Cooperative Control For Heterogeneous Multi-agent Systems And Its Application To DC Microgrids

Cooperative control of multi-agent systems has drawn much attention due to its great potential applications in industry,such as the formation control of unmanned aircrafts and power dispatch in a microgrid.In practice,multi-agent systems are naturally with heterogeneities and the communication among them are basically non-continuous.Therefore,it is of great significance in both theoretical research and industrial applications to investigate the cooperation of heterogeneous multi-agent systems with non-continuous communication.To this end,this dissertation aims to propose a hierarchical control design framework for several heterogeneous multi-agent systems with different dynamics and dimensions and to enhance the theoretical analysis on the cooperation of heterogeneous systems.Furthermore,the developed theoretical methods will be applied to the DC microgrid.The main contexts are listed as follows:For general linear heterogeneous multi-agent systems with uncontrollable parts,hierarchical control design framework is proposed.The optimal control design method is employed to generate the trajectories of agents in the upper layer,whereas the output regulation method is used in the lower layer to design the tracking controller of each agent such that it drives agent to follow the trajectory in the upper layer.By the stability analysis tools of switching systems,a rigorous proof on the cooperation of heterogeneous multiagent systems under the proposed controllers is given,where they eventually achieve the output consensus.For continuous-time heterogeneous multi-agent systems,impulsive cooperation controller is designed based on the hierarchical control design framework mentioned above.Compared to the existing methods,our method greatly reduces the communication burden during cooperation.By the stability analysis tools of hybrid systems,an upper bound of the average interacted time is derived for achieving the output formation-containment of heterogeneous systems,where the leaders achieve the desired formation and the followers get into the convex hull spanned by the leaders.For discrete-time heterogenous multi-agent systems,differentially private consensus controller is designed based on the hierarchical control design framework mentioned above.The differential privacy indexes are given for describing the measure of initial states of agents based on the two aspects,which are from the communication data and from the convergent value respectively.By the stability analysis of discrete-time systems,it reveals that the convergent rate of the proposed algorithm is related to the dynamics of agents,topology of network,and the damping rate of added noises.It also reveals that the relationship between network topology and the degree of protection.The discrete-time heterogenous systems eventually achieve average output consensus.For a DC microgrid system,under the hierarchical control design framework,a distributed current sharing controller is designed based on impulsive cooperation method,where it achieves an accurate and desired current sharing via discrete-time information of currents from neighbors.Moreover,the DC bus voltage quickly converges to the rated value.By the stability analysis tools of hybrid systems,an upper bound of the average interacted time of current information is derived.Furthermore,both simulation and expremental tests are carried out to demonstrate the effectiveness of the proposed method.For a DC microgrid system with unknown model parameters,under the hierarchical control design framework,a distributed game-based current sharing controller is designed based on reinforcement learning.Detailed steps are given on how to use artificial networks to implement reinforcement learning algorithm.The proposed controller not only merely employs the discrete-time manner in a communication network,but also ensures all the distributed generation units to fall into a Nash equilibrium.The simulation results illustrate that the performance enhances with the increasing episode of learning.For a DC microgrid system with unknown model parameters,under the hierarchical control design framework,a distributed optimal power dispatch controller is designed based on reinforcement learning.Firstly,the original problem is divided into an optimization problem of power dispatch and a control problem of power sharing,which are solved by distributed optimization algorithm and reinforcement learning,respectively.The proposed method can adaptively compensate the mismatch between the real power demand and the optimized ones,and it can also adaptively match the change of real-time power loss on lines.