CN103313354A - Heterogeneous network selection method based on weight vectors of four kinds - Google Patents

Abstract

The invention relates to a heterogeneous network selection method based on weight vectors of four kinds. According to the method, an integrated weight vector is obtained through integrating four network attribute weights, and the rationality of the integrated weight vector is guaranteed through compatibility check; and then, the network selection is carried out by being combined with an SAW method. The method has the advantages that the current network status is taken into full account and satisfactory quality of service can be provided according to type of service. The method comprises the following specific steps of giving objective parameters, calculating objective network attribute weight vectors through an entropy method and a standard deviation method, giving subjective parameters, respectively obtaining respective subjective network attribute weight vectors through an AHP (Analytic Hierarchy Process) and a G-1 method, combining the weight vectors of four kinds by using a group decision making theory so as to generate a new weight vector, then, obtaining the integrated weight vector W, and finally, carrying out network selection by being combined with the SAW method, wherein the four methods represent four decision makers.

Description

Heterogeneous network system of selection based on four kinds of weight vectors

Technical field

The present invention relates to a kind of heterogeneous network system of selection based on four kinds of weight vectors, belong to communication technical field.

Background technology

Mobile communication has experienced from the first generation analog cellular system 1G of analog voice business is provided, adopt the second generation mobile communication system 2G of Digital Modulation to the process of the 3-G (Generation Three mobile communication system) 3G take CDMA as mainstream technology, development along with radio communication, following communication network will be the heterogeneous network that comprises various wireless access technologys, 2G for example, 3G, UMTS, WLAN, WiMAX, WiFi etc., the often overlapped covering of these dissimilar wireless networks, there is not unified standard interface, and they are in coverage, access rate, capacity, business and mobility characteristics, the aspects such as network quality exist than big difference, respectively have superiority again, such as between heterogeneous network, realizing seamless access by many interface terminations, the user can select the service quality of UMTS to obtain, perhaps select WiMAX to obtain high data rate, perhaps choose WLAN is to obtain lower cost, this just need to take full advantage of the advantage of each network, so the fusion of heterogeneous network is the development trend of future network communication.

It is the committed step that realizes the network integration that heterogeneous network is selected.And multiple attributive decision making method MADM (Multiple Attributes Decision Making) is one of effective method in the heterogeneous network system of selection.Classical MADM method comprises simple weighted method SAW, multiplication index weights method MEW, the sort method TOPSIS near ideal value, gray relative analysis method GRA, VIKOR method and superseded back-and-forth method ELECTRE etc.These multiple attributive decision making methods all can relate to the multiattribute weight vectors, and the vector of multiattribute weight generally will be considered the objective attribute of network also should take into full account user preference and type of service.Objective weighted model commonly used comprises entropy power method EW(Entropy Method) and dispersion method etc.; Classical subjective enabling legislation comprises analytic hierarchy process AHP (Analytic Hierarchy Process) and G-1 method.

A lot of methods integrate subjective and objective weight carries out the network selection." a kind of heterogeneous network selection algorithm of multiple attribute decision making (MADM) " that " radio engineering " the 39th volume in 2009 the 1st periodical carries, author: Wang Kang, Ceng Zhimin, Feng Chunyan, Zhang Tiankui (Beijing University of Post ﹠ Telecommunication's communication network comprehensive technology research institute), propose to adopt AHP to ask subjective weight, entropy power method is asked objective weight, then with the two linear weighted function, provides a kind of network selecting method in conjunction with the SAW method, but the method does not provide the method for asking of linear weighted function coefficient, and namely weight vectors is definite too random." 6377/2010:213-220 " document " A Novel AHP and GRA Based Handover Decision Mechanism in Heterogeneous Wireless Networks " that is numbered that " Information Computing and Applications Lecture Notes in Computer Science " publishes proposes a kind of AHP in conjunction with the network selecting method of GRA, it is larger that this method is used complexity, and do not have enough theory support.Although and above-mentioned two kinds of subjective and objective taking into account of method, the objective circumstances of network can be considered and user preference can be taken into account, can take into full account type of service.The document " Network selection based on multiple attribute decision making and group decision making for heterogeneous wireless networks " that is numbered " 19(5): 92-98 " of publishing " Chinese post and telecommunications journal " in October, 2012 adopts AHP to ask subjective weight, entropy power method is asked objective weight, then the two is carried out Group Decision, obtain a kind of balance algorithm, but when only adopting two policymaker, the policymaker participates in quantity very little, so that Group Decision can not take full advantage of the characteristics that it gathers a plurality of policymaker's wisdom.

Summary of the invention

Technical problem: the purpose of this invention is to provide a kind of heterogeneous network system of selection based on four kinds of weight vectors, the multiple attribute weight vector that the method utilizes multiple attributive decision making method to obtain, theoretical based on Group Decision, obtain a geometric generalization weight vector thereby they are made up.If this result does not satisfy the compatibility requirement, then need to revise the judgement matrix of subjective enabling legislation, again calculate, satisfy the comprehensive weight vector that compatibility requires until obtain one, then carry out network in conjunction with the SAW method and select, the method can provide gratifying service quality for different service types.

The present invention is directed to the deficiencies in the prior art, proposed a kind of method of improved combining weights based on Group Decision.At first obtain respectively separately subjective network attribute weight vector by G-1 method and AHP method, then utilize Information Entropy and dispersion method to calculate separately objective network attribute.Four kinds of methods have represented four policymaker, two subjective decision persons, and two objective making decision persons, policymaker's quantity is moderate and representative, has also avoided simultaneously the problem of said method.

Technical scheme: the present invention is a kind of heterogeneous network system of selection based on four kinds of weight vectors, wisdom by comprehensive a plurality of policymaker, consider the network objective attribute, user preference and type of service, and can provide more gratifying network to select according to type of service.

