CN103313354B

CN103313354B - Based on the heterogeneous network system of selection of four kinds of weight vectors

Info

Publication number: CN103313354B
Application number: CN201310192317.5A
Authority: CN
Inventors: 朱琦; 张硕
Original assignee: Nanjing Post and Telecommunication University
Current assignee: Nanjing Post and Telecommunication University
Priority date: 2013-05-22
Filing date: 2013-05-22
Publication date: 2016-04-13
Anticipated expiration: 2033-05-22
Also published as: CN103313354A

Abstract

The invention is a heterogeneous network selection method based on four weight vectors. The method obtains a comprehensive weight vector by synthesizing four network attribute weights, and ensures its rationality through compatibility inspection. Then combine the SAW method for network selection. This method fully considers the current network conditions and can provide satisfactory service quality according to the business type. The specific steps are as follows: First, assign objective parameters to calculate the objective network attribute weight vector by entropy method and standard deviation method, assign subjective parameters to obtain respective subjective network attribute weight vectors by AHP method and G-1 method, four kinds The method represents four decision makers, uses the group decision-making theory to combine four weight vectors to generate a new weight vector, thereby obtaining a comprehensive weight vector W, and finally combines the SAW method for network selection.

Description

A Heterogeneous Network Selection Method Based on Four Weight Vectors

技术领域technical field

本发明涉及一种基于四种权值向量的异构网络选择方法，属于通信技术领域。The invention relates to a heterogeneous network selection method based on four weight vectors, belonging to the technical field of communication.

背景技术Background technique

移动通信经历了从提供模拟语音业务的第一代模拟蜂窝系统1G、采用数字调制的第二代移动通信系统2G到以CDMA为主流技术的第三代移动通信系统3G的过程，随着无线通信的不断发展，未来的通信网络将会是包含各种无线接入技术的异构网络，例如2G，3G，UMTS，WLAN,WiMAX，WiFi等，这些不同类型的无线网络往往相互重叠覆盖，没有统一的标准接口，而且它们在覆盖范围、接入速率、容量、业务和移动性特点、网络质量等方面存在较大差别，又各有优势，比如通过多接口终端在异构网络间实现无缝接入，用户可以选择UMTS以获得好的服务质量，或者选择WiMAX以获得高数据速率，或者选择WLAN以获得较低的成本，这就需要充分利用各个网络的优势，所以异构网络的融合是未来网络通信的发展趋势。Mobile communication has gone through the process from the first generation analog cellular system 1G providing analog voice services, the second generation mobile communication system 2G using digital modulation to the third generation mobile communication system 3G with CDMA as the mainstream technology. With continuous development, the future communication network will be a heterogeneous network containing various wireless access technologies, such as 2G, 3G, UMTS, WLAN, WiMAX, WiFi, etc. These different types of wireless networks often overlap and cover each other, and there is no unified standard interfaces, and they have great differences in coverage, access rate, capacity, service and mobility characteristics, network quality, etc., and each has its own advantages, such as seamless connection between heterogeneous networks through multi-interface terminals Income, users can choose UMTS to obtain good service quality, or choose WiMAX to obtain high data rate, or choose WLAN to obtain lower cost, which needs to make full use of the advantages of each network, so the integration of heterogeneous networks is the future The development trend of network communication.

异构网络选择是实现网络融合的关键步骤。而多属性决策方法MADM(MultipleAttributesDecisionMaking)是异构网络选择方法中最有效的方法之一。经典的MADM方法包括简单加权方法SAW、乘法指数权值方法MEW、接近理想值的排序方法TOPSIS、灰色关联分析法GRA、VIKOR法和淘汰选择法ELECTRE等。这些多属性决策方法都会涉及多属性权重向量，多属性权重的向量，一般要考虑网络的客观属性，也应当充分考虑用户偏好以及业务类型。常用的客观赋权法包括熵权法EW（EntropyMethod）和标准离差法等；经典的主观赋权法包括层次分析法AHP(AnalyticHierarchyProcess)和G-1法。Heterogeneous network selection is a key step in realizing network convergence. The multi-attribute decision-making method MADM (Multiple Attributes Decision Making) is one of the most effective methods in heterogeneous network selection methods. Classical MADM methods include simple weighting method SAW, multiplicative index weight method MEW, sorting method TOPSIS close to ideal value, gray relational analysis method GRA, VIKOR method and elimination selection method ELECTRE, etc. These multi-attribute decision-making methods all involve multi-attribute weight vectors and multi-attribute weight vectors. Generally, the objective attributes of the network should be considered, and user preferences and business types should also be fully considered. Commonly used objective weighting methods include entropy weight method EW (EntropyMethod) and standard deviation method; classic subjective weighting methods include AHP (Analytic Hierarchy Process) and G-1 method.

很多方法将主客观权重综合起来进行网络选择。《无线电工程》2009年第39卷第1期刊载的“一种多属性决策的异构网络选择算法”，作者：王康，曾志民，冯春燕，张天魁（北京邮电大学通信网络综合技术研究所），提出采用AHP求主观权重，熵权法求客观权重，然后将二者线性加权，结合SAW方法给出一种网络选择方法，但此方法并未给出线性加权系数的求法，即权重向量的确定太过随意。《InformationComputingandApplicationsLectureNotesinComputerScience》出版的编号为“6377/2010:213-220”文献“ANovelAHPandGRABasedHandoverDecisionMechanisminHeterogeneousWirelessNetworks”提出一种AHP结合GRA的网络选择方法，这种方法应用复杂度较大，并且没有足够的理论支撑。并且上述两种方法虽然主客观兼顾，能考虑网络的客观情况并能兼顾用户偏好，但能充分考虑业务类型。《中国邮电学报》2012年十月出版的编号为“19（5）：92-98”的文献“Networkselectionbasedonmultipleattributedecisionmakingandgroupdecisionmakingforheterogeneouswirelessnetworks”采用AHP求主观权重，熵权法求客观权重，然后对二者进行群组决策，得到一种权衡算法，但只采用两个决策者时，决策者参与数量太少，使得群组决策并不能充分利用其集合多个决策者智慧的特点。Many methods combine subjective and objective weights for network selection. "A heterogeneous network selection algorithm for multi-attribute decision-making" published in "Radio Engineering" Volume 39, Issue 1, 2009, Authors: Wang Kang, Zeng Zhimin, Feng Chunyan, Zhang Tiankui (Beijing University of Posts and Telecommunications Institute of Communication Network Technology) , it is proposed to use AHP to calculate the subjective weight, and the entropy weight method to calculate the objective weight, and then linearly weight the two, and combine the SAW method to give a network selection method, but this method does not give the method of linear weighting coefficient, that is, the weight vector Definitely too casual. The document "A Novel AHP and GRA Based Handover Decision Mechanism in Heterogeneous Wireless Networks" published by "Information Computing and Applications Lecture Notes in Computer Science" with the number "6377/2010:213-220" proposes a network selection method combining AHP and GRA. This method is complex in application and does not have sufficient theoretical support. Moreover, although the above two methods are both subjective and objective, and can consider the objective situation of the network and user preferences, they can fully consider the type of business. The document "Network selection based on multiple attribute decision making and group decision making for heterogeneous wireless networks" published in October 2012 by "China Journal of Posts and Telecommunications" with the serial number "19 (5): 92-98" uses AHP to find the subjective weight and the entropy weight method to find the objective weight, and then makes group decisions on the two. A trade-off algorithm is obtained, but when only two decision makers are used, the number of decision makers participating is too small, so that group decision-making cannot make full use of its characteristics of gathering the wisdom of multiple decision makers.

发明内容Contents of the invention

技术问题：本发明的目的是提供一种基于四种权值向量的异构网络选择方法，该方法利用多属性决策方法得到的多种属性权重向量，基于群组决策理论，对它们进行组合从而获取一个几何综合权值向量。该结果如果不满足相容性要求，则需要修改主观赋权法的判决矩阵，再次计算，直到得到一个满足相容性要求的综合权重向量，然后结合SAW法进行网络选择，该方法可以为不同业务类型提供令人满意的服务质量。Technical problem: The purpose of the present invention is to provide a heterogeneous network selection method based on four weight vectors. This method uses multiple attribute weight vectors obtained by the multi-attribute decision-making method, and combines them based on the group decision-making theory. Get a vector of geometric synthesis weights. If the result does not meet the compatibility requirements, it is necessary to modify the decision matrix of the subjective weighting method and calculate again until a comprehensive weight vector that meets the compatibility requirements is obtained, and then combined with the SAW method for network selection. This method can be used for different The business type provides a satisfactory quality of service.

本发明针对现有技术的不足，提出了一种改进的基于群组决策的组合权重的方法。首先通过G-1法和AHP法分别得到各自的主观网络属性权值向量，然后利用熵值法和标准离差法计算出各自的客观网络属性。四种方法代表了四个决策者，两个主观决策者，两个客观决策者，决策者数量适中而且具有代表性，也同时避免了上述方法的问题。Aiming at the deficiencies of the prior art, the present invention proposes an improved method for combining weights based on group decision-making. Firstly, the respective subjective network attribute weight vectors are obtained by the G-1 method and the AHP method, and then the respective objective network attributes are calculated by using the entropy method and the standard deviation method. The four methods represent four decision makers, two subjective decision makers, and two objective decision makers. The number of decision makers is moderate and representative, and at the same time avoids the problems of the above methods.

