Fully Distributed Event-triggered Adaptive Control of Uncertain Heterogeneous Multi-agent Systems


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date6 Jul 2020


Distributed coordination of multi-agent systems has attracted much attention due to its broad application in many areas including formation control, distributed sensor networks, flocking and so on. The consensus problem is a fundamental problem in distributed coordination of multi-agent systems. A key feature of distributed consensus is that each agent can only interact with its neighbors to achieve agreement on certain variables of common interest such as states or outputs. In practice, most of distributed consensus strategies are often implemented digitally in a time-triggered manner with periodic sampling. In order to handle the worst scenario, the sampling period often has to be small enough and thus tends to be conservative. To avoid unnecessary sampling, event-triggered consensus strategies have been developed for multi-agent systems. Most of them need to know the exact values of agent parameters and/or some global graph information. However, the agents may contain uncertainties or unknown parameters in many practical circumstances, and the global graph information is difficult to obtain in large scale systems. Therefore, fully distributed event-triggered control for uncertain heterogeneous multi-agent systems is a research subject of great practical and theoretical significance, and thus deserves further investigation.

This thesis will investigate the consensus problem of uncertain heterogeneous multi-agent systems. The main results of this thesis can be summarized as follows.

Firstly, the leader-following output consensus problem is studied for continuous-time heterogeneous linear multi-agent systems with parameter uncertainties. By using the internal model principle, two novel distributed event-triggered control strategies are developed with a dynamic event-triggering mechanism. It should be noted that when designing the controllers, no knowledge of global graph information is needed. In addition, the inter-event interval can be prolonged and Zeno behavior can be excluded with the proposed event-triggering mechanism.

Secondly, the leader-following output consensus problem is investigated for a class of unknown heterogeneous continuous-time linear minimum-phase multi-agent systems. A distributed event-triggered model reference adaptive control strategy is developed based on a distributed event-triggered adaptive observer. Meanwhile, a distributed adaptive event-triggering mechanism is designed so that each agent can determine when to broadcast its information to its neighbors. By using the proposed adaptive control strategy, the outputs of all the agents synchronize to the output of the leader asymptotically. In addition, the proposed control strategy could deal with unknown continuous-time multi-agent systems with individual relative degrees and do not need any global graph information.

Thirdly, the leader-following output consensus problem is addressed for unknown heterogeneous discrete-time linear minimum-phase multi-agent systems in the presence of external disturbances. Two novel distributed robust discrete-time model reference adaptive control laws with and without event-triggered communication are developed respectively. Under the proposed controllers, the tracking errors between the outputs of all the agents and the output of the leader converge to a residual set. Moreover, the tracking errors will converge to zero asymptotically when the disturbances are absent. In addition, the proposed controllers could deal with the unknown discrete-time multi-agent systems with individual relative degrees and do not need any global graph information.

Finally, the output consensus problem is studied for unknown heterogeneous discrete-time multi-agent systems subject to unmodeled dynamics. A robust model reference adaptive controller is developed based on an event-triggered internal reference model. Then, a dynamic event-triggering mechanism is designed to reduce the data transmission between neighboring agents. It is shown that if the unmodeled dynamics satisfy certain conditions, the output consensus errors will converge to a residual set. Moreover, in the absence of unmodeled dynamics, the consensus errors will approach zero asymptotically. In addition, the proposed controller is able to handle the agents with individual unknown nominal dynamics and unmodeled dynamics, and avoid the knowledge of any global information.