In the heterogeneous network system of selection based on four kinds of weight vectors, at first to obtain the network attribute weight.Multiple attribute decision making (MADM) generally all can relate to determining of weight vectors, and definite method of traditional weight vectors can be divided into subjective enabling legislation and objective weighted model.Subjective enabling legislation comprises AHP, G-1 method and Delphi method etc.; Objective weighted model comprises Information Entropy, dispersion method, CRITIC method etc.Two class enabling legislations emphasize particularly on different fields and characteristics.Therefore, in order to consider the factors such as type of service, user preference, network objective attribute, and take full advantage of the Group Decision balance and gather the characteristics of a plurality of policymaker's conclusions, this method obtains the multiattribute weight vectors with comprehensive multiple subjective and objective weight.Wherein subjective enabling legislation adopts AHP and G-1 method, and objective weighted model adopts Information Entropy and dispersion method.After using four kinds of enabling legislations to obtain four groups of weight vectors, obtain a comprehensive weight vector by Group Decision, then weigh the reasonability that its compatibility is judged the comprehensive weight vector.The comprehensive weight vector that satisfies compatibility carries out network in connection with the SAW method and selects.

Objective weighted model adopts Information Entropy and dispersion method.Under heterogeneous network model of the present invention, the property parameters that network relates in selecting is available bandwidth, hardware circuit solution, packet delay, packet jitter, packet loss and every bit expense, and wherein: B represents bandwidth, R and represents hardware circuit solution, D and represent packet delay, J and represent that packet jitter, L represent packet loss, C represents every bit expense.Can go out according to the network attribute calculation of parameter objective weight of attribute.

Heterogeneous network model of the present invention has comprised N kind heterogeneous network.The network attribute matrix of hypothesis the method is in Information Entropy and the dispersion method:

R=(x _Ij) _{N * 6}, i.e. N heterogeneous network number, 6 kinds of network attributes.

x _IjJ the attribute that represents i network, as:

x _I1Represent i bandwidth of network;

x _I2The hardware circuit solution that represents i network;

x _I3The time delay that represents i network;

x _I4Represent the shake of i network;

x _I5The packet loss that represents i network;

x _I6The every bit expense that represents i network;

And 1＜=i＜=N, 1＜=j＜=6.

Generally speaking, attribute is divided into benefit type and cost type, and the attribute of benefit type is the bigger the better, and gets $x_{\max }^j=\max (x_{1j},x_{2j},...,x_{\mathrm{Nj}});$

The attribute of cost type is the smaller the better, gets $x_{\min }^j=\min (x_{1j},x_{2j},...,x_{\mathrm{Nj}}).$

For B and R, they belong to benefit type attribute, and the standardization formula is:

$r_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},1\≤j\≤2$

For D, J, L and C,, they belong to cost type attribute, and the standardization formula is:

$r_{\mathrm{ij}}=\frac{x_{\max }^i+x_{\mathrm{mim}}^i-x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},3\≤j\≤6$

Can obtain the weight vectors of attribute by Information Entropy:

$W^{\mathrm{EW}}=(w_1^{\mathrm{EW}},w_2^{\mathrm{EW}},w_3^{\mathrm{EW}},w_3^{\mathrm{EW}},w_4^{\mathrm{EW}}{w}_5^{\mathrm{EW}}{w}_6^{\mathrm{EW}})$

The weight vectors that utilizes dispersion method to calculate is:

$W^{\mathrm{\σ}}=(w_1^{\mathrm{\σ}},w_2^{\mathrm{\σ}}{,w}_3^{\mathrm{\σ}},w_4^{\mathrm{\σ}}{,w}_5^{\mathrm{\σ}}{,w}_6^{\mathrm{\σ}})$

The present invention is subjective, and enabling legislation adopts AHP method and G-1 method.In the judgment matrix of AHP each element be an Attribute Relative in another ratio with the weight of the attribute under the rule layer, AHP can be adjudicated matrix and be designated as B, then the B matrix is write as following form:

$B=\left[\begin{array}{cccc}{b}_{11}& {\mathrm{<figure-callout\; id="12"\; label="b"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">b</figure-callout>}}_{12}& ...& b_{1n}\\ b_{21}& b_{22}& ...& b_{2n}\\ ...& ...& ...& ...\\ b_{n1}& b_{n2}& ...& b_{\mathrm{nn}}\end{array}\right]=\left[\begin{array}{cccc}{\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1/w_1& {\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1/w_2& <figure-callout\; id="1"\; label="...\; w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">.</figure-callout>..& w_{}{}_1/w_n\\ w_2/w_1& w_2/w_2& ...& w_2/w_n\\ ...& ...& ...& ...\\ w_n/w_1& w_n/w_2& ...& w_n/w_n\end{array}\right]$

Wherein: n is network attribute parameter sum, b _IjRepresent that i Attribute Relative is in the significance level of j attribute.About how to confirm b _IjValue, Saaty etc. suggestion reference numerals 1-9 and reciprocal as scale.The implication that table 1 has been listed the 1-9 scale represents the individual character, experience and knowledge based on the optant and the parameter of the significance level of carrying out is selected.

From psychological perspective, classification can surmount people's judgement too much, has both increased the difficulty of doing judgement, and therefore false data easily is provided again.The correctness of people's judged result under various different scales that the people such as Saaty have also used comparison, experimental result also show, adopt the scale of 1-9 the most suitable.

Table 1.Saaty scale

Sequence number	Importance rate	b ijAssignment
			1	I, j two elements are of equal importance	1
2	The i element is more important a little than j element	3
			3	The i element is obviously more important than j element	5
4	The i element is strongly more important than j element	7
			5	The i element is extremely more important than j element	9
6	The i element is slightly more inessential than j element	1/3
			7	The i element is obviously more inessential than j element	1/5
8	The i element is strongly more inessential than j element	1/7
			9	The i element is extremely more inessential than j element	1/9

According to the difference of type of service, the judgement matrix of subjective enabling legislation can change.Judgment matrix in the AHP method is A ^AHP, the judgment matrix of G-1 method is G ^G-1Therefore under every kind of type of service, in conjunction with the relative importance between each attribute in the practical application, and in conjunction with the Saaty scale, can obtain A ^AHPAnd G ^G-1, the dissimilar judgement matrix of AHP and G-1 method provides in table 2 and table 3 respectively.