技术方案：本发明即一种基于四种权值向量的异构网络选择方法，通过综合多个决策者的智慧，考虑网络客观属性，用户偏好和业务类型，并根据业务类型可以提供更令人满意的网络选择。Technical solution: The present invention is a heterogeneous network selection method based on four weight vectors. By synthesizing the wisdom of multiple decision makers, considering the objective attributes of the network, user preferences and service types, and according to the service types, it can provide more impressive Satisfied network choice.

在基于四种权值向量的异构网络选择方法中，首先要获取网络属性权重。多属性决策一般都会涉及权重向量的确定，传统的权重向量的确定方法可以分为主观赋权法和客观赋权法。主观赋权法包括AHP、G-1法和Delphi法等；客观赋权法包括熵值法、标准离差法、CRITIC法等。两类赋权法各有侧重和特点。因此，为了综合考虑业务类型、用户偏好、网络客观属性等因素，并充分利用群组决策权衡和集合多个决策者结论的特点，本方法将综合多种主客观权重来得到多属性权重向量。其中主观赋权法采用AHP和G-1法，客观赋权法采用熵值法和标准离差法。使用四种赋权法得到四组权重向量后，通过群组决策获取一个综合权重向量，然后衡量其相容性来判定综合权重向量的合理性。满足相容性的综合权重向量将结合SAW法进行网络选择。In the heterogeneous network selection method based on four weight vectors, the network attribute weights should be obtained first. Multi-attribute decision-making generally involves the determination of weight vectors. Traditional methods for determining weight vectors can be divided into subjective weighting methods and objective weighting methods. Subjective weighting methods include AHP, G-1 method, and Delphi method; objective weighting methods include entropy method, standard deviation method, CRITIC method, etc. The two types of empowerment laws have their own emphases and characteristics. Therefore, in order to comprehensively consider factors such as business types, user preferences, and network objective attributes, and make full use of the characteristics of group decision-making trade-offs and the aggregation of conclusions of multiple decision makers, this method will synthesize various subjective and objective weights to obtain multi-attribute weight vectors. The subjective weighting method adopts AHP and G-1 method, and the objective weighting method adopts entropy method and standard deviation method. After using the four weighting methods to obtain four sets of weight vectors, a comprehensive weight vector is obtained through group decision-making, and then its compatibility is measured to determine the rationality of the comprehensive weight vector. The comprehensive weight vector that satisfies compatibility will be combined with the SAW method for network selection.

客观赋权法采用熵值法和标准离差法。在本发明的异构网络模型之下，网络选择中涉及的属性参数为可用带宽、峰值传输速率、包时延、包抖动、包丢失与每比特费用，其中：B代表带宽、R代表峰值传输速率、D代表包时延、J代表包抖动、L代表包丢失、C代表每比特费用。可以根据网络属性参数计算出属性的客观权重。The objective weighting method uses the entropy method and the standard deviation method. Under the heterogeneous network model of the present invention, the attribute parameters involved in network selection are available bandwidth, peak transmission rate, packet delay, packet jitter, packet loss and cost per bit, wherein: B represents bandwidth, R represents peak transmission Rate, D stands for packet delay, J stands for packet jitter, L stands for packet loss, and C stands for cost per bit. The objective weight of the attribute can be calculated according to the network attribute parameters.

本发明的异构网络模型包含了N种异构网络。熵值法和标准离差法中假设该方法的网络属性矩阵为：The heterogeneous network model of the present invention includes N types of heterogeneous networks. In the entropy method and the standard deviation method, it is assumed that the network attribute matrix of this method is:

R＝(x_ij)_N×6，即N个异构网络数目，6种网络属性。R=(x _ij ) _N×6 , that is, the number of N heterogeneous networks and 6 types of network attributes.

x_ij表示第i个网络的第j个属性，如：x _ij represents the jth attribute of the i-th network, such as:

x_i1表示第i个网络的带宽；x _i1 represents the bandwidth of the i-th network;

x_i2表示第i个网络的峰值传输速率；x _i2 represents the peak transmission rate of the i-th network;

x_i3表示第i个网络的时延；x _i3 represents the delay of the i-th network;

x_i4表示第i个网络的抖动；x _i4 represents the jitter of the i-th network;

x_i5表示第i个网络的丢包率；x _i5 represents the packet loss rate of the i-th network;

x_i6表示第i个网络的每比特费用；x _i6 represents the cost per bit of the i-th network;

并且1<=i<=N，1<=j<=6。And 1<=i<=N, 1<=j<=6.

一般而言，属性分为效益型和成本型，效益型的属性越大越好，取 $x_{\max}^{j} = \max (x_{1 j}, x_{2 j}, . . ., x_{Nj});$ Generally speaking, attributes are divided into benefit type and cost type. The larger the benefit type attribute is, the better it is. $x_{\max}^{j} = \max (x_{1 j}, x_{2 j}, . . ., x_{Nj});$

成本型的属性越小越好，取 $x_{\min}^{j} = \min (x_{1 j}, x_{2 j}, . . ., x_{Nj}) .$ The smaller the attribute of the cost type, the better. Take $x_{\min}^{j} = \min (x_{1 j}, x_{2 j}, . . ., x_{Nj}) .$

对于B和R，它们属于效益型属性，标准化处理公式为：For B and R, they are benefit attributes, and the standardized processing formula is:

${r r}_{ij ij} = = \frac{{x x}_{ij ij}}{{x x}_{max max}^{j j} + + {x x}_{min min}^{j j}},, 11 \leq \leq j j \leq \leq 22$

对于D，J，L和C，，它们属于成本型属性，标准化处理公式为：For D, J, L, and C, they are cost-type attributes, and the normalization formula is:

${r r}_{ij ij} = = \frac{{x x}_{max max}^{i i} + + {x x}_{mim mime}^{i i} - - {x x}_{ij ij}}{{x x}_{max max}^{j j} + + {x x}_{min min}^{j j}},, 33 \leq \leq j j \leq \leq 66$

通过熵值法可以得到属性的权重向量：The weight vector of the attribute can be obtained by the entropy method:

${W W}^{EW EW} = = (({w w}_{11}^{EW EW},, {w w}_{22}^{EW EW},, {w w}_{33}^{EW EW},, {w w}_{33}^{EW EW},, {w w}_{44}^{EW EW} {w w}_{55}^{EW EW} {w w}_{66}^{EW EW}))$

利用标准离差法计算出的权重向量为：The weight vector calculated by the standard deviation method is:

${W W}^{σ σ} = = (({w w}_{11}^{σ σ},, {w w}_{22}^{σ σ} {,, w w}_{33}^{σ σ},, {w w}_{44}^{σ σ} {,, w w}_{55}^{σ σ} {,, w w}_{66}^{σ σ}))$

本发明主观赋权法采用AHP法以及G-1法。AHP的判断矩阵中每一个元素都是一个属性相对于另一个同准则层下的属性的权重的比例，可以将AHP判决矩阵记为B，则B矩阵写成如下形式：The subjective weighting method of the present invention adopts the AHP method and the G-1 method. Each element in the judgment matrix of AHP is the weight ratio of an attribute relative to another attribute under the same criterion layer. The AHP judgment matrix can be recorded as B, and the B matrix can be written as follows:

$B B = = [\begin{matrix} {b b}_{1111} & {b b}_{1212} & . . . . . . & {b b}_{11 n no} \\ {b b}_{21 twenty one} & {b b}_{22 twenty two} & . . . . . . & {b b}_{22 n no} \\ . . . . . . & . . . . . . & . . . . . . & . . . . . . \\ {b b}_{n no 11} & {b b}_{n no 22} & . . . . . . & {b b}_{nn n} \end{matrix}] = = [\begin{matrix} {w w}_{11} / / {w w}_{11} & {w w}_{11} / / {w w}_{22} & . . . . . . & {w w}_{11} / / {w w}_{n no} \\ {w w}_{22} / / {w w}_{11} & {w w}_{22} / / {w w}_{22} & . . . . . . & {w w}_{22} / / {w w}_{n no} \\ . . . . . . & . . . . . . & . . . . . . & . . . . . . \\ {w w}_{n no} / / {w w}_{11} & {w w}_{n no} / / {w w}_{22} & . . . . . . & {w w}_{n no} / / {w w}_{n no} \end{matrix}]$

其中：n是网络属性参数总数，b_ij表示第i个属性相对于第j个属性的重要程度。关于如何确定b_ij的值，Saaty等建议引用数字1-9及其倒数作为标度。表1列出了1-9标度的含义来表示基于选择者的个性、经验和知识而进行的重要程度的参数选择。Among them: n is the total number of network attribute parameters, b _ij represents the importance of the i-th attribute relative to the j-th attribute. Regarding how to determine the value of b _ij , Saaty et al. suggested quoting the numbers 1-9 and their reciprocals as scales. Table 1 lists the meaning of a scale of 1-9 to represent the importance of parameter selection based on the selector's personality, experience, and knowledge.

