CN103108382A - Heterogeneous network multi-attribute decision-making method based on network analytic hierarchy process - Google Patents

Abstract

The invention discloses a heterogeneous network multi-attribute decision-making method based on a network analytic hierarchy process. According to the method, when computing network attributes weight, attributive interaction and feedback in a dynamic network are considered, influence on network selection of a target network is also considered, physical truths for decision problems of a heterogeneous network are well fit, and a low-delay and low-quivering network can be selected under a real-time voice service. Concrete steps comprise that weight is firstly calculated; attributive factors affecting the network selection and the target network are divided into a functional group, a cost group and a scheme group; mutual relation between intra-groups and between-group elements and mutual relation between groups are set up; judgment matrixes of pairwise comparison according to the analytic hierarchy process are set up, feature vectors are obtained, submatrixes are formed, and all the submatrixes form a non-weighted hypermatrix, and an ultimate hypermatrix and ANP weight are obtained through weighting and exponentiation operating of the hypermatrix; and then the ANP weight and normalization network parameters obtain network power functional values, the network power functional values are sorted, and the largest utility value is selected to serve as the target network.

Description

A kind of heterogeneous network multiple attributive decision making method of analytic hierarchy process (AHP) Network Based

Technical field

The present invention relates to a kind of analytic hierarchy process (AHP) ANP(heterogeneous network Network Based) the heterogeneous network multiple attributive decision making method, belong to communication technical field.

Background technology

Along with the development of wireless communication technology, will be the situation of various radio coexistences future, and heterogeneous network is the development trend of next generation wireless network.The unified technology that formed by certain advanced wireless technology of contemplated 3-G (Generation Three mobile communication system), the network of unified management in the past, 4G merges the collaborative work on the IP platform of various dissimilar wireless access technologys with the heterogeneous communication system that forms, and how existing network being carried out choose reasonable is the major issue of studying at present.Under isomery UNE environment, isomerism and the otherness of network are larger, at first the user needs according to network the present situation, network to be selected under the netinit state, after selecting to finish, change along with the user geographical position, the variation of business, and the variation of network itself, not necessarily current time is best for original network performance of selecting, at this moment need network is reselected, vertically switch in heterogeneous network, network is initial all to be needed network is adjudicated and decision-making during with vertical switching.

In the heterogeneous network communication system of various wireless access technology collaborative works, when network is selected, between network, the RSS index does not have comparativity, although can be respectively the different handoff threshold value of network settings, is only inadequate based on this factor decision-making.For the satisfied QoS of user and minimizing service cost are provided, except considering to receive the intensity of signal, also need the many factors such as the QoS relevant according to network, application, user and terminal, consumer taste, service rate, safe class, network parameter that the network performance in whole heterogeneous network is comprehensively judged, this is a typical Multiple Attribute Decision Problems.Multiple attribute decision making (MADM) is network selecting method classical in heterogeneous network, wherein objective making decision mechanism comprises and approaches ideal solution ranking method (TOPSIS), entropy power method (EW), gray analysis method (GRA) etc., and subjective decision mechanism comprises simple weighted method (SAW), exponential weighting method (MEW) and analytic hierarchy process (AHP) (AHP) etc.AHP, SAW and EW algorithm are the methods of comparatively commonly using, and are mainly the weights that calculates each network attribute, then by network parameter, network performance are analyzed and then selected best objective network.

The AHP algorithm is that widely use a kind of is used for the method for Determining Weights, substantially satisfies in scope of subjective the level certification to things.But analytic hierarchy process (AHP) is based on following three hypothesis carries out decision-making:

1. decision system is divided into some levels, the upper strata element plays dominating role to lower floor's element, is separate between the element of same level.But in fact, generally all have dependence between the element of each layer inside, also there is the effect of anti-domination in lower floor to the upper strata simultaneously.

2. decision problem can be divided into many levels, and the upper strata element plays control action to lower floor's element, and is separate between the element of same level, do not have inner interdependency.Such as preference between each index on same indicator layer is independent.Influence each other and often exist between some index in actual decision problem.

3. at all levels just exists top-down influence between adjacent two levels, does not consider that lower floor is to the reaction on upper strata.Influencing each other between non-adjacent level do not considered.And lower floor's element have reaction to upper strata unit in actual decision-making.

But supposing all to have with some actual decision problem for three deviates from, do not consider in theory interaction and the feedback of each key element in complex dynamic systems, although tried to achieve subjective weight from quantitative and qualitative analysis, not the result that meets most truth, its improvement is necessary.

Summary of the invention

Technical problem: the multiple attributive decision making method that the purpose of this invention is to provide analytic hierarchy process (AHP) ANP Network Based in a kind of heterogeneous network, the method is by considering interaction and the feedback between each property element in the complex dynamic network selective system, also has objective network to impact and the feedback of attribute, the decision-making technique of dependent reponse system has been proposed, make decision-making more tally with the actual situation, can select preferably the objective network that time delay is low, shake is little under the real-time voice business.

Technical scheme: the present invention is the multiple attributive decision making method of a kind of analytic hierarchy process (AHP) ANP Network Based in heterogeneous network, considers the network attribute internal relations, and the correlation between network attribute and objective network, has realized the decision-making that more tallies with the actual situation.

