基于神经网络的多智能体量化迭代学习控制

Quantized iterative learning control for multi-agent system based on neural network

迭代学习控制是实现多智能体一致性跟踪的有效技术手段。针对未知动态的多智能体系统在受到量化误差影响的情况下,研究多智能体迭代学习控制一致性跟踪实现问题。基于神经网络拟合方法,提出了分布式量化迭代学习控制算法,在范数意义上对算法进行收敛性分析,推导出其收敛的充分条件,并对充分条件作出了可行性分析。理论分析得出,迭代学习控制器中增益矩阵的设计不仅与神经网络参数矩阵有关,还与多智能体的网络拓扑结构有关。实验仿真结果表明,量化误差会降低迭代学习控制算法的收敛速度,且量化精度越小,收敛速度越快;但随着迭代次数的增加,跟踪误差渐近收敛为零,所提算法仍能实现多智能体的一致性跟踪。

Iterative learning control(ILC)is one of the effective technical methods to achieve consensus tracking of multi-agent system(MAS).Consensus tracking problem of MAS with unknown dynamics when ILC of MAS is affected by quantization error is studied.Based on the neural network fitting method,a distributed quantized-ILC algorithm is proposed.The convergence of the algorithm is analyzed in the sense of norms,and the sufficient condition is derived.A feasibility analysis is made for the sufficient condition.Theoretical analysis shows that the design of the gain matrix in the iterative learning controller is not only related to the neural network parameter matrix,but also to the network topology of the multi-agent.The simulation results show that the quantization error will reduce the convergence speed of the ILC algorithm,and the smaller the quantization scale,the faster the convergence speed;however,as the number of iterations increases,the tracking error asymptotically converges to zero,and the proposed algorithm can still guarantee the consensus tracking of MAS.