Cooperative Consensus Control Of Stochastic Multi-Agent System Based On Probability Density Compensation

Most multi-agent systems are stochastic systems with unknown disturbance and unmodeled non-linear parts.Therefore,it is of great theoretical and practical significance to study the consensus of these kind of systems.In this paper,a new concept of distribution consensus is proposed for stochastic nonlinear multi-agent system,that is,the error does not need to converge to zero,but follows the desired distribution function.Aiming at this kind of consensus problem in the sense of probability,a new control framework is proposed,which makes full use of the strong robustness of sliding mode control to nonlinear system and the advantage of probability density function control to compensate the uncertainty in the system.A control strategy combining sliding mode stability control with probability density function compensation control is designed.The specific work is as follows:Firstly,the paper designs a suitable control protocol based on the output error probability density function graph,combined with sliding mode control and probability density function compensator(PDF compensator).The sliding mode controller is the core part to ensure the stability of the whole system,and PDF compensator is used to compensate for random changes and reduce the chattering effect of sliding mode control.In order to realize the real-time control,the feedforward compensator is modeled by radial basis function neural network,and the optimal control law is calculated by iterative training of the weights of the radial basis function network.The effectiveness of the method is verified by simulation of three kinds of multi-agent systems with different communication topology.It is shown that PDF compensator can greatly improve the consensus effect of the stochastic multi-agent system.Secondly,under the sliding mode PDF compensation fusion control,the consensus in the sense of probability is achieved by further reducing the output error entropy of the multi-agent system.In order to further optimize the PDF compensator,the minimum entropy criterion combined with radial basis function neural network iterative training is used to obtain the optimal weight,the PDF compensator is further optimized to minimize the output error entropy of each agent and optimize the control effect.The simulation results verify the effectiveness of the method.Finally,the distributed formation control problem is considered when not all agents can obtain global information.In the leader-follower architecture,only the neighbor node information is used to construct a distributed controller combining sliding mode and PDF compensation to achieve formation consensus in the sense of probability.The simulation results show the advantages and good control effect of the proposed method.