Studies on Optimization Control of Networked Systems based on Local Models

我们研究了基于局部/不确定模型的复杂/大型/网络化系统的控制方法。我们考虑的情况下，局部的模型在空间和/或时间的假设和时间延迟存在于子系统之间的通道。针对空间局部性和非线性，提出了一种利用线性变参数（LPV）系统分析与综合技术的方法;针对时间局部性，提出了一种模型预测控制方法;针对时间和空间均局部性，提出了一种基于JIT（Just-In-Time）建模和模型预测控制方案的方法。这些方法是通过解决复杂/大型/网络化系统的两种类型的控制问题而开发的。首先，我们将已知的LPV系统的控制方法推广到电液伺服系统，这类系统含有非线性、不确定性和时滞，并设计了一种增益调度H ∞控制器。其次，针对高炉这一复杂的非线性物理现象，提出了一种基于高炉过程数据库的在线建模和预测控制方案。对于时滞，我们采用基于模型的输入/状态/输出时滞区间预测，并将时滞系统的控制问题转化为LPV系统的控制问题，从而解决了一些依赖于参数的线性矩阵不等式（LMI）问题。本文还讨论了基于控制理论的含时滞非线性网络的拥塞控制问题。

We studied methods to control complex/large/networked systems based on local/uncertain models. We considered the cases where locality of models in space and/or time is assumed and time delays exist in the channels between subsystems. For locality in space and nonlinearity, we proposed an approach using analysis and synthesis techniques of linear parameter varying (LPV) systems ; for locality in time, we proposed a model predictive control approach ; for locality in both space and time, we proposed an approach based on Just-In-Time(JIT) modeling and model predictive control scheme. These approaches are developed through solving two types of control problems for complex/large/networked systems. First, we extended the known method for LPV systems to electro-hydraulic servosystems, which contain nonlinear elements, uncertainties and time delays, and designed a gain scheduled H-infinity controller. Second, we focused our attention to blast furnace which presents very complicated/nonlinear physical phenomena and is difficult to built models, and proposed a new online modeling and predictive control scheme based on a large scale database consisting of blast furnace process data. As for time delays, we conquered this issue by using input/state/output prediction on delay time interval based on models and by transforming the control problems for time delay systems to the ones for LPV systems, which leads to solving some parameter dependent linear matrix inequalities (LMIs). We also discussed a congestion control of nonlinear networks including time delay based on control theory.