从心理学观点来看，分级太多会超越人们的判断能力，既增加了作判断的难度，又容易因此而提供虚假数据。Saaty等人还用实验方法比较了在各种不同标度下人们判断结果的正确性，实验结果也表明，采用1-9的标度最为合适。From a psychological point of view, too many classifications will exceed people's ability to judge, which not only increases the difficulty of making judgments, but also easily provides false data. Saaty et al. also used experimental methods to compare the correctness of people's judgment results under various scales. The experimental results also showed that the scale of 1-9 is the most appropriate.

表1.Saaty标度Table 1. Saaty scale

序号serial number 重要性等级importance level b_ij赋值b _ij assignment 11 i,j两元素同等重要i and j elements are equally important 11 22 i元素比j元素稍微重要The i element is slightly more important than the j element 33 33 i元素比j元素明显重要The i element is significantly more important than the j element 55 44 i元素比j元素强烈重要The i element is strongly more important than the j element 77 55 i元素比j元素极端重要The i element is extremely more important than the j element 99 66 i元素比j元素稍不重要The i element is slightly less important than the j element 1/31/3 77 i元素比j元素明显不重要The i element is significantly less important than the j element 1/51/5 88 i元素比j元素强烈不重要The i element is stronger and less important than the j element 1/71/7 99 i元素比j元素极端不重要The i element is extremely less important than the j element 1/91/9

根据业务类型的不同，主观赋权法的判决矩阵会发生变化。AHP法中的判断矩阵为A^AHP，G-1法的判断矩阵为G^G-1。因此在每种业务类型之下，结合实际应用中各个属性之间的相对重要性，并结合Saaty标度，可以得到A^AHP和G^G-1，AHP和G-1法的不同类型判决矩阵分别在表2和表3中给出。Depending on the type of business, the judgment matrix of the subjective empowerment method will change. The judgment matrix in the AHP method is A ^AHP , and the judgment matrix in the G-1 method is G ^G-1 . Therefore, under each business type, combined with the relative importance of each attribute in the actual application, and combined with the Saaty scale, the different types of decision matrices of the A ^AHP and G ^G-1 , AHP and G-1 methods can be obtained respectively It is given in Table 2 and Table 3.

AHP法确定主观权重：对应于每种业务类型，结合表2，根据AHP计算权重步骤，特定业务类型条件下，可以算出AHP法确定的主观权重：AHP method to determine subjective weight: corresponding to each type of business, combined with Table 2, according to the AHP calculation weight steps, under the condition of a specific business type, the subjective weight determined by the AHP method can be calculated:

${W W}^{AHP AHP} = = (({w w}_{11}^{AHP AHP},, {w w}_{22}^{AHP AHP},, {w w}_{33}^{AHP AHP},, {w w}_{44}^{AHP AHP},, {w w}_{55}^{AHP AHP},, {w w}_{66}^{AHP AHP}))$

G-1法确定主观权重分为三个步骤，首先根据某个评价标准对所有评价指标进行重要性排序，然后给定排序后相邻指标的重要程度比值，最后计算各个指标的权重。某种特定业务类型下的G-1法确定的主观权值向量为：The G-1 method is divided into three steps to determine the subjective weight. First, the importance of all evaluation indicators is sorted according to a certain evaluation standard, and then the importance ratio of the adjacent indicators after sorting is given, and finally the weight of each indicator is calculated. The subjective weight vector determined by the G-1 method under a certain type of business is:

${W W}^{G G - - 11} = = (({w w}_{11}^{G G - - 11},, {w w}_{22}^{G G - - 11},, {w w}_{33}^{G G - - 11},, {w w}_{44}^{G G - - 11} {w w}_{55}^{G G - - 11} {w w}_{66}^{G G - - 11}))$

群组决策通过集中群体成员智慧来发挥群体决策的优势，因此所选群体成员应当具有代表性并要保证一定数量。本方法所选的四种决策成员分别为层次分析法、熵值法、G-1法和标准离差法，四种决策方法包括两种主观赋权法、两种客观赋权法，数量适中，通过群组决策获取网络属性权重参数不仅能利用多个决策者参与时群组决策的优势，更能充分考虑用户要求、业务类型以及网络状况。Group decision-making takes advantage of group decision-making by concentrating the wisdom of group members, so the selected group members should be representative and a certain number must be guaranteed. The four decision-making members selected by this method are AHP, entropy value method, G-1 method and standard deviation method. The four decision-making methods include two subjective weighting methods and two objective weighting methods, and the number is moderate , obtaining network attribute weight parameters through group decision-making can not only take advantage of group decision-making when multiple decision makers participate, but also fully consider user requirements, business types and network conditions.

设A＝（a_ij）,B＝（b_ij）和C＝（c_ij）均为n阶正互反矩阵，由A得到的排序向量为W＝（w₁,w₂,...,w_n）^T，矩阵W＝(w_i/w_j)称为A的特征矩阵。定义A和B的乘积C(A,B)＝e^T*A*B*e为A,B的相容度，其中e^T=(1,1,…1)。Assume that A=(a _ij ), B=(b _ij ) and C=(c _ij ) are all positive and reciprocal matrices of order n, and the sorting vector obtained from A is W=(w ₁ ,w ₂ ,..., w _n ) ^T , the matrix W=(w _i /w _j ) is called the characteristic matrix of A. Define the product C(A,B)=e ^T *A*B*e of A and B as the compatibility of A and B, where e ^T =(1,1,...1).

为了方便，一般取其对数作为相容度，记为：一般的，LC（A,B）≥0，如果LC（A,B）=0，则A,B完全相容。For convenience, the logarithm is generally taken as the compatibility degree, which is recorded as: Generally, LC (A, B) ≥ 0, if LC (A, B) = 0, then A, B are completely compatible.

四种赋权法基于同一个网络参数矩阵，由这四种赋权法确定的排序向量分别为W^EW,W^σ,W^AHP和W^G-1，为方便计算，记：The four weighting methods are based on the same network parameter matrix, and the sorting vectors determined by these four weighting methods are W ^EW , W ^σ , W ^AHP and W ^G-1 respectively. For the convenience of calculation, remember:

${W W}^{ew ew} = = {W W}^{11} = = (({w w}_{11}^{11},, {w w}_{22}^{11},, {w w}_{33}^{11},, {w w}_{44}^{11},, {w w}_{55}^{11},, {w w}_{66}^{11}))$

${W W}^{σ σ} = = {W W}^{22} = = (({w w}_{11}^{22},, {w w}_{22}^{22},, {w w}_{33}^{22},, {w w}_{44}^{22} {,, w w}_{55}^{22},, {w w}_{66}^{22}))$

${W W}^{AHP AHP} = = {W W}^{33} = = (({w w}_{11}^{33},, {w w}_{22}^{33},, {w w}_{33}^{33},, {w w}_{44}^{33},, {w w}_{55}^{33},, {w w}_{66}^{22}))$

${W W}^{G G - - 11} = = {W W}^{44} = = (({w w}_{11}^{44},, {w w}_{22}^{44},, {w w}_{33}^{44},, {w w}_{44}^{44},, {w w}_{55}^{44},, {w w}_{66}^{44}))$

四个排序向量对应的特征矩阵分别为A^EW,A^σ,A^AHP和A^G-1。几何综合平均向量W＝(w₁,w₂,w₃,w₄,w₅,w₆)是对数意义下使得A^EW,A^σ,A^AHP和A^G-1与综合特征矩阵W＝(w_i/w_j)最相容的向量，即使取最小值的向量。如果W＝(w₁,w₂,w₃,w₄,w₅,w₆)使得P取最小值，那么应该有 The feature matrices corresponding to the four sorting vectors are A ^EW , A ^σ , A ^AHP and A ^G-1 . Geometric comprehensive mean vector W=(w ₁ ,w ₂ ,w ₃ ,w ₄ ,w ₅ ,w ₆ ) is logarithmic sense so that A ^EW ,A ^σ ,A ^AHP and A ^G-1 and comprehensive feature matrix W= (w _i /w _j ) most compatible vector, even if Vector of minimum values. If W=(w ₁ ,w ₂ ,w ₃ ,w ₄ ,w ₅ ,w ₆ ) makes P take the minimum value, then there should be

因为解公式得：because Solve the formula to get:

${w w}_{t t} = = (({w w}_{t t}^{EW EW} {w w}_{t t}^{σ σ} {w w}_{t t}^{AHP AHP} {w w}_{t t}^{G G - - 11})) / / {Σ Σ}_{t t = = 11}^{66} {(({w w}_{t t}^{EW E W} {w w}_{t t}^{σ σ} {w w}_{t t}^{AHP AHP} {w w}_{t t}^{G G - - 11}))}^{11 / / 44} = = (({w w}_{t t}^{11} {w w}_{t t}^{22} {w w}_{t t}^{33} {w w}_{t t}^{44})) / / {Σ Σ}_{t t = = 11}^{66} {(({w w}_{t t}^{11} {w w}_{t t}^{22} {w w}_{t t}^{33} {w w}_{t t}^{44}))}^{11 / / 44}$

结合上述公式，得到四种权重向量的几何综合平均向量：Combining the above formulas, the geometric comprehensive mean vector of the four weight vectors is obtained:

W＝(w₁,w₂,w₃,w₄,w₅,w₆)W＝(w ₁ ,w ₂ ,w ₃ ,w ₄ ,w ₅ ,w ₆ )

综合特征矩阵：Comprehensive feature matrix:

A＝[(w_i/w_j)],1≤i,j≤6A＝[(w _i /w _j )], 1≤i, j≤6

该方法把指标SI（A,B）=C(A,B)/n²称为矩阵A,B的相容性指标。In this method, the index SI (A, B) = C (A, B)/n ² is called the compatibility index of the matrix A, B.

