【作者】 郭亚锋；

【导师】 李少远；

【作者基本信息】 上海交通大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 在网络控制系统中,通常对象和控制器通过通信网络进行连接,该网络可能同时也为其他应用提供信息传输服务。网络控制系统的主要优点是成本低,重量轻,可靠性高,易于安装维护等。因此,网络控制系统在许多领域得到应用,如移动传感器网络、制造系统、遥操作机器人、航天系统等等。可是,引入网络会出现一些新问题:例如网络时延、数据丢包和信号量化等。本文主要针对存在网络时延和(或)数据丢包的网络控制系统,对它的分析和设计进行了深入的探讨和研究。研究了具有随机丢包的网络控制系统的可镇定性和可检测性问题,其中随机丢包建模为当前常用的独立同分布伯努利序列。考虑了两种常用补偿丢失信号的方法:用“零信号”和用“最新收到的信号”。首先讨论了用“零信号”的网络控制系统,给出了此类网络控制系统可镇定的几个充要条件。并提出了可镇定程度的概念,它有助于进一步分析网络传输概率与网络控制系统的可镇定性之间的关系。由于镇定和检测之间的对偶性,这些结果或概念可方便地推广到网络控制系统的可检测问题。对于单输入单输出(SISO)系统,给出了保证系统可镇定或可检测所需要的最小网络传输概率的解析结果,结果表明镇定或检测同一个(SISO)系统,所需要的最小传输概率相等,并且最小传输概率只与被控对象的不稳定极点有关。从理论上证明了最小传输概率与系统的不稳定程度呈严格的正相关,即系统越不稳定,镇定或观测它所需要的传输概率越大。随后给出了用“最新收到的信号”的SISO网络控制系统可镇定的必要条件。从理论上证明了,用“最新收到的信号”补偿丢包对系统稳定性的影响不比用“零信号”的效果更好。另外,研究了基于状态观测器的状态反馈控制器在网络中最佳位置设定问题。当这类控制器配置在最佳网络节点上时,通过选择适当的控制器参数,可以使闭环系统在保证稳定的情况下,能容忍的整个网络的丢包率最大。通过仿真例子验证了所提结果的有效性。对同时存在网络时延和数据包丢失的离散时间网络控制系统进行了稳定性分析,提出了新的时延相关的稳定判据。与现有的一些结果相比,本文提出的稳定判据主要优点是具有较低的保守性,这主要是在推导过程中利用了凸集上凸函数的性质,不需要像现有一些方法那样把时变时延和其上下界的差同时放大为时延的上下界之差,因此减少了保守性的引入。本文通过理论证明和例子比较都验证了所提稳定判据的优越性。在提出的稳定判据基础上,还给出了状态反馈控制器的设计方法。并对当前求解控制器常用的CCL算法提出了改进。最后,通过仿真示例验证了所提出的控制器设计方法也具有较低的保守性。对于同时存在网络时延和数据包丢失的离散时间网络控制系统,提出了一种新的网络预测控制方法。和现有的网络预测控制方法相比,本文提出网络预测控制方法具有如下优点:本文方法能同时有效补偿网络时延和随机丢包的影响,利用本文方法所得到的闭环系统可镇定的条件大大降低;实现了状态反馈控制器和状态预测器满足分离原理,从而可以方便地设计保闭环系统稳定的网络预测控制器。通过理论证明和仿真例子验证了所提方法的优越性。考虑了网络控制系统中的H∞滤波问题。在这种情形下,系统的测量信号通过存在随机时延的网络传输到滤波器,滤波器通过受到网络时延的测量信号估计系统的被估输出,这里随机时延满足独立同分布的伯努利序列。首先提出了一种新的滤波误差系统表示形式,其优点是把具有时滞状态的系统转化为无时滞的随机参数系统,从而避免利用Lyapunov-Krasovskii泛函推导系统的H∞滤波性能,减少了保守性的引入,并且证明了提出的性能分析判据比现有的结果保守性低。在滤波器的设计方法上,提出了一种新的参数化设计方法,从理论上和例子对比都验证了比现有设计方法的保守性低。另外,通过引入松弛变量的策略,又提出了进一步改进的滤波器设计方法,并通过例子进行了验证。更多还原