In the multiple attributive decision making method of analytic hierarchy process (AHP) ANP Network Based, at first be divided into function group C with affecting the attribute factor of network selection and all factors of objective network in heterogeneous network ₁, cost group C ₂With scheme group C ₃, C ₁Be divided into available bandwidth c ₁₁With total bandwidth c ₁₂, C ₂Be divided into packet delay c ₂₁, packet jitter c ₂₂, packet loss c ₂₃With price c ₂₄, C ₃The objective network that comprises has UMTS-1, UMTS-2, and WLAN-1, WLAN-2, WiWAX-1 and WiMAX-2 correspond respectively to element c ₃₁, c ₃₂, c ₃₃, c ₃₄, c ₃₅And c ₃₆, use C _iRepresent i element set, c _ikK the element that represents i element set, K _iRepresent the element number in i element set, i=1,2,3, k=1 ..., K _i, K ₁=2, K ₂=4, K ₃=6, three groups of 12 property elements altogether.

Build the network hierarchical structure model, phase-split network is selected decision problem and the target of problem, at first whole heterogeneous network selective system is divided into two parts, first is key-course, comprise decision objective and decision rule, wherein require independent between each decision rule and arranged by destination layer.Second portion is network layer, and it is comprised of the element of all controlled preparative layer dominations, and its inside is the formed network configuration of interactional element, i.e. C clear and definite according to the place scene ₁Function group, C ₂Cost group and C ₃The structural relation that influences each other and feed back of scheme group.

The judgment matrix of element between group in the structure group.In network configuration, if having interdependence and mutual feedback relationship between element, can a given main criterion and one criterion, two elements carry out the comparison of influence degree to inferior criterion under main criterion, can indirectly obtain judgment matrix thus.Three groups that are divided into are according to interdependence and the feedback relationship of hierarchical structure, first to element set C ₁Interior element c ₁₁And c ₁₂Respectively with each element c _ikCompare in twos during as criterion, set up the judgment matrix of 12 2*2 dimensions, each value reflection element c in matrix ₁₁And c ₁₂For element c _ikSignificance level.The same analytic hierarchy process (AHP) of using is with each element c _ikFor the criterion time-division other to element set C ₂Interior element and C ₃Interior element relatively obtains the judgment matrix of 12 4*4 dimensions and the judgment matrix of 12 6*6 dimensions in twos.Suppose element set C _iIn K _iIndividual element is with element set C _jIn k element be that the judgment matrix that criterion is set up is B=(b _mn), m, n=1,2 ..., K _i, b _mnThe significance level of m relative n the element of element in i group when reflection is organized k element as criterion take j, j=1,2,3.

The judgment matrix of organizing element between interior group is asked characteristic vector and eigenvalue of maximum.Above-mentioned judgment matrix is B=(b _mn), its characteristic vector be G=(g (1) .., g (m) ..., g (K _i)) ^T, wherein M, n=1,2 ..., K _i, b _mnMatrix element in the expression judgment matrix.When the dimension of judgment matrix B of structure is K _i2 o'clock, judgment matrix not necessarily satisfies consistency, its maximum characteristic root λ _maxAlso just no longer equal matrix dimension K _i, need to calculate maximum characteristic root by following formula:

B is judgment matrix, and G is matrix characteristic vector, (BG) _mM vector value for gained vector (BG).

Build the judgment matrix between element set.Take objective network as main criterion, use respectively analytic hierarchy process (AHP) with element set C ₁, C ₂, C ₃As inferior criterion to C ₁, C ₂, C ₃Three element set compare in twos, obtain the judgment matrix of 3 3*3 dimensions.Suppose that this class judgment matrix is E=(e _mn), e _mnExpression is with element set C _jBe the relative significance level of m relative n the element set of element set of criterion, j=1,2,3, m, n=1 ..., K _c, K _cElement set number and K that representative is compared in twos _c=3.

Calculate characteristic vector and the eigenvalue of maximum of the judgment matrix between element set.Be E=(e from the judgment matrix between upper known elements group _mn), its characteristic vector be H=(h (1) ..., h (m) ..., h (K _c)) ^T, wherein

M, n=1 ..., K _c, e _mnBe the matrix element in judgment matrix E, its maximum characteristic root also can calculate according to formula:

E is judgment matrix, and H is matrix characteristic vector, (EH) _mM vector value for gained vector (EH).

All judgment matrixs are done consistency check.The index of conformity of judgment matrix represents with Consistency Ratio CR, and CR=CI/RI, RI are the mean random coincident indicator, and the RI in the same level analytic approach is consistent, and the RI under different matrix dimensions has different definite values, coincident indicator CI=(λ _max-K _i)/(K _i-1).When CR＜0.1, the consistency level that judgment matrix is described is acceptable; Work as CR〉0.1, illustrate that there is self-contradictory situation in the previous judgement of policymaker, or things self has circulation law or symmetry, at this moment need judgment matrix is adjusted again.

Utilize the weight Vector Groups to become not weighting hypermatrix.Suppose element set C _iIn K _iIndividual element is based on element set C _jIn k element c _jkThe normalization characteristic vector of the judgment matrix B that sets up

Obtain, it is expressed as $G_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}={(g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(1\right),...g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(m\right)...,g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(K_i\right))}^T,$ Wherein

Be vector In m vector value, i, j=1,2,3, element c _jkSuccessively from c _j1Arrive

Calculate respectively its characteristic vector

The vectorial permutation that draws is filled into submatrix W _ijIn, namely

W _ijThe sub-block of hypermatrix, if C _iMiddle element is not to C _jMiddle element exerts an influence, W _ij=0.In final network layer element set and all W that between group, element influences each other and consists of _ijMatrix forms not weighting hypermatrix W=(W together _ij), i, j=1,2,3.

