Towards A Practical Model Predictive Control Framework for Networked and Distributed Dynamic Systems

网络和分布式动态系统在许多新兴的工业应用中扮演着重要的角色，如网络物理系统、智能电网、智能交通系统、分布式过程、传感器网络和多代理系统。在控制系统中加入通信网络可以带来许多明显的优势，但也给控制系统的设计者带来了巨大的挑战，主要是因为通信网络的影响和约束。 模型预测控制(MPC)是工业应用中对复杂系统进行高性能和高性价比控制的主要范例。其固有的预测能力使其成为网络和分布式动态系统控制的一个很好的候选者。然而，经典的和鲁棒的预测控制公式不能系统地处理网络和分布式动态系统中普遍存在的随机测量噪声、随机约束、参数不确定性和外部扰动。 这项研究计划将解决网络和分布式系统控制面临的最重要的挑战。这将需要进行基础研究，以克服目前MPC的局限性。提出的研究方案将把实际的网络诱导约束、统计不确定性和概率约束以及控制性能指标纳入到实际的预测控制框架中。它将开发新的随机和自适应预测控制方法和快速实时优化算法。我们将研究两类一般类型的控制问题：(1)网络动态系统(单个对象通过网络被控制)的预测控制；(2)多智能体系统(通过通信网络连接多个智能体)的分布式预测控制。根据本申请中提出的研究计划建立的MPC分析和综合框架将在实验系统上进行测试和验证，例如：(1)四旋翼的网络控制系统，(2)机器人机械手的网络控制系统，以及(3)包括多个四旋翼和移动机器人的多智能体系统。 所提出的研究方案将填补理论与实践之间的空白；它将为控制工程师分析和综合网络和分布式动态系统的随机、自适应和快速预测控制提供新的工具；它将建立一个解决实际约束的新的统一范式。此外，该计划将通过与技术和行业相关的研究培训使研究生大大受益。

Networked and distributed dynamic systems play essential roles in many emerging industrial applications such as cyber-physical systems, smart grid, intelligent transportation systems, distributed process, sensor networks, and multi-agent systems. The insertion of communication networks into a control system can bring many distinct advantages; but it also presents tremendous challenges to control system designers mainly because of the effects and constraints induced by communication networks. Model predictive control (MPC) is the leading paradigm for high-performance and cost-effective control of complex systems in industrial applications. Its inherent prediction ability makes it a good candidate for the control of networked and distributed dynamic systems. Yet, the classical and robust MPC formulations cannot systematically deal with the random measurement noise, stochastic constraints, parametric uncertainties, and exogenous disturbances that are ubiquitous in networked and distributed dynamical systems. This research program will address the most important challenges facing the control of networked and distributed systems. It will entail fundamental research to overcome the current limitations of MPC. The proposed research program will incorporate practical network-induced constraints, statistic uncertainties and probabilistic constraints, and control performance specifications into the practical MPC framework. It will develop new stochastic and adaptive MPC methods and fast real-time optimization algorithms. Two general types of control problems will be investigated: (1) MPC for networked dynamic systems (with a single plant to be controlled over network), and (2) Distributed MPC for multi-agent systems (with multiple agents connected over a communication network). The MPC analysis and synthesis framework that will be established from the research program proposed in this application will be tested and validated on experimental systems, e.g., (1) the networked control system of a quadrotor, (2) the networked control system of a robot manipulator, and (3) a multi-agent system including multiple quadrotors and mobile robots. The proposed research program will fill the gap between theory and practice; it will provide control engineers with new tools for analysis and synthesis of the stochastic, adaptive and fast MPC for networked and distributed dynamic systems; it will establish a novel unified paradigm addressing practical constraints. Furthermore, the program will greatly benefit graduate students through technology- and industry-relevant research training.