CN105228185B - Method used for identifying identity of fuzzy redundant node in communication network - Google Patents

Abstract

The invention belongs to the field of structure research of complex networks, and in particular relates to a method used for identifying identity of a fuzzy redundant node in a communication network. The method comprises the following steps of obtaining communication network data; distinguishing the communication network data into a known ascertained part and a fuzzy and uncertain part; showing a sending side and a receiving side having occurred in the obtained communication information and a sending side and a receiving side which are vacant or undefined possibly caused by information loss by using a placeholder respectively; establishing a preliminary connection relationship graph; then clustering the placeholders by adopting a spectral clustering algorithm based on the attributes of the placeholders and the topologic characteristics of the placeholders in the connection relationship graph; determining which placeholders practically show the same node; and combining the placeholders which are clustered in the same group into a communication node so as to achieve the goal of identifying and removing fuzzy and redundant nodes and realize reconstruction of the communication network.

Description

A kind of method for recognizing fuzzy redundant node identity in communication network

Technical field

The invention belongs to the structural research field of complex network, and in particular to a kind of fuzzy superfluous in communication network for recognizing The method of remaining node identities, it is adaptable to the reduction of network topology structure, Denoising Problems in communication network.

Background technology

For the reconstruct of communication network topology structure is by get imperfect or even containing noisy communication number According to being analyzed, so as to restore as close possible to real network structure.Communication network reconstruction includes three aspects：Lack Lose the reduction of link；The reduction of disappearance, the communication node hidden；And identity ambiguous, redundant node distinguishing and remove.Herein Identification problem for obscuring node in the 3rd aspect, i.e. communication network is studied.

Under the epoch of information explosion, particularly complex electromagnetic environment, because the utilization of the means such as interference, camouflage, we In the communication information for getting, the disappearance of unavoidable existence information, even omission, mistake, also have substantial amounts of redundancy noise, for example The communication information form that we get is usually：[call duration time, sender, recipient, data package size], but be because doing Disturb, the factor such as noise, be likely to result in packet packet loss, get lines crossed, situations such as sender recipient is indefinite or wrong.Part Communication data may be lost or damage so that the data source of communication data that recipient obtains there may be mistake, originally from same Any a plurality of communication for sending may be mistaken as from two sending even more than different nodes.This causes receiving point accurate Judge to expect state of communication source etc..If can produce in a network with the presence of this problem of multiple nodes in a communication network The substantial amounts of litura of life, the communication of whole network can therefore suffer from affecting.If the source node of clear these fuzzy nodes can not be differentiated (node that fuzzy node should be represented), the communication efficiency of network and accuracy rate can all be affected.In order to maintain communication network Stability, it is necessary to fuzzy node is distinguished, the association situation between communication network interior joint is returned to into ideal as far as possible State.

Some researchs have been carried out with regard to the reconstruction of complex network topologies both at home and abroad at present, is concentrated mainly on In terms of disappearance link prediction, the research with regard to the correlation of nodes information is just at the early-stage.It is existing a small amount of with regard to communication The research of nodes information relates only to lack the prediction of node, mainly has two kinds of thinkings：

The first is from the angle of network reconfiguration while predicting disappearance node and Lian Bian.Such as Leskovec etc. [1] combines the phase Maximum framework and Kronecker graph models is hoped, KronEM algorithms are proposed, model parameter is estimated first with the network observed, Then model prediction lack part being used, then model parameter being estimated with the network structure of prediction, such iteration, until parameter restrains.

Second is that directly prediction lacks node, Wen-Xu Wang et al. [2] using compressed sensing based on game theoretic Network reconfiguration is carried out in network, it is believed that reconstruct the abnormal point point implicit with certain occurs and is connected, so as to predict missing point in network In position.Author defines a kind of game mechanism first in a network, and individuality has two kinds of selections, cooperation or betrayal (S= { cooperation, Defense }), using the individual policy update of Fermi rule simulations.Compressed sensing is initially used in signal In recovery, complete signal is restored by the partial data of the sparse signal for getting.For based on game theoretic evolved network, Through the game of many wheels, because link probability matrix is a sparse matrix, meet compressed sensing definition, according to the plan that individuality is often taken turns The income omited and get and the relation of chain matrice, are just melted into the form of similar Y=Φ X, and Y represents measured value, Φ tables Show calculation matrix, X represents actual signal.In the link probability matrix for finally obtaining, if probability is close to 1 has been considered as side phase Even, it is close to 0 and is considered as no side and is connected.The threshold value of author's setting is 0.1.For the game of many wheels, in theory every time with compression sense Know that the link probability matrix for reconstructing should be the same, because the game mechanism in text does not change network linking situation, but If the connection of certain several node is in each wheel and different and do not correspond with network topological characteristic itself, so that it may judge These abnormity points are connected with missing point, so that it is determined that missing point position in a network.