Utilize the weight Vector Groups to become weighting matrix.Because weighting matrix W is not the row normalizing, need to multiply by coefficient to guarantee last weighting hypermatrix row normalizing.According to three element set function group C ₁, cost group C ₂With scheme group C ₃Based on one of them element set C _jThe judgment matrix that relatively obtains in twos is

Obtain characteristic vector

After, it is expressed as

a _1j, a _2j, a _3jDifference representation feature vector value, j=1,2,3, three vectorial permutations are filled in weighting matrix A, namely $A=(H^{\left(C_1\right)},H^{\left(C_2\right)},H^{\left(C_3\right)})=\left(a_{\mathrm{ij}}\right),$ i，j=1,2,3；

Calculate the weighting hypermatrix.Weighting hypermatrix M=(M _ij), its submatrix M _ij=a _ij* W _ij, i, j=1,2,3.When the limit hypermatrix of weighting hypermatrix exists, weighting hypermatrix M is carried out power operation S=M 2k+1 time ^2k+1, limit hypermatrix S=MS can appear; Otherwise need adjust judgment matrix and recomputate.Limit of utilization hypermatrix S calculates the ANP weights W of network attribute _anpWeight vectors W with network objectives _net

With benefit type and the cost type respectively normalization of network attribute parameter according to attribute, utilize the method for cost function to calculate the network attribute index according to W at last _anpThe cost function value of each network after assigning weight, replace valency functional value maximum as objective network.

The present invention with the content application of Analytic Network Process method ANP in network selection procedures, consider influencing each other and feedback relationship and impact and the feedback relationship of network objectives on selecting between each property element, set up the multiattribute decision-making of comprehensive evaluation model of each index relation.

The heterogeneous network multiple attributive decision making method of analytic hierarchy process (AHP) Network Based of the present invention specifically comprises:

A. the foundation of the network structure model of network selective system: the network layer inside of whole heterogeneous network selective system is the formed network configuration of interactional element, is divided into function group C with affecting attribute factor that network selects and all factors of objective network ₁, cost group C ₂With scheme group C ₃, C ₁Be divided into available bandwidth c ₁₁With total bandwidth c ₁₂, C ₂Be divided into packet delay c ₂₁, packet jitter c ₂₂, packet loss c ₂₃With price c ₂₄, C ₃The objective network that comprises has UMTS-1, UMTS-2, and WLAN-1, WLAN-2, WiWAX-1 and WiMAX-2 correspond respectively to element c ₃₁, c ₃₂, c ₃₃, c ₃₄, c ₃₅And c ₃₆, C _iRepresent i element set, c _ikK the element that represents i element set, K _iThe element number that represents i element set, i=1,2,3, k=1 ..., K _i, K ₁=2, K ₂=4, K ₃=6, three groups of 12 elements altogether;

B. the foundation of the judgment matrix of element between group in the group: to above three groups, according to interdependence and the feedback relationship of hierarchical structure, utilize analytic hierarchy process (AHP) first with element set C ₁Interior element c ₁₁And c ₁₂Respectively with each element c _ikCompare in twos during as criterion, obtain the judgment matrix of 12 2*2 dimensions, each value reflection element c in matrix ₁₁And c ₁₂For element c _ikSignificance level, set up respectively element set C with analytic hierarchy process (AHP) equally ₂Interior element and C ₃Interior element is respectively with each element c _ikThe judgment matrix that compares in twos during for criterion obtains respectively the judgment matrix of 12 4*4 dimensions and the judgment matrix of 12 6*6 dimensions, supposes above judgment matrix B=(b _mn) expression, b _mnThe relative significance level of relative n the element of m element under certain criterion that represents with digital 1-9 or its inverse;

C. the foundation of the judgment matrix of element set: take objective network as main criterion, equally according to analytic hierarchy process (AHP), respectively with C ₁, C ₂, C ₃Compare in twos as three element set of inferior criterion structure, obtain the judgment matrix of 3 3*3 dimensions, suppose that judgment matrix is expressed as E=(e _mn), e _mnThat relative n the element set of m element set that represent with digital 1-9 or its inverse is at element set C _jRelative significance level under criterion;

D. the judgment matrix of the calculating of the characteristic vector of judgment matrix and characteristic value: step b is B=(b _mn), use the summation method calculate its characteristic vector as G=(g (1) ..., g (m) ..., g (K _i)) ^T, wherein

M, n=1,2 ..., K _i, i, j=1,2,3, its maximum characteristic root is

B is judgment matrix, and G is the judgment matrix characteristic vector, (BG) _mM vector value for gained vector (BG); The Consistency Ratio of judgment matrix B represents with CR, and CR=CI/RI, RI are the mean random coincident indicator, and CI is coincident indicator and CI=(λ _max-K _i)/(K _i-1), if CR〉0.1, need to adjust again to comparing matrix, return to step b; For the judgment matrix E between the element set that draws in step c, method is the same obtains its characteristic vector and eigenvalue of maximum, and does consistency check, if CR〉0.1 item return to step c and adjust matrix;