一般地，A与W具有相容性，但是，A和W的相容性完全由A决定，而当认为A和W具有满意的相容性。因此为了便于判断，该方法取作为相容性指标的边界值，边界值在表5中给出。Generally, A and W are compatible, however, the compatibility of A and W is completely determined by A, and when A and W are considered to have satisfactory compatibility. Therefore, for the convenience of judgment, the method takes As the boundary value of the compatibility index, the boundary value is given in Table 5.

当衡量A和B是否相容时，当时该方法认为A和B具有令人满意的相容性。When measuring whether A and B are compatible, when When the method considers A and B to be satisfactorily compatible.

为保证组合向量的合理性，该方法对合成权重进行相容性检验，即判断矩阵A是否与A^EW,A^σ,A^AHP，A^G-1分别具有满意的相容性，它们的相容性指标分别为：In order to ensure the rationality of the combined vectors, this method performs a compatibility test on the combined weights, that is, to judge whether the matrix A has satisfactory compatibility with A ^EW , A ^σ , A ^AHP , and A ^G-1 respectively, and their compatibility Sex indicators are:

${SI Si}_{11} = = SI Si ((A A,, {A A}^{EW E W})) = = C C ((A A,, {A A}^{EW E W})) / / 66^{22} {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j} / / {w w}_{j j})) (({w w}_{j j}^{EW E W} / / {w w}_{i i}^{EW E W})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{EW E W})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{EW E W} / / {w w}_{j j})) / / 66^{22}$

${SI Si}_{22} = = SI Si ((A A,, {A A}^{σ σ})) = = C C ((A A,, {A A}^{σ σ})) / / 66^{22} = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{σ σ} / / {w w}_{i i}^{σ σ})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{σ σ})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{σ σ} / / {w w}_{j j})) / / 66^{22}$

${SI Si}_{33} = = SI Si ((A A,, {A A}^{G G - - 11})) = = C C ((A A . . {A A}^{G G - - 11})) / / 66^{22} = = {Σ Σ}_{j j = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{G G - - 11} / / {w w}_{i i}^{G G - - 11})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{G G - - 11})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{G G - - 11} / / {w w}_{j j})) / / 66^{22}$

${SI Si}_{44} = = SI Si ((A A,, {A A}^{AHP AHP})) = = C C ((A A,, {A A}^{AHP AHP})) / / 66^{22}$

当四个相容性指标同时小于同阶相容性指标临界值则说明合成的权重符合相容性要求。利用群组决策得到的权值向量来进行网络选择，即获取权重之后，结合SAW方法，每个网络的性能函数可以表示为： When the four compatibility indicators are less than the critical value of compatibility indicators of the same order It means that the synthesized weight meets the compatibility requirements. Use the weight vector obtained by the group decision to select the network, that is, after obtaining the weight, combined with the SAW method, the performance function of each network can be expressed as:

最佳网络为： $F^{*} = \arg \max_{i &Element; N} \underset{j &Element; 6}{Σ} w_{j} r_{ij .}$ The best network is: $f^{*} = \arg \max_{i &Element; N} \underset{j &Element; 6}{Σ} w_{j} r_{ij .}$

本发明将群组决策的内容运用到网络选择过程中，考虑了网络客观属性，用户需求和业务类型。The invention applies the content of group decision-making to the network selection process, and considers the objective attributes of the network, user needs and service types.

有益效果：Beneficial effect:

1.采用群组决策理论获取综合网络属性权重时，采用四个决策者，数量适中并分别具有代表性，综合权重通过相容性理论检验保证其合理性，即权重的确定明确而又有充分理论支撑，并正确充分利用组合理论，网络选择过程又不至于过分复杂。1. When using group decision-making theory to obtain the weight of comprehensive network attributes, four decision makers are used, the number of which is moderate and each is representative. The comprehensive weight is tested by compatibility theory to ensure its rationality, that is, the determination of the weight is clear and sufficient Theoretical support, and the correct and full use of combination theory, the network selection process will not be overly complicated.

2.网络选择过程中通过综合多个决策者智慧达到考虑网络客观属性、用户偏好以及业务类型的影响。并且能根据业务类型的不同通过给定不同的判决矩阵为不同业务类型下的用户提供满意的QoS。2. In the network selection process, the influence of network objective attributes, user preferences, and service types is considered by integrating the wisdom of multiple decision makers. And it can provide satisfactory QoS for users under different business types by giving different decision matrices according to different business types.

附图说明Description of drawings

图1为本发明的网络选择框图。Fig. 1 is a network selection block diagram of the present invention.

图2为本发明的方法具体流程图。Fig. 2 is a specific flowchart of the method of the present invention.

图3为会话类业务的时延仿真图。Fig. 3 is a time delay simulation diagram of conversational services.

图4为会话类业务的抖动仿真图。Fig. 4 is a jitter simulation diagram of conversational services.

图5为会话类业务的吞吐量仿真图。FIG. 5 is a simulation diagram of throughput of conversational services.

图6为流媒体业务的抖动仿真图。Fig. 6 is a jitter simulation diagram of a streaming media service.

图7为流媒体业务的丢包率仿真图。FIG. 7 is a simulation diagram of a packet loss rate of a streaming media service.

图8为流媒体业务的吞吐量仿真图。Fig. 8 is a simulation diagram of the throughput of the streaming media service.

图9为交互型业务的丢包率仿真图。FIG. 9 is a simulation diagram of a packet loss rate of an interactive service.

图10为交互型业务的吞吐量仿真图。Fig. 10 is a simulation diagram of the throughput of the interactive service.

图11为交互型业务的代价仿真图。Fig. 11 is a simulation diagram of the cost of an interactive service.

图12为背景类业务的代价仿真图。Fig. 12 is a simulation diagram of the cost of background services.

图13为背景类业务的丢包率仿真图。FIG. 13 is a simulation diagram of the packet loss rate of background services.

具体实施方式detailed description

下面结合附图对发明的技术方案进行详细说明：Below in conjunction with accompanying drawing, the technical scheme of invention is described in detail:

本发明的思路是将群组决策和相容性理论运用到解决异构网络垂直切换过程中网络选择的各属性参数的权重，在此权重基础上对网络进行排序并选择代价函数最大的网络如图1所示，首先赋予客观参数通过熵值法和标准离差法计算出客观网络属性权值向量，赋予主观参数通过AHP法和G-1法分别得到各自的主观网络属性权值向量，四种方法代表了四个决策者，利用群组决策理论将四种权值向量组合生成新的权值向量，从而获取一个综合权值向量W，最后结合SAW法进行网络选择。The idea of the present invention is to apply the group decision-making and compatibility theory to solve the weight of each attribute parameter selected by the network during the vertical handover process of the heterogeneous network, sort the network on the basis of this weight, and select the network with the largest cost function, such as As shown in Figure 1, first assign objective parameters to calculate the objective network attribute weight vector by entropy method and standard deviation method, assign subjective parameters to obtain respective subjective network attribute weight vectors by AHP method and G-1 method respectively, four This method represents four decision makers, using the group decision-making theory to combine four weight vectors to generate a new weight vector, thereby obtaining a comprehensive weight vector W, and finally combining the SAW method for network selection.

整个网络切换过程采用群组决策方法实现网络选择的详细流程图见图2。The detailed flow chart of network selection using group decision-making method in the whole network switching process is shown in Fig. 2 .

一、客观权重的确定1. Determination of objective weight

本发明异构网络模型包含了N种异构网络，网络选择中涉及的属性参数为可用带宽、峰值传输速率、包时延、包抖动、包丢失与每比特费用，本方法中B代表带宽、R代表峰值传输速率，D代表包时延、J代表包抖动、L代表包丢失、C代表每比特费用。假设熵值法和标准离差法中的标准化的网络属性矩阵为：The heterogeneous network model of the present invention includes N kinds of heterogeneous networks. The attribute parameters involved in network selection are available bandwidth, peak transmission rate, packet delay, packet jitter, packet loss and cost per bit. In this method, B represents bandwidth, R stands for peak transmission rate, D stands for packet delay, J stands for packet jitter, L stands for packet loss, and C stands for cost per bit. Assume that the standardized network attribute matrix in the entropy method and standard deviation method is:

R＝(x_ij)_N×6，即N个异构网络数目，6种网络属性。其中：R=(x _ij ) _N×6 , that is, the number of N heterogeneous networks and 6 types of network attributes. in:

x_i6表示第i个网络的每比特费用，并且1<=i<=N，1<=j<=6。x _i6 represents the cost per bit of the i-th network, and 1<=i<=N, 1<=j<=6.