Dynamic behaviors of the J.P.Bagrow et al. [3] on network, for fixed network structure, analog simulation Information flow situation on network, the information flow speed for then passing through each node with prediction algorithm prediction judge information flow Speed, judges neighbours of the abnormal point as missing point occur, so that it is determined that missing point position in a network.

Although prior art can carry out preliminary judgement, the algorithm of Leskovec to lacking node in communication network Computation complexity is especially high, in real network and does not apply to, and the method for compressed sensing is only applicable to net known to mechanism of Evolution Algorithm more satisfactoryization of network, J.P.Bagrow et al., can only be tested on emulation data set.Above three algorithm is all The implicit node of possible disappearance in preliminary analyses network, without carrying out to possible redundant node, fuzzy node in network point Analysis and identification.

List of references in text：

[1]M.Kim,J.Leskovec.The Network Completion Problem:Inferring Missing Nodes and Edges in Networks.SIAM International Conference on Data Mining(SDM) 2011.

[2]Wen-Xu Wang,Ying-Cheng Lai,Celso Grebogi,and Jieping Ye.Network Reconstruction Based on Evolutionary-Game Data via Compressive Sensing.PHYSICAL REVIEW X 1,021021(2011).

[3]J.P.Bagrow,S.Desu,M.R.Frank,N.Manukyan,L.Mitchell,A.Reagan, E.E.Bloedorn,L.B.Booker,L.K.Branting,M.J.Smith,B.F.Tivnan,C.M.Danforth, P.S.Dodds,and J.C.Bongard,Shadow networks:Discovering hidden nodes with models of information flow.In Proceedings of CoRR.2013.

The content of the invention

For in communication network because the factor such as interference, noise be likely to occur packet packet loss, get lines crossed, sender's reception The fuzzy node that causes of situations such as side is indefinite or wrong, redundant node are identified, the invention enables the communication for reconstructing Network topology is close to its real structure.Concrete technical scheme is as follows：

A kind of method for recognizing fuzzy redundant node identity in communication network, comprises the following steps：

(1) communication network data is obtained, communication network data is first distinguished known determination part G_k=＜ V_k,E_k＞ and Fuzzy uncertain part, each uncertain node is represented with a placeholder；Build preliminary annexation figure G_a=＜ V_a,E_a＞, wherein, ＜ V_a,E_a＞, ＜ V_k,E_k＞ is connected respectively graph of a relation G_a、G_kIncluded in node set and Lian Bian Set, calculates placeholder number | V_p|=| V_a|-|V_k|；V_kRepresent known and determine node, V_pRepresent placeholder, i.e., uncertain mould Paste node, V_aRepresent whole nodes；

(2) Gauss distance calculating method is adopted, calculates preliminary annexation figure G_aIn all nodes between association square Battle arrayWherein | V_a| for node set V_aIn element number；Represent real number field,Represent in real number field | V_a| Row, | V_a| the set of column matrix.

(3) defineFor diagonal matrix, wherein, diagonal element D_iiFor the i-th row element sum of incidence matrix C, I=1 .., | V_a|, bySet up corresponding matrixWhereinEach of expression diagonal matrix D Element is made even the inverse of root；

(4) assume that the source node number that fuzzy node should be represented is known, be designated as h, the maximum for finding h L is special Vector is levied, matrix is constitutedThe h characteristic vector of wherein L is respectively the row of Q；

(5) normalized matrix Q makes every behavior unit length, is designated as matrix Q '；

(6) remove the known row for determining node of the middle correspondences of matrix Q ', retain the row of correspondence placeholder in matrix, obtain matrixNow, one placeholder of every a line correspondence in matrix Q "；

(7) Q " in row using k-mediods clusterings into h class；

(8) placeholder in same class is merged into into a node, the node being merged into is former with the placeholder of the apoplexy due to endogenous wind The neighbours for coming are connected.