E. the formation of weighting hypermatrix not: suppose element set C _iIn K _iIndividual element is based on element set C _jIn k element c _jkThe characteristic vector of the judgment matrix B that sets up

Obtain in steps d, i, j=1,2,3, k=1 ..., K _j, it is expressed as $G_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}={(g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(1\right),...g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(m\right)...,g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(K_i\right))}^T,$

The expression vector In m vector value, successively from c _j1Arrive

Calculate respectively its characteristic vector

The vectorial permutation that draws is filled into submatrix W _ijIn, namely

In final network layer element set and all W that between group, element influences each other and consists of _ijMatrix forms not weighting hypermatrix W=(W together _ij), i, j=1,2,3;

F. the formation of weighting matrix: with element set C _jAs inferior criterion to C ₁, C ₂, C ₃Three element set are relatively set up judgment matrix in twos

Its characteristic vector is expressed as by asking with same method in steps d

a _1j, a _2j, a _3jDifference representation feature vector value, j=1,2,3, vectorial permutation is filled in weighting matrix A, $A=(H^{\left(C_1\right)},H^{\left(C_2\right)},H^{\left(C_3\right)})=\left(a_{\mathrm{ij}}\right),$ i，j=1,2,3；

G. the formation of weighting hypermatrix: weighting matrix W not is weighted processes to guarantee the row normalizing, obtain weighting hypermatrix M=(M _ij), its submatrix M _ij=a _ij* W _ij, weighting hypermatrix M is carried out power operation S=M 2k+1 time ^2k+1, until limit hypermatrix S=MS occurs, also calculated characteristics is vectorial again otherwise need to adjust judgment matrix, limit of utilization hypermatrix S calculates the ANP weights W of network attribute _anpWeight vectors W with network objectives _net

H. network is selected: network parameter is distinguished normalization according to benefit type and the cost type of attribute, then calculate the cost function value of each network with the cost function method, to its sequence, replace the network of valency functional value maximum as objective network.

Beneficial effect:

1. when it determines dependence between index by qualitative analysis, have certain subjectivity, but with the network objectives characteristic as Consideration, can estimate preferably objective network.Make decision-making more tally with the actual situation, gained weight result is more reasonable.

2. used the scheme group when hypermatrix in Analytic Network Process method ANP is set up, namely comprised the weight of these 6 network objectives in final ANP weight, can analyze the good and bad situation of roughly performance of these objective networks from the weight result.

3. for the situation of real-time voice business, the average delay, shake that uses the network sequencing selection result of Analytic Network Process method ANP all less than other as SAW, MEW, EW, GRA and TOPSIS algorithm, and the mean handoff number of whole network is minimum, illustrates that the ANP algorithm can select suitable network according to actual conditions preferably.

Description of drawings

Fig. 1 is that whole network switching process adopts the ANP algorithm to realize the schematic flow sheet that network is selected.

Fig. 2 be in the network selective system property element grouping relation and influence each other, the schematic diagram of feedback relationship.

Fig. 3 is the network selection result figure that the ANP Weight algorithm is inscribed in the timesharing of dynamic network selection portion.

Embodiment

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

Thinking of the present invention is that Analytic Network Process method ANP is applied to the weight that solves each property parameters that in the heterogeneous network vertical handover procedure, network is selected, and on this weight basis, network is sorted and selects the network of cost function maximum.

The overview flow chart that whole network switching process adopts the ANP algorithm to realize that network is selected is seen accompanying drawing 1.

One, the network attribute weight of analytic hierarchy process (AHP) ANP Network Based is found the solution

As shown in Figure 2, the heterogeneous network multiple attributive decision making method of analytic hierarchy process (AHP) ANP Network Based is divided into function group C with affecting the attribute factor of network selection and all factors of objective network in heterogeneous network ₁, cost group C ₂With scheme group C ₃, C ₁Be divided into available bandwidth c ₁₁With total bandwidth c ₁₂, C ₂Be divided into packet delay c ₂₁, packet jitter c ₂₂, packet loss c ₂₃With price c ₂₄, C ₃The objective network that comprises has UMTS-1, UMTS-2, and WLAN-1, WLAN-2, WiWAX-1 and WiMAX-2 correspond respectively to element c ₃₁, c ₃₂, c ₃₃, c ₃₄, c ₃₅And c ₃₆, use C _iRepresent i element set, c _ikK the element that represents i element set, K _iRepresent the element number in i element set, i=1,2,3, k=1 ..., K _i, K ₁=2, K ₂=4, K ₃=6, always have 12 property elements.

Take the real-time voice business as scene, build the network hierarchical structure model.At first whole heterogeneous network selective system is divided into two parts, first is key-course, comprises decision objective and decision rule, wherein requires independent between each decision rule and arranged by destination layer.Second portion is network layer, and it is comprised of the element of all controlled preparative layer dominations, and its inside is the formed network configuration of interactional element, namely sets up C according to the place scene ₁Function group, C ₂Cost group and C ₃The structural relation that influences each other and feed back of scheme group, shown in accompanying drawing 2, arrow represents its correlation, do not produce between each objective network in the scheme group influence each other, all have interactional relation in twos between other attributes.

