基于事件驱动的分布式一致性控制研究

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	基于事件驱动的分布式一致性控制研究
作者	马宏文1,2
学位类别	工学博士
答辩日期	2017-05-24
授予单位	中国科学院大学
授予地点	北京
导师	刘德荣
关键词	一致性, 分布式控制算法, 事件驱动机制, 多智能体系统, 神经网络
中文摘要	随着控制系统复杂度的增加和规模的扩大，设计单一系统的控制算法已不能满足当前需求。近年来分布式一致性控制成为解决这一问题的有效方法，其重点讨论如何设计分布式控制器使多智能体系统最终趋于一组共同的目标。多智能体系统的状态可以是物理空间中的地理位置，也可以是社交网络中的舆论观点，甚至是智能电网中的电压幅度和频率。采用分布式控制算法解决实际应用问题，具有很强的鲁棒性和可靠性，且具有较高的求解效率，但其中不可避免需要考虑到子系统间的通信问题。如何在保证系统性能的同时减少通信次数以节约信道资源，并降低能量损耗的问题成为目前研究所面临的挑战。事件驱动技术是一种基于驱动函数触发控制器改变控制输出的机制。其有别于传统周期性控制需要不断更新控制输出的特性，应用于分布式一致性控制中，可以有效地减少智能体间通信的频次，提高信道资源的利用率。基于事件驱动的分布式一致性控制算法，既充分利用了多个体间协调控制所具有的聚合优势，亦降低了通信资源的消耗，保证在低能耗下实现对复杂系统的控制。本论文包含以下工作及贡献： 1. 研究了一类含时延和外部噪声干扰的非线性多智能体系统，设计了一种分布式鲁棒自适应一致性控制算法。使用径向基神经网络学习未知非线性系统的特性。通过Lyapunov-Krasovskii 泛函和Young 氏不等式，消除时延所带来的负面影响，进而保证了多智能体系统的稳定性。因为外部噪声干扰会降低系统的性能，所以引入一个鲁棒控制项来处理噪声干扰。最后，将设计方法推广到高阶系统的情况，并通过仿真例子和多机械臂实例验证了该分布式一致性控制算法的有效性。 2. 研究了非线性时延多智能体系统利用事件驱动和神经网络技术达到连通度保持的一致性问题。借助分而治之的思想，分布式控制器被分成五个部分来设计。事件驱动机制的引入减少了智能体之间的通信次数，节约能耗。通过设计一个合适的事件驱动函数，Zeno 现象得以排除从而提高了多智能体系统的稳定性。把势能函数的阈值参数设定为合适的值可以保证多智能体系统在趋于一致性过程中的连通度。 3. 研究了两种群组一致性的控制方法，分别是集中式事件驱动控制算法和分布式事件驱动控制算法。在分布式事件驱动控制算法中，通过计算相应参数的最大值和最小值，驱动函数得以简化从而减少了对系统存储空间的占用，这种简化对于片上系统有限的存储空间来说尤其重要。 Zeno现象在集中式的情况下可以被排除。 4. 研究了一种异质非线性多智能体系统的输出一致性问题，在多智能体系统和参考输出信号之间增加了两层虚拟层。在第二层虚拟层中引入事件驱动机制来减少智能体间的通信次数，使得所有智能体可以跟踪上外部参考信号。考虑了含输入时延的情况以适应网络通信中数据丢包和信道阻塞对多智能体系统的影响。针对多智能体系统的异质性，为每个智能体设计了一套内部模型用于表征其非线性特性。使用FitzHughNagumo(FHN)模型验证了算法的有效性。
英文摘要	With the increase of the complexity and the expansion of the scale of the control systems, the design of control algorithms for single systems cannot meet the current requirements. In recent years, the distributed consensus control algorithm has become an effective method to solve this problem. It focuses on how to design a distributed controller to make the multi-agent systems reach a common goal. The goal of a multi-agent system can be the location of the physical space, the opinion of the social network, and the voltage amplitude and frequency in the smart grid. The distributed control algorithm is used to solve the practical application problems, which has strong robustness and reliability, and has high problem-solving efficiency. However, it is inevitable to take the communication problem between subsystems into account. How to reduce the number of communications while ensuring system performance, in order to save channel resources and reduce energy loss, becomes the challenge of the current research. Event-triggered technique is the mechanism that triggers a controller to change control output based on an event-triggered function. It is different from the characteristics of traditional periodic control which needs to constantly update the control output. When combined with the distributed consensus control algorithm, it can effectively reduce the frequency of communication between agents to improve the utilization of channel resources. The event-triggered distributed consensus control algorithm not only makes use of the advantages of multiple inter-body coordination control, but also reduces the consumption of communication resources and ensures the control objective of complex systems under low energy consumption. This thesis mainly contains the following contributions. 1. A class of nonlinear multi-agent systems with time delay and external noises is studied, and a distributed robust adaptive consensus control algorithm is designed. By using the radial basis function neural networks, the characteristics of the unknown nonlinear systems are learned. The negative effects caused by the time delay can be eliminated by the Lyapunov-Krasovskii functional and Young’s inequality, and the stability of the multi-agent systems is ensured. Because external noises can degrade the performance of the systems, a robust control is introduced to handle this problem. Finally, the design method is extended to the case of high-order systems, and the effectiveness of the distributed consensus control algorithm is verified by simulation examples and multi-manipulator examples. 2. Consensus of connectivity preservation in time-delayed nonlinear multiagent systems is investigated by using event-triggered technique. With the idea of divide and conquer, the distributed controller is divided into five parts. The introduction of event-triggered mechanisms reduces the number of communications between agents and saves energy consumption. By designing a suitable event-triggered function, the Zeno phenomenon is eliminated to improve the stability of the multi-agent systems. Setting the threshold parameter of the potential energy function to a proper value ensures the connectivity of the multi-agent systems in the process that tends to the consensus point. 3. Two control methods of group consensus are studied. They are centralized event-triggered control algorithm and distributed event-triggered control algorithm. In the distributed event-triggered control algorithm, by calculating the maximum and minimum values of the corresponding parameters, the event-triggered function is simplified to reduce the occupancy of the system memory space. This simplification is particularly important for the limited storage space of the on-chip system. Zeno phenomenon can be excluded in the case of centralized control. 4. Output consensus of a heterogeneous nonlinear multi-agent system is investigated, and a two-fold virtual layer is added between the multi-agent system and the reference output signal. In the second virtual layer, eventtriggered mechanism is introduced to reduce the number of communication between agents, so that all the agents can track the external reference signal. To accommodate the impact of data packet loss and channel congestion in multi-agent systems, input time delay is considered. Aiming at the heterogeneity of multi-agent systems, an internal model is designed for each agent to characterize its nonlinearity. The validity of the algorithm is verified by the FitzHugh-Nagumo (FHN) model.
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/14616]
专题	毕业生_博士学位论文
作者单位	1.中国科学院自动化研究所 2.中国科学院大学
推荐引用方式 GB/T 7714	马宏文. 基于事件驱动的分布式一致性控制研究[D]. 北京. 中国科学院大学. 2017.