Fund Project:Project supported by the National Natural Science Foundation of China (Grant Nos. 61573077, U1808205), the Fundamental Research Fund for the Central Universities of Ministry of Education of China (Grant No. N2023022), and the Natural Science Foundation of Hebei Province, China (Grant Nos. F2016501023, F2017501041)
Received Date:03 November 2020
Accepted Date:30 December 2020
Available Online:07 July 2021
Published Online:20 July 2021
Abstract:It is undisputed that complex networks are used to describe the interaction between large-scale complex systems. Different edges have different effects on network controllability. When some edges in a network are attacked or destroyed, the network controllability may be affected very little; when some other edges are attacked, network controllability may be affected very greatly, even results in the uncontrollability of the network. Which edges failure will affect the network controllability? To solve this problem, according to the node classification and edge classification, the concept of quasi-critical edge set is proposed, and the judgment theorem of quasi-critical edge set is given in this paper. In order to study the influence of quasi-critical edge set on the network controllability, the failure model of quasi-critical edge set is proposed, and the network controllability is quantified by the ratio of the number of driver nodes to the number of network nodes. In this failure model, the quasi-critical edge set with the minimum number of edges is removed first, thus destroying the network controllability quickly. By analyzing the failure model of quasi-critical edge set, the failure curve of quasi-critical edge set is obtained. It is found that the failure curve is a piecewise linear function and that the maximum (initial) slope of failure curve is related to the average degree of network. In addition, the failure of quasi-critical edge set has the greatest influence on network controllability. A comparison among the failure of quasi-critical edge set, random failure, degree failure, and betweenness failure verifies that the failure of quasi-critical edge set has the greatest damage to the network controllability in both model networks (ER random network, BA scale-free network, random triangle network and random rectangle network) and real networks in 26 different fields. For some of real networks, such as cancer cell networks, terrorist communication networks and other networks that are harmful to human beings, the failure model of quasi-critical edge set can provide a reference attack method. Keywords:complex network/ structural controllability/ quasi-critical edge set/ edge failure model
本节在4种模型网络(ER随机网络、BA无标度网络、随机三角形网络和随机矩形网络)和26种实际网络中, 将类关键边集失效(failure of quasi-critical edge set, FQ)和常见的3种边失效方式进行对比, 观察不同失效方式下的网络能控性$ n_{\rm D} $, 分析类关键边集失效后对网络能控性的影响. 常见的边失效方式有以下几种. 1)边随机失效: 随机删除网络中的一条边. 2)基于度的边失效: 删除网络中度最大的一条边. 边度的定义为边两端节点度的几何均数[37]. 对于边$ a_{ij} $, 其边度可以表示为$ \sqrt{k_{i} \cdot k_{j}} $, 其中$ k_{i} $为节点$ i $的入度, $ k_{j} $为节点$ j $的出度. 3)基于介数的边失效: 删除网络中介数最大的一条边. 边的介数定义为网络中所有的最短路径中经过边的数量比例. 对于边$ a_{ij} $, 其介数可以表示为$\displaystyle\sum\nolimits_{\text {all}, l, m ; l \neq m \atop\{l, m\} \neq\{i, j\}} \dfrac{N_{l m}\left(a_{i j}\right)}{N_{l m}}$, 其中$ N_{l m} $表示节点$ l $和节点$ m $之间的最短路径的条数, $ N_{l m}\left(a_{i j}\right) $表示节点$ l $和节点$ m $之间的最短路径经过边$ a_{ij} $的条数. 根据Lu和Li[22]的结论, 基于重新计算的失效方式通常比基于初始计算的失效方式更能损害网络的能控性. 因此, 对于以上3种失效方式, 和本文提出的类关键边集失效模型一样, 在每次边失效后都需要对网络中的度或介数重新计算. 需要注意的是, 基于随机、度、介数的失效方式每次只攻击一条边, 而我们提出的类关键边集失效模型虽然每次移除的边数不同, 但由于最终分析的是网络边移除比例$ p $对网络能控性$ n_{\rm D} $的影响, 因此每次移除边数的多少对最终的结论无影响. 34.2.1.ER 随机网络 -->
4.2.1.ER 随机网络
本节将生成节点数分别为300, 500, 700, $ \left\langle { k_{\rm in}} \right\rangle = \left\langle {k_{ {\rm out}}} \right\rangle = 2 $的ER随机网络. 考虑到网络的能控性与网络平均度(边的密度)具有相关性[38], 又生成节点总数分别为300, 500, 700, $\langle {k_{\rm in} }\rangle \!=\! \langle {k_{ {\rm out}}} \rangle \!=\! 6$的ER随机网络, 按照以上4种边失效方式(类关键边集失效、随机失效、按度失效和按介数失效)将网络中边移除, 记录边失效比例$ p $与网络能控性$ n_{\rm D} $的曲线, 如图8(a)—图8(f)所示. 图 8 不同节点总数和平均度的随机网络在4种边失效方式下网络能控性$n_{\rm D}$的变化 (a)节点总数$N=300$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机网络; (b)节点总数$N=500$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机网络; (c)节点总数$N=700$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$ 的随机网络; (d)节点总数$N=300$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$ 的随机网络; (e)节点总数$N=500$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机网络; (f)节点总数$N=700$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机网络 Figure8. The change of controllability $n_{\rm D}$ of random networks with different number of nodes and average degree under four kinds of edge failure modes: (a) A random network with number of nodes $N=300$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (b) a random network with number of nodes $N=500$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (c) a random network with number of nodes $N=700$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (d) a random network with number of nodes $N=300$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (e) a random network with number of nodes $N=500$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (f) a random network with number of nodes $N=700$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$.
