really exists some typos. Besides, I have also made some corrections based on my friends’ suggestions. Again thanks for your carefulness and tolerance.Finally, we would like to thank the referees again for the careful reading of our paper. In addition, we have revised the manuscript carefully and believe that the new version is much better than the old one. Hope the revised version is acceptable.Best wishes,Y our name,May 28, 2015.。

基于Multi-Agent的编队对空防御方法

基于Multi-Agent的编队对空防御方法

( . i lt nT an n e tr D l nWa s i a e f L v , l n 1 6 , ia 1 S mu ai r i i gC n e , a i rh pAc d myo A Na y Dai 1 0 Chn ; o a P a 1 8 2 T c n l g p r n , ae l eMa i me rc ig& C nr l n p  ̄ n f h n , in y n 2 4 3 , h: ) . e h o o y De a t t S t l t r i a kn me i t T o t l gDe a me t C i a Ja g i 1 4 l C ia o i o n
o tm ia i n p r o e n t e e d p i z t u p s s i h n .Co o mp r o t e ta ii n l g n t l o i m t t t t e t o e ,t e M AS a e t h r d to a e e i a g rt c h wi s a i ma h ma i m d l h h c c
Ab t a t Ac o d n t f a u e o Ai- e e s wa f r t s ra e o ma i n c n t u t ir p we d srb t n sr c : c r ig o e t r s f rd f n e r a e o u f c f r to , o sr c f e o r iti u i o o tm ia i n s l to a e n M u t— e t s s e .Ai— e e s r a e i n f t e mo t i o a t f r s。 p i z to o u i n b s d o liAg n y t m r d f n e wa f r s o e o h s mp r n o m i mo e n t n d r n v l b tl s t d n fr p we it i u i n s r t g s n a e n M u t- g n y t m o s l e fr p we y a i a a a te ,su y o e o r d srb t ta e y u i g b s d o i o li A e t s s e t o v e o r d n m c i

一个基于离散事件仿真的Multi_Agent经济仿真模型_宣慧玉

一个基于离散事件仿真的Multi_Agent经济仿真模型_宣慧玉

文章编号:100220411(2002)012084205一个基于离散事件仿真的M u lti 2A gen t 经济仿真模型宣慧玉 高宝俊 李 群 冯 平(西安交通大学管理学院 710049)摘 要:本文根据复杂系统及M ulti 2A gent 仿真原理构建了一个简单的三部门经济仿真模型.在模型中采用了学习算法模拟了经济个体的智能行为,并应用离散事件仿真技术实现了代理人之间的交互与通信问题及系统的动态模拟,系统模拟了企业在亏损时退出与不退出市场这两种策略下的经济运作状况,说明了亏损企业不退出市场不仅将引起更多企业的亏损,而且会使宏观经济指标下降.α关键词:经济模型;M ulti 2A gent ;离散事件仿真中图分类号:F 224.0 文献标识码:AA M UL T I -AGENT ECONOMY SI M ULAT I ON MOD EL BASED OND ISCRETE EVENTS SI M ULAT I ONXUAN H u i 2yu GAO B ao 2jun L I Q un FEN G P ing(S chool of M anag e m en t X i ’an J iaotong U n iversity 710049)Abstract :T he paper build a model fo r a si m p le th ree 2secto r econom y ,to w h ich the comp lex system idea and the m ulti 2agent modeling m ethod are app lied .T he econom ic individuals’intelligent behavi o r and the rela 2ti ons betw een them are si m ulated .A lso the m arket response is si m ulated ,respectively to the quitting the m arket and staying on of bad 2perfo r m ance companies .T he result analysis show s that the staying on behavi o r w ill cause mo re o ther companies to fail and the m acroeconom ic index to decrease .Keywords :econom ic model ,m ulti 2agent ,discrete event si m ulati on1 引言(I n troduction )复杂系统与复杂科学正被人们广泛重视.复杂系统与简单系统相比具有一些根本性的差异,主要差异之一就是在复杂系统内部包含了一些中等数目的具有自主行为的个体,这些个体数目既不是少数的几个,又不是数目巨大但具有相同的行为规则可以用统计方法求解的巨大群体.这些个体具有智能性及自适应性,他们具有各自不同的行为规则、目标及学习(适应)能力,同时这些个体处于复杂的外部环境中,影响他们行为的因素非常多,而他们只能得到其中一部分信息,并且环境的变化的影响又具有不确定性和难以量化的特点,所以表现为他们的行为又往往带有随机性. 复杂系统的这些特性决定了复杂系统的研究方法.对复杂系统中的智能个体行为及整个系统的运行过程的描述及求解尚不能采用现有的数学工具.计算机仿真技术的应用则是描述及求解复杂系统的有效尝试.将复杂系统中的智能个体整合为多个A 2gen t ,那么M u lti 2A gen t 仿真技术就必然成为复杂系统研究的一种重要形式.从仿真个体行为得到整个系统的总体结果也是M u lti 2A gen t 仿真的显著特点. 经济系统的复杂性表现为不确定性、不可预测性和个体的智能性.经济系统是一个基于过程的、自组织的和不断演进的系统,目前San ta Fe In stitu te ,Stanfo rd ,M IT ,Ch icago 和其它机构的经济学家采用以仿真方法为代表的新的研究方法来扩展传统经济学的简单静态均衡方法[2].而对经济个体行为及整个经济系统运行过程的模拟则是真实经济系统在计算机上的再现.用它研究经济个体行为对系统的影响,预测在外部环境变化时的经济运行情况则是经济研究的有力工具.第31卷第1期2002年2月 信息与控制Info r m ati on and Contro lV o l .31,N o.1 Feb.,2002 α收稿日期:2001-01-03 基金项目:国家自然科学基金资助项目(批准号79970011)2 A spen模型与本文模型的实现方法(A s-pen and the m ethodology of the m odel i n question)美国Sandia国家实验室的研究人员开发了一种美国经济的仿真模型,这种新型基于A gen t的仿真模型叫做A sp en.A spen模型引起了经济学界的广泛重视.最初的模型[4]模拟简单的市场经济、检验信息流的算法和代理人决策的逻辑顺序.1996年5月开始开发新的模型,并应用于研究过度经济和美国宏观经济.在A spen模型中,经济中的决策者称为“代理人”.A sp en模型中有与现实世界相对应的各种类型的代理人:政府、居民、企业、银行等.“代理人”模拟了其在真实世界里对应的对象的各种日常行为. A spen模型除了具有一般的M u lti2A gen t模型的特点——通过模拟微观个体的行为而得出宏观的总量指标以外,最显著的特征是智能性与并行性.对于单个的代理人的模拟,除了定义其一般的行为规则,考虑了个体行为的随机性之外,A spen模型的企业定价行为采用了Sandia的学习算法[3],这是其智能性的一个表现. 在A spen模型中“消息传递机制”与并行计算机体系结构是实现代理人行为之间的交互的关键,而且行动越复杂,需要传递相应的消息的结构也越复杂.事件(“代理人”的行动)发生的时间顺序是实现“消息传递机制”的关键.在A SPEN模型中,连续的时间被划分为离散的时间段(天),每一天又被划分为11个阶段,代理人的行动分布于这些不同的时间段来执行.在每天的每一阶段内,每一个“代理人”都会获取消息,并且基于其当时的状态来采取相互独立的、可行的行动[3]. 要建立一个M u lti2A gen t经济仿真模型,至少两点是必不可少的,一是模型首先应当合理地定义经济中各微观个体的属性与行为;二是模型应当有一套机制来实现模型中的代理人之间的交互与通信问题. 本文采用离散事件仿真技术来建立M u lti2A2 gen t仿真模型.建模的第一步是建立各部门即“代理人”的面向对象模型,定义各“代理人”的属性与事件.之后,采用离散事件仿真技术来解决代理人之间的交互与通信问题及系统的动态模拟;由于M u lti2 A gen t系统是一个离散系统,系统状态的变化是由于在一系列离散的特定时刻发生的事件而引起的,所以采用离散事件仿真的仿真时钟来模拟系统运行中的时间变化与一系列的特定时刻;将代理人的行为作为事件,分析模型中各“代理人”的事件之间的相互关系,确定事件之间的引发关系,按照事件之间的先后顺序形成一张反映事件发生的时间、条件、次序和事件之间相互关系的事件表.模型建立之后,仿真时钟的推进与事件表的处理就是M u lti2A gen t仿真模型的运行过程.3 模型中的代理人(The agen ts descr iption i n the m odel)本文的三部门经济仿真模型包括三类“代理人”(A gen t):一个政府、若干个企业和大量的居民.由于模型的目的是研究企业不同经济行为,即当亏损时退出与不退出市场对整个宏观经济和经济中的其它企业的经营状况造成的影响,因此,根据研究目标模型中将经济系统简化为10个“厂商”代理人,一个政府,并引入1200个“居民”代理人以代表“大量”的居民. 模型假定整个经济只生产一种产品,且厂商的生产只使用劳动这一种生产要素.居民在各企业或政府部门工作来取得收入,当企业因亏损而退出生产时,该企业原有的职工从政府处领取一定数额的救济金.居民的消费数额取决于其当期收入和边际消费倾向,并有一定的随机因素;居民的消费决策即具体选择哪一个厂商的产品所依据主要是各厂商产品的价格——价格越低的厂商的产品被选取的概率越大. 厂商在每一仿真时钟内都要决定其产量、产品的价格并根据其经营状况来决定是否退出生产.产量决策依据的是前一段时间的销量、当前的库存量和生产能力.厂商在定价时要考虑到其销售量的变化趋势、利润的变化趋势和行业平均价格的变化趋势.当厂商亏损时,应当决定是否退出生产:若不退出,则其工人的工资率将降低,若退出,则其职工将从政府处领取一定数额的救济金.模型中政府部门的作用是解决一部分居民的就业、向居民和政府征税和向失业的居民发放救济金. 对任何单个的代理人的模拟应当解决两方面的问题:代理人的数量特征的表示和行为的模拟.建立各代代理人的面向对象模型后,对象的属性用来表示代理人的数量特征,如居民代理人的收入,厂商代理人的产品的价格等,模型中称之为代理人的属性;而对象的方法则用来表示代理人决策行为的一般规581期宣慧玉等:一个基于离散事件仿真的M ulti2A gent经济仿真模型则,如厂商的定价决策,居民的消费决策等. 同一种类的代理人的同一种行为的定义是相同的,由于代理人的行为定义了代理人的决策的一般的规则,而行为处理的结果依赖于其输入——代理人在此行为发生时的状态,因此行为的规则反映了代理人的理性——基于决策时的状态,按照一定的原则做出决策.行为的定义除了定义代理人决策的一般规则之外,引入了大量的随机因素,因此代理人的理性又是不完全的,即代理人的行为有一定的随机性.代理人应当能够根据其自身的状态、环境的变化和历史的经验做出合理的决策,即代理人具有自学习能力或者说代理人的行为具有智能型的特点.虽然同一种代理人具有相同的决策规则,但是由于其决策时所处的状态不同(相应的属性的值不同)、学习的结果不同、并且存在着随机因素,因此其决策的结果也是不同的,这样就反映了经济中代理人行为的多样性.4 代理人智能行为的仿真(The I n tell igen t Behav ior of Agen ts)对于可以由简单规则定义的确定型的行为(如政府征税)或随机型的行为(如居民消费),一般可以表示为简单的等式,此时用统计方法求出定义规则的等式中的参数就可以确定这一类代理人的行为. 对于那些代理人的智能行为则需要借助人工智能领域的算法,如本文模型中定义厂商的定价决策行为的自动学习算法.该算法主要借鉴了机器学习方法中的分类器系统,并且根据实际问题作了不少的改动. 企业定价的自动学习算法的具体思路是:首先,企业要确定四种趋势:(1)近来本企业产品价格的升或降;(2)销售量的升或降;(3)利润的升或降;(4)企业价格相对于社会平均价格的高或低.企业根据这四种趋势来判断自己处于16种市场状态中的哪一种,从而产生市场状态的一条消息.这里的消息和规则的条件都用长度为4的二进制字符串表示,其中的字符(1或0)依次表示了企业价格、销售量、利润的升降及企业价格与社会平均价格比较的结果. 消息产生后,算法找到与之匹配的规则,从而得到这种市场状态下提高价格,降低价格和维持价格不变三种对策的概率.随后,根据这三种对策的概率求得所采用的对策是提高价格、降低价格还是维持价格不变.每次价格的变动为一个定值∃P.最后,企业根据改变后的价格计算其利润额.若该对策使得利润增加,则增加该对策对应的概率值,若是降低了利润就减少该对策对应的概率值. 通过对三种对策的概率值进行调整,企业实现了定价的渐进学习.这里的企业具有相当的理性,同时又受到其它随机因素的影响.企业的理性体现在效果最好的对策(也即概率值最大的对策)最有可能被企业选用.但是考虑到存在市场变化等随机或不可预测的因素,企业也可能选用其它的对策. 本文的企业定价自动学习算法能够积累以往的经验,通过多次学习而逐步使结果得到优化,成功地实现了企业定价的自动学习.算法得到的价格与理论计算法的价格基本相符,不依赖于初始价格,并且算法不因为参数的改变而失效,具有智能性和有效性.算法适用于多个企业代理人.并且对于不同政策、环境下的企业定价,自动学习算法也是适用的.5 代理人行为间的引发机制(The tr igger m echan is m of agen t behav iors)实现了各“代理人”对象的定义,即定义了各种代理人具有那些表征其特性的特征值(属性)、各代理人具有什么行为仅仅是实现了建模的第一步,下一步是确定各代理人的行为在什么时候、什么条件下激发,以及行为的发生会对自身、其它代理人有什么影响.这些问题的核心就是“如何使模型运动起来”.模型的动态运行是以事件引发机制和事件表为基础的. 如前文所述,A sp en模型是通过“消息传递机制”与并行计算机来实现代理人行为之间的交互的.本文研究不同于A spen模型,代理人的行为的表示及行为之间的引发机制是应用离散事件系统仿真技术的,它将代理人的行为作为系统的事件,并将这些事件分为原发事件与后续事件两种类型;代理人之间的关系就体现在事件之间的引发关系上,事件的发生使系统的状态发生变化,这样系统运行的过程就是事件发生、处理的过程.5.1 事件与事件之间的关系 系统中的事件(代理人的行为)分为原发事件和后续事件两种类型.原发事件是在特定时刻主动发生,而不需由其它事件激发才发生的事件;后续事件是指事件自身不能主动发生,必须由其它事件激发才能发生的事件. 事件引发机制要决定的是各事件之间的关系,事件之间有三种关系:引发关系、信息流动关系、时序关系.68信 息 与 控 制31卷  原发事件和后续事件之间是一种引发关系.对于事件之间的引发关系,原发事件一定在后续事件之前发生,而后续事件在引发其发生的原发事件之后,其它任何事件之间发生;如居民的“消费”(原发事件)和厂商的“销售”事件(后续事件)是引发关系,在一个居民的“消费”事件发生后,在处理其它事件之间,应当先处理由此“消费”事件而引发的那个厂商的“销售”事件. 信息流动关系是指一个事件的发生所需要的信息(事件的参数)是其它事件执行的结果;如居民的“消费”事件发生时,需要各厂商的供应量(产量和库存量)和产品的价格的信息,而这些信息分别是厂商的“产量决策”和“定价”事件执行的结果.对于事件之间是信息传递关系的情况,提供信息的事件应当先处理. 时序关系是指各事件的发生在时间上的顺序关系,如居民的“消费”事件必须在其“取得收入”事件发生之后才可以发生. 有些事件无严格的先后关系,如政府“支付工资”与“发放救济金”两个事件就无严格的关系.系统中存在许多这样的关系,在建模时可根据处理的方便来安排其先后次序. 图1说明了模型中表示代理人行为的各个事件及其相互间的引发关系、信息传递关系和时序关系.5.2 代理人行为之间的交互 代理人行为之间的关系就是事件之间的关系,事件表是反映事件之间的关系的事件集合.它是实现代理人行为之间的交互的基础,本文的模型是在离散事件系统仿真中的事件表的基础上,修改了初始时间表的定义、事件的插入、删除等操作来实现行为之间的交互的.事件表是在分析清楚模型中事件之间的引发关系之后,模型中要处理的事件按照发生的先后顺序的一种先进先出的线性表. 在每一个仿真时钟开始时,首先建立一个初始事件表对象,它只包括模型中所有原发事件,并且是按照时序、信息流动关系顺序排列的.之后处理事件表中的事件,若当前处理的事件是原发事件,则在其处理过程中将产生后续事件并被插入到事件表的当前事件之后其它事件之前;若当前事件是后续事件,则处理完毕直接进入下一个事件的处理;若事件表中下一个事件为空,则释放事件表对象,本次仿真结束.事件处理时,无论当前事件使原发事件还是后续事件,在事件处理完毕之后都删除此事件,这样在仿真运行的过程中,任何时候处理的事件都是事件表中的第一个事件;而在每一个仿真时钟的仿真结束条件是事件表为空.在一个仿真时钟内,事件表中的内容总是从初始事件表变为空事件表,而一个时钟内仿真的过程也就是事件表动态变化的过程.图1 模型中的事件及其相互关系F ig .1 Events and relati ons betw een them6 仿真结果(Si m ula tion result )应用本文建立的模型研究了亏损企业退出或保留在经济中对整个宏观经济(社会平均价格、总消费量)和经济中其它企业的经济效益(利润率、产品价格和销售量)的影响.模拟初始时假定10个厂商的经营状况各不相同,有高利润、保本和亏损三类企业,各企业所雇佣的劳动力数量互不相同,其劳动边781期宣慧玉等:一个基于离散事件仿真的M ulti 2A gent 经济仿真模型际产出率互不相同,因而各企业的最大产量也不同;居民的工资率取决于其工作的部门(政府或企业)和企业的状态(盈利或亏损),详细的初始数据见文献[8]. 在亏损企业分别采用退出与不退出两种策略的情况下,整个行业的平均价格的变化趋势如下图所示,图2中横轴表示仿真时钟,纵轴表示行业平均价格.在不退出的情况下,价格不断降低,最终稳定与一个很低的价格;同时行业内的亏损企业也越来越多;而在退出的情况下,最终平均价格却是稳定与一个较高的水平,而且亏损企业的数目及亏损程度都比不退出的情况要少.(a) 亏损企业退出生产(b ) 亏损企业不退出生产图2 两种策略下的模拟结果F ig .2 Si m ulati on results under tw o different strategies产生这种结果的原因在于:当亏损企业不退出生产时,企业亏损和企业减亏的某些措施(减少支付给工人的工资、奖金和职工下岗待业等)导致了工人收入的减少和预期收入的减少,因此总需求降低,但此时由于亏损企业不退出生产,因此整个供给水平并没有明显的下降,这将导致行业平均价格的降低,而价格的降低又会使一部分处在亏损边缘的企业亏损,从而导致价格的进一步下降和总需求的进一步降低,形成恶性循环,最终形成通货紧缩.而在亏损企业退出生产的情况下,虽然居民的收入降低,但与此同时总供给水平也下降,因此不会导致价格的进一步下降和亏损状况的进一步恶化.7 结论(Conclusion )基于A gen t 的仿真是研究复杂经济系统的一个有利工具,而目前国内这方面的研究大多是基于一些集成的开发工具,如Sw ar m .本文的研究不是基于任何M A S 系统开发工具,而是在原理上对M A S 建模方面的一次尝试. 在本文的研究过程中,我们认为代理人的行为的模拟和代理人之间的交互的实现是建立多A gen t 仿真模型的难点.模拟的结果往往依赖于代理人的行为规则,模型中代理人的行为规则越接近于自然系统中对应的个体的行为,模拟结果越接近于真实.因此,一个好的模型中的代理人应当具有一定的智能,本文的企业定价自动学习算法是模拟代理人的智能行为的一个尝试.本文提出的基于离散事件仿真的通过分析事件之间的引发机制和事件表对象的方法能够较好地解决代理人之间的通信问题.目前,我们正在以上两个方面在进一步完善模型. 对真实经济系统的合理简化与整合及经济个体行为的真实模拟是经济仿真可行性及使用化的关键,这个研究及困难又有意义,本文仅仅是经济系统仿真的开始与尝试.参 考 文 献(R eferences )1 约翰・L ・斯蒂着,王千祥,权利宁译.虚实世界——计算机仿真如何改变科学疆域.上海科技教育出版社,19982 W B rian A rthur .Comp lexity and the Econom y .Science 1999,284:107~1093 R J P ryo r ,N Basu ,T Q uint .D evelopm ent of A spen :A M icro 2analytic Si m ulati onM odel of the U .S .Econom y .Sandia R epo rt #SAND 9620434,Sandia N ati onal L abo rato ries ,A lbuquerque ,NM ,O ctober 1996,F rom www .sandia .gov4 Basu J ,P ryo r T ,Q uint ,T A rno ld .A spen :A M icro si m ulati onM odel of Econom y .Sandia R epo rt #SAND 96-2459,SandiaN ati onal L abo rato ries ,A lbuquerque ,NM ,O ctober 1996,F rom www .sandia .gov5 H el m ut L o rek ,M ichael Sonnenschein .O bject 2o riented Suppo rtFo r M odeling and Si m ulati on O f Individual -o riented Eco logicalM odels ",Eco logicalM odelling 108(1998)77~966 Zobrist ,Geo rge W .O bject 2o riented Si m ulati on :R eusability ,A dap tability ,M aintainability .N ew Yo rk :IEEE P ress ,19977 Go ldberg D E :Genetic A lgo rithm s in Search ,Op ti m izati on andM ach ine L earning ,A ddison 2W esley ,19898 高宝俊.微观模拟模型的研究与实践.西安交通大学硕士学位论,20009 李善同,高嘉陵著.微观分析模拟模型及其应用.机械工业出版社,199910 余永定.打破通货紧缩的恶性循环.经济研究,1999,(7)作者简介宣慧玉(1942-),博士生导师.研究领域为计算机仿真、人工智能等.高宝俊(1976-),博士生.研究领域为仿真模型、计量经济.李 群(1977-),硕士生.研究领域为仿真模型.88信 息 与 控 制31卷 。

