"Decentralized Regulation, Tracking, and Disturbance Rejection for Multi-agent Systems: A Targeting Approach"

点击翻译按钮获取中文摘要

This research deals with the control of multi-agent systems, that is, systems that are composed of multiple interacting subsystems. We focus on problems in which control is introduced through separate control agents that influence the system agents. Examples of socio-economic systems of this type include police officers trying to control a crowd (where the system agents are people in the crowd and the control agents are officers), military leaders directing a field of vehicles (where the system agents are vehicles and the control agents are leaders), and environmentalists trying to entice companies to adopt "green" technology (where the system agents are companies and the control agents are activists). Examples of technological systems of this type include distributed energy systems, in which thousands of small sources of electricity, rather than a few large ones, are connected to the power grid (where the system agents are generators and the control agents are voltage/phase/power controllers) and automated traffic systems (where the system agents are traffic lights and the control agents are traffic controllers). All of these systems can, in principle, be modelled mathematically and control rules for the control agents can be devised. Due to the complexity of the systems in question, a decentralized framework is appealing. However, even decentralized controllers can quickly become overwhelmingly complicated to design and, in practice, to deploy. We aim to deal with this complexity by using a new approach we have developed called targeting, wherein each of the control agents focuses attention on a single, carefully chosen, system agent. The major contribution of this research is that rigorous theory, based on both existing knowledge and the knowledge created through this proposal, is used to determine how many control agents should be used, where they should be placed, what their respective targets should be, and what control signals should be used to successfully control all of the system agents. At present, such information is often unavailable scientifically, leading to dependence on intuition or experience alone. A total of 10 graduate students will be partially supported through this funding.