【Abstract】 In a networked control system (NCS), plant and controller are typically connected via a communication network which may be shared with other applications. The main advantages of NCSs are low cost, reduced weight, simple installation and maintenance, and high reliability. Consequently, NCSs have been applied in a broad range of areas such as mobile sensor networks, manufacturing systems, teleportation of robots, aircraft systems, etc. however,the usage of the network may lead to some new problems, such as network delay, data packet loss, signal quantization, etc. In this paper, the analysis and design of networked control systems with network delay and (or) packet dropout are considered. The main contents are as follows:Stabilizability and detectability of NCSs with random packet dropout are investigated, where the network packet-loss is modeled as an i.i.d. Bernoulli process. Two commonly used modes compensating for signal-loss—using a zero signal and using the latest available signal are compared. First NCSs with the“zero signal”mode is considered. Several necessary and sufficient conditions for stabilizablity are established. A notion of stabilizable degree is proposed, which help to further analyze the relationship between network transmission probability and stabilizability of the networked control systems. Due to the duality between stabilizability and detectability, the above results and notions about the stabilizability can readily be extended to solve the detectabiliy. For the SISO systems, an explicit formula for the least required transmission probability for the purpose of stabilizing or detecting a networked control system is derived. The result shows that the least required transmission probabilities are equal each other when stabilizing or detecting a networked control system. And it is merely dependent on the unstable poles of the system. It is theoretically validated that the least required transmission probability is strictly positive interrelated with the degree of instability of the system. That is, the more the degree of instability of a system is, the more the required transmission probability will be. Then a necessary condition for stabilizablity of NCSs with the“latest available signal”mode is proposed. It is theoretically proved that the“latest available signal”mode can not better compensate the effect of loss data than the“zero signal”one. In addition, optimizing location of state-observer based controller in the network is considered. When the controller is located the optimal node, by selecting appropriate controller parameter, the closed-loop system can be stable with maximum loss probability of the network. The efficiency of the proposed results is illustrated with some simulation examples.The stability of the discrete-time networked control systems is analyzed when there are both network delay and packet loss in the network. A new delay-dependent stability criterion is proposed. Compared with the some existing results, the advantage of the proposed stability condition lies in its less conservativeness. Differently to pervious methods, by utilizing the properties of convex function, the differences of the delay and its upper and lower bound need not simultaneously be enlarged as the differences of its upper and lower bound. Thus the conservativeness of the result is reduced. Both theoretical proof and example show the advantage of the proposed stability criterion. Based on the stability criterion, the design method of the state-feedback controller is provided. Moreover, the CCL algorithm has been improved. Simulation example is given to show that the proposed design method of controller is also less conservativeness than the existing ones.A new networked predictive control method is proposed for the discrete-time networked control systems, which simultaneously contain network delay and packet loss. Compared with the existing methods, the proposed method has the following merits. The proposed method can more efficiently compensate the effect of network delay and packet dropout. Thus the stabilizability condition of closed-loop systems is very easily satisfied. Moreover, by the proposed method, the state-feedback controller and state predictor can be separately designed, so the networked predictive controller can easily be design to guarantee the stability of closed-loop system. By theoretical analysis and simulation example, the advantages of the proposed method are verified.The H∞filtering in the networked control systems is considered. Under this case, the measured signals of the system are transmitted to the filter through the network, which are inevitably subjected to the effect of the network delay. The network delays are assumed to satisafy an i.i.d. Bernoulli process. A new form of the filtering error system is proposed, which translates the delay system into a delay-free system. Therefore, Lyapunov-Krasovskii is not required when the H∞filtering performance is analyzed so that the conservativeness of result of performance analysis is reduced. Moreover, the proposed H∞performance criterion is theoretically proved that it is less conservative than the existing one. Based on this performance criterion, a new parametric design method of the filter is proposed. It also is proved that it is less conservative than the existing one. In addition, by introducing some slack variables, an improved filter design is presented and its efficiency is illustrated by an example.更多还原