Consensus in Multi-Agent Systems With

Consensus in Multi-Agent Systems With

Consensus in Multi-Agent Systems With Second-Order Dynamics and Sampled Data Wenwu Yu,Member,IEEE,Lei Zhou,Xinghuo Yu,Fellow,IEEE,Jinhu Lü,Fellow,IEEE,andRenquan Lu,Member,IEEEAbstract—This paper studies second-order consensus in multi-agent systems with sampled position and velocity data.A distributed linear consensus protocol with second-order dynamics isfirst designed,where both sampled position and velocity data are utilized.A necessary and sufficient condition based on the sampling period,the coupling gains,and the spectra of the Lapla-cian matrix,is established for reaching consensus of the system in this setting.It is found that second-order consensus in such a multi-agent system can be achieved by appropriately choosing the sampling period determined by a polynomial with order three. In particular,second-order consensus cannot be reached for a sufficiently large sampling period while it can be reached for a sufficiently small one under some conditions.Then,the coupling gains are carefully designed under the given network structure and the sampling period.Furthermore,the consensus regions are characterized for the spectra of the Laplacian matrix.On the other hand,second-order consensus in delayed undirected networks with sampled position and velocity data is then dis-cussed.A necessary and sufficient condition is also given,by which appropriate sampling period can be chosen to achieve consensus in multi-agent systems.Finally,simulation examples are given to verify and illustrate the theoretical analysis.Index Terms—Algebraic graph theory,coupling gain,consensus region,multi-agent system,second-order consensus,sampling period.Manuscript received November01,2012;accepted December07,2012.Date of publication December20,2012;date of current version October14,2013. This work was supported by the National Natural Science Foundation of China under Grant Nos.61104145,561025017,61025017,61203148,and11072254, in part by the Natural Science Foundation of Jiangsu Province of China under Grant No.BK2011581,in part by the Research Fund for the Doctoral Program of Higher Education of China under Grant No.20110092120024,in part by the Information Processing and Automation Technology Prior Discipline of Zhe-jiang Province-Open Research Foundation under Grant No.20120802,in part by the Fundamental Research Funds for the Central Universities of China,in part by the Australian Research Council(ARC)Future Fellowships under Grant FT0992226,and in part by the Discovery Scheme under Grant DP130104765. Paper no.TII-12-0754.W.Yu is with the Department of Mathematics,Southeast University, Nanjing210096,China and also with the School of Electrical and Computer Engineering,RMIT University,Melbourne VIC3001,Australia(e-mail: wenwuyu@,wwyu@).L.Zhou is with the Department of Mathematics,Southeast University,Nan-jing210096,China and also with the School of Electrical Engineering,South-east University,Nanjing210096,China(e-mail:cole66@).X.Yu is with the School of Electrical and Computer Engineering,RMIT Uni-versity,Melbourne VIC3001,Australia,and also with the School of Automa-tion,Southeast University,Nanjing210096,China(e-mail:yu@.au). J.Lüis with the Institute of Systems Science,Academy of Mathematics and Systems Science,Chinese Academy of Sciences,Beijing100190,China (e-mail:jhlu@).R.Lu is with the Institute of Information and Control,Hangzhou Dianzi Uni-versity,Hangzhou310018,China(e-mail:rqlu@).Color versions of one or more of thefigures in this paper are available online at .Digital Object Identifier10.1109/TII.2012.2235074I.I NTRODUCTIONC OLLECTIVE behavior of networks in the multi-agentsystems has been a topic of significant interest in recent years due to the studies of animal group behaviors and applica-tions in biological systems,sensor networks[26],Unmanned Air Vehicle(UA V)formations,robotic teams,and underwater vehicles.Convergence to a common state is usually called a consensus or an agreement problem.Consensus problems have been in-vestigated in thefield of computer science,particularly in au-tomata theory and distributed computation for quite a long time. In many cooperative multi-agent systems,a group of agents need to agree upon a certain quantity of interest by sharing in-formation with their locally connected neighbors.Through the information exchange,the whole agents may coordinate so as to achieve a certain global criterion of common interest.As one of the most typical collective behaviors,consensus usually refers to reaching an agreement among a group of autonomous agents [1],[2],[10],[11],[18],[19],[23],[24],[25].To study consensus in multi-agent systems,algebraic graph theory[9]has been recognized as a useful method.By analyzing the algebraic properties of the corresponding graph matrices, some conditions have been established in which the agents of the multi-agent systems can reach an agreement.In[16],the authors discussed thefirst-order consensus problems for net-works of dynamic agents withfixed and switching topologies. The relationship between the algebraic connectivity(or Fiedler eigenvalue[5])of the network and the performance(or the con-vergence speed)of a linear consensus protocol was also estab-lished.Thefirst-order consensus problem has attracted a lot of attention and plenty of nice results have been given[14].In [21],a distributed coordinated scheme with local information exchange was described for multiple vehicle systems.By taking into account the general case where informationflow may be unidirectional,a necessary and/or sufficient conditions under which consensus can be achieved in the context of unidirec-tional information exchange were given.Recently,there is a growing interest focusing on the con-sensus algorithms where all the agents are governed by second-order dynamics.In[22],some basic theoretical analysis was car-ried out for this case,where for each agent,the second-order dynamics are governed by the position and velocity terms of the agents and the asymptotic velocity is constant.A necessary and sufficient condition was given to ensure second-order con-sensus and it was found that both the real and imaginary parts of the eigenvalues of the Laplacian matrix of the corresponding network play key roles in reaching consensus.1551-3203©2012IEEEIn practise,it is quite difficult to measure the continuous in-formation transmission due to the unreliability of information channels,the capability of transmission bandwidth of networks, etc.Thus,it is more practical to apply sampled-data control, which has been widely studied recently and applied in many real-world systems,such as radar tracking systems,power sys-tems,and temperature control.It has been found that sampled-data control has many good properties,such as robustness and low cost.Recently,many results have been established for in-vestigating the second-order consensus in multi-agent systems with sampled data.For example,some conditions were derived for multi-agent systems with sampled control by using zero-order holds or direct discretization[3],[6],[15],[29].On the other hand,consensus of continuous-time multi-agent systems with time-varying topology and sampled control was discussed in[7].Communication delays were considered in multi-agent systems with sampled control in[8],[28].However,in most of the aforementioned works[3],[6],[15], [29],it is difficult to see how the network structure,sampling period,and control gains can affect the network dynamics since the authors mainly showed that consensus can be reached if the network parameters were in some derived regions,which are hard to apply.In order to further reveal how the sampling period can affect the consensus problem,in[27],a distributed linear consensus protocol with second-order dynamics was designed, where both the current and some sampled past position data were utilized.It was found that second-order consensus in such a multi-agent system cannot be reached without any sampled position data under the given protocol while it can be achieved by appropriately choosing the sampling period.However,the current information is usually unavailable,and therefore,in this paper,more sampled data and no current data will be introduced into the protocol,that is,both sampled position and velocity data will be used instead,which is memoryless since only in-formation at some particular time instants is needed.By com-pletely using the sampled data instead of the current informa-tion,one can utilize less information and save energy.Until now, there are no established results working on necessary and suffi-cient conditions for second-order consensus in continuous-time multi-agent systems with sampled data.The main contributions of this paper include the design of two protocols together with some necessary and sufficient condi-tions derived for reaching second-order consensus in the multi-agent system by using both sampled position and velocity data, which have not been investigated so far.Furthermore,the de-rived necessary and sufficient conditions are theoretically ana-lyzed to clearly show how the network structure and the parame-ters can affect the network dynamics,which is still a challenging problem nowadays.Specifically,the sampling period,the cou-pling gains,and the spectra of the Laplacian matrix for reaching second-order consensus in the multi-agent system are carefully designed and discussed.The rest of the paper is organized as follows.In Section II, some preliminaries on graph theory and model formulation are given.The main results about second-order consensus in multi-agent systems with sampled position and velocity data are pre-sented in Section III.In Section IV,second-order consensus in delayed undirected networks with sampled position and velocity data is discussed.In Section V,numerical examples are given to illustrate the theoretical analysis.Conclusions arefinally drawn in Section VI.II.P RELIMINARIESIn this section,some basic concepts and results about alge-braic graph theory[9]are introduced.Let be a weighted digraph(or directed graph) of order,with the set of nodes,set of directed edges,and a weighted adjacency matrix.An edge in network is denoted by ,where and are called the terminal and initial nodes, respectively,which means that node can receive information from node.The adjacency elements associated with the edges of the graph are positive,i.e.,.Moreover, we assume,.A directed path from node to node in is a sequence of edges of the form in the di-rected network with distinct nodes,.A root is a node having the property that for each node dif-ferent from,there is a directed path from to.A directed tree is a directed graph,in which there is exactly one root and every node except for this root itself has exactly one parent.A directed spanning tree is a directed tree consisting of all the nodes and some edges in.A directed graph contains a directed spanning tree if one of its subgraphs is a directed spanning tree.The second-order consensus protocol in multi-agent dynam-ical systems is described by[19],[20]and[22]:(1) where and are the position and velocity states of the th agent(node),respectively,andare the coupling gains,and is the coupling configuration matrix representing the topological structure of the network and thus is the weighted adjacency matrix of the network.The Laplacian matrix is defined by(2) which satisfies the diffusion property that.For notational simplicity,is considered throughout the paper, but all the obtained results can be easily generalized to the case with by using the Kronecker product operations[13].In the above case,all the current position and velocity states have to be utilized,but in the real situation,agents in the sys-tems usually communicate with each other in some certain time intervals.Therefore,in order to utilize less information and saveYU et al.:CONSENSUS IN MULTI-AGENT SYSTEMS WITH SECOND-ORDER DYNAMICS AND SAMPLED DATA2139energy,it is desirable to use sampled data instead of the current data[27]:(3) where are the sampling instants satisfying,and and are the coupling gains. For simplicity,assume that,where is the sampling period.Definition1:Second-order consensus in multi-agent system (3)is said to be achieved if for any initial conditions, Because of(2),system(3)can be equivalently rewritten as follows:(4) The following notations will be used throughout the paper.Let and be the real and imaginary parts of the complex number,be the eigen-values of the Laplacian matrix,be the-dimensional identity(zero)matrix,be the vector with all entries being1(0).Lemma1:([21]):The Laplacian matrix has a simple eigen-value0and all the other eigenvalues have positive real parts if and only if the directed network has a directed spanning tree. Lemma2:([12]):The Kronecker product has the fol-lowing properties:for any matrices A,B,C,and D with appro-priate dimensions,(1);(2).Lemma3:([17]):Given a complex coefficient polynomial of order two as follows:where,,,and are real constants.Then,is stable if and only if and.Lemma4:([17]):Given a real coefficient polynomial of order three as follows:Then,is stable if and only if,,,are positive and .III.S ECOND-O RDER C ONSENSUS IN M ULTI-A GENTD YNAMICAL S YSTEMS W ITH S AMPLED P OSITION ANDV ELOCITY D ATALet,,,and .Then,system(4)can be rewritten as(5) Note that a solution of an isolated node of system(5)satisfies(6) where is the state vector.Letand rewrite system(5)in a matrix form:(7) where is the Kronecker product[13].Let be the Jordan form associated with the Laplacian matrix ,i.e.,,where is a nonsingular matrix.By Lemma2,one has(8) where.If the graph is undirected,then is symmetric and is a diagonal matrix with real eigenvalues. However,when is directed,some eigenvalues of may be complex and,where.........