Research On Containment Control Protocol Of Second-order Multi-agent Systems

Distributed cooperative control is the basic guarantee to realize the formation task of large-scale multi-agent systems,and the information interaction between agents is the precondition of the distributed cooperative control.For some specific application scenarios,such as handling of dangerous goods,forest fire rescue,etc.,containment control method provides an efficient solution.The so-called containment control is a control method for multi-leader multi-follower system structure,which drives all followers to converge into the convex hull of leaders.The research on containment control methods for multi-agent systems has achieved fruitful results,but there are still some weak points and problems to be broken through.For example,theoretical research often ignores some realistic network model and system dynamics model;there are much more stringent requirements on network communication topology structure for the realization of containment control;while the interference induced by communication network and some uncertainty factors are not fully considered yet;in practical complex environment,the process noise for the dynamic system and communication noise between agents are inevitable;the system state can not be measured directly or expensive to obtain;and the demand of more reasonable and efficient design methods for containment control strategy is increasing nowadays.To solve the above-mentioned problems,this thesis takes multi-agent systems as the main research object,takes the cooperative behaviors as the main control target,takes feedback containment control,model predictive control and surrounding control as the main control means,and comprehensively considers the second-order system model,network communication topology model,non-ideal network communication environment,sensor network,system performance requirements,system parameter constraints and stability analysis methods,and so on,and especially foucuses on the design and improvement principles of the cooperative control algorithm,the construction of distributed finite impulsive response(DFIR)filtering algorithm,the theoretical framework of stability analysis,and the judgment criteria for containment realization in the case of network communication resource restriction and system interference.More specifically,the following aspects are discussed in this thesis for the multi-agent cooperative control problem and DFIR filtering problem:(1)Based on distributed feedback control protocol and quantized communication algorithm,the cooperative containment control problem of second-order multi-agent systems is studied.A directed graph is considered and a pair of matrix norm and vector norm is designed.Accounting for the limitation of the finite bandwidth channels,quantized communication topology based on the encoding-decoding strategy is designed,in which the quantizers only have finite quantization levels and it is independent on the initial state of agents.Moreover,the quantizer and controller are jointly designed only using the estimated value of the neighbors' state information to ensure the system stability with less communication resource.The relationship between the quantization levels and sampling interval is established to guarantee that all the quantizers are not saturated,and thus ensure the asymptotic stability of the system.In addition,the quantization level constrained by the sufficient condition of system stability is greatly reduced,and it is not affected by the dimension and number of agents.Finally,all followers are guaranteed to converge asymptotically to the convex hull composed of leaders.(2)Deal with the problem of distributed model predictive control(DMPC)for second-order multi-agent systems under event-triggered technique and logarithm quantized communication based on a directed topological graph.Considering the limitation of communication bandwidth,a new bounded logarithm quantized communication strategy is proposed to pre-process the information before its transmission,thus reducing the influence of quantization error on the final convergence state.In order to decrease the frequency of control law update and reduce the power consumption,a distributed event-triggered rule is designed to decide when to transmit the information and when to optimize the model predictive control optimization problem,in which trigger function synthesizes three factors,namely,predictive step,saturation of quantizer and event-triggered error related with quantized error.The optimal control sequence of DMPC guides the update of controller between two triggering instants.The relationship among the quantization level,event-triggered parameters and communication topology is established.Conditions are presented to ensure that all leaders asymptotically converge to a designed formation configuration,while all followers reach to the convex hull of them.(3)Develop a tube-based distributed model predictive full containment control(TDMPFCC)algorithm for leader-following systems with bounded disturbance and dynamic leaders,and this algorithm employs knowledge regarding the constraints on states and control inputs to extrapolate their admissible values in the entire predictive horizon.Kalman filter(KF)is introduced for the output feedback control,resulting in estimated error along with which the disturbance is involved in the time-varying tubes to construct the tightened constraints.For each follower,by penalizing the control difference with its neighbors and the deviation of the states from the convex hull produced by its neighbors' corresponding states,a TDMPFCC problem subject to the tightened constraints is optimized only utilizing the local nominal state and control sequences.The TDMPFCC algorithm leads to a stable full containment state by applying the optimal solution,based on which,the feasibility and stability are proved by designing proper distributed terminal controller and constraints.(4)For the traditional containment control algorithm,how to make the follower converge into the convex hull of leaders is mainly studied before,but the behavior anaysis of leaders is ignored,just solving some simple tasks.In order to achieve more complex and diverse tasks,the target pursuit problem based on containment control strategy is studied.Firstly,based on the sensor network constructed by leaders,a distributed iterative finite impulse response(DIFIR)consensus filter for target-leader systems is addressed,thus reducing the communication and computation resource while resulting in more robustness.This filtering algorithm is expressed by a computationally efficient iterative algorithm,for which a distributed measurement model is involved wherein not only the neighbors' estimates for target are applied,but also the directed measurement data of target is used.Applying this DIFIR strategy,it is shown that the target's estimates by all leaders reach H? consensus,whose value is the local unbiased estimates of the target.Then the result is extended to the system whose target has unknown inputs.Then,utilizing model predictive control algorithm,surrounding control algorithm,finite impulsive response(FIR)filtering algorithm and consensus feedback control method,the hybrid switching control-filtering strategy is designed to realize the leader's regional patrol,the pursuit,surrounding and transportation for target individual,and the corresponding states for followers are the tracking of leaders,containment behaviors,assisting in the capture and cooperative transportation.Finally,the security of the pointed area is guaranteed.