The judgment matrix of element between group in the structure group.In network configuration, if having interdependence and mutual feedback relationship between element, can a given main criterion and one criterion, two elements are carried out the comparison of influence degree under main criterion based on inferior criterion, can indirectly obtain judgment matrix thus.The analytic hierarchy process (AHP) method is as follows: the relation in analytical system between each fundamental, set up the hierarchical structure of whole system; In inferior about last layer to each key element of same level, the importance of a certain criterion compares in twos, Judgement Matricies; Calculated by judgment matrix and be compared key element for the relative weighting of this criterion.The judgment matrix form that obtains in analytic hierarchy process (AHP) is $D=\left[\begin{array}{ccc}{d}_{11}& ...& d_{1t}\\ ...& ...& ...\\ d_{t1}& ...& d_{\mathrm{tt}}\end{array}\right]=\left(d_{\mathrm{mn}}\right),$ M, n=1 ..., t, d _mnBe m Attribute Relative in the significance level of n attribute under a certain criterion, t is the number altogether of the attribute that compares in twos.About the d in the how to confirm judgment matrix _mnValue, Saaty etc. suggestion reference numerals 1-9 and reciprocal as scale represents that with 1 two factors compare and have equal importance, the former is slightly more important than the latter in 3 expressions, the former is obvious more important than the latter in 5 expressions, and the former is strong more important than the latter in 7 expressions, and the former is extremely more important than the latter in 9 expressions, 2, the median of 4,6, the 8 above-mentioned adjacent judgements of expression, its inverse represents that m attribute not as the significance level of n attribute, is 1 on the leading diagonal of matrix.Utilize analytic hierarchy process (AHP) with three groups being divided into interdependence and feedback relationship according to hierarchical structure, first to element set C ₁Interior element c ₁₁And c ₁₂Respectively with each element c _ikCompare in twos during as criterion, obtain the judgment matrix of 12 2*2 dimensions, each value reflection element c in matrix ₁₁And c ₁₂For element c _ikSignificance level, set up respectively element set C with analytic hierarchy process (AHP) equally ₂Interior element and C ₃Interior element is respectively with each element c _ikThe judgment matrix that compares in twos during for criterion obtains respectively the judgment matrix of 12 4*4 dimensions and the judgment matrix of 12 6*6 dimensions, supposes element set C _iIn K _iIndividual element is based on element set C _jIn the judgment matrix set up of k element be expressed as B=(b _mn), m, n=1,2 ..., K _i, b _mnThe significance level of m relative n the element of element in i group when reflection is organized k element as criterion take j.

The judgment matrix of organizing element between interior group is asked characteristic vector and eigenvalue of maximum.The judgment matrix of gained is B=(b as mentioned above _mn), its characteristic vector be G=(g (1) ..., g (m) ..., g (K _i)) ^T, wherein

$g\left(m\right)=\frac{\underset{n=1}{\overset{K_i}{\mathrm{\Σ}}}(b_{\mathrm{mn}}/\underset{m=1}{\overset{K_i}{\mathrm{\Σ}}}{b}_{\mathrm{mn}})}{K_i},$ m，n=1,2，...,K _i（1）

b _mnBe the matrix element in judgment matrix.When the dimension of judgment matrix B of structure is K _i2 o'clock, judgment matrix not necessarily satisfies consistency, its maximum characteristic root λ _maxAlso just no longer equal matrix dimension K _i, need to calculate maximum characteristic root by formula:

${\mathrm{\λ}}_{\max }=\frac{1}{K_i}\underset{m=1}{\overset{K_i}{\mathrm{\Σ}}}\frac{{\left(\mathrm{BG}\right)}_m}{g\left(m\right)}---\left(2\right)$

B is judgment matrix, and G is matrix characteristic vector, (BG) _mM vector value for gained vector (BG).

Build the judgment matrix between element set.Take objective network as main criterion, use respectively analytic hierarchy process (AHP) with element set C ₁, C ₂, C ₃As inferior criterion to C ₁, C ₂, C ₃Three element set compare in twos, obtain the judgment matrix of 3 3*3 dimensions.Suppose that judgment matrix is E=(e _mn), e _mnAlso to represent based on element set C with 1-9 or its inverse _jThe relative significance level of m relative n the element set of element set, j=1,2,3, m, n=1 ..., K _c, K _cRepresentative element group number and K _c=3, criterion C _jThere are 3, therefore the judgment matrix of 3 such forms is arranged.

Calculate characteristic vector and the eigenvalue of maximum of the judgment matrix between element set.Judgment matrix between element set is E=(e _mn), its characteristic vector be H=(h (1) ..., h (m) ..., h (K _c)) ^T, wherein

$h\left(m\right)=\frac{\underset{n=1}{\overset{K_c}{\mathrm{\Σ}}}(e_{\mathrm{mn}}/\underset{m=1}{\overset{K_c}{\mathrm{\Σ}}}{e}_{\mathrm{mn}})}{K_c},$ m，n=1,2，...,Kc （3）

Its maximum characteristic root also can calculate according to formula:

${\mathrm{\λ}}_{\max }=\frac{1}{K_c}\underset{m=1}{\overset{K_c}{\mathrm{\Σ}}}\frac{{\left(\mathrm{EH}\right)}_m}{h\left(m\right)}---\left(4\right)$

E is judgment matrix, and H is matrix characteristic vector, (EH) _mM vector value for gained vector (EH).

