Research On Multi - Agent System Predictive Cooperative Control

Cooperative control of multi-agent systems has attracted researchers’ wide atten-tion and plentiful investigations with its distributed characteristic and robustness for decades. With the rapid development of computing and communication technologies, cooperative control of multi-agent systems has more extensive applications, such as formation control of unmanned air vehicles, wireless sensor networks, autonomous multi-robot systems, etc. Inspired by the predictive intelligence of natural bio-groups, and thanks to the prevalence of model predictive control (MPC) in industry as well as its fruitful theoretical results, investigators initiated incorporating the predictive mech-anism into multi-agent systems in recent years, and they gave exploratory results on solving the consensus problem by using the model predictive control method. This study not only helps people understand the important role natural bio-groups’ predic-tive intelligence is playing in their coordinated behaviors, but also provides innovative approaches to solve cooperative control problems. Therefore, this thesis addresses sev-eral cooperative control problems based on model predictive control, aiming to acquire more understanding of predictive cooperative control from the aspects of control algo-rithm design, theoretical analysis and numerical simulation.This thesis utilizes the model predictive control method to study cooperative con-trol of multi-agent systems, focusing on consensus and flocking control, which are two of the most widely concerned problems. The main contents are organized as follows:·We study the consensus problem by applying the model predictive control method, focusing on periodic-sampled and continuous-time multi-agent sys-tems. This part of contents is fully presented in chapter3. We first introduce a distributed MPC weighted-average consensus protocol for a periodic-sampled multi-agent system, prove that the weighted-average consensus is asymptotically reached as long as the network is jointly connected, and numerically verify the effectiveness of the MPC consensus protocol as well as its advantages in increas-ing the consensus convergence speed and expanding the feasible range of the sampling interval. We further propose a distributed MPC weighted-average con-sensus protocol for a continuous-time multi-agent system, and provide the corre-sponding proof for convergence, as well as numerical simulation results, which include the protocol’s effectiveness verification and effects of network topologies on the consensus performance. ●Apart from consensus, flocking control is another commonly concerned problem in multi-agent systems. Therefore, we address the flocking control problem by applying the model predictive control method in chapter4. We first design a centralized MPC flocking protocol, assuming that each agent obtains the global information of the whole network. We further propose a decentralized MPC flocking protocol, where each agent can only obtain the local information of its neighbors. Finally, we give numerical examples to verify the feasibility of the MPC flocking protocols, and further study the effects of three parameters (the sampling interval, the prediction horizon and the control horizon) as well as the MPC flocking protocols’advantages compared with the routine flocking proto-col, from the viewpoints of the convergence speed and the final lattice regularity.●The information of both the position and velocity of agents are required in most existing flocking protocols including our work in chapter4. By incorporating the impulsive control, we further study the MPC flocking of a multi-agent system based on position measurements only in chapter5. We first propose a central-ized impulsive MPC flocking protocol, and then develop a feasible sequential-negotiation based distributed impulsive MPC flocking protocol. We prove that both the centralized and distributed impulsive MPC flocking protocols lead to a stable flock by using geometric properties of the optimal path followed by in-dividual agents, and provide numerical examples to illustrate their effectiveness and advantages in aspects of the convergence speed, the communication cost, and the final lattice regularity.