Further, incidence matrix is calculated using Gauss distance calculating method in step (2)Process It is as follows：

Define d_iFor preliminary annexation figure G_aVectors of the interior joint i to the shortest path length of other all nodes, together Reason, d_jFor preliminary annexation figure G_aVectors of the interior joint j to the shortest path length of other all nodes, then

Parameter σ is the standard deviation of Gauss range formula, | | | | expression asks vector field homoemorphism computing, e to represent math constant, i.e., The truth of a matter of natural logrithm, C_ijThe element of corresponding i-th row jth column position in representing matrix C.Because lower (the net of the bigger sensitivity of σ Network increase and decrease a line between node Gauss distance affect it is less), σ more sluggishness is higher, when value is between 3-5 discrimination compared with Good (in embodiment, the general values of σ are 4).

The beneficial effect obtained using the present invention：The network topology knot that the present invention is caused for data in communication network disappearance Structure reconstructs imperfect inaccurate problem, to communication network in fuzzy node distinguished and merged, to cause weight as far as possible The communication network topology that structure goes out is close to its real structure, at present both at home and abroad also without for Fuzzy Redundancy section in identification communication network The related research of point problem, by the experiment test present invention for the identification of Fuzzy Redundancy node can reach higher standard True rate (more than 75%).By the present invention for the identification and merging of Fuzzy Redundancy node, it is possible to increase communication network topology is tied The accuracy of structure reconstruct, is further to carry out Analytic Network Process, communication process analysis, Business Process Analysis on a communication network Support Deng infallible data is provided.

Description of the drawings

Fig. 1 is correct diagram of communications networks and contains Fuzzy Redundancy nodal communication network figure contrast schematic diagram；

Fig. 2 is Clustering Effect schematic diagram；

Fig. 3 is the inventive method flow chart；

Fig. 4 is artificial network original graph；

Fig. 5 is network after emulation data set fractionation；

Fig. 6 is open Network data set Football primitive network figures.

Specific embodiment

Below, with reference to the drawings and specific embodiments, the invention will be further described.

The present invention is to remove what a part determined for the basic ideas for obscuring redundant node identification problem in communication network Outside node, for the sender, recipient occurred in the communication information for getting and probably due to information lose and vacancy Or indefinite sender, recipient are represented with a placeholder respectively, a preliminary annexation figure, Ran Houyi is built These placeholders are gathered by the topological characteristic according to the attribute of these placeholders and in annexation figure using spectral clustering Class, determines which placeholder actually represents same node, and placeholder of the cluster in same group is merged into a communication Node, so as to reach the target for recognizing and removing fuzzy, redundant node, realizes the reconstruct of communication network.

During the structure original to communication network of the communication data by getting is reduced, because noise, The factors such as interference, cause the imperfect or even mistake of communication data, there are some uncertain nodes in the network for constructing unavoidably, Multiple redundant nodes may be easily mistaken for into, as shown in figure 1, during figure (a) represents that correct diagram of communications networks, figure (b) represent network Containing Fuzzy Redundancy node schematic diagram, in figure No. 6, No. 7 nodes are Fuzzy Redundancy node, by the algorithm in the present invention to this A little redundancies do not know node and are clustered, and identify its real identity, as shown in Fig. 2 so as to restore as close possible to true Real communication network topology structure.

First below the gatherer process of data is illustrated.It is common using emulation data set and true social activity Network data set With verification algorithm effectiveness.The thinking of checking is to concentrate a part of node of random selection to be torn open from emulation data and public data It is divided into multiple nodes (i.e.：Redundant node), if the node after fractionation can be identified the corresponding node of its original by algorithm, Prove that algorithm is effective.For the subgraph for obtaining, ξ node is randomly selected from whole nodes as experiment point set V_m, V_mIt is mould Then this ξ point split into several by source node that paste node should be represented respectively, is constituted fuzzy set of node, is used occupy-place Symbol v' ∈ V_pReplace, V_pPlaceholder collection is represented, this just constitutes preliminary annexation figure G_a.The node for so splitting out is some Individual one source node of its real representation, what is splitted out is all represented with placeholder, and whole data set processing procedure is exactly data set structure Process is made, a training set for embodiment is constructed.The purpose of algorithm is placeholder (the i.e. Fuzzy Redundancy for splitting out Node) go back to cluster, find out which fuzzy node actually represents same source node.

