大规模网络化多智能体系统递阶一致性容错控制方法研究

A large-scale scientific facility plays a vital role in the key basic research. The architecture of the control system with several hundred thousand nodes is layered and networked, which leads to long convergent time, communication constraints and low fault tolerance. To solve these problems, this project studies the hierarchical consensus and fault-tolerant control method for the large-scale networked control system with several hundred thousand nodes. Besides, the issue of long convergent time will be settled by developing a hierarchical decomposition algorithm based on the analysis of convergence time measure and network topology. Meanwhile, the networked predictive scheme is introduced to make active compensation for communication delay, data loss and data disorder. The fault-tolerant controller based on a multi-variate statistical analysis and a residual generator is designed, which can overcome the influence of model uncertainties on the hierarchical consensus control system. The proposed method will be applied into the automatic alignment system of the SG-III laser fusion facility, which will raise the number of control nodes from 1000 to 100000 order of magnitude, reduce the convergent time by 20% and achieve a 80% successful fault tolerance rate. Finally, this project will build the theoretical and experimental foundations for the hierarchical consensus and fault-tolerant control systems of large-scale laser facilities.

大科学装置控制系统是在基础科学研究中发挥着重要作用，其节点规模可达十万量级，具有多层次、网络化的特点，导致了收敛时间长、网络通信约束、故障容错率低等问题。为解决这些问题，本项目面向大规模网络化多智能体系统主要研究递阶一致性控制算法。通过递阶分解算法解决在控制过程中收敛时间过长的问题，利用收敛时间测度来分析递阶一致性分解算法的可行性条件；通过网络预测机制来主动补偿通信受限特性（如：通信延时、数据丢包、数据错序等）对控制性能的影响；结合残差产生器、多元统计分析方法设计容错控制器，克服因模型不确定性而导致故障容错率低的问题。该方法将应用于神光-III激光装置的光路准直控制中，可支持的控制节点数从数千量级上升到十万量级，收敛时间降低20%，容错成功率不低于80%。最终，为面向大型激光装置的网络化递阶一致性容错控制方法的研究奠定一个理论、实验和应用的基础。

本项目针对大科学装置网络化控制系统中存在的收敛时间长、网络通信约束、故障容错率低等问题，开展了大规模网络化多智能体系统递阶一致性容错控制方法的研究工作。项目组利用收敛时间测度因子来分析递阶控制拓扑结构与收敛时间的关系，建立满足特定收敛时间条件下递阶一致性控制结构的设计准则；针对网络通信约束问题（如通信延时、数据丢包等），利用代数图论和非负矩阵的性质设计了一种一致性容错控制方法，有效克服了通信受限对控制性能的影响；结合残差产生器、多元特征量优选等方法设计容错控制器，克服因模型不确定性而导致故障容错率低的问题。该方法应用于大型激光装置的光路准直控制中，可支持的控制点数从数千量级上升到十万量级，收敛时间降低20%，容错成功率高于80%。项目成果可以为大型激光装置的递阶一致性容错控制方法研究提供理论和应用基础。

内容获取失败，请点击重试