(9)in which are the eigenvalues of the Laplacian matrix,with multiplicity,and. Let,,,and.Note that if the network contains a directed spanning tree,then by Lemma1,0is a simple eigenvalue of the Laplacian matrix, so(10)2140IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS,VOL.9,NO.4,NOVEMBER2013Theorem1:([27]):Suppose that the network contains a directed spanning tree.Then,seocnd-order consensus in system (3)can be reached if and only if,in(8)(11) Corollary1:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3)can be reached if and only if the following systems are asymp-totically stable:(12)Proof:This can be approved by using the similar analysis in[27].Until now,it is still very hard to check conditions(11)and (12)in Theorem1and Corollary1which do not reveal how network structure affects the consensus behavior.Next,a the-orem is derived to ensure consensus depending on the control gains,spectra of the Laplacian matrix,and the sampling period. Theorem2:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3)can be reached if and only if(13) and(14)Proof:It suffices to prove that system(12)is asymptot-ically stable if and only if the conditions(13)and(14)are satisfied.Form(12),it follows that(15) By simple calculation,one obtains.Inte-grating both sides of(15)from to,one has(16)where.Let.It is easy to see that is bounded on.So,forand,one has(17) Since is bounded when,if and only if all eigenvalues of satisfy.Let,one has(18) Let.Then,(18)can be transformed to(19) It is well known that in(18)if and only ifin(19).Therefore,if and only if all the roots in(19) have negative real parts.By Lemma3,(19)is stable if and only if(20) and(21) Therefore,if and only if(13)and(14)are satisfied. By Corollary1,second-order consensus in system(3)is reached if and only if(13)and(14)are satisfied.Though a necessary and sufficient condition for second-order consensus in multi-agent system(3)is established in Theorem 2,it is still difficult to see how to design the network parame-ters in system(3)to achieve this goal.In particular,for a given network,one can design appropriate parameters,,and such that the conditions(13)and(14)in Theorem2are satis-fied.On the other hand,since the condition(14)holds for all ,one can apply the consensus regions[4],[25] to further simplify the condition(14)to get a more applicable result.Next,the design for choosing the sampling period,the cou-pling gains and,and the spectrum of the Laplacian matrix are discussed in Sections III-A–III-C.A.Selection of the Sampling PeriodThe sampling period plays a key role for reaching con-sensus in multi-agent system(13).It is still unclear from The-orem2about how to choose appropriate sampling period and this subsection aims to solve this problem.Different from the re-sults in[27]where the hyperbolic functions and trigonometric functions on the sampling period were derived,the function on the sampling period may be simple in the condition(14) of Theorem2.Note that the inequality(14)can be equivalently written as(22)YU et al.:CONSENSUS IN MULTI-AGENT SYSTEMS WITH SECOND-ORDER DYNAMICS AND SAMPLED DATA2141Corollary1:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3)can be reached if and only if(13)and(22)hold.Actually,from(22),it is easy to choose the appropriate sam-pling period since the left hand of inequality(22)is a poly-nomial of with order3.Let(23),and.It is well known that has at most three roots indicates that can be easily solved.Then,a corollary for how to choose the sampling period is given.Theorem3:Suppose that the network contains a di-rected spanning tree.Then,second-order consensus in system(3)can be reached if and only if the sampling pe-riod.For a general sampling period,one can apply Theorem3to check if it is useful for reaching consensus in multi-agent system (3).However,is a small or a large sampling period can work? Corollary3:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3) cannot be reached for a sufficiently large sampling period while it can be reached for a sufficiently small sampling period if(13)and(24) hold for all.Proof:If the sampling period is sufficiently large or larger than,the condition(22)or(13)is not satisfied which indicates that consensus cannot be reached in system(3). While the sampling period is sufficiently small,condition (22)holds if(24)is satisfied.If the network is undirected,a simplified result can be obtained.Corollary4:Suppose that the network is undirected and connected.Then,second-order consensus in system(3)can be reached if and only if(25)Proof:Since the network is undirected,andfor all due to the symmetric Laplacian matrix.Then,(22)is equivalent to(26) Combining the condition(13),one concludes that second-order consensus in system(3)can be reached if and only if(25) holds.B.Design of the Coupling GainsThough the sampling period can be appropriately chosen from Theorem3,it is still unknown for how to design the coupling gains and under the given sampling period. Multiplying on both sides of(14),then inequality(14)can be equivalently written as(27) Theorem4:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3)can be reached if and only if(28)and(29)where, .Proof:From(14)and(22),it is easy to obtain that.Then,(27)can be rewritten as(30)The left hand side of(30)has two roots,that is,(31) By solving(30),one obtains that or.Since is contradicts with the condition in (13),onefinally has that for all.Given the network structure,one canfirst choose the cou-pling gain satisfying the condition(29)and thenfind the ap-propriate parameter such that(28)holds under the given for reaching consensus in multi-agent system(3).If the network is undirected,a similar simplified result can be obtained since in Theorem4.Corollary5:Suppose that the network is undirected and connected.Then,second-order consensus in system(3)can be reached if and only if(32) Note that the condition(25)in Corollary4and the condi-tion(32)in Corollary5are the same.Since and ,one can choose the parameter such that can obtain its maximum value such that consensus in system(3)can be reached.Corollary6:Suppose that the network is undirected and connected.By designing,second-order consensus2142IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS,VOL.9,NO.4,NOVEMBER2013in system(3)can be reached if and only if the sampling period is less than a maximum value,that is,(33)C.Consensus Region for the SpectrumSince the condition(14)holds for all,one can find a stable consensus region[4],[25]as follows:(34) Then,the problem is transformed tofinding if all the nonzero eigenvalues of the Laplacian matrix lie in the stable consensus region,i.e.,for all.Theorem5:Suppose that the network contains a directed spanning tree.Then,second-order consensus in system(3)can be reached if and only if(13)holds and for all .IV.S ECOND-O RDER C ONSENSUS IN D ELAYED U NDIRECTED N ETWORKS W ITH S AMPLED P OSITION AND V ELOCITY D ATA In some real situations,the input time delays always exist which cannot be ignored.When the time delays is introduced into the protocol,one can consider the system as follows:(35)where is the time delay.Let,,,,and .Then,system(35)can be rewritten as(36) Let and rewrite system(36)in a matrix form:(37)where is the Kronecker product[13].Let be the Jordan form associated with the Laplacian matrix,i.e.,, where is a nonsingular matrix.By Lemma2,one has(38) where.Here,we simply discuss the undirected networks,i.e.,is symmetric and is a diagonal matrix with real eigenvalues.By the same calculation,we can also get that the asymptotical behavior of the system(35)is dominated by the stability of the system as follows:(39) Theorem6:Suppose that the undirected network is con-nected and.Then,second-order consensus in system (35)can be reached if and only if(40) and(41) where,,and.Proof:From(39),if follows that(42) Integrating both sides of(42)from to,one has(43)Let and. It is easy to see that and are both bounded on.YU et al.:CONSENSUS IN MULTI-AGENT SYSTEMS WITH SECOND-ORDER DYNAMICS AND SAMPLED DATA2143Fig.1.Position and velocity states of agents,where,,and.So,for,and ,one hasand. Let and.One hasThen,if and only if,i.e.,all eigenvalues of satisfy.Let.One has(44) It is easy to see that,has a eigenvalue.Let,,,and.Onefinally gets(45)It is well known that in(44)if and only if. Therefore,if and only if all the roots in(45)have negative real parts.By Lemma4,(45)is stable if and only if,,,, and.By solving thefirst two polynomials,one obtains the condi-tion(41).Therefore,if and only if(40)and(41)are satisfied.By Corollary1,second-order consensus in system(35) can be reached if and only if(40)and(41)are satisfied. Remark1:In Theorem6,a necessary and sufficient condi-tion for second-order consensus in the multi-agent system(35) with the time delay is established.For a given network,one can design appropriate parameters,,,and such that the con-ditions(40)and(41)in Theorem6are satisfied.V.S IMULATION E XAMPLESIn this section,some examples are given to verify the theo-retical analysis of this paper.A.Second-Order Consensus in a Multi-Agent System With an Undirected TopologyConsider the multi-agent system(3)with an undirected topology,where,,and .By simple calculation,one has,,,and.From Corollary4,multi-agent system(3) can reach second-order consensus if and only if.It is easy to see that second-order consensus in system(3)can be reached if for a sufficiently small as in Corollary2144IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS,VOL.9,NO.4,NOVEMBER2013Fig.2.Position and velocity states of agents,whereand.Fig.3.States of versus the sampling period ,,3,4.3and ,while it cannot be reached for .The position and velocity states of all the agents are shown in Fig.1.According to Corollary 6,one can choose so as to get a maximum value for that can tolerate the sampling period.The position and velocity states of all the agents areshown in Fig.2where second-order consensus can be reached if while it cannot be reached when .B.Second-Order Consensus in a Multi-Agent System With a Directed Topology1)Section of the Sampling Period:Consider the multi-agent system (3)with a directed topology,where,,and.By simplecalculation,one has ,,,and .From Theorem 3,multi-agent system (3)can reach second-order consensus if and only if.By (23),one has and.Then,second-order con-sensus can be reached in multi-agent system (3)if and only if.Consider the sampling period as a variableof .In Fig.3,it is easy to see that second-order consensus in the system can be reached if while itcannot be reached forFig.4.Position and velocity states of agents,whereand.Fig.5.Position and velocity states of agents,whereand.Fig.6.Position and velocity states of agents,whereand.where .The position and velocity states of all the agents are shown in Fig.4.2)Design of the Coupling Gains:Consider the same ex-ample in this subsection except that the sampling periodis fixed.The aim to choose the appropriate coupling gains and such that second-order consensus in multi-agent system (3)can be reached.By Theorem 4,one knows that it can be achieved if and .Then,one can firstYU et al.:CONSENSUS IN MULTI-AGENT SYSTEMS WITH SECOND-ORDER DYNAMICS AND SAMPLED DATA2145fix and obtains that second-order consensus can be reached if and only if.The position and velocity states of all the agents are shown in Fig.5where it is easy to see that second-order consensus in system(3)can be reached if while it cannot be reached for.C.Second-Order Consensus in Delayed Undirected Networks Consider the multi-agent system(35)with an undirected topology,where,,, and.By simple calculation,one has,, ,.From Theorem6,one knows that second-order consensus in multi-agent system(35)can be reached ifwhile it cannot be reached if.The position and velocity states of all the agents are shown in Fig.6.VI.C ONCLUSIONSIn this paper,second-order consensus in multi-agent systems with sampled position and velocity data has been studied.A distributed linear consensus protocol with second-order dynamics has been designed,where both sampled position and velocity data have been utilized.A necessary and sufficient condition based on the sampling period,the coupling gains, and the spectra of the Laplacian matrix,has been established for reaching consensus of the system in this setting.On the other hand,second-order consensus in delayed undirected networks with sampled position and velocity data has also been discussed.A necessary and sufficient condition was given,by which appropriate sampling period can be chosen to achieve the consensus.In the future,nonuniform sampling intervals and multiple time delays will be introduced into the protocol which are more challenging.Many interesting works can be also found in these multi-agent systems,which deserve further investigation.R EFERENCES[1]M.Cao,A.S.Morse,and B.D.O.Anderson,“Reaching a consensusin a dynamically changing environment:A graphical approach,”SIAMJ.Contr.Optimizat.,vol.47,no.2,pp.575–600,Feb.2008.[2]Y.Cao,W.Yu,W.Ren,and G.Chen,“An overview of recent progressin the study of distributed multi-agent coordination,”IEEE Trans.r.,vol.9,no.1,pp.427–438,2013.[3]Y.Cao and W.Ren,“Multi-vehicle coordination for double-integratordynamics underfixed undirected/directed interaction in a sampled-datasetting,”Int.J.Robust Nonlinear Contr.,vol.20,pp.981–1000,2010.[4]Z.Duan,G.Chen,and L.Huang,“Disconnected synchronized regionsof complex dynamical networks,”IEEE Trans.Autmo.Contr.,vol.54,pp.845–849,2009.[5]M.Fiedler,“Algebraic connectivity of graphs,”Czechoslovak Math.J.,vol.23,pp.298–305,1973.[6]Y.Gao,L.Wang,G.Xie,and B.Wu,“Consensus of multi-agent sys-tems based on sampled-data control,”Int.J.Contr.,vol.82,no.12,pp.2193–2205,2009.[7]Y.Gao and L.Wang,“Sampled-data based consensus of continuous-time multi-agent systems with time-varying topology,”IEEE Trans.Auto.Contr.,vol.56,no.5,pp.1226–1231,2011.[8]Y.Gao and L.Wang,“Consensus of multiple dynamic agents withsampled information,”IET Control Theory Appl.,vol.4,no.6,pp.945–956,2010.[9]C.Godsil and G.Royle,Algebraic Graph Theory.New York,NY,USA:Springer-Verlag,2001.[10]Y.Hong,G.Chen,and L.Bushnell,“Distributed observers design forleader-following control of multi-agent networks,”Automatica,vol.44,pp.846–850,2008.[11]Y.Hong,J.Hu,and L.Gao,“Tracking control for multi-agent con-sensus with an active leader and variable topology,”Automatica,vol.42,pp.1177–1182,2006.[12]R.A.Horn and C.R.Johnson,Matrix Analysis.Cambridge,U.K.:Cambridge Univ.Press,1985.[13]R.A.Horn and C.R.Johnson,Topics in Matrix Analysis.Cambridge,U.K.:Cambridge Univ.Press,1991.[14]A.Jadbabaie,J.Lin,and A.S.Morse,“Coordination of groups of mo-bile autonomous agents using nearest neighbor rules,”IEEE Trans.Auto.Contr.,vol.48,no.6,pp.985–1001,Jun.2003.[15]H.Liu,G.Xie,and L.Wang,“Necessary and sufficient conditions forsolving consensus problems of double-integrator dynamics via sam-pled control,”Int.J.Robust Nonlinear Contr.,vol.20,pp.1706–1722,2010.[16]R.Olfati-Saber and R.M.Murray,“Consensus problems in networksof agents with switching topology and time-delays,”IEEE Trans.Auto.Contr.,vol.49,no.9,pp.1520–1533,Sep.2004.[17]P.C.Parks and V.Hahn,Stability Theory.Englewood Cliffs,NJ,USA:Prentice Hall,1993.[18]J.Qin,B.Yu,and S.Hirche,“Stationary consensus of asynchronousdiscrete-time second-order multi-agent systems under switchingtopology,”IEEE rmat.,vol.8,no.4,pp.986–994,2012.[19]W.Ren,“On consensus algorithms for double-integrator dynamics,”IEEE Trans.Autom.Contr.,vol.58,no.6,pp.1503–1509,2008.[20]W.Ren and E.Atkins,“Distributed multi-vehicle coordinated controlvia local information exchange,”Int.J.Robust Nonlinear Contr.,vol.17,no.10–11,pp.1002–1033,2007.[21]W.Ren and R.W.Beard,“Consensus seeking in multiagent systemsunder dynamically changing interaction topologies,”IEEE Trans.Autom.Contr.,vol.50,no.5,pp.655–661,May2005.[22]W.Yu,G.Chen,and M.Cao,“Some necessary and sufficient condi-tions for second-order consensus in multi-agent dynamical systems,”Automatica,vol.46,no.6,pp.1089–1095,2010.[23]W.Yu,G.Chen,and M.Cao,“Consensus in directed networks ofagents with nonlinear dynamics,”IEEE Trans.Auto.Contr.,vol.56,no.6,pp.1436–1441,2011.[24]W.Yu,G.Chen,and J.Lü,“On pinning synchronization of complexdynamical networks,”Automatica,vol.45,no.2,pp.429–435,2009.[25]W.Yu,G.Chen,W.Ren,J.Kurths,and W.X.Zheng,“Distributedhigher-order consensus protocols in multi-agent dynamical systems,”IEEE Trans.Circuits Syst.I,Reg.Papers,vol.58,no.8,pp.1924–1932,2011.[26]W.Yu,G.Chen,Z.Wang,and W.Yang,“Distributed consensusfil-tering in sensor networks,”IEEE Trans.Systems,Man,Cybern.B,vol.39,no.6,pp.1568–1577,2009.[27]W.Yu,W.Zheng,G.Chen,W.Ren,and J.Cao,“Second-order con-sensus in multi-agent dynamical systems with sampled position data,”Automatica,vol.47,no.7,pp.1496–1503,2011.[28]Y.Zhang and Y.Tian,“Consensus of data-sampled multi-agent sys-tems with random communication delay and packet loss,”IEEE Trans.Autom.Contr.,vol.55,no.4,pp.939–943,2010.[29]H.Zhao,S.Xu,and D.Yuan,“Consensus of data-sampled multi-agentsystems with Markovian switching topologies,”Asian J.Contr.,vol.14,no.5,pp.1366–1373,2012.Wenwu Yu(S’07–M’12)received the B.Sc.degreein information and computing science and the M.Sc.degree in applied mathematics from the Departmentof Mathematics,Southeast University,Nanjing,China,in2004and2007,respectively,and thePh.D.degree from the Department of ElectronicEngineering,City University of Hong Kong,HongKong,China,in2010.Currently,he is an Associate Professor in theResearch Center for Complex Systems and NetworkSciences,Department of Mathematics,Southeast University,China.His research interests include multiagent systems,nonlinear dynamics and control,complex networks and systems,neural networks, cryptography,and communications.。

