Optimal leader selection for controllability and robustness in multi-agent networks
Optimal leader selection for controllability and robustness in multi-agent networks
复制标题
多智能体网络中可控性和鲁棒性的最佳领导者选择
DOI：
--
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Naomi Ehrich Leonard
中科院分区：
文献类型：
--
作者：
Katherine E. Fitch;Naomi Ehrich Leonard
研究了多智能体网络中的两个最优领导者选择问题。最优领导者集合是使线性动态网络的性能最大化的m > 0个领导者的集合。在可控性问题中，每个领导者被确定为一个控制输入，并通过平均可控性和可达子空间体积来衡量性能。在鲁棒性问题中，每个领导者响应于外部信号，线性动态是有噪声的，并且性能由稳态系统误差来测量。之前，我们证明了鲁棒性的最佳领导集最大化了网络图中的联合中心。在本文中，我们展示了如何控制的最佳领导人集也取决于图的措施，包括信息中心的领导人和特征向量的图拉普拉斯算子。我们探讨了一个基本的权衡最佳领导人的选择可控性和鲁棒性，我们概述了一个分布式算法的选择对领导树。
Two optimal leader selection problems are examined for multi-agent networks. The optimal leader set is the set of m > 0 leaders that maximizes performance of a linear dynamic network. In the problem for controllability, each leader is identified with a control input, and performance is measured by average controllability and reachable subspace volume. In the problem for robustness, each leader responds to an external signal, the linear dynamics are noisy, and the performance is measured by the steady-state system error. Previously, we showed that the optimal leader set for robustness maximizes a joint centrality in the network graph. In this paper, we show how the optimal leader set for controllability depends also on measures of the graph, including information centrality of leaders and eigenvectors of the graph Laplacian. We explore a fundamental trade-off between optimal leader selection for controllability and for robustness, and we outline a distributed algorithm for the selection of a pair of leaders in trees.