The AHP method is determined subjective weight: corresponding to every kind of type of service, and associative list 2, according to AHP Determining Weights step, under the particular traffic type condition, can calculate the subjective weight that the AHP method is determined:

$W^{\mathrm{AHP}}=(w_1^{\mathrm{AHP}},w_2^{\mathrm{AHP}},w_3^{\mathrm{AHP}},w_4^{\mathrm{AHP}},w_5^{\mathrm{AHP}},w_6^{\mathrm{AHP}})$

The G-1 method determines that subjective weight is divided into three steps, at first according to certain evaluation criterion all evaluation indexes is carried out importance ranking, and the significance level ratio of adjacent index after the given ordering then calculates the weight of each index at last.The subjective weight vector that G-1 method under certain particular traffic type is determined is:

$W^{G-1}=(w_1^{G-1},w_2^{G-1},w_3^{G-1},w_4^{G-1}{w}_5^{G-1}{w}_6^{G-1})$

Group Decision is by concentrating group member wisdom to bring into play the advantage of group decision-making, and therefore selected group member should be representative and will be guaranteed some.The selected four kinds of decision-making members of this method are respectively analytic hierarchy process (AHP), Information Entropy, G-1 method and dispersion method, four kinds of decision-making techniques comprise two kinds of subjective enabling legislations, two kinds of objective weighted models, quantity is moderate, the advantage of Group Decision more can take into full account customer requirements, type of service and network condition when obtaining the network attribute weight parameter and can not only utilize a plurality of policymaker to participate in by Group Decision.

If A=(a _Ij), B=(b _Ij) and C=(c _Ij) being the positive reciprocal matrix in n rank, the ordering vector that is obtained by A is W=(w ₁, w ₂..., w _n) ^T, matrix W=(w _i/ w _j) be called the eigenmatrix of A.The product C (A, B) of definition A and B=e ^T* A*B*e is A, the compatible degree of B, wherein e ^T=(1,1 ... 1).

For convenient, generally get its logarithm as compatible degree, be designated as: General, LC(A, B) 〉=0, if LC(A, B)=0, A then, B is fully compatible.

Four kinds of enabling legislations are based on same network paramter matrix, and the ordering vector of being determined by these four kinds of enabling legislations is respectively W ^EW, W ^σ, W ^AHPAnd W ^G-1, calculate note for convenient:

$W^{\mathrm{ew}}={\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">W</figure-callout>}}^1=(w_1^1,w_2^1,w_3^1,w_4^1,w_5^1,w_6^1)$

$W^{\mathrm{\&sigma;}}=W^2=(w_1^2,w_2^2,w_3^2,w_4^2{,w}_5^2,w_6^2)$

$W^{\mathrm{AHP}}=W^3=(w_1^3,w_2^3,w_3^3,w_4^3,w_5^3,w_6^2)$

$W^{G-1}=W^4=(w_1^4,w_2^4,w_3^4,w_4^4,w_5^4,w_6^4)$

Four ordering vector characteristic of correspondence matrixes are respectively A ^EW, A ^σ, A ^AHPAnd A ^G-1Geometric generalization average vector W=(w ₁, w ₂, w ₃, w ₄, w ₅, w ₆) be so that A under the logarithm meaning ^EW, A ^σ, A ^AHPAnd A ^G-1With comprehensive characteristics matrix W=(w _i/ w _j) the most compatible vector, even

Get the vector of minimum value.If W=is (w ₁, w ₂, w ₃, w ₄, w ₅, w ₆) so that P gets minimum value, should have so

Because

Solution formula gets:

$w_t=\left(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}{\left(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}\right)}^{1/4}=\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}{\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)}^{1/4}$

In conjunction with above-mentioned formula, obtain the geometric generalization average vector of four kinds of weight vectors:

W＝(w ₁,w ₂,w ₃,w ₄,w ₅,w ₆)

The comprehensive characteristics matrix:

A＝[(w _i/w _j)],1≤i,j≤6

The method is index S I(A, B)=C (A, B)/n ²Be called matrix A, the compatibility index of B.

Usually, A and W have compatibility, and still, the compatibility of A and W determines by A fully, and works as

Think that A and W have satisfied compatibility.Therefore for the ease of judging, the method is got

As the boundary value of compatibility index, boundary value provides in table 5.

When weighing A and B when whether compatible, when The time the method think that A and B have gratifying compatibility.

Be to guarantee the reasonability of mix vector, the method is carried out consistency check to synthetic weight, namely judgment matrix A whether with A ^EW, A ^σ, A ^AHP, A ^G-1Have respectively satisfied compatibility, their compatibility index is respectively:

${\mathrm{<figure-callout\; id="1"\; label="SI"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">SI</figure-callout>}}_1=\mathrm{SI}(A,A^{\mathrm{EW}})=C(A,A^{\mathrm{EW}})/6^2\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j/w_j)(w_j^{\mathrm{EW}}/w_i^{\mathrm{EW}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{EW}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{EW}}/w_j)/6^2$

${\mathrm{SI}}_2=\mathrm{SI}(A,A^{\mathrm{\&sigma;}})=C(A,A^{\mathrm{\&sigma;}})/6^2=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{\mathrm{\&sigma;}}/w_i^{\mathrm{\&sigma;}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{\&sigma;}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{\&sigma;}}/w_j)/6^2$

${\mathrm{SI}}_3=\mathrm{SI}(A,A^{G-1})=C(A.A^{G-1})/6^2=\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{G-1}/w_i^{G-1})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{G-1})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{G-1}/w_j)/6^2$

${\mathrm{SI}}_4=\mathrm{SI}(A,A^{\mathrm{AHP}})=C(A,A^{\mathrm{AHP}})/6^2$

When four compatibility indexes simultaneously less than same order compatibility index critical value

Then the synthetic weight of explanation meets the compatibility requirement.The weight vector that utilizes Group Decision to obtain carries out network to be selected, and namely obtains after the weight, and in conjunction with the SAW method, the performance function of each network can be expressed as:

Optimum network is: $F^{*}=\arg \underset{i\&Element;N}{\max }\underset{j\&Element;6}{\mathrm{\&Sigma;}}{w}_j{r}_{\mathrm{ij}.}$

The present invention in network selection procedures, has considered the network objective attribute, user's request and type of service with the content application of Group Decision.