All judgment matrixs are done consistency check.The index of conformity of judgment matrix represents with Consistency Ratio CR:

CR＝CI/RI （5

RI is the mean random coincident indicator, and the RI in the same level analytic approach is consistent, and the RI of different matrix dimensions has different definite values, and the computing formula of coincident indicator CI is:

CI＝(λ _max-K _i)/(K _i-1) （6）

When CR＜0.1, the consistency level that judgment matrix is described is acceptable; Work as CR〉0.1, illustrate that there is self-contradictory situation in the previous judgement of policymaker, or things self has circulation law or symmetry, at this moment need judgment matrix is adjusted.

Utilize the weight Vector Groups to become not weighting hypermatrix.Suppose element set C _iIn K _iIndividual element is based on element set C _jIn k element c _jkThe characteristic vector of the judgment matrix B that sets up Obtain, it is expressed as $G_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}={(g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(1\right),...g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(m\right)...,g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(K_i\right))}^T,$

The expression vector In m vector value, i, j=1,2,3, element c _jkSuccessively from c _j1Arrive

Calculate respectively its characteristic vector

The vectorial permutation that draws is filled into submatrix W _ijIn, namely

W _ijThe sub-block of hypermatrix, if C _iMiddle element is not to C _jMiddle element exerts an influence, W _ij=0.In final network layer element set and all W that between group, element influences each other and consists of _ijMatrix forms not weighting hypermatrix W=(W together _ij), i, j=1,2,3.These 9 sub-matrix W _ijBe constructed as follows:

$W_{11}=\left[\begin{array}{cc}{g}_{11}^{\left(c_{11}\right)}\left(1\right)& g_{11}^{\left(c_{12}\right)}\left(2\right)\\ g_{11}^{\left(c_{11}\right)}\left(2\right)& g_{11}^{\left(c_{12}\right)}\left(2\right)\end{array}\right]---\left(7\right)$

$W_{12}=\left[\begin{array}{cccc}{g}_{12}^{\left(c_{21}\right)}\left(1\right)& g_{12}^{\left(c_{22}\right)}\left(1\right)& g_{12}^{\left(c_{23}\right)}\left(1\right)& g_{12}^{\left(c_{24}\right)}\left(1\right)\\ g_{12}^{\left(c_{21}\right)}\left(2\right)& g_{12}^{\left(c_{22}\right)}\left(2\right)& g_{12}^{\left(c_{23}\right)}\left(2\right)& g_{12}^{\left(c_{24}\right)}\left(2\right)\end{array}\right]---\left(8\right)$

$W_{13}=\left[\begin{array}{cccccc}{g}_{13}^{\left(c_{31}\right)}\left(1\right)& g_{13}^{\left(c_{32}\right)}\left(1\right)& g_{13}^{\left(c_{33}\right)}\left(1\right)& g_{13}^{\left(c_{34}\right)}\left(1\right)& g_{13}^{\left(c_{35}\right)}\left(1\right)& g_{13}^{\left(c_{36}\right)}\left(1\right)\\ g_{13}^{\left(c_{31}\right)}\left(2\right)& g_{13}^{\left(c_{32}\right)}\left(2\right)& g_{13}^{\left(c_{33}\right)}\left(2\right)& g_{13}^{\left(c_{34}\right)}\left(2\right)& g_{13}^{\left(c_{35}\right)}\left(2\right)& g_{13}^{\left(c_{36}\right)}\left(2\right)\end{array}\right]---\left(9\right)$

$W_{21}=\left[\begin{array}{cc}{g}_{21}^{\left(c_{11}\right)}\left(1\right)& g_{21}^{\left(c_{12}\right)}\left(1\right)\\ g_{21}^{\left(c_{11}\right)}\left(2\right)& g_{21}^{\left(c_{12}\right)}\left(2\right)\\ g_{21}^{\left(c_{11}\right)}\left(3\right)& g_{21}^{\left(c_{12}\right)}\left(3\right)\\ g_{21}^{\left(c_{11}\right)}\left(4\right)& g_{21}^{\left(c_{12}\right)}\left(4\right)\end{array}\right]---\left(10\right)$

$W_{22}=\left[\begin{array}{cccc}{g}_{22}^{\left(c_{21}\right)}\left(1\right)& g_{22}^{\left(c_{22}\right)}\left(1\right)& g_{22}^{\left(c_{23}\right)}\left(1\right)& g_{22}^{\left(c_{24}\right)}\left(1\right)\\ g_{22}^{\left(c_{21}\right)}\left(2\right)& g_{22}^{\left(c_{22}\right)}\left(2\right)& g_{22}^{\left(c_{23}\right)}\left(2\right)& g_{22}^{\left(c_{24}\right)}\left(2\right)\\ g_{22}^{\left(c_{21}\right)}\left(3\right)& g_{22}^{\left(c_{22}\right)}\left(3\right)& g_{22}^{\left(c_{23}\right)}\left(3\right)& g_{22}^{\left(c_{24}\right)}\left(3\right)\\ g_{22}^{\left(c_{21}\right)}\left(4\right)& g_{22}^{\left(c_{22}\right)}\left(4\right)& g_{22}^{\left(c_{23}\right)}\left(4\right)& g_{22}^{\left(c_{24}\right)}\left(4\right)\end{array}\right]---\left(11\right)$