随机拓扑下离散多智能体事件触发一致性

随机拓扑下离散多智能体事件触发一致性

随机拓扑下离散多智能体事件触发一致性作者:赵阳解静曹洒来源:《青岛大学学报(工程技术版)》2024年第01期摘要:針对离散时间多智能体跟踪不稳定的问题,本文研究离散多智能体系统的事件触发一致性控制问题,通过马尔可夫跳变拓扑结构实现各智能体间的信息交互,设计了一种基于动态响应的事件触发条件,给出了马尔可夫跳变控制协议,构造带有转移概率的离散Lyapunov函数,得到所有智能体是均方一致性的充分条件。

数值算例验证了所提方法的有效性,证明了本结论可用于解决随机拓扑下离散多智能体的跟踪不一致问题。

关键词:离散多智能体系统;随机切换拓扑;马尔可夫链;事件触发;均方一致性中图分类号: TP13文献标识码: A离散多智能体具有自主性强、距离范围内的容错率高、抗干扰能力强、系统强耦合及强不确定性等特征[1],适用于描述机器人协调技术及群集运动等[2-7]实际工程问题。

对于多智能体系统的拓扑结构,张圆圆等人[8]研究了无向联通拓扑结构图下的多智能体系统;尉晶波等人[9]解决了拓扑切换下的多智能体协同输出调节问题。

但有关离散多智能体系统的文献大多集中在固定拓扑和切换拓扑上[10-13],随机切换拓扑结构的成果较少。

随机切换拓扑结构能够更直观地表示智能体之间的信息交换问题,因此本文将重点考虑基于马尔可夫链的随机切换拓扑结构[14-15]。

事件触发控制在资源节约方面具有显著优势,可有效减少通信次数。

陈侠等人[16]使用动态事件触发机制研究了网络攻击一致性问题;XIE D S等人[17]研究了具有事件触发策略的领导者-追随者一致性控制;XUE S S等人[18]研究了分布式事件触发一致性问题。

目前在马尔可夫跳变拓扑条件下的事件触发结果并不多,还有许多问题需要研究。

基于此,本文考虑马尔可夫跳变拓扑下离散多智能体系统的事件触发一致性问题,利用线性矩阵不等式技术[19]给出均方一致性的充分条件,避免事件触发时间序列对邻域内其他智能体信息的持续监控,并说明如何避免Zeno现象,通过数值算例验证了所提方法的有效性。

一致性编队控制基本知识梳理

一致性编队控制基本知识梳理
生成树 有向图
节点

通路
无向图
Laplacian矩阵( Laplacian matrix) 连通 强连通 准强连通
L [lij ]
j i aij , if i=j lij aij , if if i j
0, i 1, 2,..., n
i ui x n ui aij (t )( x j (t ) xi (t )) j 1
xi (t 1) xi (t ) vi (t )t
(t 1) (t ) r
突破
Jadbabaie A, Lin J, Morse A S. Coordination of Groups of Mobile Autonomous A-gents Using Nearest Neighbor Rules [J]. IEEE Transactionson Automatic Control.2003, 48 (6): 988–1001.
i ui x n ui aij (t )( x j (t ) xi (t )) j 1
代数图论
矩阵理论
图论基础知识
Laplace矩阵表示 基本概念
邻接,i) A [aij ] aij 0, if (j,i)
应用
编队避障
队形变换
队形保持
一致性理论 重构
集结
解散
。。。
拓展
二阶乃至高阶系统:Wen G, Zhao Y, Duan Z, et al. Containment of HigherOrder Multi-Leader Multi-Agent Systems: A Dynamic Output Approach[J]. IEEE Transactions on Automatic Control, 2016, 61(4): 1135-1140. 离散系统,带通讯延迟:Ren W, Beard R W, Atkins E M, et al. Information consensus in multivehicle cooperative control[J]. IEEE Control Systems Magazine, 2007, 27(2): 71-82 事件驱动: Li L, Ho D W C, Zou Y, et al. Event-trigged control for discretetime multi-agent networks[C]//Control Conference (ASCC), 2013 9th Asian. IEEE, 2013: 1-6.