本节将生成节点数分别为300, 500, 700 的BA无标度网络. 设置初始网络为$ 4 $个节点和$ 4 $条边, 且首尾连接. 为了研究度的影响, 分别使新节点作为弧头连接的节点数$ m_{\rm in} $等于新节点作为弧尾连接的节点数$ m_{\rm out} $等于1或2[39]. 按照以上4种边失效方式(类关键边集失效、随机失效、按度失效和按介数失效)将网络中边移除, 记录边失效比例$ p $与网络能控性$ n_{\rm D} $的曲线, 如图9(a)—图9(f)所示. 图 9 不同节点总数和平均度的无标度网络在4种边失效方式下网络能控性$n_{\rm D}$的变化 (a) 节点总数$N=300$, $m_{\rm in}=m_{\rm {out}}=1$ 的无标度网络; (b) 节点总数$N=500$, $m_{\rm in}=m_{\rm {out}}=1$的无标度网络; (c) 节点总数$N=700$, $m_{\rm in}=m_{\rm {out}}=1$的无标度网络; (d) 节点总数$N=300$, $m_{\rm in}=m_{\rm {out}}=3$的无标度网络; (e) 节点总数$N=500$, $m_{\rm in}=m_{\rm {out}}=3$ 的无标度网络; (f) 节点总数$N=700$, $m_{\rm in}=m_{\rm {out}}=3$的无标度网络 Figure9. The change of controllability $n_{\rm D}$ of scale-free networks with different number of nodes and average degree under four kinds of edge failure modes: (a) A scale-free network with number of nodes $N=300$ and $m_{\rm in}=m_{\rm {out}}=1$; (b) a scale-free network with number of nodes $N=500$ and $m_{\rm in}=m_{\rm {out}}=1$; (c) a scale-free network with number of nodes $N=700$ and $m_{\rm in}=m_{\rm {out}}=1$; (d) a scale-free network with number of nodes $N=300$ and $m_{\rm in}=m_{\rm {out}}=3$; (e) a scale-free network with number of nodes $N=500$ and $m_{\rm in}=m_{\rm {out}}=3$; (f) a scale-free network with number of nodes $N=700$ and $m_{\rm in}=m_{\rm {out}}=3$.
本节生成节点总数分别为300, 500, 700, 平均度为2的随机三角形网络, 又生成节点总数分别为300, 500, 700, 平均度为6的随机三角形网络, 选取4种失效方式分别对网络中的边进行移除, 记录能控性变化$ n_{\rm D} $与边失效比例$ p $的曲线, 如图10(a)—图10(f)所示. 图 10 不同节点总数和平均度的随机三角形网络在4种边失效方式下网络能控性$n_{\rm D}$的变化 (a) 节点总数$N=300$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机三角形网络; (b) 节点总数$N=500$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$ 的随机三角形网络; (c) 节点总数$N=700$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机三角形网络; (d) 节点总数$N=300$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机三角形网络; (e) 节点总数$N=500$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机三角形网络; (f) 节点总数$N=700$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机三角形网络 Figure10. The change of controllability $n_{\rm D}$ of random triangle networks with different number of nodes and average degree under four kinds of edge failure modes: (a) A random triangle network with number of nodes $N=300$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (b) a random triangle network with number of nodes $N=500$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (c) a random triangle network with number of nodes $N=700$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (d) a random triangle network with number of nodes $N=300$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (e) a random triangle network with number of nodes $N=500$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (f) a random triangle network with number of nodes $N=700$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$.
本节生成节点总数分别为300, 500, 700, 平均度为2的随机矩形网络, 又生成节点总数分别为300, 500, 700, 平均度为6的随机矩形网络, 选取4种失效方式分别对网络中的边进行移除记录能控性变化$ n_{\rm D} $与边失效比例$ p $的曲线, 如图11(a)—图11(f)所示. 图 11 不同节点总数和平均度的随机矩形网络在四4边失效方式下网络能控性$n_{\rm D}$的变化 (a) 节点总数$N=300$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机矩形网络; (b) 节点总数$N=500$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机矩形网络; (c) 节点总数$N=700$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$的随机矩形形网络; (d) 节点总数$N=300$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机矩形网络; (e) 节点总数$N=500$, 平均度$ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$ 的随机矩形网络; (f) 节点总数$N=700$, 平均度$\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$的随机矩形网络 Figure11. The change of controllability $n_{\rm D}$ of random rectangle networks with different number of nodes and average degree under four kinds of edge failure modes: (a) A random rectangle network with number of nodes $N=300$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (b) a random rectangle network with number of nodes $N=500$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (c) a random rectangle network with number of nodes $N=700$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=2$; (d) a random rectangle network with number of nodes $N=300$ and average degree $\left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (e) a random rectangle network with number of nodes $N=500$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$; (f) a random rectangle network with number of nodes $N=700$ and average degree $ \left\langle {k_{\rm in}} \right\rangle= \left\langle { k_{ {\rm out}}} \right\rangle=6$.