$W_{23}=\left[\begin{array}{cccccc}{g}_{23}^{\left(c_{31}\right)}\left(1\right)& g_{23}^{\left(c_{32}\right)}\left(1\right)& g_{23}^{\left(c_{33}\right)}\left(1\right)& g_{23}^{\left(c_{34}\right)}\left(1\right)& g_{23}^{\left(c_{35}\right)}\left(1\right)& g_{23}^{\left(c_{36}\right)}\left(1\right)\\ g_{23}^{\left(c_{31}\right)}\left(2\right)& g_{23}^{\left(c_{32}\right)}\left(2\right)& g_{23}^{\left(c_{33}\right)}\left(2\right)& g_{23}^{\left(c_{34}\right)}\left(2\right)& g_{23}^{\left(c_{35}\right)}\left(2\right)& g_{23}^{\left(c_{36}\right)}\left(2\right)\\ g_{23}^{\left(c_{31}\right)}\left(3\right)& g_{23}^{\left(c_{32}\right)}\left(3\right)& g_{23}^{\left(c_{33}\right)}\left(3\right)& g_{23}^{\left(c_{34}\right)}\left(3\right)& g_{23}^{\left(c_{35}\right)}\left(3\right)& g_{23}^{\left(c_{36}\right)}\left(3\right)\\ g_{23}^{\left(c_{31}\right)}\left(4\right)& g_{23}^{\left(c_{32}\right)}\left(4\right)& g_{23}^{\left(c_{33}\right)}\left(4\right)& g_{23}^{\left(c_{34}\right)}\left(4\right)& g_{23}^{\left(c_{35}\right)}\left(4\right)& g_{23}^{\left(c_{36}\right)}\left(4\right)\end{array}\right]---\left(12\right)$

$W_{31}=\left[\begin{array}{cc}{g}_{31}^{\left(c_{11}\right)}\left(1\right)& g_{31}^{\left(c_{12}\right)}\left(1\right)\\ g_{31}^{\left(c_{11}\right)}\left(2\right)& g_{31}^{\left(c_{12}\right)}\left(2\right)\\ g_{31}^{\left(c_{11}\right)}\left(3\right)& g_{31}^{\left(c_{12}\right)}\left(3\right)\\ g_{31}^{\left(c_{11}\right)}\left(4\right)& g_{31}^{\left(c_{12}\right)}\left(4\right)\\ g_{31}^{\left(c_{11}\right)}\left(5\right)& g_{31}^{\left(c_{12}\right)}\left(5\right)\\ g_{31}^{\left(c_{11}\right)}\left(6\right)& g_{31}^{\left(c_{12}\right)}\left(6\right)\end{array}\right]---\left(13\right)$

$W_{32}=\left[\begin{array}{cccc}{g}_{32}^{\left(c_{21}\right)}\left(1\right)& g_{32}^{\left(c_{22}\right)}\left(1\right)& g_{32}^{\left(c_{23}\right)}\left(1\right)& g_{32}^{\left(c_{24}\right)}\left(1\right)\\ g_{32}^{\left(c_{21}\right)}\left(2\right)& g_{32}^{\left(c_{22}\right)}\left(2\right)& g_{32}^{\left(c_{23}\right)}\left(2\right)& g_{32}^{\left(c_{24}\right)}\left(2\right)\\ g_{32}^{\left(c_{21}\right)}\left(3\right)& g_{32}^{\left(c_{22}\right)}\left(3\right)& g_{32}^{\left(c_{23}\right)}\left(3\right)& g_{32}^{\left(c_{24}\right)}\left(3\right)\\ g_{32}^{\left(c_{21}\right)}\left(4\right)& g_{32}^{\left(c_{22}\right)}\left(4\right)& g_{32}^{\left(c_{23}\right)}\left(4\right)& g_{32}^{\left(c_{24}\right)}\left(4\right)\\ g_{32}^{\left(c_{21}\right)}\left(5\right)& g_{32}^{\left(c_{22}\right)}\left(5\right)& g_{32}^{\left(c_{23}\right)}\left(5\right)& g_{32}^{\left(c_{24}\right)}\left(5\right)\\ g_{32}^{\left(c_{21}\right)}\left(6\right)& g_{32}^{\left(c_{22}\right)}\left(6\right)& g_{32}^{\left(c_{23}\right)}\left(6\right)& g_{32}^{\left(c_{24}\right)}\left(6\right)\end{array}\right]---\left(14\right)$