一般而言，属性分为效益型和成本型，效益型的属性越大越好：取 $x_{\max}^{j} = \max (x_{1 j}, x_{2 j}, . . ., x_{Nj});$ Generally speaking, attributes are divided into benefit-type and cost-type. The larger the benefit-type attribute, the better: take $x_{\max}^{j} = \max (x_{1 j}, x_{2 j}, . . ., x_{Nj});$

成本型的属性越小越好：取 $x_{\min}^{j} = \min (x_{1 j}, x_{2 j}, . . ., x_{Nj}) .$ The smaller the attribute of the cost type, the better: take $x_{\min}^{j} = \min (x_{1 j}, x_{2 j}, . . ., x_{Nj}) .$

${r r}_{ij ij} = = \frac{{x x}_{ij ij}}{{x x}_{max max}^{j j} + + {x x}_{min min}^{j j}},, 11 \leq \leq j j \leq \leq 22 - - - - - - ((11))$

对于D，J，L和C，它们属于成本型属性，标准化处理公式为：For D, J, L, and C, they are cost-type attributes, and the normalization formula is:

${r r}_{ij ij} = = \frac{{x x}_{max max}^{i i} + + {x x}_{min min}^{i i} - - {x x}_{ij ij}}{{x x}_{max max}^{j j} + + {x x}_{min min}^{j j}},, 33 \leq \leq j j \leq \leq 66 - - - - - - ((22))$

熵值法确定客观权重：The entropy method determines the objective weight:

(1)标准化(1) Standardization

${\overset{&OverBar; &OverBar;}{r r}}_{ij ij} = = {r r}_{ij ij} / / {Σ Σ}_{i i = = 11}^{66} {r r}_{ij ij},, 11 \leq \leq j j \leq \leq 66 - - - - - - ((33))$

(2)确定信息熵(2) Determine information entropy

${H h}_{i i} = = - - K K {Σ Σ}_{j j = = 11}^{66} {\overset{&OverBar; &OverBar;}{r r}}_{ij ij} 11 n no {\overset{&OverBar; &OverBar;}{r r}}_{ij ij},, 11 \leq \leq i i \leq \leq N N - - - - - - ((44))$

这里：K=1/lnNHere: K=1/lnN

(3)计算权重向量(3) Calculate the weight vector

${w w}_{j j}^{EW E W} = = \frac{11 - - {H h}_{i i}}{66 - - {Σ Σ}_{j j = = 11}^{66} {H h}_{i i}},, 11 \leq \leq i i \leq \leq N N - - - - - - ((55))$

${W W}^{EW EW} = = (({w w}_{11}^{EW E W},, {w w}_{22}^{EW E W},, {w w}_{33}^{EW EW},, {w w}_{44}^{EW EW},, {w w}_{55}^{EW E W},, {w w}_{55}^{EW EW},, {w w}_{66}^{EW EW}))$

标准离差法确定客观权重：标准离差法和熵值法计算原理相似。一般在某个网络中一个指标的标准差与指标值的变异程度成正比，如果标准差越大，则该指标的变异程度越大，提供的信息量越大，在评价中发挥作用越大，其权重也越大。反之，则权重越小。本发明计算六个网络统一属性标准差公式为：Standard deviation method to determine the objective weight: the standard deviation method and the entropy value method have similar calculation principles. Generally, the standard deviation of an index in a network is proportional to the degree of variation of the index value. If the standard deviation is larger, the degree of variation of the index is greater, the amount of information provided is greater, and the role played in the evaluation is greater. Its weight is also greater. On the contrary, the weight is smaller. The present invention calculates the six network uniform attribute standard deviation formulas as:

${σ σ}_{j j} = = \sqrt{\frac{{Σ Σ}_{i i = = 11}^{N N} {(({r r}_{ij ij} - - {\overset{&OverBar; &OverBar;}{r r}}_{j j}))}^{22}}{N N}},, {\overset{&OverBar; &OverBar;}{r r}}_{j j} = = \frac{{Σ Σ}_{i i = = 11}^{N N} {r r}_{ij ij}}{66},, 11 \leq \leq j j \leq \leq 66 - - - - - - ((66))$

利用标准离差法计算N个异构网络的各指标权重公式如下：The formula for calculating the weight of each index of N heterogeneous networks using the standard deviation method is as follows:

${w w}_{j j}^{σ σ} = = \frac{{σ σ}_{j j}}{{Σ Σ}_{j j = = 11}^{66} {σ σ}_{j j}},, j j = = 1,2 1,2,, . . . . . .,, 66 - - - - - - ((77))$

则网络属性矩阵R＝（r_ij）_N×6利用标准离差法计算出的权重向量为：Then the network attribute matrix R=(r _ij ) _N×6 The weight vector calculated by using the standard deviation method is:

${W W}^{σ σ} = = (({w w}_{11}^{σ σ},, {w w}_{22}^{σ σ},, {w w}_{33}^{σ σ},, {w w}_{44}^{σ σ},, {w w}_{55}^{σ σ},, {w w}_{66}^{σ σ})) . .$

二、主观权重的确定2. Determination of subjective weight

本发明主观赋权法采用AHP法以及G-1法。根据业务类型的不同，主观赋权法的判决矩阵会发生变化。AHP法中的判断矩阵为A^AHP，G-1法的判断向量为G^G-1，AHP和G-1法的不同类型判决矩阵分别在表2和表3中给出。表中判决矩阵的数据是根据专家给定根据业务类型的不同对应参数之间相对重要性给出的，并结合Saaty标度依次给出的。The subjective weighting method of the present invention adopts the AHP method and the G-1 method. Depending on the type of business, the judgment matrix of the subjective empowerment method will change. The judgment matrix in the AHP method is A ^AHP , and the judgment vector in the G-1 method is G ^G- 1 . The different types of judgment matrices of the AHP and G-1 methods are given in Table 2 and Table 3, respectively. The data of the decision matrix in the table is given by experts according to the relative importance of different corresponding parameters of the business type, and combined with the Saaty scale.

表2为AHP方法中不同业务类型对应的判断矩阵Table 2 shows the judgment matrix corresponding to different business types in the AHP method

表3为G-1方法中不同业务类型对应的判断矩阵Table 3 is the judgment matrix corresponding to different business types in the G-1 method

业务类型business type r₂ r ₂ r₃r₄r₅ r ₃ r ₄ r ₅ r₆ r ₆ 排序状况Sort status 会话类业务Conversational business 1.21.2 1.611.21.611.2 1.21.2 R>J>B>C>P>LR>J>B>C>P>L 流媒体业务streaming business 1.81.8 111111 11 L>B>P>R>J>CL>B>P>R>J>C 交互类业务Interactive business 11 1.811.41.811.4 11 L>C>R>J>B>PL>C>R>J>B>P 背景类业务background business 1.21.2 1.4111.411 1.21.2 C>L>J>R>B>RC>L>J>R>B>R

因此在每种业务类型之下，A^AHP和G^G-1分别与表2和表3中判决矩阵对应。AHP法确定主观权重：对应于每种业务类型，结合表2，根据AHP计算权重步骤，特定业务类型条件下，可以算出AHP法确定的主观权重：Therefore, under each service type, A ^AHP and G ^G-1 correspond to the decision matrices in Table 2 and Table 3 respectively. AHP method to determine subjective weight: corresponding to each type of business, combined with Table 2, according to the AHP calculation weight steps, under the condition of a specific business type, the subjective weight determined by the AHP method can be calculated:

G-1法确定主观权：G-1法确定主观权重分为三个步骤，首先根据某个评价标准对所有评价指标进行重要性排序，然后给定排序后相邻指标的重要程度比值，最后计算各个指标的权重。具体步骤如下：G-1 method to determine the subjective weight: G-1 method to determine the subjective weight is divided into three steps. First, the importance of all evaluation indicators is sorted according to a certain evaluation standard, and then the importance ratio of the adjacent indicators after the sorting is given, and finally Calculate the weight of each indicator. Specific steps are as follows:

（1）序关系的确定(1) Determination of sequence relationship

对所有评价指标相对于某评价标准重要程度排序，则该异构网络模型下的六种属性在每种业务类型下的重要性排序情况如下：Ranking the importance of all evaluation indicators relative to a certain evaluation standard, the importance ranking of the six attributes under the heterogeneous network model under each business type is as follows:

会话类业务：R>J>B>C>P>LConversational services: R>J>B>C>P>L

流媒体业务：L>B>P>R>J>CStreaming business: L>B>P>R>J>C

交互类业务：L>C>R>J>B>PInteractive business: L>C>R>J>B>P

背景类业务：C>L>J>R>B>RBackground business: C>L>J>R>B>R

（2）相邻属性相对重要性程度判断(2) Judgment on the relative importance of adjacent attributes

在某种评价标准的重要程度排序之下相邻属性之间重要程度之比为：Under the importance ranking of a certain evaluation standard, the importance ratio between adjacent attributes is:

w_k-1/w_k＝r_k,2≤k≤6(8)w _k-1 /w _k ＝r _k ,2≤k≤6(8)

r_k赋值可参照表4，表4来自于G-1法作者为郭亚军的著作《综合评价理论、方法与应用》，科学出版社2007年5月。作者在介绍G-1法时定义了该r值的赋值规则。The assignment of r _k can refer to Table 4. Table 4 comes from the book "Comprehensive Evaluation Theory, Method and Application" by Guo Yajun, the author of G-1 method, Science Press, May 2007. The author defined the assignment rules for the value of r when introducing the G-1 method.

