量化状态信息下多智能体Gossip算法及分布式优化

Multi-agent Gossip Consensus Algorithm with Quantized Data and Distributed Optimizing

基于量化状态信息的异步随机Gossip算法大多以均匀选择概率的时间模型为基础,未充分考虑网络拓扑结构对局部信息传递的影响。为此,该文提出了一种以非均匀选择概率为时间模型的改进算法。首先给出了非均匀选择概率下的多智能体系统时间模型,在随机性量化策略下给出了一致性误差的收敛性质;并讨论了量化精度和概率化权重矩阵第2大特征值对一致性误差收敛速度的影响,进而利用投影次梯度给出了选择概率的分布式优化方法。仿真结果表明,该基于量化状态信息的算法可通过选择概率的分布式优化,提高一致性误差的收敛速度。

As the traditional quantized asynchronous randomized gossip consensus algorithm is based on uniform selection probability time mode, the impact of network topology on local information transfer is not been fully considered. Thus, an improved quantized asynchronous randomized gossip consensus algorithm with non-uniform selection probability is proposed in this paper. Firstly, the asynchronous time model with non-uniform selection probability is proposed. Then the convergence of the algorithm is analyzed with randomized quantized information. The impact of the quantization resolution and the second largest eigenvalue of the probabilistic weighted matrix on convergence rate is also discussed. Furthermore, this paper proposes an optimization algorithm for selection probabilities with projection subgradient method in a distributed manner. The numerical example indicates that, the proposed algorithm improves the convergence rate by optimizing selection probabilities of agents.