有向图下的多智能体系统事件触发预设时间实用一致性

有向图下的多智能体系统事件触发预设时间实用一致性

有向图下的多智能体系统事件触发预设时间实用一致性DOI :10.19557/ki.1001-9944.2024.05.005付璐,陈霞,郭晓亚(青岛理工大学信息与控制工程学院,青岛266525)摘要:该文基于事件触发控制,研究了有向图下的多智能体系统的预设时间实用一致性问题。

首先为每个智能体提出了一个基于时变函数的事件触发控制输入,并构造合适的动态触发条件,使得系统可以在完全预设的时间内达到实用一致性。

该文提出了一个基于李雅普诺夫函数的预设时间稳定性充分条件,借助此条件给出了多智能体系统的预设时间实用一致性分析。

此外,该文从理论上严格排除了整个时间区间上的Zeno 行为。

最后,给出数值仿真实例验证了该文结果的正确性和有效性。

关键词:多智能体系统;事件触发控制;预设时间实用一致性中图分类号:TP13文献标识码:A文章编号:1001鄄9944(2024)05鄄0020鄄06Event 鄄triggered Predefined 鄄time Practical Consensus for Multi 鄄agent Systems Under Directed GraphFU Lu ,CHEN Xia ,GUO Xiaoya(School of Information and Control Engineering ,Qingdao University of Technology ,Qingdao 266525,China )Abstract :In this paper ,we study the predefined 鄄time practical consensus problem for multi 鄄agent systems under di 鄄rected graphs based on event 鄄triggered control.An event 鄄triggered control input based on a time 鄄varying function is firstly proposed for each agent ,and a suitable dynamic triggering condition is constructed so that the system can achieve practical consensus at a completely predefined time.A sufficient condition for the predefined 鄄time stability based on the Lyapunov function is proposed in this paper ,with the help of which the analysis of the multi 鄄agent system predefined 鄄time practical consensus is given.In addition ,this paper rigorously excludes Zeno behavior over the whole time interval from theory.Finally ,a numerical simulation example is given to verify the correctness and validity of the results of this paper.Key words :multi 鄄agent system ;event 鄄triggered control ;predefined 鄄time practical consensus收稿日期:2023-12-08;修订日期:2024-04-15基金项目:国家自然科学基金项目(61703225);山东省自然科学基金项目(ZR2022MF297)作者简介:付璐(1998—),女,硕士,研究方向为多智能体系统的事件触发预设时间一致性;陈霞(通信作者)(1986—),女,博士,副教授,研究方向为事件触发控制、多智能体系统的协调控制等。

多个体协调控制问题综述_闵海波

多个体协调控制问题综述_闵海波

多个体协调控制问题综述_闵海波第38卷第10期自动化学报Vol.38,No.10 2012年10月ACTA AUTOMATICA SINICA October,2012多个体协调控制问题综述闵海波1刘源1王仕成1孙富春2摘要对多个体协调控制问题的研究现状进行综述.介绍了多个体协调控制领域的基本问题,并结合系统中网络与动力学不确定性,对该领域当前的研究热点和前沿进行分析阐述.进一步,对工程中广为应用的Euler-Lagrange系统协调控制最新研究成果进行归纳总结.最后指出该研究领域存在的问题及今后的研究方向.关键词协调控制,一致性,时延,拓扑切换,异步网络,网络化Euler-Lagrange系统引用格式闵海波,刘源,王仕成,孙富春.多个体协调控制问题综述.自动化学报,2012,38(10):1557?1570DOI10.3724/SP.J.1004.2012.01557An Overview on Coordination Control Problem of Multi-agent SystemMIN Hai-Bo1LIU Yuan1WANG Shi-Cheng1SUN Fu-Chun2Abstract This paper presents an overview on the state-of-the-art of the coordination control problem of multi-agent systems.The fundamental problems in the coordination control?eld are introduced,and the research hotspots and frontiers are analyzed and summarized.Furthermore,some of the latest research achievements on coordination control of networked Euler-Lagrange systems are also presented.Finally,the problems in this area and the prospect of future research are summarized.Key words Coordination control,consensus,time delay,switching topology,asynchronous network,networkedEuler-Lagrange systemsCitation Min Hai-Bo,Liu Yuan,Wang Shi-Cheng,Sun Fu-Chun.An overview on coordination control problem of multi-agent system.Acta Automatica Sinica,2012,38(8):1557?1570 自上世纪80年代以来,受生物学、人类社会学研究的启发,多个体协调控制成为众多领域的研究热点[1].类似于鱼群捕食、鸟群迁徙等生物与人类社会中的群体性优势,多个体之间的协调与合作将大大提高个体行为的智能化程度,更好地完成很多单个个体无法完成的工作,并具有高效率,高容错性和内在的并行性等优点.时至今日,多个体协调控制技术已在无线传感器网络[2]、多机械臂协同装配[3]、无人机编队[4]、卫星编队[5]、集群航天器深空探测[6]等收稿日期2011-09-23录用日期2012-04-25Manuscript received September23,2011;accepted April25, 2012国家重点基础研究发展计划(973计划)(2012CB821206),国家自然科学基金(61203354,60904083,61004021),高等学校博士学科点专项科研基金资助课题(20111011321),北京市自然科学基金(4122037)资助Supported by National Basic Research Program of China(973 Program)(2012CB821206),National Natural Science Founda-tion of China(61203354,60904083,61004021),Specialized Re-search Fund for the Doctoral Program of Higher Education(2011 1011321),and Municipal Natural Science Foundation of Beijing (4122037)本文责任编委吕金虎Recommended by Associate Editor LV Jin-Hu1.第二炮兵工程大学西安7100252.清华大学计算机科学与技术系人工智能国家重点实验室北京1000841.Hi-Tech Institute of Xi an,Xi an7100252.State Key Laboratory of Intelligence Technology,Department of Computer Science and Technology,Tsinghua University,Beijing100084领域得到广泛应用.多个体协调控制的基本问题包括一致性控制、会合控制、聚结控制和编队控制.其中后三者可视为一致性控制的推广与特例.目前已有文献对一致性问题进行了系统性综述[7?8].就系统属性而言,多个体系统区别于单体系统最本质的要素在于其通过网络进行信息传递与共享,网络的不确定性(如丢包,时延等)对多个体协调控制性能的影响不容忽视,因此,该问题成为近几年协调控制领域研究的前沿和热点[9].另一方面,多个体协调控制领域的研究对象也逐渐由前期简单的一阶或高阶线性系统过渡为更一般的非线性系统,并考虑更为实际的系统不确定性等因素.其中Euler-Lagrange系统作为一类工程中应用极其广泛的典型系统受到学者的普遍关注.为此,本文力图在剖析多个体协调控制领域最新研究成果的基础上,对该领域的发展现状进行综述,并提出该领域研究的前沿性问题.本文结构安排如下:第1节针对多个体协调控制几个基本问题进行阐述,为后续讨论奠定基础.第2节围绕最近几年该领域的发展前沿—网络不确定性与系统不确定性进行综述,并对相关研究方法进行归纳总结.第3节介绍了工程中广为应用的Euler-Lagrange系统协调控制最新研究成果.第4节总结全文,并提出该领域的发展方向.1558自动化学报38卷1多个体协调控制中的基本问题1.1一致性(Consensus)控制一致性是指多个体通过信息的共享与交互,实现某种状态的趋同[7],其控制目标可描述为limt→∞x j(t)?x i(t) =0,?i,j∈I(1)其中,I为系统中个体的集合,x i为系统中第i个个体的状态.这种状态可以是卫星的姿态[10]、鱼群或鸟群的行动方向[11?12]、数据融合[13]或者分布式传感器滤波值[2].一致性问题最早在计算机科学中提出.在该领域奠基性工作中,Tsitsiklis等[14]针对分布式决策问题,研究了异步情形下的优化算法.计算机图形领域,Reynolds[15]按照自然界中鸟群的特点较早地对鸟群、鱼群等系统的群体行为进行了计算机仿真,并提出著名的Boid模型.该模型按照下面的规则进行仿真:1)中心聚结:所有个体试图靠近邻近的个体;2)防撞:所有个体与邻近个体保持适当间距,以免碰撞;3)速度匹配:所有个体试图与邻近个体的速度保持一致.1995年,Vicsek等[16]又从统计力学的角度,对文献[15]中的规则进行了简化,通常称为Vicsek模型.该模型是由N个自治的个体组成的离散时间系统,每个个体在平面中以恒定的速率运行,其角度是邻居范围内所有个体角度的平均,另外还受到一个均值为零的噪声的影响.邻居是那些与该个体的Euclidean距离小于某个给定的半径r内的所有个体的集合.Vicsek模型实际上是Boid 模型的一种特殊情形.文献[16]通过仿真给出了令人惊讶的结果:当个体密度比较大且噪声比较小时,系统中的每个个体的飞行方向趋于某个共同的角度,该现象称为同步[17].Vicsek模型不但引起了物理学家的兴趣,数学家、控制工程师和系统理论专家也试图对该群体系统的这种一致性行为给出一个严格的理论分析. Jadbabaie等[18]通过对Vicsek模型中角度的非线性更新用线性模型来近似,试图给出同步的理论分析.对一阶积分器型线性系统,其动力学模型为˙x i=u i,?i∈I(2)其中,x i∈R为第i个个体的状态,u i为施加的控制量.早期的工作中,学者们设计了几种不同的协调控制律[18?21],可统一为以下数学描述[7]:u i=?j∈N i(t)a ij(t)(x i(t)?x j(t))(3)其中,j∈N i(t)表示第i个个体的相邻个体集合, a ij(t)为时变的拓扑加权系数.控制律(3)意味着每个个体的状态不断趋同于其相邻个体的状态(可能时变).此外,由式(3)可以看出,在针对第i个个体设计分布式控制律时,仅利用其相邻的个体与其自身的状态差,而无需非相邻个体的任何信息,因此这一规则称为“近邻规则”.Ren等[21]证明了:对于一阶系统,当通信拓扑中存在一个衍生树(Spanning tree)时,该协调控制律能够保证系统达到一致性.此外,对于复杂网络系统,Chen等[22]提出以图论的视角对系统同步性进行分析,得到了有趣的结论:通信图中多余的边不仅无益于同步,相反还会破坏同步,这也揭示了通信图对于系统可同步性的基础性作用.对应于离散系统,文献[18,20?21]设计了如下控制协议:x i[k+1]=j∈N i(k)βij[k]x j[k](4)其中,j∈N i(k)∪{i}βij[k]=1,βij[k]>0,?j∈N i(k){i}.其基本思想是利用迭代算法,使系统状态趋向于一个由x1=x2=···=x n刻画的共同空间.在一阶系统结果的基础上,近几年二阶及高阶线性系统也成为研究的热点,代表性工作包括文献[23?26].以二阶线性系统为例,其动力学模型可表述为˙x i=v i,˙v i=u i,?i∈I(5)对应的一致性控制律(或控制协议)可设计为[26] u i(t)=αj∈N ia ij(x j(t)?x i(t))+βj∈N ia ij(v j(t)?v i(t)),i∈I(6)以经典控制的视角来看,该控制器本质上是一个PD 控制器,即协调控制律中既包含位置误差信息,也包含速度误差信息.所不同的是,对于多个体系统而言,其利用的是群体性误差信息,而针对单体系统的控制仅需要该单体与期望状态的误差信息.因此,相对于单体,针对多个体一致性控制的分析与综合更为复杂.事实上,二阶系统实现一致性的条件相比于一阶系统而言更为保守:除通信拓扑中需要具有一个衍生树之外,还需要控制增益α和β满足一定条件[26].此外,针对更为一般的高阶线性系统,Li 等[27]提出一种新的控制体系,并在该体系下处理多体系统的一致性与复杂网络系统的同步问题.在另一个工作中,Li等[28]针对系统的H∞和H2分布式控制问题进行研究.通过引入性能域(Performance 10期闵海波等:多个体协调控制问题综述1559 region)概念,对系统的H∞和H2性能进行了详尽的分析讨论.有关线性系统一致性更详尽的分析和综述可参见文献[7?8].近年来,针对非线性系统的一致性控制研究也取得了一定的进展.众所周知,一般非线性系统相对于线性系统在动力学特性方面要复杂得多,因此针对非线性系统一致性控制的分析和综合通常比较困难.