通信受限多个体网络的一致性和分布式优化研究

In practical applications, the information transmission among multi-agent networks usually suffers from the effects of limited bandwidth and network topology. The research on consensus and distributed optimization of multi-agent networks with communication constraints mainly focuses on how to make the agents eventually reach an agreement and to cooperatively solve a network-wide optimization problem through the design of local quantized consensus protocol for each agent in a given communication network. The main purpose of this project lies in studying the interconnection of network topology, quantized information and consensus protocol; additionally, the interest in this project also pertains to the development of efficient quantized average consensus algorithms by virtue of the broadcast nature of wireless communications and then applies the algorithms to multi-agent distributed optimization issues. Specifically, we mainly consider how each agent of directed unbalanced networks updates its states by relying solely on locally available quantized information, then we conduct convergence analysis under the corresponding quantized consensus algorithm. Moreover, we consider how to design effective and robust quantized average consensus algorithms for directed networks exploiting the broadcast nature of wireless communications. Finally, we investigate quantized consensus-based distributed optimization algorithms for multi-agent systems; we try to analyse how agents simultaneously achieve consensus and optimize a network-wide objective function in the presence of quantized interactions. The research of this project will help to better understand the evolvement of agents' states in the presence of quantized information interactions and the quantization effect on the performance of distributed optimization algorithms.

多个体网络中个体间的信息传递，通常会受到有限带宽、网络拓扑等因素影响。通信受限多个体网络的一致性与分布式优化主要关注如何设计个体的局部量化一致性协议，使所有个体形成共识并协同地解决关于整个网络的优化计算问题。本项目重点研究网络拓扑、量化信息与个体一致性协议间的关系，以及如何利用无线通信的广播特性来发展更符合实际的量化平均一致性算法，并应用于多个体分布式优化问题。具体地，项目考虑有向非平衡网络中个体如何与其邻居个体交换量化信息来调整状态，并分析量化一致性算法的收敛性；研究利用广播特性对有向网络设计有效的鲁棒量化平均一致性算法。还将考虑基于量化一致性的多个体分布式优化算法的研究，分析如何在量化通信受限下，所有个体达成一致性的同时并使关于整个网络的优化问题目标函数最优。本项目研究将有助于更好地理解多个体系统中个体状态在量化信息通信下的演化，以及量化对分布式优化算法性能的影响。

多个体网络中个体间的信息传递，通常会受到有限带宽、网络拓扑等因素影响。在课题资助下，主要研究如何设计个体的局部量化一致性协议，使所有个体形成共识并协同地解决关于整个网络的优化计算问题。本项目重点研究网络拓扑、量化信息与个体一致性协议间的关系，以及如何利用无线通信的广播特性来发展更符合实际的量化平均一致性算法，并应用于多个体分布式优化问题。具体地，项目考虑了有向非平衡网络中个体如何与其邻居个体交换量化信息来调整状态，并分析量化一致性算法的收敛性及相关的分布式优化算法的收敛性；研究了基于事件触发通信的无线传感网络时钟同步问题，并提出一种基于事件触发机制的二阶一致性算法。研究了多个体网络的无梯度优化问题，提出了一种分布式随机投影无梯度优化算法及分布式流言push-sum 无梯度算法、分布式在线条件梯度优化算法、分布式次梯度优化算法、分布式在线对偶平均优化算法等。本课题研究可应用于大数据计算、分布式机器学习、机物系统的信息处理与融合、数据隐私保护等重要领域。