【作者】 于晓明；

【导师】 蒋静坪；

【作者基本信息】 浙江大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 近年来,网络控制系统已在国防、工业控制和医学等领域中获得了广泛的应用,同时,也成为控制理论学术界的研究热点之一,并取得了一定的研究成果。与传统的点对点控制系统相比,网络控制系统具有以下主要优点：能方便地利用网络资源实现信息资源共享；能进行远程监测与远程控制；增强了控制系统的灵活性,减少了系统的布线,降低投资；易于扩展,具有更广泛的开放性。网络控制系统虽然具有以上突出的优点,但由于在控制系统中引入了通信网络,造成了不确定、时变的网络传输延时,故也存在许多有挑战性的问题,迫切需要研究解决。论文的主要工作如下：通过对大量网络传输延时的实测数据进行统计分析,了解网络传输延时的特性。介绍了3种常用的网络传输延时建模方法。同时,对应用较广的网络传输延时线性神经网络预测算法进行了仿真研究。在合理的假设条件下,将具有随机、有界网络传输延时的网络控制系统建模为时变时滞的离散时滞控制系统；利用Lyapunov-Krasovskii定理推导出具有线性矩阵不等式形式的状态反馈和输出反馈闭环网络控制系统渐近稳定和H∞稳定的充分条件；在此基础上,通过矩阵变换,将状态反馈闭环网络控制系统的稳定性充分条件转换成能使状态反馈闭环系统渐近稳定和H∞稳定的控制器的设计方法。在矩阵不等式的推导过程中,选择了合理的零等式,获取了需要的上限约束技术,并通过带约束的自由权矩阵来消除计算Lyapunov泛函的差分时产生的求和项,虽然增加了自由权矩阵的个数,但是,明显改善了计算结果的保守性。探讨了自适应控制在无刷直流电机网络调速系统中的应用。通过一阶Pade方法,将网络传输延时环节转换为惯性环节,从而得到无刷直流电机网络调速系统的近似线性数学模型；采用比较成熟的Narendra模型参考自适应控制策略设计控制器；针对网络控制系统中随机、时变的网络信息传输延时,使用带有时间戳的线性神经网络进行了在线预测,实时地获得当前采样周期的网络传输延时预测值。从而,在每一个网络传输延时不同的采样周期内,都能得到近似的线性数学模型。最后,采用对象模型已知的模型参考自适应控制方法,进行了反馈控制器的设计。深入研究了基于动态规划的网络控制系统最优状态反馈控制器的设计方法。针对网络传输延时的不确定性和时变性,本文给出了两种基于动态规划的最优状态反馈控制器设计方法。第一种方法的主要特点是：在执行器节点引入储存控制信号的缓冲器,执行器节点采用时间驱动方式,并设定执行器动作时间为τ∈{max[τ(k)],h}。这样,将不确定、时变的网络传输延时变换成确定性延时处理,·能使基于动态规划的最优状态反馈矩阵进行离线计算,简化了反馈控制器的设计。但是由于人为地加大了网络传输延时,这种设计方法使控制系统在稳定性分析和系统设计上都显得过于保守。第二种方法的主要特点是：引入带有时间戳的线性神经网络,在每一个采样周期内,对网络传输延时进行实时预测,并在每一个采样周期内,进行最优状态反馈矩阵的在线计算。这种控制方法使系统具有较高的控制精度,较好的响应特性,但是计算量较大,寻求快速的算法是今后重要的研究任务。最后,对上述两种控制策略,给出了仿真实验结果。更多还原

【Abstract】 In recent years, networked control systems have been used widely in national defense, industrial control and medicine. At the same time, networked control systems also have become one of the academic research focuses of control theory and have made some achievements. Compared with the conventional point-to-point control systems, networked control systems have the following main advantages: the cyber source can be conveniently used for sharing of information resources; remote monitoring and remote control can be carried out; control system flexibility can be enhanced, system wiring and investment can be reduced; system can be expanded easily and is open extensively. Although networked control systems have advantages above, there are many challenging problems which need be solved urgently. The main problem is uncertain and time-varying network transmission delay introduced by communication network.The main research works are as follows:The network transmission delay is learned by statistical analysis for a large number of measured network transmission delays. Three kinds of common network transmission delay modeling methods are introduced, and time stamped linear neural network prediction method of network transmission delay is simulated which is used extensively.Networked control systems with random and bounded network transmission delay are modeled as a discrete control system with time-varying time delay under reasonable assumptions. Based on the Lyapunov-Krasovskii functional, linear matrix inequality sufficient conditions which make state-feedback and output-feedback closed-loop networked control systems asymptotically stable and H-infinity stable are derived. By matrix transformation, sufficient conditions above are converted to controller design methods which make the state-feedback closed-loop control system asymptotically stable and H-infinity stable.In the derivation process of matrix inequalities, a reasonable zero-equation is adopted to obtain upper bound and the sum term of the differential of Lyapunov functional is eliminated by constrained free weighting matrix. Despite the increase in the number of free weight matrix, the system conservativeness is reduced.The model reference adaptive control in brushless DC motor networked control systems is discussed. The network transmission delay element is converted to inertia element by one step Pade method and the approximate linear mathematical model of brushless DC motor networked control systems is obtained; The common Narendra model reference adaptive control strategy is adopted to design controller; For random and time-varying network transmission delay, time stamped linear neural network is employed to access current sampling period prediction value of network transmission delay. Therefore, the approximate linear mathematical model is obtained in every sampling period with different network transmission delay. Finally, the model reference adaptive control based on known object model is used to design feedback controller.The optimal state feedback controller of networked control systems based on dynamic programming is studied deeply. For uncertain and time-varying network transmission delay, two optimal state feedback controllers based on dynamic programming are presented. First controller:Actuator node is time driven by using buffer which stores control signal, and the actuating time is set to τ∈{max[τ(k)], h), therefore, the uncertain and time-varying network transmission delay is transformed into deterministic delay. The optimal state feedback matrix based on dynamic programming can be calculated off-line, which simplify the design of state feedback controller. However, the design approach is conservative for stability analysis and system design, because network transmission delay is increased artificially. Second controller:Time stamped linear neural network is introduced to access prediction value of network transmission delay in every sampling period, and optimal state feedback matrix is calculated on-line. The control system using this method has higher precision and better performance, but large amount of calculation, the next important research task is looking for faster algorithms. At last, the simulation results of these two control strategies are given.更多还原