相对于线性系统完备的分析工具(如代数图论[29]、非负矩阵论[30]等)与结论,面向一般非线性多体系统一致性控制的研究尚处于起步阶段,目前的研究仍集中在几种特殊的系统中.例如Su等[31]和Song等[32]等针对含有非线性动力特性的二阶系统设计了一致性跟随控制器.虽然假定系统中含有未知动力学参数的非线性项,但如果将该非线性项视为扰动,该问题本质上仍是一个二阶线性系统的协调控制.而M¨u nz等[33?34]则在考虑通信时延与动态拓扑情形下,针对局部无源的一类特殊非线性系统设计了一致性控制律,其控制律仍是线性的.而Fang等[35]和Qu等[36]则直接利用非线性控制器对该问题加以解决.另外,最近几年,针对工程中广泛应用的Euler-Lagrange多体系统的一致性控制研究成为该领域的一个前沿和热点[37].总体而言,目前在非线性一致性控制领域所形成的理论体系远非成熟,该领域的研究仍然任重而道远.另外需要指出的是,以上讨论的一致性控制问题绝大多数都是针对理想通信网络情形,即假定网络中不存在通信时延且拓扑为时不变的.考虑到现实网络的非理想特性,这些问题就变得更为复杂和困难.我们将在第2节中对这些特殊问题进行更为深入的分析.1.2会合(Rendezvous)控制概括而言,会合指系统中的所有个体速度逐渐趋于零,且静止于某一位置,其控制目标可描述为limt→∞x j(t)?x i(t) =0,?i,j∈Ilim t→∞ ˙x i(t) =0,?i∈I(7)由式(7)可以看出,该问题在本质上是一致性问题的一个特例,可简单理解为终态为静止的一致性.会合控制最早由Ando等提出[38],后被Lin 等推广至同步和异步的“走–停”策略[39].针对线性系统,在通信拓扑保持连通性的条件下,Cortes等[40]放宽了文献[38]中通信拓扑连通的条件.通过引入“邻近图(Proximity graphs)”概念,Cortes等进一步降低了这些算法对于通信拓扑的保守性.在此基础上,Martinez等[41]还对这些算法进行了时间复杂度分析.针对非线性系统会合的研究也引起了学者的关注.例如,Dimarogonas等[42]针对非万向轮的小车设计了分散反馈控制器,使得由小车组成的MAS系统实现会合.Hui[43]则针对有限时间限制的会合问题,设计了分布式非平滑静态与动态输出反馈控制器.值得注意的是,学术界对于会合问题还有另一种定义.该定义要求MAS中所有个体同时达到相同位置,工程应用中包括航天器的交会[44]等.1.3聚结(Flocking)控制聚结问题在自然界中十分常见,如鸟群的迁徙,鱼群的捕食等.早期Reynolds[15]的工作即是针对聚结现象展开的.此后,针对聚结现象理论研究的研究小组包括:Toner等[45]、Shimoyama等[46]和Levine等[47],但这些研究小组都没有给出聚结行为严格的理论分析.2001年,Leonard等[48]首次将人工势场(Arti?cial potential,AP)方法引入聚结行为的理论分析中,之后该方法成为研究聚结现象的一种重要数学工具.Olfati-Saber[49]首先基于AP 建立了一个完整的理论分析框架.其基本思想是:通过建立子系统之间的局部势能函数,使得全局势能函数(所有局部函数相加值)的最小值对应于期望的聚结状态.此外,通过引入分别代表“自由聚合”、“障碍”和“共同目标”的α-agent、β-agent 和γ-agent,使得聚结可考虑外部障碍和给定的期望轨迹.利用类似的思路,Tanner等[12]针对固定和切换通信拓扑下的二阶积分器线性系统,设计了聚结控制律.该控制律展示了通信拓扑连通性与系统稳定性的关系,并从理论上证明该系统的稳定性对通信网络拓扑的切换具有鲁棒性.1.4编队(Formation)控制编队控制是多个体协调控制中的一个研究热点,其本质是一种几何构型严格的聚结控制[5,8].编队控制的目标在于通过调整个体的行为使系统实现特定几何构型的整体性位移,其数学描述为limt→∞[x j(t)?x i(t)]=x dij,?i,j∈Ilimt→∞[˙x i(t)?c(t)]=0,?i∈I(8)其中,x dij为第i个与第j个个体的期望相对位置矢量,c(t)为期望的整体位移速度.编队控制在诸如多机器人协调[50?52]、无人机(Unmanned aerial vehicle,UAVs)编队[4]及航天器编队[53]中得到广泛研究.尽管诸多应用各具特点,但编队控制亦有其共同点.例如,在绝大多数应用中,系统中的个体具有相同的动力学和相似的局部控制器构架;同时,每个个体的通信和计算能力都是受限的,且通信拓扑在编队控制中都起关键性作用.目前针对编队控制问题已有许多研究方法,大致可分为三类,即主从式、虚拟结构式和行为式.1)主从式1560自动化学报38卷主从式方法中,多个体系统中的一个或几个个体充当“主体”,其他个体充当“从体”并通过与其相邻个体的交互达到跟随主体的目的.通常情形下,主体的动力学特性可简化为式(8)中期望的速度c(t).由于每个个体仅需适应其局部环境,这种方法可自然地利用分布式模式实现.然而,这一方法内在的不足表现在其严重依赖主体的状态:一旦主体失效,整个编队也随之失效,且单个个体的稳定性并不意味着编队的稳定性.因此需要对局部控制器施加严格的稳定性条件.此外,该构架下各个体的动力学扰动具有累加性,因此基于该构架的控制律不具有可裁剪性[54].这意味着随编队中个体数目不断增大,编队不稳定的风险将越来越大.2)虚拟结构式虚拟结构是一种广为采用的编队控制方法[55].该构架下,通过对所有个体状态进行一定的代数运算生成一个虚拟个体.整个编队被视为一个刚体,而虚拟的个体就成为参考点.这样,每个个体的位置和轨迹都可通过该参考点的位置和期望的编队构型明确地计算出来.这一方法易于刻画整个编队的几何构型,并保持准确的编队.但是,由于虚拟结构方法需要集中处理数据,其仅适用于小型编队系统.近几年来,Ren等[5]在利用分布式构架实现虚拟结构式编队方面,做了大量卓有成效的工作.3)行为式行为式方法中针对每个个体预定义一个合适的控制律,使其对应于所有个体可能的状态.尽管该方法是分布式的,但难以定量分析.事实上,基于行为式的编队控制并不多见,典型的文献包括文献[56?57]等.通过上述讨论,我们可以得出编队控制属于分布式控制构架.对于第i个个体,可将其动力学表示为[58]˙x i=f(x i,u i(x i,x neighbor))(9)其中,x neighbor为通过其相邻个体收集的信息,并由通信拓扑确定,u i(x i,x neighbor)为施加在第i个个体上的局部控制器.当系统中个体的动力学相对于系统协调控制的影响可以忽略(如深空探测任务中的航天器[6])时,多个体系统的动力学便可简化为一阶或高阶积分器,可利用一致性控制丰富的研究成果对编队稳定性进行分析.与之相反,在处理诸如近地航天器、无人机编队等系统时,系统自身的动力学特性便不能被忽略,因此不能将这些系统直接简化为简单的积分器型线性系统.这种情况下,需直接利用系统实际的动力学特性,并结合一致性控制中的相关理论具体分析[50?51,53,59].2多个体协调控制中的特殊问题就系统属性而言,多个体系统区别于单体系统最本质的要素在于其通过网络进行信息传递与共享,因此,网络传输中普遍存在的丢包、时延等因素对于系统性能的影响不容忽视.此外,由于系统动力学参数未知或外部扰动引发系统动力学参数发生变化,使得多个体系统动力学存在不确定性.为此,本节将分别针对多个体系统网络不确定性和动力学不确定性这两类特殊问题进行阐述.2.1网络不确定性2.1.1时延网络多个体协调控制中的时延问题在近几年得到广泛关注.按照时延属性可将系统中存在的时延分为两种:通信时延和自时延.2.1.1.1通信网络通信时延是指多个体中个体之间在进行信息交互时,发送方的信息经一定时间滞后才到达接收方.假定系统中第i个个体与第j个个体之间的状态差为e ij,则存在时延T ij(T ij>0)情形下的误差可表述为e ij=x j(t?T ij)?x i(t)(10)因此,第i个个体在t时刻接收到的信息实际上是第j个个体在T ij时刻之前发送的.对于一阶线性系统,最为常见的分布式一致性控制律设计为[20]u i=?j∈N i(t)a ij(t)(x i(t)?x j(t?T ij))(11)其中,a ij>0.由式(11)可知,这类控制律实质上是一个分布式P类时滞控制器.区别于一般单体系统,针对该控制器的理论分析与系统通信拓扑的性质密切相关.例如,对于恒定的通信时延T ij,Moreau 等[20]通过构造Lyapunov泛函证明了:在连续有界的时间间隔T内,如果t+TtA(s)d s对应的连通图包含一个全局可达结点(A(s)为t时刻的近邻矩阵,该矩阵由通信拓扑决定),则该控制律可容许任意有界时延.M¨u nz 等[33]根据Lyapunov-Razumikhin 函数的不变集原理证明:当T ij为时变时延时,如果连接拓扑在连续有界时间间隔内的连通图含有全局可达结点,则系统渐近实现一致性.此外,针对式(11)中T ij为时变值及系统存在外部噪声干扰的情形,刘学良等[60]设计分布式H∞控制器,保证了系统的一致性.对应于连续系统,一阶离散系统的通信时延问10期闵海波等:多个体协调控制问题综述1561题也得到了广泛研究.一种典型的系统描述为[61?62]x i(k+1)=j∈N i∪{i}a ij x j(k?T ij)(12)其中,a ij>0,j∈N i∪{i},a ii≥0且nj=1a ij=1,N i∪{i}表示第i个个体相邻个体的并集,T ij>0为通信时延且T ii=0(即不存在自时延).结合延时状态的有限性,可将该系统通过状态扩维技术转化为非时滞系统:z(k+1)=Γz(k)(13)其中,z为状态x的扩维向量,Γ是一个随机矩阵,其对应的连通拓扑为有向图,Cao等[61]将其定义“时延图”.利用时延图与对应随机矩阵的关系,Wang 等[63]证明了在有界时变时延作用下,系统渐近收敛一致的充要条件是连通拓扑下含有全局可达结点. Xiao等[62]研究了在个体异步接收邻个体信息条件下的一致性问题,并得到结论:在有界通信时延下,系统渐近收敛一致的充要条件是通信拓扑中含有全局可达结点.针对二阶线性系统的分布式协调控制也取得了很多研究成果.类似于无时延情形下的一致性控制律(6),二阶一致性控制律可设计为[64] u i=j∈N i α[(x i(t)?x j(t?T ij))]+j∈N i β[(v i(t)?v j(t?T ij))](14)由式(14)可知,含通信时延的分布式控制器本质上是时滞PD类控制器.所不同的是,针对该控制器的稳定性分析不仅与通信拓扑和控制增益有关,而且与时延T ij有关[64?65].在式(14)这一基本思想下,Lin等[65]利用Lyapunov-Krasovskii定理,推导了平均一致性的线性矩阵不等式(Linear matrix inequality,LMI)条件,并证明了其所设计的分布式协调控制律可容许系统中存在任意有界的恒定通信时延.Zhu等[66]在主从式构架下,研究了T ij 为时变情形下的动态跟随控制.结果表明:当主体为系统全局可达结点且T ij有界时,通过选取适当的控制增益可实现系统的一致性跟随控制.对于一般的非主从式构架,Yang等[64]利用频域分析中的small-μ定理,推导了系统稳定性条件.通过对比可知,利用频域所得的充分性条件相比于时域分析的结果较为保守.此外,针对含通信时延的二阶离散系统一致性问题,Lin等[67]等利用模型转换和非负矩阵的性质证明其所设计的控制律可容许任意有界时延.刘德林等[68]针对具有通信时延的离散时间二阶多体系统的一致性问题,根据广义Nyquist判据和Gerschgorin圆盘定理,得到系统渐近收敛的条件.通过对含有通信时延的二阶积分器线性系统和一阶系统的对比分析,我们可以得到以下结论:1)通过设计合适的协调控制律,两种系统都可实现任意有界时延下的一致;2)为保证系统稳定性,二阶系统的控制增益与通信拓扑和时延都相关,而一阶系统的控制增益则与这两者无关.此外,针对通信时延下非线性系统协调控制的研究也取得了一定成果.例如,佘莹莹等[69]提出了基于一类连续非线性函数的有限时间一致性算法. Chopra等[70]考察了仿射非线性系统的协调控制,并应用无源性理论[71]与波变量理论[72]技术,设计了输出协调控制律.类似地,利用波变量技术与收缩定理,Wang等[73]针对网络互连的非线性系统设计了线性协调控制律.另外,Chopra等[74]和Nu?n o 等[75?76]针对遥操作协调控制进行了广泛的研究, Chung等[53,77]则针对Euler-Lagrange系统的协调控制进行了初步研究.:˙x i=f(x i)+g(x i)u iy i=h(x i)(15)2.1.1.2自时延多个体中的自时延用于刻画包括执行器时延,个体对于自身行为(状态)和邻个体行为的不同反应延迟以及通信与计算的混合延时等等.实际上,由于通信时延与自时延一般同时存在,而且系统仅存自时延是其特例,因此绝大多数文献也将“自时延”默认为两者共存的情况.本文的讨论也循此惯例.假定系统中第i个个体与第j个个体之间的状态差为e ij,则当系统同时存在通信时延T ij(T ij>0)和自时延τij(τij>0)时,e ij可表述为e ij=x j(t?T ij)?x i(t?τij)(16)可见,在t时刻该误差实际上是第i个个体和第j个个体分别在T ij时刻与τij时刻之前的状态差.对于存在自时延情形下的一阶系统,一个经典的协调控制律设计为[19]˙x i=j∈N i(t)αij(t)[x j(t?T ij)?x i(t?τij)](17)该控制律假定通信时延与自时延相同.利用代数图论、矩阵论和稳定性控制理论,文献[19]给出了系统达到平均一致性的充要条件:假定τij=τ为常值,则当τ∈[0,π2λmax(L)]时,系统达到平均一致性,其中λmax(L)是Laplacian矩阵的最大特征值.由于文献[19]针对的是固定,无向和连通的通信拓扑,因此结果相对保守.Xiao等[78]将其推广至更为一般的拓1562自动化学报38卷扑构型.Bliman等[79]则将文献[19]的结论推广至τij为时变时延以及τi j=T ji两种情形,通过偏微分方程(Partial di?erential equations,PDEs)给出了系统平均一致性的充分条件.在假定τij为Markov 链支配的随机时延情形下,Wu等[80]以随机稳定性视角研究了所设计一致性控制律的稳定性.Tian 等[81]将自时延描述为“输入时延”,针对输入时延与通信时延不同的情形进行研究,并得到结论:该类系统的可一致性条件(Consentability)与输入时延有关,而与通信时延无关.此外,Lin等[62]还将文献[19]的结果推广至二阶共连(Jointly-connected)的拓扑情形.最近,M¨u nz利用广义Nyquist理论,对相对度一和相对度二的含自时延线性系统进行了深入研究[82].在假定通信拓扑为无向图的前提下,这一方法将文献[19,79,81]等统一在同一个理论分析框架下,并得到了该条件下相同的结论.文献[82]的研究结果表明:相对于自时延,系统对于通信时延具有更高的鲁棒性,因此自时延对于系统稳定性具有较大影响.同时我们注意到,尚无针对非线性多个体含自时延的研究文献,因此目前该问题仍是多个体协调控制研究的热点和难点.对现有方法进行归纳,可将含时延多个体协调控制的研究方法分为以下三类:1)状态扩维法该方法主要针对离散系统.其主要思想是:通过状态扩维使原系统转变为部分无时延系统,然后根据代数图论和矩阵理论进行系统分析[61?63,67?68,78].2)频域法其主要思想是:通过Laplace变换将时域下的系统动力学转换至频域,然后利用Nyquist定理等频域分析方法对系统稳定性进行分析[19,64,81?82],或利用偏微分分析研究[79].3)时域Lyapunov类分析方法该方法主要针对连续系统.通过合理构造Lya-punov类函数(如。