BA scales-free networks are generated with Matlab emulation, create-rule is BA (m0, m, N, pp), before wherein m0 is growth Network node number, m are side number newly-generated when introducing new node every time, and N is the network size after increasing, and pp is initial network Situation, the value of pp have 1,2,3 three kinds, and wherein 1 represents it is all isolated；2 represent composition complete graph；3 expressions connect at random Side.Be respectively provided with different parameters generate two scales-free networks, as shown in figure 4, figure (a) be 1 corresponding parameter BA of network (4,3, 100,2)；Figure (b) is 2 corresponding parameter BA (10,6,100,3) of network.

Three points are randomly selected in experiment respectively in two networks as experimental point, choose in network 1 for 9,19,57 Number three nodes, are 69,87, No. 93 three nodes in network 2.The communication information transmission that hypothesis is carried out on that network has makes an uproar Sound, the communication existence information that these nodes are carried out are lost, so as to cause these nodes not occur in the network architecture, Carry out during data collection, being marked as multiple different identity, algorithm proposed by the present invention is exactly by uncertain node in network It is identified, restores real network node.

Find the nearest-neighbors of three experiment nodes in the two networks respectively, then experiment node is torn open respectively Point, shown in corresponding relation following table.

Node corresponding relation is tested in the emulation data of table 1

As shown in figure 5, the node after splitting is connected with the neighbours of former experiment node respectively, each is new for network after fractionation Point retains the company side of former experiment node half number, preliminary annexation figure G_a。

Public data integrates as Football networks, the data set record participate within 1998 22 football teams of world cup into Signing situation between member is national at 35, the company side in network represent certain member from country's output to another state Family.The data set is regarded as and haves no right Undirected networks, comprising 35 nodes, 118 sides.Initial network connection figure is as shown in Figure 6. Arbitrarily choose No. 3, No. 14, No. 16 three nodes as experiment node, respectively these three nodes are split, each section Point splits into four parts, splits posterior nodal point as shown in table 2 with origin node corresponding relation.

Table 2Football data sets test node corresponding relation

As shown in figure 3, being flow chart of the present invention；Specific embodiment one, by taking Football Network data sets as an example (in experiment 4) σ takes：

(1) for the experimental data set for having constructed, first distinguish out known determination part G_k=＜ V_k,E_k＞ and it is fuzzy not Determine part, (this 3 nodes constitute set V except 3,14, No. 16 nodes in 35 nodes in its Central Plains network_m) beyond 32 Individual node composition is known to determine part V_k, 12 nodes are Fuzzy Redundancy node V to 36 to No. 47 for splitting out altogether_p, each obscures Redundant node builds preliminary annexation figure G as 1 placeholder_a=＜ V_a,E_a＞, ＜ V_a,E_a＞ is respectively annexation figure G_aComprising node set and even line set, V_aIt is known determination node V_kWith Fuzzy Redundancy node V_pUnion, E_aRepresent V_a Annexation between interior joint；

(2) Gauss distance calculating method is adopted, calculates preliminary annexation figure G_aIn all nodes between association square Battle arrayWherein 44 is set V_aIn element number, that is, G_aIn the nodes that include；

(3) diagonal element D_ii(the i-th row element sum of i=1 .., 4 for incidence matrix C, bySet up phase The matrix answered

(4) the source node number h=3 that node should be represented is obscured, the maximal eigenvector for finding 3 L is (special to repeating Value indicative looks for vertical characteristic vector), constitute matrixWherein the 3 of L characteristic vectors are respectively the row of Q；

(5) standardization Q matrixes make every behavior unit length, are designated as Q '；

(6) in removing Q ' matrixes, (algorithm is each node is corresponding according to node serial number to the known row for determining node of correspondence A line of matrix, such as node i the i-th row of correspondence, is front 32 row of correspondence in the present embodiment；Correspondence placeholder in reservation matrix OK, in the present embodiment be matrix last 12 row), obtain matrixNow, Q " correspondences one per a line in matrix Placeholder；

(7) Q " in 12 rows be clustered into 3 using a kind of k-mediods methods (existing sample clustering method) Class, that is, this 12 placeholders are clustered to No. 36 to No. 47, cluster result is：

