Event-Based Distributed Control of Constrained Multi-Agent Systems

点击翻译按钮获取中文摘要

Multi-agent systems are sets of agents interacting for achieving local and global objectives. The agents have dynamics, are spatially distributed, can communicate, and are often coupled in terms of dynamics, constraints, and objectives. Typical examples include power systems, where the agents are generators, loads, and storages which are physically coupled by the power grid and informationally coupled by a communication network, transportation systems, irrigation systems, and manufacturing systems. Such systems have a high technological, economical, ecological, and social relevance and have therefore received considerable attention both in academia and industry during the last years. The control of multi-agent systems is very challenging. The challenges primarily result from an enormous computation and communication complexity. To cope with this complexity, distributed control and event-based control have been proposed in very recent years. These methods have proved very versatile and effective. Research on these methods is, however, still in the beginning. Particularly holistic methods are still missing. This is where this project starts. The objective of the project consists in holistic methods for a simultaneous design of controllers, schedulers, event generators, and communication topologies for multi-agent systems with physical constraints and informational constraints. Physical constraints result from limitations in actuators and plants. Informational constraints stem from shared and imperfect computation and communication resources comprising medium access constraints, communication and computation delays, and packet losses. Within the project two approaches are pursued: event-based distributed codesign (EBDC, Liu) and event-based distributed model predictive control (EBDMPC, Görges). EBDC is based on a robust codesign of the controllers, scheduler, event generators, and communication topology which is performed largely offline while EBDMPC relies on an optimal codesign which is conducted mostly online. The approaches are complementary in terms of complexity and performance. EBDC has a low complexity but also a comparably lower performance due to the offline codesign while EBDMPC has a high complexity but also a comparably higher performance. The approaches have strong interactions, allowing a close cooperation between the project partners. Specifically methods for distributed controller design, event generation based on optimality conditions, communication topology design, stability, feasibility, and optimality analysis, distributed plug and play control and distributed optimization will be developed synergistically wherever possible. Throughout the project the methods will be evaluated for distributed control of power systems and implemented as an open, modular, and scalable MATLAB-based toolbox.