Beneficial effect:

1. when adopting the Group Decision theory to obtain the integrated network attribute weight, adopt four policymaker, quantity is moderate and representative respectively, comprehensive weight guarantees its reasonability by the compatibility theory testing, be determining clear and definite and abundant theory support being arranged of weight, and correctly taking full advantage of combinatorial theory, network selection procedures is unlikely to again undue complexity.

2. reach the impact of considering network objective attribute, user preference and type of service by comprehensive a plurality of policymaker's wisdom in the network selection procedures.And can provide satisfied QoS by given different judgement matrix for the user under the different service types according to the difference of type of service.

Description of drawings

Fig. 1 is that network of the present invention is selected block diagram.

Fig. 2 is method particular flow sheet of the present invention.

Fig. 3 is the time delay simulation figure of session service.

Fig. 4 is the judder figure of session service.

Fig. 5 is the throughput analogous diagram of session service.

Fig. 6 is the judder figure of streaming media service.

Fig. 7 is the packet loss analogous diagram of streaming media service.

Fig. 8 is the throughput analogous diagram of streaming media service.

Fig. 9 is the packet loss analogous diagram of interactive service.

Figure 10 is the throughput analogous diagram of interactive service.

Figure 11 is the cost analogous diagram of interactive service.

Figure 12 is the cost analogous diagram of background business.

Figure 13 is the packet loss analogous diagram of background business.

Embodiment

Be elaborated below in conjunction with the technical scheme of accompanying drawing to invention:

Thinking of the present invention is that Group Decision and compatibility theory are applied to the weight that solves each property parameters that network is selected in the heterogeneous network vertical handover procedure, the network that on this weight basis network is sorted and select the cost function maximum as shown in Figure 1, at first give objective parameter and calculate objective network attribute weight vector by Information Entropy and dispersion method, give subjective parameters and obtain respectively separately subjective network attribute weight vector by AHP method and G-1 method, four kinds of methods have represented four policymaker, utilize the Group Decision theory with four kinds of weight vectors that the weight vector combination producing is new, thereby obtain a comprehensive weight vector W, carry out network in conjunction with the SAW method at last and select.

The detail flowchart that whole network switching process adopts the Group Decision method to realize that network is selected is seen Fig. 2.

One, objective weight determines

Heterogeneous network model of the present invention has comprised N kind heterogeneous network, the property parameters that network relates in selecting is available bandwidth, hardware circuit solution, packet delay, packet jitter, packet loss and every bit expense, B represents bandwidth in this method, R represents hardware circuit solution, and D represents packet delay, J and represents that packet jitter, L represent packet loss, C represents every bit expense.Suppose that the standardized network attribute matrix in Information Entropy and the dispersion method is:

R=(x _Ij) _{N * 6}, i.e. N heterogeneous network number, 6 kinds of network attributes.Wherein:

x _IjJ the attribute that represents i network, as:

x _I1Represent i bandwidth of network;

x _I2The hardware circuit solution that represents i network;

x _I3The time delay that represents i network;

x _I4Represent the shake of i network;

x _I5The packet loss that represents i network;

x _I6The every bit expense that represents i network, and 1＜=i＜=N, 1＜=j＜=6.

Generally speaking, attribute is divided into benefit type and cost type, and the attribute of benefit type is the bigger the better: get $x_{\max }^j=\max (x_{1j},x_{2j},...,x_{\mathrm{Nj}});$

The attribute of cost type is the smaller the better: get $x_{\min }^j=\min (x_{1j},x_{2j},...,x_{\mathrm{Nj}}).$

For B and R, they belong to benefit type attribute, and the standardization formula is:

$r_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},1\&le;j\&le;2---\left(1\right)$

For D, J, L and C, they belong to cost type attribute, and the standardization formula is:

$r_{\mathrm{ij}}=\frac{x_{\max }^i+x_{\min }^i-x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},3\&le;j\&le;6---\left(2\right)$

Information Entropy is determined objective weight:

(1) standardization

${\stackrel{\&OverBar;}{r}}_{\mathrm{ij}}=r_{\mathrm{ij}}/\underset{i=1}{\overset{6}{\mathrm{\&Sigma;}}}{r}_{\mathrm{ij}},1\&le;j\&le;6---\left(3\right)$

(2) determine comentropy

$H_i=-K\underset{j=1}{\overset{6}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{r}}_{\mathrm{ij}}1n{\stackrel{\&OverBar;}{r}}_{\mathrm{ij}},1\&le;i\&le;N---\left(4\right)$

Here: K=1/ln N

(3) Determining Weights vector

$w_j^{\mathrm{EW}}=\frac{1-H_i}{6-\underset{j=1}{\overset{6}{\mathrm{\&Sigma;}}}{H}_i},1\&le;i\&le;N---\left(5\right)$

Can obtain the weight vectors of attribute by Information Entropy:

$W^{\mathrm{EW}}=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^{\mathrm{EW}},w_2^{\mathrm{EW}},w_3^{\mathrm{EW}},w_4^{\mathrm{EW}},w_5^{\mathrm{EW}},w_5^{\mathrm{EW}},{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^{\mathrm{EW}})$