分布式约束满足问题研究及其进展

分布式约束满足问题研究及其进展

ISSN 1000-9825, CODEN RUXUEW E-mail: jos@Journal of Software, Vol.17, No.10, October 2006, pp.2029−2039 DOI: 10.1360/jos172029 Tel/Fax: +86-10-62562563© 2006 by Journal of Softwar e. All rights reserved.∗分布式约束满足问题研究及其进展王秦辉, 陈恩红+, 王煦法(中国科学技术大学计算机科学技术系,安徽合肥 230027)Research and Development of Distributed Constraint Satisfaction ProblemsWANG Qin-Hui, CHEN En-Hong+, WANG Xu-Fa(Department of Computer Science and Technology, University of Science and Technology of China, Hefei 230027, China)+ Corresponding author: Phn: +86-551-3602824, Fax: +86-551-3603388, E-mail: cheneh@Wang QH, Chen EH, Wang XF. Research and development of distributed constraint satisfaction problems.Journal of Software, 2006,17(10):2029−2039. /1000-9825/17/2029.htmAbstract: With the rapid development and wide application of the Internet technology, many problems ofArtificial Intelligence, for example scheduling, planning, resource allocation etc., are formally distributed now,which turn into a kind of multi-agent system problems. Accordingly, the standard constraint satisfaction problemsturn into distributed constraint satisfaction problems, which become the general architecture for resolvingmulti-agent system. This paper first briefly introduces the basic concepts of distributed CSPs, and then summarizesthe basic and the improved algorithms. Their efficiency and performance are analyzed and the typical applicationsof distributed CSPs in recent years are discussed. Finally, this paper presents the extensions of the basicformalization and the research trends in this area. Recent related work indicates that the future work will focus onthe theoretical research to present the solid theoretical foundation for the practical problems.Key words: constraint satisfaction; distributed AI; multi-agent system; search; asynchronous摘 要: 近年来,随着网络技术的快速发展和广泛应用,人工智能领域中的诸多问题,如时序安排、计划编制、资源分配等,越来越多地以分布形式出现,从而形成一类多主体系统.相应地,求解该类问题的传统约束满足问题也发展为分布式约束满足问题,分布式约束满足已经成为多主体系统求解的一般框架.首先,简要介绍了分布式约束满足问题的基本概念,总结了该问题的基本算法及其改进算法,并对这些算法的效率和性能进行了比较分析.然后,讨论了近年来分布式约束满足问题的若干典型应用;最后,给出了分布式约束满足问题基本形式的扩展和今后的研究方向.分布式约束满足问题最新研究进展表明:今后的工作将着重于面向现实问题求解的理论研究,为实际应用提供坚实的理论基础.关键词: 约束满足;分布式人工智能;多主体系统;搜索;异步中图法分类号: TP301文献标识码: A自1974年Montanari在图像处理中首先提出了约束满足问题(constraint satisfaction problems,简称CSPs)[1]∗ Supported by the National Natural Science Foundation of China under Grant No.60573077 (国家自然科学基金); the Program forNew Century Excellent Talents in University of China under Grant No.NCET-05-0549 (新世纪优秀人才支持计划)Received 2006-03-09; Accepted 2006-05-082030 Journal of Software软件学报 V ol.17, No.10, October 2006以来,约束满足作为一种重要的求解方法在人工智能与计算机科学其他领域的很多问题中都得到了广泛的应用[2],从n皇后、图染色等经典问题到时序安排、计划编制、资源分配等大型应用问题,都可以形式化为约束满足问题进行求解.正因为在人工智能领域中的广泛适用,约束满足问题在理论、实验、应用上都得以深入研究,成为人工智能中很成功的问题解决范例之一.其相关成果一直是人工智能权威期刊《Artificial Intelligence》的热点,并有多个专题对此进行讨论;国内也有很多学者致力于约束满足问题的研究,主要的工作有约束程序理论、设计与应用的研究[3,4]、约束归纳逻辑程序设计等方面的研究[5,6],以及约束满足问题的求解研究[7−9]等等.约束满足问题是由一系列变量、变量相应的值域以及变量之间的约束关系组成,目标是为这些变量找到一组或多组满足所有约束关系的赋值.回溯搜索以及约束一致性检查两种基本思想和引入它们中的各种启发式方法构成了多种约束满足问题求解算法.随着硬件和网络技术的发展,分布式计算环境快速、广泛地在各个领域中得到应用,很多人工智能问题也越来越多地处于分布式计算环境下,使得分布式人工智能成为一个十分重要的研究领域,特别是关系到人工自治Agent间需要相互协调影响的分布式问题.如在多智能体系统(multi-agent system,简称MAS)中,处于同一环境下的Agent间通常存在着某种约束,此时,为各个Agent寻找一组满足它们之间约束的动作组合的分布式人工智能应用问题都可以看作是分布式约束满足问题(distributed CSPs).分布式约束满足问题是变量以及变量间的约束都分布在不同自治Agent中的约束满足问题,每个Agent控制一个或多个变量,并试图决定这些变量的值,一般在Agent内和Agent间都存在约束关系,对变量的赋值要满足所有这些约束.正因为不同的变量和约束是由不同的Agent控制,因此,在这种情形下,将所有Agent控制的变量及相关的约束等信息集中到一个Agent,再用传统的集中式约束满足算法进行求解往往是不充分或者是不可能的,有如下几点原因[10]:(1) 生成集中控制会带来额外开销.如类似于传感网络的约束满足问题很可能自然地分布在由一些同等Agent构成的集合中.这种情况下,对问题进行集中控制就需要增加不出现在原有结构中的额外元素.(2) 信息传递的开销.在很多情况下,约束由复杂的决策过程产生,这些过程是内在于Agent并且不可能被集中控制的.集中式算法需要获得这些约束关系就要承担信息传递的开销.(3) 隐私和安全的保证.在电子商务等情况中,常出现Agent之间的约束是不能泄露给竞争者甚至也不能泄露给集中控制的战略信息的情况.此时,隐私只能在分布式方法中得到很好的保护.(4) 面对失败的鲁棒性.集中控制求解时的失败可能是致命的;而在分布式方法中,一个Agent的失败并不是致命的,其他Agent可以在忽略已失败Agent的情况下找到问题的解.比如在传感网络和基于网络的应用中,当约束求解过程正在进行而参与者可能离开时,都会产生这些问题.从上述原因可以看出:此类分布式环境中的问题需要更有效的解决方法.随着人工智能领域协作式分布问题研究的深入,Yokoo等人在文献[11]中提出了分布式约束满足问题的框架和相应算法.作为一种新的技术,它特别适用于表示及求解规模大、难度高的组合问题.所以,分布式约束满足问题成为人工智能领域的一个研究热点.本文在文献[12]对分布式约束满足问题综述的基础上,不仅介绍了分布式约束满足的问题形式和一系列求解算法,还介绍了近年来在分布式约束满足问题基本形式上的扩展和多主体系统的应用.本文第1节介绍分布式约束满足问题的定义.第2节详述一系列求解分布式约束满足问题的算法,比如异步回溯、异步Weak-commitment搜索、分布式逃逸算法等.第3节介绍相应的应用.最后总结该问题上的一些扩展和类似的工作,如开放式、分布式局部约束满足、隐私安全性等.1 分布式约束满足1.1 约束满足问题约束满足问题是在一定的值域范围内为所有变量寻找满足它们彼此间约束关系的赋值的问题,由变量、变量的值域和变量之间的约束组成.定义1(约束满足问题). 约束满足问题可以形式化为一个约束网[13],由变量的集合、每个变量的值域的集合以及变量间的约束关系的集合来定义,表示为三元组(V,D,C),其中:王秦辉等:分布式约束满足问题研究及其进展2031V是变量的集合{v1,…,v n};D是所有变量的值域的集合,D={D1,…,D n}, D i是变量v i的所有可能取值的有限域;C是变量之间的约束关系的集合C={C1,…,C m},其中每个约束包含一个V的子集{v i,…,v j}和一个约束关系R⊆D i×…×D j.约束满足方法是一种有效的问题求解方法,它为每个变量在其值域中寻找一个赋值,使得所有约束被满足.定义2(约束满足问题的解). 约束满足问题的解是分配给问题中所有变量的一组不违反任何约束的赋值.也即一组对所有变量的赋值S(v1,…,v n)={d1∈D1,…,d n∈D n},∀C r∈C都有S(v ri,…,v rj)={d ri,…,d rj}∈R r.例如,n皇后问题就是典型的约束满足问题.该问题描述为要在n×n的棋盘上摆放n个皇后,使得每一行、每一列和每条对角线上只能有一个皇后存在.图1是4皇后问题的示例以及相应的约束满足问题.(a) (b) (d)Fig.1 4 queens constraint satisfaction problem图1 4皇后及相应的约束满足问题图1(a)表示在4×4的棋盘上放置4个皇后Q1~Q4,即为变量集合;图1(b)表示任意行、列、对角线上不能同时有两个皇后,即为约束关系;图1(c)是相对应的约束满足关系网;图1(d)是该问题的一个解.1.2 分布式约束满足问题分布式约束满足问题是变量和约束都分布在不同自治Agent中的约束满足问题.在约束满足问题定义的基础上,可如下定义分布式约束满足问题:定义3(分布式约束满足问题). n个Agent表示为A1,A2,…,A n,m个变量为v1,v2,…,v m,m个变量的值域为D1, D2,…,D m,变量间的约束仍用C表示;每个Agent有一个或多个变量,每个变量v j属于一个A i,表示为belongs(v j, A i);变量间的约束关系分布在Agent内或Agent之间,当A l知道约束关系C k时,表示为Known(C k,A l).分布在Agent内的约束称为局部约束,而Agent间的约束称为全局约束,局部约束可以通过Agent的计算来处理,全局约束不仅需要Agent的计算,更需要Agent间的通信来处理,因此需要如下的通信模式假设: 假设1. Agent间的通信通过传递消息完成,当且仅当一个Agent知道对方地址时才能够传递消息给其他Agent.假设2. 传递消息的延时是随机但有限的,任何一对Agent间消息接收的顺序与消息发送的顺序是一致的.假设3. 每个Agent只知道整个问题的部分信息.分布式约束满足中的Agent与多智能体系统(MAS)中的Agent有着细小的差别[14],分布式约束满足中的Agent是遵从协作机制来执行决策行为的计算实体;MAS中的Agent自治地决定是否遵从特定的协作机制,并能以结构化的语义消息交换形式与其他Agent进行通信.在将MAS形式化为分布式约束满足问题进行求解时,并不考虑这些区别.每一个Agent负责一些变量并决定它们的值,因为还存在着Agent间的内在约束,所以,赋值必须满足这些约束.分布式约束满足问题的解的形式化定义为:定义4(分布式约束满足问题的解). 当且仅当满足下述条件时,分布式约束满足问题找到了解:∀A i,∀v j存在关系belongs(v j,A i),当v j的赋值是d j∈D j时,∀C k,∀A l,Known(C k,A l)都有C k被满足.也即此时对问题中所有变量的赋值满足Agent间及Agent内的所有约束.2032 Journal of Software 软件学报 V ol.17, No.10, October 2006图2是一个分布式约束满足问题的示例.该问题为分布式图染色问题,从黑、白、灰这3种颜色中选一种分配给Agent 中的节点变量,使得互相连接的节点颜色不同.图中每个Agent 都各有3个变量,变量之间的边就表示彼此间存在着约束关系,该问题不仅Agent 内而且Agent 间都存在着约束关系.(a) (b) (c)Fig.2 Example of distributed constraint satisfaction problem图2 分布式约束满足问题示例图2(a)表示该问题的约束关系网;图2(b)为随机分配着色的初始状态;图2(c)是该问题的一个解.1.3 分布式约束满足与并行/分布式计算的区别分布式约束满足问题看起来与求解约束满足问题的并行/分布式方法[15,16]虽然很相似,但它们从根本上是不同的.并行/分布式方法应用到约束满足问题求解中的目的是为了提高问题的求解效率,针对不同的约束满足问题可以选择任何一种合适的并行/分布式计算机体系结构将问题分而治之,取得较高的求解效率.而在分布式约束满足中,问题的变量和约束等相关信息从问题给定时就既定地分布于各个自治Agent 中,所以,研究的出发点是如何在这种固有的情形下有效地获得问题的解.比如一个大规模的n 皇后问题,可以利用分布式并行计算获得更快的求解速度.而对应到分布式的n 皇后问题,则是很多个不同的Agent 拥有数量不同的皇后,通过自我决策和Agent 间的通信协作来共同达到问题的解.2 求解分布式约束满足问题的算法在分布式约束满足问题提出的同时,Yokoo 就在文献[11]中提出了异步回溯算法.近些年来,其他的分布式约束满足求解算法也得到了进一步的研究,特别是异步Weak-commitment 搜索[17,18]和分布式逃逸算法[19]等.这些算法基本上是由约束满足问题的求解算法而来,是这些传统算法的分布式扩展.但是,因为分布式约束满足问题中Agent 之间也存在着约束关系,所以,Agent 间需要通信是与传统算法的最大区别.分布式约束满足算法有两种最基本的消息需要通信,分别是ok ?和nogood [11].定义5(ok ?消息). ok ?是指Agent 将当前的赋值信息传递给相邻Agent 的消息.定义6(nogood 消息). nogood 是用来传递约束是否发生冲突而产生新约束的消息.在文献[12]中,对各种算法的描述都做了如下假设:(1) 每个Agent 只控制一个变量;(2) 所有的约束都是二元的;(3) 每个Agent 知道所有和自己变量相关的约束.因此,可以不加区分地使用相同标识v i 表示一个Agent 及其变量,用约束网中的有向边表示约束关系,该有向边由发送消息Agent 指向接收消息Agent.对假设2和假设3都可以自然地扩展到一般情形.下面分别介绍基于回溯的异步回溯算法、基于优化迭代的分布式逃逸算法和基于混合算法的异步Weak-commitment 算法.2.1 异步回溯(AB :Asynchronous backtracking )异步回溯算法是由求解约束满足问题的回溯算法而来.所不同的是,异步回溯算法是分布式的、异步的.在异步回溯算法中,每个Agent 都有一个优先顺序,该优先顺序是预先定义好的,一般由Agent 标识的字母顺序来王秦辉等:分布式约束满足问题研究及其进展2033决定,比如,按降序或升序来决定Agent的优先序的高低.在该算法中,每个Agent除了要发送接受ok?和nogood 消息以外,还要维护一个agent_view,这是用来记录其他Agent的当前赋值的.当一个Agent接收到ok?消息时,则检查其赋值与优先顺序高的Agent的当前赋值是否满足约束关系,如果不满足约束关系产生冲突而不一致就改变自己的赋值;如果该Agent值域中没有能与高优先序Agent的赋值相一致的值,就产生一个新的约束关系(也就是nogood),并且将nogood传递给高优先序Agent,这样,高优先序Agent就可以改变自己的赋值.必须注意到:如果Agent不断地改变它们的赋值而不能达到一个稳定状态,那么它们就处于一种无限处理循环,当一个Agent的赋值导致其他Agent改变赋值而最终影响到自己时就可能产生这种无限循环.为了避免这种情况的发生,在算法中,按照标识的字母序为每个Agent定义了优先顺序,ok?只能从高优先序Agent发送给低优先序Agent.当产生nogood时,也是nogood中的优先序最低的Agent接收到nogood消息.另外,每个Agent的行动都是同时异步发生的,而且Agent间的通信是通过消息传递来进行的,所以,agent_view中可能包含有已经无用的信息.因此,每个Agent都需要产生新的nogood进行通信,新nogood的接收方也必须检查在自己的agent_view的基础上与此nogood是否有冲突.因为算法中最高优先序的Agent不会陷入无限处理循环中,文献[18]用归纳法证明了该算法是具备完全性的,也即:如果问题有解存在,那么一定能找到这个解;如果没有解存在,那么算法也会终止,不会陷入无限循环.近年来,有很多工作都对异步回溯算法进行了改进.在对算法进行扩展时,文献[20]采用了Agent的动态重排序;文献[21]引入了一致性维护;文献[22]提出了不存储nogood消息的分布式回溯算法.这些算法与基本的异步回溯算法相比只是存储nogood消息的方式不同,它们都需要在未相连的Agent之间添加通信连接来检测已经无用的消息.而文献[23]中提出了一种新的异步回溯算法来避免在初始未相连的Agent之间动态地增添新的约束,这样就可以避免将一些信息传递给不需要知道的Agent,从而提高效率.文献[24]从另一个角度提出了如何利用值聚集来减少信息阻塞以及如何利用弧一致维护来提高异步分布式下问题求解的有效性.2.2 异步Weak-Commitment搜索(AWS:Asynchronous weak-commitment search)异步回溯算法的局限在于Agent的优先顺序是预先定义好的,是静态的.如果高优先序Agent的赋值选择得不好,那么,低优先序Agent就要进行穷尽查找来修正不利的赋值.异步Weak-commitment搜索算法[17,18]的两个基本思想是:为了减少不利赋值的风险而引入了最小冲突启发;更进一步地,Agent的优先级顺序是可以动态改变的,这样,不利赋值不需要穷尽搜索就能够得到更正.最小冲突启发是指Agent选择值域中那些与其他Agent的赋值产生最少冲突的值作为自己的赋值.而为了使Agent的优先级顺序能够动态改变,特别地为Agent引入了优先值,优先值是非负整数,优先值大的Agent具有较高的优先顺序,优先值相等的Agent的优先顺序由它们所标识的字母序来决定.初始时,Agent的优先值均为0,当Agent的赋值与约束发生冲突而不一致时,该Agent的优先值就变为相邻Agent中的最大优先值再加1.与异步回溯算法相比,异步Weak-commitment搜索算法的不同在于:(1) 异步回溯中每个Agent只将变量赋值发送给约束相关的低优先级Agent;而异步Weak-commitment搜索中每个Agent将变量赋值发送给约束相关的所有Agent,无论优先级的高低;(2)ok?消息不仅用来传递Agent的当前赋值,还用来传递Agent的优先值;(3) 如果当前的赋值与agent_view不一致,则Agent用最小冲突启发来改变赋值;(4) 如果Agent不能找到与自己的agent_view一致的赋值,就发送nogood消息给其他Agent,同时改变自己的优先值.如果Agent不能生成新的nogood,那么就不改变自己的优先值,并等待下一条消息.第4步的过程是保证算法的完全性所必需的.因为优先值的改变只有在新的nogood产生时才发生,而可能的nogood的数量是有限的,优先值不可能无限地改变,所以到了某个时间之后,优先值就稳定下来,此后,过程就与异步回溯算法一样,故而算法是完全的.为了保证算法的完全性,Agent要记录所有目前已知的nogood,实际操作时,可以限制记录nogood的数目,比如每个Agent只记录固定数目的最近发生的nogood消息.正如前面所提到的假设,这些算法中的Agent都只含有一个变量,对于解决Agent含有多个变量的问题,无论是采用先让Agent找到自己局部问题的所有解后再将问题重新形式化为分布式约束满足问题来求解,还是让2034 Journal of Software软件学报 V ol.17, No.10, October 2006Agent为每个局部变量生成一个虚拟Agent再来模拟这些Agent的动作来求解,对大规模问题而言,既没有效率也不能扩展.文献[25]对异步Weak-commitment搜索算法进行了扩展,利用变量顺序来解决多个局部变量的问题,称为Multi-AWS.它的特点是Agent按顺序来改变自己变量的值,当某个变量不存在满足所有与高优先序的变量有关的约束时,就增加该变量的优先值.不断反复该过程,当Agent中所有局部变量都与高优先序变量满足约束时,就传递值改变消息给相关的Agent.2.3 分布式逃逸(DB:Distributed breakout)在最小冲突回溯等约束满足算法中的爬山(hill-climbing)搜索策略,有时会使求解过程陷入局部最小(local-minima)状态.local-minima状态就是一些约束没有被满足从而出现了冲突,但是这些冲突的数目不能通过单独改变任何一个变量的值来减少.文献[26]中提出的逃逸算法是一种跳出local-minima状态的方法,算法中为每个约束定义了权值,所有冲突约束的权值总和作为一个评估值.当陷入local-minima状态时,逃逸算法增加当前状态中冲突约束的权值,使得当前状态的评估值高于其他邻接的状态,从而跳出local-minima状态,开始新的搜索.文献[19]在此基础上通过以下两个步骤来实现分布式逃逸:(1) 始终保证评估值是逐步提高的:相邻的Agent对可能会提高的评估值进行交流,只有能够最大提高评估值的Agent才有权改变自己的值.如果两个Agent不相邻,那么它们有可能同时改变自己的值;(2) 与检测整个Agent是否陷入local-minima不同的是,每个Agent检测其是否处于quasi-local-minima状态,这是比local-minima要弱一些的条件,并且能够通过局部通信而检测到.定义7(Agent A i处于quasi-local-minimum状态). A i的赋值使部分约束产生冲突,并且A i和所有A i邻居的可能提高值均为0.在分布式逃逸算法中,相邻的Agent之间有两种消息的通信:ok?和improve. improve消息用来对可能提高的评估值进行通信,Agent在整个过程中处于wait_ok?和wait_improve两种交替状态.分布式逃逸算法有可能陷入无限循环当中,因而不能保证算法的完全性.文献[27]在对分布式逃逸算法进行扩展时,不仅提出了解决Agent有多个局部变量的Multi-DB算法,还对此算法引入了两种随机方式.一种是利用了随机跳出技术的Multi-DB+算法,另一种是在Multi-DB+中又引入随机行走的Multi-DB++算法.这些算法比其他异步方法有更好的扩展性,但有时也会有更差的性能.2.4 几种算法的比较上面介绍了几种求解分布式约束满足问题的基本算法,这些算法有各自的特点和适用性,也各有优点和局限性,表1是就上述算法在完全性及解决多局部变量方面的一个定性比较.Table 1Comparison of algorithms for solving distributed CSPs表1几种基本分布式约束满足问题求解算法的定性比较Algorithm AB AWS Multi-AWS DB Multi-DBNo NoCompleteness Yes Yes YesMulti local variables No No Yes No Yes文献[12]对基本的异步回溯(AB)、异步Weak-commitment(AWS)和分布式逃逸(DB)算法进行了性能比较.通过离散的事件模拟来评估算法的效率,其中每个Agent维护自己的模拟时钟,只要Agent执行一个计算周期,其时间就增加一个模拟时间单元.一个计算周期包括读取所有消息、执行局部计算和发送消息.假设一个消息在时间t发布,则对接收者来说,在时间t+1时可用.最后,通过解决问题所需的计算周期数量来分析算法的性能.表2是用分布式图染色问题来评测的结果,其中:Agent(变量)的数目n=60,90和120;约束的数量为m=n×2,可能的颜色数目为3.总共生成10个不同的问题,对每个问题执行10次不同的初始赋值,并且限制周期最大为1000,超过后就终止算法.表中列出了算法求解所需的平均周期和求解成功的比例.明显地,AWS算法要优于AB算法,因为在AWS算法中,不需要执行穷尽查找就能修正错误的赋值.在表3和表4比较AWS与DB算法时,Agent(变量)的数目为n=90,120和150,约束的数量分别为m=n×2王秦辉等:分布式约束满足问题研究及其进展2035和m=n×2.7两种情况,可能的颜色数仍然为3.当m=n×2时,可认为Agent间的约束是比较稀疏的;而当m=n×2.7时,则被认为是能够产生阶段跳跃的临界状态[28].Table 2 Comparison between AB and AWS表2算法AB和AWS的比较nAlgorithm60 90 120Ratio (%) 13 0 0ABCycles 917.4 --Ratio (%) 100 100 100AWSCycles 59.4 70.1 106.4Table 3Comparison between DB and AWS (m=n×2)表3算法DB和AWS的比较(m=n×2)nAlgorithm90 120 150Ratio (%) 100 100 100DBCycles 150.8 210.1 278.8Ratio (%) 100 100 100AWSCycles 70.1 106.4 159.2Table 4Comparison between DB and AWS (m=n×2.7)表4算法DB和AWS的比较(m=n×2.7)nAlgorithm90 120 150Ratio (%) 100 100 100DBCycles 517.1 866.4 1175.5Ratio (%) 97 65 29AWSCycles 1869.6 6428.4 8249.5从表3和表4中可以看出:当问题为临界困难时,DB算法要优于AWS算法.而对于一般情形,AWS算法要优于DB算法.因为在DB算法中,每个模式(wait_ok?或者wait_improve)都需要一个周期,所以每个Agent在两个周期内至多只能改变一次赋值;而在AWS算法中,每个Agent在每个周期内都可以改变赋值.如前所述,近年来的很多工作都对这些基本算法进行了改进或者扩展,在性能和效率方面也有各自的特点.文献[20]中的实验表明:对Agent进行动态重排序的ABTR(ABT with asynchronous reordering)算法的平均性能要优于AB算法,说明额外增加动态重排序的启发式消息实际上是可以提高算法效率的.文献[24]通过实验表明:采用值聚集的AAS(asynchronous aggregation search)算法的效果要稍好一些.虽然在查找第一个解时,不利用值聚集的效果会更好,但是如果解不存在,那么AAS的性能总是要优于不采用值聚集的算法,因为此时需要扩展到整个搜索空间,AAS可以减少消息序列,也就是减少存储nogoods消息的数目.更进一步地,使用了bound- consistency一致维护技术的MHDC(maintaining hierarchical distributed consistency)算法在实验中的整体性能要比AAS有很大的提高.通过图染色实验,文献[25]比较了算法Multi-AWS,AWS-AP(Agent priority)和Single-AWS,对于Agent有多个变量的情况,Multi-AWS算法在执行周期以及一致性检查数目上都要优于其他两种算法,而且随着Agent或变量的增多,其性能就越优于AWS-AP算法和Single-AWS算法.这是因为在解决Agent有多个变量的问题时,AWS-AP算法和Single-AWS算法需要增加额外的虚拟Agents来将多局部变量分布化,这样就增加了Agents 间的通信,从而使性能降低.文献[27]首先比较了Multi-DB和Multi-AWS算法.Multi-DB算法随着变量数目的增多,效率会越来越优于Multi-AWS算法,在很多情形下会有至少1个数量级的提高,但是成功率却较低.原因是由于Multi-DB算法在查找过程中固有的确定性使算法缺少了随机性.对于实验中Multi-DB++算法的效率优于Multi-DB算法,可以得出结论:多Agent搜索进程中的停滞可以通过添加随机行走来避免.。

