Multi-agent Based Federated Control of Large-scale Systems

Abstract

Large-scale systems are systems that consist of many interconnected local systems requiring a large number of state variables to control. Conventional centralized control schemes are not suitable for large-scale systems because of the complex global dynamic behavior as well as computational difficulties associated with modeling and control. In addition, implementation of traditional centralized control is problematic, since any small change in the dynamics of a local system will require a complete redesign of the controller. This research introduces the general framework of agent-based federated control, motivated by the political structure of a federal government, in which partially self-governing states are united by a federal government. Likewise, a multi-agent based federated control system is composed of local autonomous subsystems (agent based controllers) that cooperate to provide an overall (a large-scale) system behavior. In this concept, each agent-based controller maintains its own control law for its local stability. In addition, each agent has partial observations of the states of other agents, via communication networks, and executes the local control law correspondingly to satisfy the performance requirements of the overall system level. Research results are presented on the general architecture of multi-agent federated control systems and on the computation of sub-system connectivity for local and global stability. Different stability analysis methods were developed for the federated control scheme in this research. First, overall system connective stability using eigenvalue methodology via determination of the connectivity strength factors between each connected sub-system was analyzed for linear interconnected systems. In addition, Lyapunov methodology was used to evaluate system connective stability. These results were applied to stabilization control of various federated control applications. Finally, the federated control concept was extended to trajectory tracking control for nonlinear entities. Simulations were performed to show the effectiveness of the federated control algorithm throughout this research. Potential applications of multi-agent federated control exist in electric power networks, modern industrial automation, robotic systems, communication networks, economic systems, and traffic networks. This research is expected to achieve scientific and theoretical advancement in control system techniques, and satisfy the U.S. Navy's need for stability and control of future large-scale systems.