表3中各种业务类型下的r值是结合表2和表4给出的，即根据专家给定的各个业务类型下每个属性之间的相对重要性，r值赋值规则，依次给出，从而保证对于所有业务类型的主观判断保持一致。The r values under various business types in Table 3 are given in combination with Table 2 and Table 4, that is, according to the relative importance of each attribute under each business type given by experts, the r value assignment rules are given in turn , so as to ensure that the subjective judgments of all business types are consistent.

表4为G-1方法中r值赋值表Table 4 is the r value assignment table in the G-1 method

r_k r _k 说明illustrate 1.01.0 指标x_k-1和x_k有相同重要性Index x _k-1 has the same importance as x _k 1.21.2 指标x_k-1比x_k稍微重要Index x _k-1 is slightly more important than x _k 1.41.4 指标x_k-1比x_k明显重要Index x _k-1 is significantly more important than x _k 1.61.6 指标x_k-1比x_k强烈重要Index x _k-1 is strongly more important than x _k 1.81.8 指标x_k-1比x_k极端重要Index x _k-1 is extremely important than x _k

（3）权重的确定(3) Determination of weight

首先计算重要性排序后最重要的属性权重W₆，然后依次计算出后续权重。First calculate the most important attribute weight W ₆ after the importance ranking, and then calculate subsequent weights in sequence.

权重可以有以下公式确定：The weight can be determined with the following formula:

${w w}_{66} = = \frac{11}{11 + + {Σ Σ}_{k k = = 22}^{66} {Π Π}_{i i = = k k}^{66} {r r}_{i i}} - - - - - - ((99))$

由步骤（2）可以得出后续权重：Subsequent weights can be obtained from step (2):

w_k-1＝r_k×w_k,k＝6,5,...，2(10)w _k-1 =r _k ×w _k , k=6,5,...,2(10)

结合上述公式，并参照表3中r值，某种特定业务类型下的G-1法确定的主观权值向量为：Combining the above formula and referring to the r value in Table 3, the subjective weight vector determined by the G-1 method under a certain type of business is:

${W W}^{G G - - 11} = = (({w w}_{11}^{G G - - 11},, {w w}_{22}^{G G - - 11},, {w w}_{33}^{G G - - 11},, {w w}_{44}^{G G - - 11},, {w w}_{55}^{G G - - 11},, {w w}_{66}^{G G - - 11}))$

三、群组决策和相容性检验3. Group decision-making and compatibility testing

群组决策通过集中群体成员智慧来发挥群体决策的优势，因此所选群体成员应当具有代表性并要保证一定数量。本发明所选的四种决策成员分别为层次分析法、熵值法、G-1法和标准离差法，四种决策方法包括两种主观赋权法两种客观赋权法，数量适中，通过群组决策获取网络属性权重参数不仅能利用多个决策者参与时群组决策的优势，更能充分考虑用户要求、业务类型以及网络状况。接下来将介绍运用群组决策和相容性理论组合四种权重向量。Group decision-making takes advantage of group decision-making by concentrating the wisdom of group members, so the selected group members should be representative and a certain number must be guaranteed. Four kinds of decision-making members selected by the present invention are respectively AHP, entropy value method, G-1 method and standard deviation method, four kinds of decision-making methods include two kinds of subjective weighting methods and two kinds of objective weighting methods, the quantity is moderate, Obtaining network attribute weight parameters through group decision-making can not only take advantage of group decision-making when multiple decision makers participate, but also fully consider user requirements, business types and network conditions. Next, we will introduce the combination of four weight vectors using group decision-making and compatibility theory.

设A＝（a_ij）,B＝（b_ij）和C＝（c_ij）均为n阶正互反矩阵，由A得到的排序向量为W＝（w₁,w₂,...,w_n）^T，矩阵W＝(w_i/w_j)称为A的特征矩阵。定义A和B的乘积C(A,B)＝e^T*A*B*e为A,B的相容度，其中e^T=(1,1,…1)。为了方便，一般取其对数作为相容度，记为一般的，LC（A,B）≥0，如果LC（A,B）=0，则A,B完全相容。Assume that A=(a _ij ), B=(b _ij ) and C=(c _ij ) are all positive and reciprocal matrices of order n, and the sorting vector obtained from A is W=(w ₁ ,w ₂ ,..., w _n ) ^T , the matrix W=(w _i /w _j ) is called the characteristic matrix of A. Define the product C(A,B)=e ^T *A*B*e of A and B as the compatibility of A and B, where e ^T =(1,1,...1). For convenience, the logarithm is generally taken as the compatibility degree, which is denoted as Generally, LC (A, B) ≥ 0, if LC (A, B) = 0, then A, B are completely compatible.

${W W}^{σ σ} = = {W W}^{22} = = (({w w}_{11}^{22},, {W W}_{22}^{22},, {w w}_{33}^{22},, {w w}_{44}^{22},, {w w}_{55}^{22},, {w w}_{66}^{22}))$

${W W}^{AHP AHP} = = {W W}^{33} = = (({w w}_{11}^{33},, {w w}_{22}^{33},, {w w}_{33}^{33},, {w w}_{44}^{33},, {w w}_{55}^{33},, {w w}_{66}^{33}))$

四个排序向量对应的特征矩阵分别为A^EW,A^σ,A^AHP和A^G-1。几何综合平均向量W＝(w₁,w₂,w₃,w₄,w₅,w₆)是对数意义下使得A^EW,A^σ,A^AHP和A^G-1与综合特征矩阵W＝(w_i/w_j)最相容的向量，即使取最小值的向量。如果W＝(w₁,w₂,w₃,w₄,w₅,w₆)使得P取最小值，那么应该有：The feature matrices corresponding to the four sorting vectors are A ^EW , A ^σ , A ^AHP and A ^G-1 . Geometric comprehensive mean vector W=(w ₁ ,w ₂ ,w ₃ ,w ₄ ,w ₅ ,w ₆ ) is logarithmic sense so that A ^EW ,A ^σ ,A ^AHP and A ^G-1 and comprehensive feature matrix W= (w _i /w _j ) most compatible vector, even if Vector of minimum values. If W=(w ₁ ,w ₂ ,w ₃ ,w ₄ ,w ₅ ,w ₆ ) makes P take the minimum value, then there should be:

$\frac{&PartialD; &PartialD; P P}{&PartialD; &PartialD; t t} = = 0,1 0,1 \leq \leq t t \leq \leq 66 - - - - - - ((1111))$

因为解公式得：because Solve the formula to get:

${w w}_{t t} = = (({w w}_{t t}^{EW E W} {w w}_{t t}^{σ σ} {w w}_{t t}^{AHP AHP} {w w}_{t t}^{G G - - 11})) / / {Σ Σ}_{t t = = 11}^{66} (({w w}_{t t}^{EW E W} {w w}_{t t}^{σ σ} {w w}_{t t}^{AHP AHP} {w w}_{t t}^{G G - - 11} {))}^{11 / / 44} = = (({w w}_{t t}^{11} {w w}_{t t}^{22} {w w}_{t t}^{33} {w w}_{t t}^{44})) / / {Σ Σ}_{t t = = 11}^{66} {(({w w}_{t t}^{11} {w w}_{t t}^{22} {w w}_{t t}^{33} {w w}_{t t}^{44}))}^{11 / / 44} - - - - - - ((1212))$

综合特征矩阵：A＝[(w_i/w_j)],1≤i,j≤6Comprehensive characteristic matrix: A＝[(w _i /w _j )], 1≤i, j≤6

该方法把指标SI（A,B）=C(A,B)/n²称为矩阵A,B的相容性指标。一般地，A与W具有相容性，但是，A和W的相容性完全由A决定，而当认为A和W具有满意的相容性。因此为了便于判断，该方法取作为相容性指标的边界值，边界值在表5中给出。表5是《系统工程理论与实践》2000年2月（第2期）刊载的“相容性与群组决策”，中国人民大学王莲芬，其中计算得出的，本发明只需使用表5中数据检验方法过程量的合理性，因此在此直接引用其数据。In this method, the index SI (A, B) = C (A, B)/n ² is called the compatibility index of the matrix A, B. Generally, A and W are compatible, however, the compatibility of A and W is completely determined by A, and when A and W are considered to have satisfactory compatibility. Therefore, for the convenience of judgment, the method takes As the boundary value of the compatibility index, the boundary value is given in Table 5. Table 5 is "compatibility and group decision-making" published in "System Engineering Theory and Practice" in February 2000 (No. 2), Wang Lianfen, Renmin University of China, where calculated, the present invention only needs to use the information in Table 5 The data test the rationality of the process quantity of the method, so its data is directly quoted here.