相关主题
  1. 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
  2. 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
  3. 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。

0005-1098/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. /10.1016/j.automatica.2013.11.023
Article history: Received 3 January 2013 Received in revised form 11 September 2013 Accepted 10 October 2013 Available online xxxx Keywords: Event-based consensus Multi-agent systems General linear systems
Huang, 2012; Wieland, Sepulchre, & Allgöwer, 2011; Zhang & Tian, 2012). In practice, autonomous agents such as mobile robots are often equipped with digital microprocessors which coordinate the data acquisition, communication with other agents, and control actuation. Thus, it is necessary to implement controllers on a digital platform. In other words, consensus protocols can only be updated at discrete time instants. A commonly used approach in the present literature is time-scheduled control, which might be conservative in terms of the number of control updates, since the constant sampling period has to guarantee stability in the worst-case scenario. In order to overcome the conservativeness of time-scheduled control, event-based control is proposed in Åström (2008) and Åström and Bernhardsson (2002), where control updates are determined by certain events that are triggered depending on the agents’ behavior. The event-based control algorithm has been adopted for control engineering applications (Lunze & Lehmann, 2010; Mazo & Tabuada, 2011; Tabuada, 2007; Wang & Lemmon, 2011) such as wireless networks and networked control systems. Event-based control strategies appear to be suitable for cooperative control tasks of multi-agent systems. Following the ideas proposed in Tabuada (2007), Dimarogonas et al. developed a decentralized event-based strategy to determine control updates in Dimarogonas, Frazzoli, and Johansson (2012) and Dimarogonas and Johansson (2009a,b). A limitation of the control strategies presented in Dimarogonas et al. (2012) and
article
info
abstract
In this paper, the event-based consensus problem of general linear multi-agent systems is considered. Two sufficient conditions with or without continuous communication between neighboring agents are presented to guarantee the consensus. The advantage of the event-based strategy is the significant decrease of the number of controller updates for cooperative tasks of multi-agent systems involving embedded microprocessors with limited on-board resources. The controller updates of each agent are driven by properly defined events, which depend on the measurement error, the states of its neighboring agents and an arbitrarily small threshold. It is shown that the controller updates for each agent only trigger at its own event time instants. A simulation example is presented to illustrate the theoretical results. © 2013 Elsevier Ltd. All rights reserved.
1. Introduction Multi-agent systems have been used to solve efficiently a variety of problems, such as search and rescue, exploration and monitoring tasks. A particular focus of multi-agent coordination is consensus, which requires all agents to achieve the desired common goal using only neighboring information. The consensus problem has been extensively studied in the literature (Cao, Yu, Ren, & Chen, 2012; Ren & Beard, 2008; Shamma, 2007) and many references therein. It should be mentioned that most of the previous papers on consensus are devoted to single-integrators and double-integrators (Gao & Wang, 2011; Jadbabaie, Lin, & Morse, 2003; Li & Jiang, 2009; Liu, Guan, Shen, & Feng, 2012; Popov & Werner, 2012; Ren & Beard, 2005; Zhu & Cheng, 2010), just to name a few, while some authors have also studied multi-agents with higher-dimensional linear models (Ma & Zhang, 2010; Su &
✩ The work is supported partly by National Natural Science Foundation of China under Grant 61004042, 61074026, partly by the U.S. National Science Foundation grants DMS-0906659 and ECCS-1230040 and partly by Research Grants Council of Hong Kong City U-113311. The work was done when the first author was visiting the Polytechnic Institute of New York University. This paper has not been presented at any IFAC meeting. This paper was recommended for publication in revised form by Associate Editor Riccardo Scattolini under the direction of Editor Frank Allgöwer. E-mail addresses: zhuwei@ (W. Zhu), zjiang@ (Z.-P. Jiang), megfeng@.hk (G. Feng). 1 Tel.: +86 23 6247 1164; fax: +86 2362471795.
相关文档
最新文档