3 placeholder cluster result of table

36 38 39 43 47

37 40 41 42

44 45 46

(8) placeholder in same class is merged into into a node, the neighbours original with the placeholder of the apoplexy due to endogenous wind are connected. The result that cluster is obtained, that is, the corresponding relation such as following table for clustering the source node and Fuzzy Redundancy node for obtaining：

Table 4 tests node corresponding relation

Node serial number	36 38 39 43 47	37 40 41 42	44 45 46
				Obtain corresponding source node serial number	3	14	16

The accuracy of cluster is 9/12=0.75, in illustrating that the algorithm effectively can identify network as can be seen from the results The true identity of the uncertain node of redundancy.

Specific embodiment two, is tested with emulation data according to the step of the present invention.Using the algorithm pair in the present invention Uncertain node in two networks that emulation is obtained is analyzed, cluster result such as following table：

Table 5 emulates data prediction result

Node serial number	101	102	103	104	105	106	107	108	109	110	Accuracy
												Network 1	9	57	19	9	57	57	57	57	19	19	0.8
Network 2	69	69	69	69	93	93	87	87	87	87	1

In table, secondary series, the 3rd row obscure the corresponding packet of redundant node, that is, corresponding original network after respectively clustering In experiment node, it can be seen that for network 1, algorithm predicts accuracy reaches 0.8, and the prediction accuracy of network 2 reaches 1.0, illustrate that the algorithm can effectively identify the true identity of the uncertain node of redundancy in network.

It is more than that the present invention is exemplarily described, it is clear that the realization of the present invention is not subject to the restrictions described above, As long as employ the various improvement that technical solution of the present invention is carried out, or it is not improved will the present invention design and technical scheme it is direct Using other occasions, within the scope of the present invention.

Claims (2)

1. it is a kind of for recognizing the method for obscuring redundant node identity in communication network, it is characterised in that to comprise the following steps：

(1) communication network data is obtained, communication network data is first distinguished known determination part G_k=＜ V_k,E_k＞ and fuzzy Uncertain part, each uncertain node is represented with a placeholder；Build preliminary annexation figure G_a=＜ V_a,E_a ＞, wherein, ＜ V_a,E_a＞, ＜ V_k,E_k＞ is connected respectively graph of a relation G_a、G_kIncluded in node set and even line set, Calculate placeholder number | V_p|=| V_a|-|V_k|；V_kRepresent known and determine node, V_pRepresent placeholder, i.e., uncertain fuzzy section Point, V_aRepresent whole nodes；

(2) Gauss distance calculating method is adopted, calculates preliminary annexation figure G_aIn all nodes between incidence matrixWherein | V_a| for node set V_aIn element number,Represent real number field；

(3) defineFor diagonal matrix, wherein, diagonal element D_iiFor the i-th row element sum of incidence matrix C, i= 1,..,|V_a|, bySet up corresponding matrixWhereinRepresent each element of diagonal matrix D Make even the inverse of root；

(4) assume that the source node number that fuzzy node should be represented is known, be designated as h, find the maximum feature of h L to Amount, constitutes matrixThe h characteristic vector of wherein L is respectively the row of Q；

(5) normalized matrix Q makes every behavior unit length, is designated as Q '；

(6) remove the known row for determining node of the middle correspondences of matrix Q ', retain the row of correspondence placeholder in matrix, obtain matrixNow, one placeholder of every a line correspondence in matrix Q "；

(7) Q " in row using k-mediods clusterings into h class；

(8) placeholder in same class is merged into into a node, the neighbours original with the placeholder of the apoplexy due to endogenous wind are connected.

2. as claimed in claim 1 a kind of for recognizing the method for obscuring redundant node identity in communication network, its feature exists In using Gauss distance calculating method calculating incidence matrix in step (2)Process it is as follows：

Define d_iFor preliminary annexation figure G_aVectors of the interior joint i to the shortest path length of other all nodes, in the same manner, d_j For preliminary annexation figure G_aVectors of the interior joint j to the shortest path length of other all nodes, then

$C_{ij}=e^{\frac{-\left|\right|d_i-d_j|{|}^2}{2{\mathrm{\σ}}^2}}$

Parameter σ is the standard deviation of Gauss range formula, | | | | vector field homoemorphism computing is asked in expression.