Dispersion method is determined objective weight: dispersion method is similar with the Information Entropy Computing Principle.Generally the standard deviation of an index is directly proportional with the degree of variation of desired value in certain network, if standard deviation is larger, then the degree of variation of this index is larger, and the amount of information that provides is larger, plays a role in evaluation larger, and its weight is also larger.Otherwise then weight is less.The present invention calculates six networks and unifies the poor formula of attribute and be:

${\mathrm{\&sigma;}}_j=\sqrt{\frac{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{(r_{\mathrm{ij}}-{\stackrel{\&OverBar;}{r}}_j)}^2}{N}},{\stackrel{\&OverBar;}{r}}_j=\frac{\underset{i=1}{\overset{\mathrm{<figure-callout\; id="6"\; label="N\; r\; ij"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">\; <figure-callout\; id="1"\; label="N\; r\; ij"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">N</figure-callout>\; </figure-callout>}}{\mathrm{\&Sigma;}}}{r}_{\mathrm{ij}}{}_{}{}_{}}{6},1\&le;j\&le;6---\left(6\right)$

Utilize each index weights formula of N heterogeneous network of dispersion method calculating as follows:

$w_j^{\mathrm{\&sigma;}}=\frac{{\mathrm{\&sigma;}}_j}{\underset{j=1}{\overset{6}{\mathrm{\&Sigma;}}}{\mathrm{\&sigma;}}_j},j=\mathrm{1,2},...,6---\left(7\right)$

Network attribute matrix R=(r then _Ij) _{N * 6}The weight vectors that utilizes dispersion method to calculate is:

$W^{\mathrm{\&sigma;}}=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^{\mathrm{\&sigma;}},w_2^{\mathrm{\&sigma;}},w_3^{\mathrm{\&sigma;}},w_4^{\mathrm{\&sigma;}},w_5^{\mathrm{\&sigma;}},{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^{\mathrm{\&sigma;}}).$

Two, subjective weight determines

The present invention is subjective, and enabling legislation adopts AHP method and G-1 method.According to the difference of type of service, the judgement matrix of subjective enabling legislation can change.Judgment matrix in the AHP method is A ^AHP, the judgement vector of G-1 method is G ^G-1, the dissimilar judgement matrix of AHP and G-1 method provides in table 2 and table 3 respectively.The data of judgement matrix are to provide according to relative importance between the different corresponding parameter of type of service according to the expert is given in the table, and provide successively in conjunction with the Saaty scale.

Table 2 is judgment matrix corresponding to different service types in the AHP method

Table 3 is judgment matrix corresponding to different service types in the G-1 method

Type of service	r 2	r 3r 4r 5	r 6	The ordering situation
					Session service	1.2	1.611.2	1.2	R>J>B>C>P>L
Streaming media service	1.8	111	1	L>B>P>R>J>C
					Interaction service
	1	1.811.4	1	L>C>R>J>B>P
					Background business	1.2	1.411	1.2	C>L>J>R>B>R

So under every kind of type of service, A ^AHPAnd G ^G-1Corresponding with judgement matrix in table 2 and the table 3 respectively.The AHP method is determined subjective weight: corresponding to every kind of type of service, and associative list 2, according to AHP Determining Weights step, under the particular traffic type condition, can calculate the subjective weight that the AHP method is determined:

$W^{\mathrm{AHP}}=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^{\mathrm{AHP}},w_2^{\mathrm{AHP}},w_3^{\mathrm{AHP}},w_4^{\mathrm{AHP}},w_5^{\mathrm{AHP}},{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^{\mathrm{AHP}})$

The G-1 method is determined subjective power: the G-1 method determines that subjective weight is divided into three steps, at first according to certain evaluation criterion all evaluation indexes is carried out importance ranking, and the significance level ratio of adjacent index after the given ordering then calculates the weight of each index at last.Concrete steps are as follows:

(1) order relation determines

With respect to certain evaluation criterion significance level ordering, then the importance ranking situation of six kinds of attributes under every kind of type of service under this heterogeneous network model is as follows to all evaluation indexes:

Session service: R〉J〉B〉C〉P〉L

Streaming media service: L〉B〉P〉R〉J〉C

Interaction service: L〉C〉R〉J〉B〉P

Background business: C〉L〉J〉R〉B〉R

(2) adjacent relative significance of attribute degree is judged

The ratio of significance level is between the adjacent attribute under the significance level of certain evaluation criterion sorts:

w _k-1/w _k＝r _k,2≤k≤6 (8)

r _kAssignment can be with reference to table 4, and table 4 comes from G-1 method author and is Guo second place's works " comprehensive evaluation theory, method and application ", Science Press in May, 2007.The author has defined the assignment rule of this r value when introducing the G-1 method.

R value in the table 3 under the various types of traffic is that associative list 2 and table 4 provide, namely according to the relative importance between each attribute under given each type of service of expert, r value assignment rule provides successively, thereby guarantees to be consistent for the subjective judgement of all types of service.

Table 4 is r value assignment table in the G-1 method

r k	Explanation
		1.0	Index x k-1And x kEqual importance is arranged
1.2	Index x k-1Compare x kImportant a little
		1.4	Index x k-1Compare x kObviously important
1.6	Index x k-1Compare x kStrongly important
		1.8	Index x k-1Compare x kExtremely important

(3) weight determines

At first calculate most important attribute weight W after the importance ranking ₆, then calculate successively follow-up weight.

Weight can have following formula to determine:

${\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6=\frac{1}{1+\underset{k=2}{\overset{6}{\mathrm{\&Sigma;}}}\underset{i=k}{\overset{6}{\mathrm{\&Pi;}}}{r}_i}---\left(9\right)$

Can draw follow-up weight by step (2):

w _k-1＝r _k×w _k,k＝6,5,...，2 (10)

In conjunction with above-mentioned formula, and with reference to r value in the table 3, the subjective weight vector that the G-1 method under certain particular traffic type is determined is:

$W^{G-1}=(w_1^{G-1},w_2^{G-1},w_3^{G-1},w_4^{G-1},w_5^{G-1},w_6^{G-1})$

Three, Group Decision and consistency check

Group Decision is by concentrating group member wisdom to bring into play the advantage of group decision-making, and therefore selected group member should be representative and will be guaranteed some.The selected four kinds of decision-making members of the present invention are respectively analytic hierarchy process (AHP), Information Entropy, G-1 method and dispersion method, four kinds of decision-making techniques comprise two kinds of objective weighted models of two kinds of subjective enabling legislations, quantity is moderate, the advantage of Group Decision more can take into full account customer requirements, type of service and network condition when obtaining the network attribute weight parameter and can not only utilize a plurality of policymaker to participate in by Group Decision.Next will introduce and use Group Decision and four kinds of weight vectors of the theoretical combination of compatibility.

