Model Predictive Control For Multi-agent Systems

In recent years, the theory of distributed multi-agent systems has been intensively investigated by many scientists and engineers due to its ability to help them understand coordinate dynamic behaviors in biology groups, and also due to their wide application capabilities in engineering. Each agent could communicate and interact with its neighbors in order to achieve cooperative behaviors, whose inner dynamic mechanism is a bottleneck in the control problems of multi-agent systems. Based on model predictive control algorithm,this thesis fully explores the key effects of agent intelligence on group cooperative behaviors, and studies some control schemes for multi-agent systems with discrete-time dynamics from the perspective of basic theory. The main achievements are listed as below:The consensus control problems of networks with single- and double-integrator dynamics are studied. Since it is always difficult to find a suitable Lyapunov function to guarantee stability in the distributed system, this thesis proposes an interactive algorithm to achieve the exact optimal solution to the optimization problem based on model predictive control, and derives the sufficient conditions to guarantee the systems’ convergence and stability by using the properties of stochastic matrices. The provided control schemes could deal with the input constraints, and significantly enlarge the valid domain of some parameters, such as the sampling period and control horizon, which facilitates the tuning operation of system parameters.The control schemes for agents’ flocking behaviors are studied. This thesis addresses two collaboration behaviors of multi-agent systems, i.e., the Cucker-Smale flocking and lattice formation control, which are both based on model predictive control algorithm. On one hand, it solves the control problem of Cucker-Smale flocking with input constraints, which shows an advantage of model predictive control handling complex constraints. On the other hand, it proposes a model predictive control scheme for lattice formation, which thus achieves the control for lattice formation with discrete-time dynamics.Conclusions and outlooks are both made. Model predictive control algorithm is deemed to solve more control problems in multi-agent systems, such as the control problems with time-varying delays or packet losses, position-only information exchange, and the consensus control schemes under randomly switching interaction topologies.