Optimal design of stochastic networked control systems

本文的研究目的是加深对不确定性和信息约束下的网络化协同控制系统的理解。随机网络控制系统出现在许多实际环境中(如能源/电网中的分散和自适应功率调节、汽车防撞应用、传感器网络和经济学理论)。然而，关于这类系统，还有许多基本和实际的方面有待了解。这项拟议的研究将有三个主要的技术和实际贡献：(I)发展一种关于信息约束下分散动态系统性能的定量理论。这将导致实际系统的设计原则和各种应用的基础设施设计。(2)通过事件驱动通信实现复杂关联系统的随机稳定性，并发展这类系统稳定性的随机漂移准则。这对于分散的系统很重要，因为只有在资源可用时才能进行通信。(Iii)当控制器能够通过部分信息或训练序列学习系统的动态时，产生近似最优控制策略。这在控制器的认知和学习方面尤其重要，以适应他们生活和行动的未知环境。

The goal of the proposed research is to advance the understanding of networked and cooperative control systems under uncertainty and information constraints. Stochastic networked control systems arise in many practical settings (such as in decentralized and adaptive power regulation in the energy/power grid, automotive applications for collision avoidance, sensor networks and in economics theory). However, there are many fundamental and practical aspects yet to be understood regarding such systems. The proposed research will have three main technical and practical contributions: (i) Development of a quantitative theory on the performance of decentralized dynamic systems given information constraints. This will lead to design principles for practical systems and infrastructure design for a large variety of applications. (ii) Stochastic stability of complex connected systems through event-driven communications and development of stochastic drift criteria for stability of such systems. This is important for decentralized systems which can only communicate when resources are available. (iii) Generation of approximately optimal control policies when controllers are able to learn the dynamics of a system through partial information or training sequences. This is especially important in cognition and learning aspects of controllers for adapting to an unknown environment in which they live and act.