If A=(a _Ij), B=(b _Ij) and C=(c _Ij) being the positive reciprocal matrix in n rank, the ordering vector that is obtained by A is W=(w ₁, w ₂..., w _n) ^T, matrix W=(w _i/ w _j) be called the eigenmatrix of A.The product C (A, B) of definition A and B=e ^T* A*B*e is A, the compatible degree of B, wherein e ^T=(1,1 ... 1).For convenient, generally get its logarithm as compatible degree, be designated as

General, LC(A, B) 〉=0, if LC(A, B)=0, A then, B is fully compatible.

Four kinds of enabling legislations are based on same network paramter matrix, and the ordering vector of being determined by these four kinds of enabling legislations is respectively W ^EW, W ^σ, W ^AHPAnd W ^G-1, calculate note for convenient:

$W^{\mathrm{ew}}={\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">W</figure-callout>}}^1=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^1,w_2^1,w_3^1,w_4^1,w_5^1,{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^1)$

$W^{\mathrm{\&sigma;}}=W^2=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^2,W_2^2,w_3^2,w_4^2,w_5^2,{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^2)$

$W^{\mathrm{AHP}}=W^3=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^3,w_2^3,w_3^3,w_4^3,w_5^3,{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^3)$

$W^{G-1}=W^4=({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_1^4,w_2^4,w_3^4,w_4^4,w_5^4,{\mathrm{<figure-callout\; id="6"\; label="w"\; filenames="HDA00003231111400031.png"\; state="\{\{state\}\}">w</figure-callout>}}_6^4)$

Four ordering vector characteristic of correspondence matrixes are respectively A ^EW, A ^σ, A ^AHPAnd A ^G-1Geometric generalization average vector W=(w ₁, w ₂, w ₃, w ₄, w ₅, w ₆) be so that A under the logarithm meaning ^EW, A ^σ, A ^AHPAnd A ^G-1With comprehensive characteristics matrix W=(w _i/ w _j) the most compatible vector, even Get the vector of minimum value.If W=is (w ₁, w ₂, w ₃, w ₄, w ₅, w ₆) so that P gets minimum value, should have so:

$\frac{\&PartialD;P}{\&PartialD;t}=\mathrm{0,1}\&le;t\&le;6---\left(11\right)$

Because

Solution formula gets:

$w_t=\left(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}{)}^{1/4}=\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}{\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)}^{1/4}---\left(12\right)$

In conjunction with above-mentioned formula, obtain the geometric generalization average vector of four kinds of weight vectors:

W＝(w ₁,w ₂,w ₃,w ₄,w ₅,w ₆)

Comprehensive characteristics matrix: A=[(w _i/ w _j)], 1≤i, j≤6

The method is index S I(A, B)=C (A, B)/n ²Be called matrix A, the compatibility index of B.Usually, A and W have compatibility, and still, the compatibility of A and W determines by A fully, and works as

Think that A and W have satisfied compatibility.Therefore for the ease of judging, the method is got

As the boundary value of compatibility index, boundary value provides in table 5.Table 5 is " compatibility and Group Decision " of publication " system engineering theory with put into practice " in February, 2000 (the 2nd phase), the Wang Lian of the Renmin University of China is fragrant, wherein calculate, the present invention only need use the reasonability of data checking method process variable in the table 5, therefore directly quotes its data at this.

Table 5 is the S.I. critical value

When weighing A and B when whether compatible, when The time the method think that A and B have gratifying compatibility.

For guaranteeing the reasonability of mix vector, synthetic weight is carried out consistency check, namely whether judgment matrix A is and A ^EW, A ^σ, A ^AHPAnd A ^G-1Have respectively satisfied compatibility, their compatibility index is respectively:

${\mathrm{<figure-callout\; id="1"\; label="SI"\; filenames="HDA00003231111400022.png,HDA00003231111400031.png"\; state="\{\{state\}\}">SI</figure-callout>}}_1=\mathrm{SI}(A,A^{\mathrm{EW}})=C(A,A^{\mathrm{EW}})/6^2\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j/w_j)(w_j^{\mathrm{EW}}/w_i^{\mathrm{EW}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{EW}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{EW}}/w_j)/6^2---\left(13\right)$

${\mathrm{SI}}_2=\mathrm{SI}(A,A^{\mathrm{\&sigma;}})=C(A,A^{\mathrm{\&sigma;}})/6^2=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{\mathrm{\&sigma;}}/w_i^{\mathrm{\&sigma;}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{\&sigma;}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{\&sigma;}}/w_j)/6^2---\left(14\right)$

${\mathrm{SI}}_3=\mathrm{SI}(A,A^{G-1})=C(A,A^{G-1})/6^2=\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{G-1}/w_i^{G-1})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{G-1})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{G-1}/w_j)/6^2---\left(15\right)$

${\mathrm{SI}}_4=\mathrm{SI}(A,A^{\mathrm{AHP}})=C(A,A^{\mathrm{AHP}})/6^2---\left(16\right)$

When four compatibility indexes simultaneously less than same order compatibility index critical value

Then the synthetic weight of explanation meets the compatibility requirement.

