Event Triggered Unknown Networked Control System Design by using Adaptive Dynamic Programming

该项目将为网络控制系统(NCS)开发新的通用控制设计，例如智能电网和现代汽车系统所需的新型分布式控制系统。与许多传统的NCS系统不同，这些系统从一开始就设计为随着时间的推移使性能最大化，充分考虑了网络中的通信延迟和意外事件，以及非线性、连续随机干扰和此PI在过去已解决的其他问题。他打算开发严格的数学证明，证明这些新设计将始终导致稳定的运行。他还将通过他现有的工业-大学智能维护中心，维护一条立即投入实际应用的管道。这些设计的目的是一般性的、基于学习的和大规模并行的，在某种程度上也可以帮助我们理解哺乳动物的大脑如何执行更传统类型的设计范围之外的任务。这些设计将是之前由NSF资助的工作的扩展，由Lewis和Liu编辑的《RLADP手册》中描述了这一点。与线性规划一样，ADP是一类挑战，而不仅仅是一种特定的方法，尽管成功的方法之间存在联系。ADP包括了在面对非线性和随机扰动的情况下针对一般多阶段优化问题的所有方法。为了处理一般的非线性任务，本项目将包括神经网络，它已被证明比更传统的系统更有效地逼近一般的非线性函数。这里的主要新奇之处在于能够将这些功能与通信延迟和随机离散事件的存在相结合，这在大多数NC中都是一个重要的实际问题。

This project will develop new universal control designs for Network Control Systems (NCS) such as the new distributed control systems needed for the smart grid and modern automotive systems. Unlike many traditional NCS systems, these will be designed from the start so as to maximize performance over time, fully accounting for communication delays in networks and unexpected events, as well as nonlinearities, continuous random disturbances and other issues which this PI has addressed in the past. He intends to develop rigorous mathematical proofs that these new designs will always lead to stable operation. He will also maintain a pipeline to immediate practical applications, through his existing Industry-University Center on Intelligent Maintenance. The designs are intended to be general, learning-based and massively parallel, in a way which may also help us understand how mammal brains can perform tasks beyond the scope of more traditional types of design.These designs will be an extension of work previously funded by NSF, described in the Handbook of RLADP, edited by Lewis and Liu. ADP, like linear programming, is a class of challenges, not just one specific method, though there are relations between the successful methods. ADP includes all the methods aimed at the general problem of multistage optimization in the face of nonlinearity and stochastic disturbance. In order to cope with general nonlinear tasks, this project will include neural networks which have been proved to be more effective as approximators of general nonlinear functions than more traditional systems. The chief novelty here is the ability to combine these capabilities with the presence of communication delays and random discrete events, which are an important practical issue in most NCS.