表5为S.I.临界值Table 5 is the S.I. critical value

为保证组合向量的合理性，对合成权重进行相容性检验，即判断矩阵A是否和A^EW,A^σ,A^AHP和A^G-1分别具有满意的相容性，它们的相容性指标分别为：In order to ensure the rationality of the combined vectors, a compatibility test is carried out on the combined weights, that is, to judge whether the matrix A has satisfactory compatibility with A ^EW , A ^σ , A ^AHP and A ^G-1 respectively, and their compatibility index They are:

${SI Si}_{11} = = SI Si ((A A,, {A A}^{EW E W})) = = C C ((A A,, {A A}^{EW E W})) / / 66^{22} {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j} / / {w w}_{j j})) (({w w}_{j j}^{EW E W} / / {w w}_{i i}^{EW E W})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{EW E W})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{EW E W} / / {w w}_{j j})) / / 66^{22} - - - - - - ((1313))$

${SI Si}_{22} = = SI Si ((A A,, {A A}^{σ σ})) = = C C ((A A,, {A A}^{σ σ})) / / 66^{22} = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{σ σ} / / {w w}_{i i}^{σ σ})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{σ σ})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{σ σ} / / {w w}_{j j})) / / 66^{22} - - - - - - ((1414))$

${SI Si}_{33} = = SI Si ((A A,, {A A}^{G G - - 11})) = = C C ((A A,, {A A}^{G G - - 11})) / / 66^{22} = = {Σ Σ}_{j j = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{G G - - 11} / / {w w}_{i i}^{G G - - 11})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{G G - - 11})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{G G - - 11} / / {w w}_{j j})) / / 66^{22} - - - - - - ((1515))$

${SI Si}_{44} = = SI Si ((A A,, {A A}^{AHP AHP})) = = C C ((A A,, {A A}^{AHP AHP})) / / 66^{22} - - - - - - ((1616))$

当四个相容性指标同时小于同阶相容性指标临界值则说明合成的权重符合相容性要求。When the four compatibility indicators are less than the critical value of compatibility indicators of the same order It means that the synthesized weight meets the compatibility requirements.

四、网络选择4. Network selection

利用群组决策得到的满足相容性要求的几何综合权值向量来进行网络选择，即获取权重之后，结合SAW方法，每个网络的性能函数可以表示为：Use the geometric comprehensive weight vector obtained by group decision-making to meet the compatibility requirements for network selection, that is, after obtaining the weights, combined with the SAW method, the performance function of each network can be expressed as:

$F f = = {Σ Σ}_{j j = = 11}^{66} {w w}_{j j} {r r}_{ij ij},, 11 \leq \leq i i \leq \leq N N - - - - - - ((1717))$

最佳网络为：The best network is:

${F f}^{* *} = = arg arg \underset{i i &Element; &Element; N N}{max max} \underset{j j &Element; &Element; 66}{Σ Σ} {w w}_{j j} {r r}_{ij ij} - - - - - - ((1818))$

综上所述，本发明的有益之处通过仿真结果给出：In summary, the benefits of the present invention are provided by simulation results:

仿真中的异构网络分别是WLAN、UMTS与WIMAX，每种类型包含两个网络。本发明在仿真中与现有技术比较：在仿真图中用“EW”代表《无线电工程》2009年第39卷第1期刊载的“一种多属性决策的异构网络选择算法”，作者：王康，曾志民，冯春燕，张天魁（北京邮电大学通信网络综合技术研究所）中的方法，仿真图中用“GRA”代表《InformationComputingandApplicationsLectureNotesinComputerScience》出版的编号为“6377/2010:213-220”文献“ANovelAHPandGRABasedHandoverDecisionMechanisminHeterogeneousWirelessNetworks”中的方法。网络选择中涉及的属性参数为可用带宽（AvailableBandwidth,B）、峰值传输速率（PeakDataRate,R）、包时延（PacketDelay,D）、包抖动（PacketJitter,J）、包丢失（PacketDelay,D）与每比特费用（CostPerBit,C），表6所示。而表6中三种异构网络，即UMTS、WLAN和WiMAX，它们的六种属性的分布范围是根据实际应用中统计的网络属性数据给出的。The heterogeneous networks in the simulation are WLAN, UMTS and WIMAX, each type contains two networks. The present invention is compared with the prior art in simulation: "EW" is used in the simulation diagram to represent "a heterogeneous network selection algorithm for multi-attribute decision-making" published in "Radio Engineering", Volume 39, Issue 1, 2009, author: Wang Kang, Zeng Zhimin, Feng Chunyan, Zhang Tiankui (Beijing University of Posts and Telecommunications Institute of Comprehensive Technology of Communication Networks) in the method, "GRA" is used in the simulation figure to represent "Information Computing and Applications Lecture Notes in Computer Science" published as "6377/2010:213-220" literature " Method in ANovelAHPandGRABasedHandoverDecisionMechanisminHeterogeneousWirelessNetworks". The attribute parameters involved in network selection are available bandwidth (AvailableBandwidth, B), peak transmission rate (PeakDataRate, R), packet delay (PacketDelay, D), packet jitter (PacketJitter, J), packet loss (PacketDelay, D) and Cost per bit (CostPerBit, C), as shown in Table 6. The distribution ranges of the six attributes of the three heterogeneous networks in Table 6, namely UMTS, WLAN, and WiMAX, are given according to the statistical network attribute data in practical applications.

表6为网络属性参数Table 6 is the network attribute parameters

属性参数attribute parameter UMTS1UMTS1 UMTS2UMTS2 WLAN1WLAN1 WLAN2WiMAX1WiMAX2WLAN2WiMAX1WiMAX2 B(MHz)B(MHz) 0.1-20.1-2 0.1-20.1-2 1-111-11 1-541-601-601-541-601-60 R(Mbps)R (Mbps) 22 22 1111 546060546060 D(ms)D(ms) 25-5025-50 25-5025-50 100-150100-150 100-15060-10060-100100-15060-10060-100 J(ms)J(ms) 5-105-10 5-105-10 10-2010-20 10-203-103-1010-203-103-10 L(per10⁶)L(per10 ⁶ ) 20-8020-80 20-8020-80 20-8020-80 20-8020-8020-8020-8020-8020-80 C(price)C(price) 0.60.6 0.80.8 0.10.1 0.050.50.40.050.50.4

仿真中选择4种业务类型来衡量算法性能，具体仿真结果如图3-图13所示。In the simulation, four business types are selected to measure the performance of the algorithm, and the specific simulation results are shown in Figure 3-Figure 13.

图3-图5给出了会话类业务的性能。会话类业务的语音通信要求低时延以及较低带宽，而视频通信要求低时延以及足够带宽，因此表2和表3的判决矩阵中会话类业务更看重时延抖动状况。从图中可以看出，对比其他两种方法，本发明方法可以提供最优的时延和抖动性能以及令人满意的带宽，从而可以满足用户的QoS要求。Figure 3-Figure 5 shows the performance of conversational services. Voice communication for conversational services requires low latency and low bandwidth, while video communication requires low latency and sufficient bandwidth. Therefore, in the decision matrices in Table 2 and Table 3, conversational services pay more attention to delay and jitter. It can be seen from the figure that compared with the other two methods, the method of the present invention can provide optimal time delay and jitter performance as well as satisfactory bandwidth, thereby meeting the QoS requirements of users.

图6-图8给出了流媒体业务的性能。流媒体业务要求较高的误码率并允许一定时延，对带宽要求较高，因此表2和表3的判决矩阵中会话类业务更看重吞吐量状况。从图中可以看出，对比其他两种方法，本发明方法可以提供最佳抖动状态，令人满意的丢包率以及最优的带宽性能，从而可以满足用户的QoS要求。Figures 6-8 show the performance of streaming media services. Streaming media services require a high bit error rate and allow a certain delay, and have high bandwidth requirements. Therefore, in the decision matrices in Table 2 and Table 3, conversational services pay more attention to throughput. It can be seen from the figure that compared with the other two methods, the method of the present invention can provide the best jitter state, satisfactory packet loss rate and optimal bandwidth performance, thereby meeting the QoS requirements of users.

图9-图11给出了交互型业务的性能。交互类业务对误码率有一定要求，相对较低时延以及相对较高的数据下行速率，因此表2和表3的判决矩阵中会话类业务更看重丢包率状况。从图中可以看出，对比其他两种方法，本发明方法可以在最低代价前提下提供最佳丢包率性能以及较高的吞吐量，从而可以满足用户的QoS要求。Figures 9-11 show the performance of interactive services. Interactive services have certain requirements on the bit error rate, relatively low delay and relatively high data downlink rate, so the conversational services in the decision matrix in Table 2 and Table 3 pay more attention to the packet loss rate. It can be seen from the figure that compared with the other two methods, the method of the present invention can provide the best packet loss rate performance and higher throughput under the premise of the lowest cost, thereby meeting the QoS requirements of users.

图12-图13给出了背景类业务的性能。背景类业务对时延要求很低，有较高的误码率要求，因此表2和表3的判决矩阵中会话类业务更看重丢包率状况。从图中可以看出，对比其他两种方法，本发明方法可以在最低代价前提下提供最佳丢包率性能，从而可以满足用户的QoS要求。因此，根据上述仿真结果并结合各种业务类型的特点，可以得出结论，本方法可以根据业务类型提供令用户满意的QoS。Figures 12-13 show the performance of background services. Background services have very low latency requirements and high bit error rate requirements. Therefore, in the decision matrices in Table 2 and Table 3, conversational services pay more attention to the packet loss rate. It can be seen from the figure that compared with the other two methods, the method of the present invention can provide the best packet loss rate performance under the premise of the lowest cost, thereby meeting the QoS requirements of users. Therefore, according to the above simulation results and combining the characteristics of various business types, it can be concluded that this method can provide QoS that satisfies users according to the business types.