Four, network is selected

The geometric generalization weight vector that satisfies the compatibility requirement that utilizes Group Decision to obtain carries out network to be selected, and namely obtains after the weight, and in conjunction with the SAW method, the performance function of each network can be expressed as:

$F=\underset{j=1}{\overset{6}{\mathrm{\&Sigma;}}}{w}_j{r}_{\mathrm{ij}},1\&le;i\&le;N---\left(17\right)$

Optimum network is:

$F^{*}=\arg \underset{i\&Element;N}{\max }\underset{j\&Element;6}{\mathrm{\&Sigma;}}{w}_j{r}_{\mathrm{ij}}---\left(18\right)$

In sum, usefulness of the present invention provides by simulation result:

Heterogeneous network in the emulation is respectively WLAN, UMTS and WIMAX, and every type comprises two networks.The present invention in emulation compared with the prior art: " a kind of heterogeneous network selection algorithm of multiple attribute decision making (MADM) " that in analogous diagram, carries with " EW " representative " radio engineering " the 39th volume in 2009 the 1st periodical, author: Wang Kang, Ceng Zhimin, Feng Chunyan, method among the Zhang Tiankui (Beijing University of Post ﹠ Telecommunication's communication network comprehensive technology research institute), the method in " 6377/2010:213-220 " document " ANovel AHP and GRA Based Handover Decision Mechanism in Heterogeneous Wireless Networks " of being numbered of publishing with " GRA " representative " Information Computing and Applications Lecture Notes in Computer Science " in the analogous diagram.The property parameters that network relates in selecting is available bandwidth (Available Bandwidth, B), hardware circuit solution (Peak Data Rate, R), packet delay (Packet Delay, D), packet jitter (Packet Jitter, J), packet loss (Packet Delay, D) with every bit expense (Cost Per Bit, C), shown in the table 6.And three kinds of heterogeneous networks in the table 6, i.e. UMTS, WLAN and WiMAX, the distribution of their six kinds of attributes is to provide according to the network attribute data of adding up in the practical application.

Table 6 is the network attribute parameter

Property parameters	UMTS1	UMTS2	WLAN1	WLAN2WiMAX1WiMAX2
					B(MHz)	0.1-2	0.1-2	1-11	1-541-601-60
R(Mbps)	2	2	11	546060
					D(ms)	25-50	25-50	100-150	100-15060-10060-100
J(ms)	5-10	5-10	10-20	10-203-103-10
					L(per10 6)	20-80	20-80	20-80	20-8020-8020-80
C(price)	0.6	0.8	0.1	0.050.50.4

Select 4 kinds of types of service to come the measure algorithm performance in the emulation, concrete simulation result such as Fig. 3-shown in Figure 13.

Fig. 3-Fig. 5 has provided the performance of session service.The voice communication of session service requires low time delay and lower bandwidth, and video communication requires low time delay and enough bandwidth, so session service is more valued the delay variation situation in the judgement matrix of table 2 and table 3.As can be seen from the figure, contrast other two kinds of methods, the inventive method can provide optimum delay time and jitter performance and gratifying bandwidth, thereby can satisfy user's qos requirement.

Fig. 6-Fig. 8 has provided the performance of streaming media service.The error rate that streaming media service is had relatively high expectations also allows a fixed response time, and is higher to bandwidth requirement, so session service is more valued the throughput situation in the judgement matrix of table 2 and table 3.As can be seen from the figure, contrast other two kinds of methods, the inventive method can provide the optimal jitter state, gratifying packet loss and optimum bandwidth performance, thus can satisfy user's qos requirement.

Fig. 9-Figure 11 has provided the performance of interactive service.Interaction service has certain requirement to the error rate, relatively low time delay and relatively high data downstream speed, so session service is more valued the packet loss situation in the judgement matrix of table 2 and table 3.As can be seen from the figure, contrast other two kinds of methods, the inventive method can provide best packet loss performance and higher throughput under the lowest costs prerequisite, thereby can satisfy user's qos requirement.

Figure 12-Figure 13 has provided the performance of background business.Background business is very low to delay requirement, and higher error rate requirement is arranged, so session service is more valued the packet loss situation in the judgement matrix of table 2 and table 3.As can be seen from the figure, contrast other two kinds of methods, the inventive method can provide best packet loss performance under the lowest costs prerequisite, thereby can satisfy user's qos requirement.Therefore, according to above-mentioned simulation result and in conjunction with the characteristics of various types of traffic, can reach a conclusion, this method can provide according to type of service and make customer satisfaction system QoS.

The present invention uses the Group Decision theory that multiple weight vectors in the multiple attribute decision making (MADM) is made up, and has proposed a kind of network selecting method.When the method adopts the Group Decision theory to obtain the integrated network attribute weight, adopt four policymaker, quantity is moderate and representative respectively, comprehensive weight guarantees its reasonability by the compatibility theory testing, be determining clear and definite and abundant theory support being arranged of weight, and correctly taking full advantage of combinatorial theory, network selection procedures is unlikely to again undue complexity.Reach the impact of considering network objective attribute, user preference and type of service by comprehensive a plurality of policymaker's wisdom in the network selection procedures, can be made overall plans in the aspect that additive method can't be taken into account, facts have proved, the method can provide satisfied QoS by given different judgement matrix for the user under the different service types according to the difference of type of service.

