具有复杂动力学的多智能体系统一致性控制及其应用

CORC > 自动化研究所 > 中国科学院自动化研究所 > 毕业生 > 博士学位论文

题名	具有复杂动力学的多智能体系统一致性控制及其应用
作者	程龙
学位类别	工学博士
答辩日期	2009-06-02
授予单位	中国科学院研究生院
授予地点	中国科学院自动化研究所
导师	侯增广
关键词	多智能体系统一致性自适应控制神经网络工业机器臂 Multi-agent system consensus adaptive control neural networks industrial manipulator
其他题名	Consensus Control of Multi-Agent Systems with Complex Dynamics and Related Applications
学位专业	控制理论与控制工程
中文摘要	多智能体系统一致性控制是当前多智能体协调领域的代表性问题，也是其它分布式控制和估计的研究基础。作为一种新兴的交叉学科问题，多智能体一致性控制能够解释自然界和人类社会中的一些自组织现象和群集行为，同时在工业和国防等领域也具有很强的实际应用前景。当前大部分对多智能体系统一致性控制的研究工作都假设智能体具有确定性的一阶/二阶积分动力学模型，然而实际应用中往往存在强非线性、不确定性和外界扰动等因素，基于简单智能体模型的控制算法不能很好的解决这些问题。因而研究具有更复杂动力学模型的多智能体系统一致性控制在理论上和实践上都有重要价值。本文在综述当前研究现状的基础上，以矩阵论、自适应控制、神经网络等为主要工具，研究了具有复杂动力学特性的多智能体系统一致性控制。论文的主要贡献包括： 1. 针对具有一般连续时间线性时不变动力学模型的多智能体系统，提出了一种基于观测器的一致性算法，给出了闭环多智能体系统可达到一致状态的充分必要条件和具体的控制算法设计策略。该算法不要求智能体的状态完全可测，并能保证各智能体的状态最终都收敛到一个静态的一致值。所得结果可以推广解决具有离散时间线性时不变动力学的多智能体系统一致性问题。 2. 在满足“不确定性参数可线性化”假设下，针对一类具有不确定性动力学的多智能体系统，提出了一种分布式自适应一致性算法，并证明了闭环多智能体系统的稳定性。利用backstepping技术把所得结论推广到具有更高阶动力学模型的智能体的情况。最后在具有不确定运动学/动力学的多机器臂系统中验证了所提出的一致性算法的有效性。 3. 提出了一种基于神经网络的鲁棒自适应算法，解决了一类动力学具有不确定性和外部扰动的多智能体系统一致性问题。克服了“不确定性参数可线性化”限制，避免了回归矩阵的推导与计算。证明了闭环多智能体系统的稳定性，并指出通过恰当设定算法参数，多智能体系统的一致性误差可以减小到任意指定的范围内。最后推广所得结果，解决了具有时变连通拓扑的多智能体系统一致性问题。 4. 研究了具有不确定性动力学的多智能体系统leader-following一致性问题，提出了基于神经网络的鲁棒自适应控制器。所提出的一致性算法仅利用邻居智能体的信息，并且leader智能体的状态是可随时间变化的。证明了所有following智能体都能以给定的精度跟踪leader智能体的状态轨迹。最后扩展所得结果，分析了多智能体系统具有时变拓扑的情况。
英文摘要	Consensus control of multi-agent systems is a representative problem in the field of multi-agent coordination, and is also the basis of many other decentralized control and estimation problems. As a novel interdisciplinary subject, the consensus control of multi-agent systems can explain many self-organization phenomena and swarm behaviors in the nature and human being society, and also has wide practical applications in the industry and national defence. So far most research on the multi-agent consensus problem has to assume that the agent has the deterministic first-order/second-order integrator dynamic model, however, strong nonlinearity, uncertainty and external disturbance are inevitable in the practical applications, and the consensus algorithm based on the simple agent dynamics cannot solve these problems. Therefore, the study on the consensus problem of multi-agent systems with complex dynamics has the significant meaning both in theory and in practice. On the basis of review and summary of the corresponding research on the multi-agent consensus problem, this thesis employs the matrix theory, adaptive control and neural networks as the major tools, studies the consensus problem of multi-agent systems with complex dynamics. The major contributions of this thesis are as follows: 1. An observer-based consensus algorithm is proposed for the multi-agent system with the general continuous-time time-invariant linear dynamics. The necessary and sufficient conditions on the consensus of the closed-loop linear multi-agent system are given, and specific controller design strategy is provided. The proposed algorithm does not need the complete state information of agents, and can guarantee all the agents' states can be convergent to a static common value. Finally, the obtained results are extended to the case where agents have the discrete-time time-invariant linear dynamics. 2. Under the assumption of ``linearity-in-parameters'', a decentralized adaptive consensus algorithm is proposed for a class of multi-agent systems with uncertain dynamics. Theoretical analysis shows the stability of the closed-loop multi-agent system. In addition, the proposed algorithm is extended to the case of agents with higher-order dynamics by the backstepping technique. At last, the effectiveness of proposed method is illustrated by the consensus control of multi-manipulator systems with uncertain kinematics and dynamics. 3. A neural-network-based robust adaptive algorith...
语种	中文
其他标识符	200618014628022
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/6210]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	程龙. 具有复杂动力学的多智能体系统一致性控制及其应用[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2009.