本发明运用群组决策理论对多属性决策中多种权重向量进行组合，提出了一种网络选择方法。该方法采用群组决策理论获取综合网络属性权重时，采用四个决策者，数量适中并分别具有代表性，综合权重通过相容性理论检验保证其合理性，即权重的确定明确而又有充分理论支撑，并正确充分利用组合理论，网络选择过程又不至于过分复杂。网络选择过程中通过综合多个决策者智慧达到考虑网络客观属性、用户偏好以及业务类型的影响，即能将其他方法无法兼顾的方面统筹兼顾，实践证明，该方法能根据业务类型的不同通过给定不同的判决矩阵为不同业务类型下的用户提供满意的QoS。The invention uses group decision-making theory to combine multiple weight vectors in multi-attribute decision-making, and proposes a network selection method. When this method adopts group decision-making theory to obtain comprehensive network attribute weights, four decision makers are used, the number of which is moderate and representative, and the comprehensive weights are tested by compatibility theory to ensure their rationality, that is, the determination of weights is clear and sufficient. Theoretical support, and the correct and full use of combination theory, the network selection process will not be overly complicated. In the process of network selection, the influence of network objective attributes, user preferences, and service types can be considered by integrating the wisdom of multiple decision makers, that is, the aspects that cannot be considered by other methods can be considered as a whole. Practice has proved that this method can pass according to different service types. Different decision matrices are set to provide satisfactory QoS for users under different service types.

Claims

1. A heterogeneous network selection method based on four weight vectors, characterized in that the method comprises the following steps:

a. Establishment of network parameter matrix: The heterogeneous network model of this method contains N kinds of heterogeneous networks, and the selected network attributes include bandwidth (B), peak transmission rate (R), packet delay (D), and packet jitter (J ), packet loss rate (L) and cost per bit (C);

Assuming that the network attribute matrix of this method is R=(x _ij ) _N×6 , that is, the number of N heterogeneous networks and 6 network attributes, x _ij represents the jth attribute of the i-th network, and x _i1 represents the i-th network The bandwidth of a network, x _i2 represents the peak transmission rate of the i-th network, x _i3 represents the delay of the i-th network, x _i4 represents the jitter of the i-th network, x _i5 represents the packet loss rate of the i-th network, x _i6 represents the cost per bit of the i-th network, and 1<=i<=N, 1<=j<=6;

b. Standardization of network attribute parameters: First, standardize the network attribute matrix. Generally speaking, attributes are divided into benefit type and cost type. The larger the benefit type attribute, the better: take The smaller the attribute of the cost type, the better: take

x_{\min}^{j} = m i no (x_{1 j}, x_{2 j}, ..., x_{N j});

For bandwidth (B) and peak transmission rate (R), they are benefit-type attributes, and the standardized processing formula is:

r_{i j} = \frac{x_{i j}}{x_{m a x}^{j} + x_{m i no}^{j}}, 1 \leq j \leq 2;

For packet delay (D), packet jitter (J), packet loss rate (L) and cost per bit (C), they are cost-type attributes, and the standardized processing formula is:

r_{i j} = \frac{x_{m a x}^{i} + x_{m i no}^{i} - x_{i j}}{x_{m a x}^{j} + x_{m i no}^{j}}, 3 \leq j \leq 6;

After standardization, all attribute parameters will fall between 0 and 1, and the network parameter matrix after standardization is: R=(r _ij ) _N×6 ;

c. Determination of four sets of weight vectors: Combine the standardized network attribute matrix with the entropy weight method and standard deviation method to calculate two sets of objective weight vectors, respectively: and

W^{σ} = W^{2} = (w_{1}^{2}, w_{2}^{2}, w_{3}^{2}, w_{4}^{2}, w_{5}^{2}, w_{6}^{2});

According to the judgment matrix table 2 corresponding to different business types in the given AHP method:

Combined with the network attribute matrix, a set of subjective weight vectors is calculated by using the AHP method: Similarly, according to the judgment matrix Table 3 corresponding to different business types in the given G-1 method:

business type r ₂ r ₃ r ₄ r ₅ r ₆ Sort status 2 --> Conversational business 1.2 1.6 1 1.2 1.2 R>J>B>C>P>L streaming business 1.8 1 1 1 1 L>B>P>R>J>C Interactive business 1 1.8 1 1.4 1 L>C>R>J>B>P background business 1.2 1.4 1 1 1.2 C>L>J>R>B>R

Where r _k ,2≤k≤6 is the ratio of the importance of adjacent attributes under the order of importance;

Another subjective weight vector is calculated using the G-1 method:

W^{G - 1} = W^{4} = (w_{1}^{4}, w_{2}^{4}, w_{3}^{4}, w_{4}^{4}, w_{5}^{4}, w_{6}^{4});

d. Determine the comprehensive weight vector and compatibility index: Let A=(a _ij ), B=(b _ij ) and C=(c _ij ) be n-order positive and reciprocal matrices, and the sorting vector obtained by A is: W=(w ₁ ,w ₂ ,...,w _n ) ^T , the matrix W=(w _i /w _j ) is called the characteristic matrix of A; define the product C(A,B) of matrix A and matrix B= e ^T *A*B*e is the compatibility of matrix A and B, where e ^T =(1,1,…1); the index SI(A,B)=C(A,B)/n ² is called is the compatibility index of matrix A and B;

Generally, A and W are compatible, however, the compatibility of A and W is completely determined by A, and when It is considered that A and W have satisfactory compatibility; in order to obtain the geometric comprehensive weight vector, the calculated four sets of weight vectors are used for group decision-making, combined with the formula:

w_{t} = (w_{t}^{E. W} w_{t}^{σ} w_{t}^{A h P} w_{t}^{G - 1}) / Σ_{t = 1}^{6} {(w_{t}^{E. W} w_{t}^{σ} w_{t}^{A h P} w_{t}^{G - 1})}^{1 / 4} = (w_{t}^{1} w_{t}^{2} w_{t}^{3} w_{t}^{4}) / Σ_{t = 1}^{6} {(w_{t}^{1} w_{t}^{2} w_{t}^{3} w_{t}^{4})}^{1 / 4}

Calculate the geometric comprehensive mean vector W and its characteristic matrix A; the four sorting vectors determined in step c are W ^EW , W ^σ , W ^AHP and W ^G-1 , and their corresponding characteristic matrices can be determined according to the definition of characteristic matrix are A ^EW , A ^σ , A ^AHP and A ^G-1 respectively, and then determine the compatibility indexes of these four feature matrices and the geometric comprehensive evaluation vector W according to the definition of compatibility indexes as follows:

{SI Si}_{11} = = S S I I ((A A,, {A A}^{E E. W W})) = = C C ((A A,, {A A}^{E E. W W})) / / 66^{22} = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{E E. W W} / / {w w}_{i i}^{E E. W W})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{E E. W W})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{E E. W W} / / {w w}_{j j})) / / 66^{22}

{SI Si}_{22} = = S S I I ((A A,, {A A}^{σ σ})) = = C C ((A A,, {A A}^{σ σ})) / / 66^{22} = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{σ σ} / / {w w}_{i i}^{σ σ})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{σ σ})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{σ σ} / / {w w}_{j j})) / / 66^{22}

{SI Si}_{33} = = S S I I ((A A,, {A A}^{G G - - 11})) = = C C ((A A,, {A A}^{G G - - 11})) / / 66^{22} = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{M m} (({w w}_{i i} / / {w w}_{j j})) (({w w}_{j j}^{G G - - 11} / / {w w}_{i i}^{G G - - 11})) = = {Σ Σ}_{i i = = 11}^{M m} (({w w}_{i i} / / {w w}_{i i}^{G G - - 11})) {Σ Σ}_{j j = = 11}^{M m} (({w w}_{j j}^{G G - - 11} / / {w w}_{j j})) / / 66^{22}

SI ₄ ＝SI(A,A ^AHP )＝C(A,A ^AHP )/6 ² , respectively calculate the characteristic matrix A ^EW , A ^σ , A ^AHP and A ^G-1 of the characteristic matrix A and the four weight vectors capacitive index;

e. Compatibility test: for the convenience of judgment, take Boundary value as a compatibility indicator;

When measuring whether A and B are compatible, when When the method considers that A and B have satisfactory compatibility; in this step, it will be judged whether these indexes are less than the critical value of the same-order compatibility index at the same time table 5:

If the compatibility index is not met, it is necessary to modify the subjective judgment matrix under this business type and return to step c; if it is satisfied, it will enter the network selection link;

f. Network selection: In the previous steps, the geometric comprehensive weight vectors satisfying the compatibility index of the six attributes are calculated. In this step, the geometric comprehensive weight vectors are combined with the SAW method to calculate the cost function value of each network, and they are sorted , replace the network with the largest cost function value as the target network, so as to obtain the optimal network number in this network state.