Claims (1)

1. heterogeneous network system of selection based on four kinds of weight vectors is characterized in that the method may further comprise the steps:

A. the foundation of network paramter matrix: the heterogeneous network model of this method comprises N kind heterogeneous network, and the network attribute of choosing comprises bandwidth (B), hardware circuit solution (R), packet delay (D), packet jitter (J), packet loss (L) and every bit expense (C);

The network attribute matrix of supposing the method is R=(x _Ij) _{N * 6}, i.e. N heterogeneous network number, 6 kinds of network attributes, x _IjJ the attribute that represents i network, and x _I1Represent i bandwidth of network, x _I2The hardware circuit solution that represents i network, x _I3The time delay that represents i network, x _I4Represent the shake of i network, x _I5The packet loss that represents i network, x _I6The every bit expense that represents i network, and 1＜=i＜=N, 1＜=j＜=6;

B. network attribute standard parameter: at first the network attribute matrix is carried out standardization, generally speaking, attribute is divided into benefit type and cost type, and the attribute of benefit type is the bigger the better: get

The attribute of cost type is the smaller the better: get $x_{\min }^j=\min (x_{1j},x_{2j},...,x_{\mathrm{Nj}});$

For bandwidth (B) and hardware circuit solution (R), they belong to benefit type attribute, and the standardization formula is: $r_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},1\&le;j\&le;2;$

For packet delay (D), packet jitter (J), packet loss (L) and every bit expense (C), they belong to cost type attribute, and the standardization formula is: $r_{\mathrm{ij}}=\frac{x_{\max }^i+x_{\min }^i-x_{\mathrm{ij}}}{x_{\max }^j+x_{\min }^j},3\&le;j\&le;6;$

The all properties parameter all will drop between 0 and 1 after the standardization, and the network paramter matrix after the standardization is: R=(r _Ij) _{N * 6}

C. four groups of weight vectors is definite: this standardized network attribute matrix is calculated two groups of objective weight vectors in conjunction with entropy power method and dispersion method respectively, be respectively: $W^{\mathrm{ew}}=W^1=(w_1^1,w_2^1,w_3^1,w_4^1,w_5^1,w_6^1)$ With $W^{\mathrm{\&sigma;}}=W^2=(w_1^2,W_2^2,w_3^2,w_4^2,w_5^2,w_6^2);$

According to judgment matrix (table 2) corresponding to different service types in the given AHP method, utilize the AHP method to calculate one group of subjective weight vectors in conjunction with the network attribute matrix: $W^{\mathrm{AHP}}=W^3=(w_1^3,w_2^3,w_3^3,w_4^3,w_5^3,w_6^3),$ Same, according to judgment matrix (table 3) corresponding to different service types in the given G-1 method, utilize the G-1 method to calculate another kind of subjective weight vectors: $W^{G-1}=W^4=(w_1^4,w_2^4,w_3^4,w_4^4,w_5^4,w_6^4);$

D. determine comprehensive weight vector and compatibility index: establish A=(a _Ij), B=(b _Ij) and C=(c _Ij) being the positive reciprocal matrix in n rank, the ordering vector that is obtained by A is: W=(w ₁, w ₂..., w _n) ^T, matrix W=(w _i/ w _j) be called the eigenmatrix of A; The product C (A, B) of definition matrix A and matrix B=e ^T* A*B*e is matrix A, the compatible degree of B, wherein e ^T=(1,1 ... 1); Index S I(A, B)=C (A, B)/n ²Be called matrix A, the compatibility index of B;

Usually, A and W have compatibility, and still, the compatibility of A and W determines by A fully, and works as

Think that A and W have satisfied compatibility; For obtaining the geometric generalization weight vectors, four groups of weight vectors that calculate are carried out Group Decision, in conjunction with formula:

$w_t=\left(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}{\left(w_t^{\mathrm{EW}}{w}_t^{\mathrm{\&sigma;}}{w}_t^{\mathrm{AHP}}{w}_t^{G-1}\right)}^{1/4}=\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)/\underset{t=1}{\overset{6}{\mathrm{\&Sigma;}}}{\left(w_t^1{w}_t^2{w}_t^3{w}_t^4\right)}^{1/4}$ Calculate geometric generalization average vector W with and eigenmatrix A; Four ordering vectors determining among the step c are respectively W ^EW, W ^σ, W ^AHPAnd W ^G-1, can determine that according to the definition of eigenmatrix their characteristic of correspondence matrixes are respectively A ^EW, A ^σ, A ^AHPAnd A ^G-1, then determine that according to the definition of compatibility index the compatibility index of these four eigenmatrixes and geometric generalization evaluation vector W is respectively:

${\mathrm{SI}}_1=\mathrm{SI}(A,A^{\mathrm{EW}})=C(A,A^{\mathrm{EW}})/6^2\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j/w_j)(w_j^{\mathrm{EW}}/w_i^{\mathrm{EW}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{EW}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{EW}}/w_j)/6^2$

${\mathrm{SI}}_2=\mathrm{SI}(A,A^{\mathrm{\&sigma;}})=C(A,A^{\mathrm{\&sigma;}})/6^2=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{\mathrm{\&sigma;}}/w_i^{\mathrm{\&sigma;}})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{\mathrm{\&sigma;}})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{\mathrm{\&sigma;}}/w_j)/6^2$

${\mathrm{SI}}_3=\mathrm{SI}(A,A^{G-1})=C(A.A^{G-1})/6^2=\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_j)(w_j^{G-1}/w_i^{G-1})=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_i/w_i^{G-1})\underset{j=1}{\overset{M}{\mathrm{\&Sigma;}}}(w_j^{G-1}/w_j)/6^2$

${\mathrm{SI}}_4=\mathrm{SI}(A,A^{\mathrm{AHP}})=C(A,A^{\mathrm{AHP}})/6^2,$ The eigenmatrix A of difference calculated characteristics matrix A and four kinds of weight vectors ^EW, A ^σ, A ^AHPAnd A ^G-1Compatibility index;

E. consistency check: for the ease of judging, get

Boundary value as compatibility index;

When weighing A and B when whether compatible, when

The time the method think that A and B have gratifying compatibility; To judge that in this step whether these indexs are simultaneously less than same order compatibility index critical value

(table 5);

If do not satisfy compatibility index, then need to revise the subjectivity judgement matrix under this kind type of service, turn back to step c; If satisfy, will enter network and select link;

F. network is selected: former steps have calculated the geometric generalization weight vectors that satisfies compatibility index of six attributes, in this step, the geometric generalization weight vectors calculates the cost function value of each network in conjunction with the SAW method, to its ordering, the network of replacement valency functional value maximum is as objective network, thereby the optimum network that obtains under this network state is numbered.

Images