$W_{33}=\left[\begin{array}{cccccc}{g}_{33}^{\left(c_{31}\right)}\left(1\right)& g_{33}^{\left(c_{32}\right)}\left(1\right)& g_{33}^{\left(c_{33}\right)}\left(1\right)& g_{33}^{\left(c_{34}\right)}\left(1\right)& g_{33}^{\left(c_{35}\right)}\left(1\right)& g_{33}^{\left(c_{36}\right)}\left(1\right)\\ g_{33}^{\left(c_{31}\right)}\left(2\right)& g_{33}^{\left(c_{32}\right)}\left(2\right)& g_{33}^{\left(c_{33}\right)}\left(2\right)& g_{33}^{\left(c_{34}\right)}\left(2\right)& g_{33}^{\left(c_{35}\right)}\left(2\right)& g_{33}^{\left(c_{36}\right)}\left(2\right)\\ g_{33}^{\left(c_{31}\right)}\left(3\right)& g_{33}^{\left(c_{32}\right)}\left(3\right)& g_{33}^{\left(c_{33}\right)}\left(3\right)& g_{33}^{\left(c_{34}\right)}\left(3\right)& g_{33}^{\left(c_{35}\right)}\left(3\right)& g_{33}^{\left(c_{36}\right)}\left(3\right)\\ g_{33}^{\left(c_{31}\right)}\left(4\right)& g_{33}^{\left(c_{32}\right)}\left(4\right)& g_{33}^{\left(c_{33}\right)}\left(4\right)& g_{33}^{\left(c_{34}\right)}\left(4\right)& g_{33}^{\left(c_{35}\right)}\left(4\right)& g_{33}^{\left(c_{36}\right)}\left(4\right)\\ g_{33}^{\left(c_{31}\right)}\left(5\right)& g_{33}^{\left(c_{32}\right)}\left(5\right)& g_{33}^{\left(c_{33}\right)}\left(5\right)& g_{33}^{\left(c_{34}\right)}\left(5\right)& g_{33}^{\left(c_{35}\right)}\left(5\right)& g_{33}^{\left(c_{36}\right)}\left(5\right)\\ g_{33}^{\left(c_{31}\right)}\left(6\right)& g_{33}^{\left(c_{32}\right)}\left(6\right)& g_{33}^{\left(c_{33}\right)}\left(6\right)& g_{33}^{\left(c_{34}\right)}\left(6\right)& g_{33}^{\left(c_{35}\right)}\left(6\right)& g_{33}^{\left(c_{36}\right)}\left(6\right)\end{array}\right]---\left(15\right)$

Utilize the weight Vector Groups to become weighting matrix.According to three element set C ₁, C ₂And C ₃Based on one of them element set C _jThe judgment matrix that work compares in twos

Obtain characteristic vector After, it is expressed as

a _1j, a _2j, a _3jDifference representation feature vector value, j=1,2,3, three vectorial permutations are filled in weighting matrix A, namely

$A=(H^{\left(C_1\right)},H^{\left(C_2\right)},H^{\left(C_3\right)})=\left[\begin{array}{ccc}{a}_{11}& a_{12}& a_{13}\\ a_{21}& a_{22}& a_{23}\\ a_{31}& a_{32}& a_{33}\end{array}\right]=\left(a_{\mathrm{ij}}\right),$ i，j=1,2,3 （16）

Because the not weighting matrix W that draws is not the row normalizing, need to be with the weighting matrix coefficient of corresponding sub-block Matrix Multiplication with correspondence, to guarantee last hypermatrix row normalizing.

Final weighting hypermatrix M=(M _ij), its submatrix:

M _ij=a _ij*W _ij，i，j=1,2,3（17）

Wherein aij is take the key-course element as criterion, and the importance of 3 element set of network layer is compared and value in the ordering vector that draws.If the limit hypermatrix of weighting hypermatrix exists, the weighting hypermatrix is carried out power operation, i.e. S=M2 2k+1 time ^k+1, limit hypermatrix S=MS can appear; Otherwise needing to adjust matrix recomputates.The limit of utilization hypermatrix calculates the ANP weights W of network attribute _anpWeight vectors W with network objectives _netW _anpRepresented the weight that in the network, each attribute is assigned with, W _netWhat represent is the weight of each objective network, and its network in general performance of the larger expression of weight is better, and selecteed chance is also larger in network selection procedures.Find W according to final result _netC in vector ₃₅And c ₃₆Weight maximum, namely the network in general performance of WiWAX-1 and WiMAX-2 is more excellent.

Two, the network based on the ANP weight sorts and selects

Each attribute of each network has corresponding parameter value or codomain when network is selected, and codomain is divided into 20 sections, and between the setting district, segment number is 1 to 21.Get at random a certain section when network is initial as the initial network parameter from 20 sections, the state variation of hypothesis network is followed markoff process in the time backward, and namely sometime network attribute parameter value is relevant with the value of previous moment.Make that P=0.4 is the probability that network parameter changes, P/2=0.2 is respectively the probability that transfers the last period and rear a section to, and 1-P=0.6 is the probability that network parameter remains unchanged, and can set up emulation according to the scene of real-time voice business.Be available bandwidth according to each attribute of gained, total bandwidth, packet delay, packet jitter, bag packet loss and required weights W corresponding to price _anp, calculate respectively the cost function value of each network, to its sequence and selection.

In considering multiple attribute and carrying out the process of decision-making, due to the difference of describing mode, dimension and value characteristics between attribute, the decision matrix that at first need to consist of for multiple decision attribute and value thereof adopts diverse ways that it is standardized.For the decision attribute of Real-valued, will carry out respectively normalization according to benefit type or cost type property parameters.

For N attribute of the network of the L in the network selective system, suppose x _pqRepresent p network about the property value of q attribute, decision matrix is X=(x _pq), p=1 wherein ..., L, q=1 ..., N.For benefit type real number attribute, can adopt formula (18) to carry out normalization:

$r_{\mathrm{pq}}=\frac{x_{\mathrm{pq}}-\min \left(x_{\mathrm{pq}}\right)}{\max \left(x_{\mathrm{pq}}\right)-\min \left(x_{\mathrm{pq}}\right)}---\left(18\right)$

For cost type real number attribute, can adopt (19) formula to carry out normalization:

$r_{\mathrm{pq}}=\frac{\max \left(x_{\mathrm{pq}}\right)-x_{\mathrm{pq}}}{\max \left(x_{\mathrm{pq}}\right)-\min \left(x_{\mathrm{pq}}\right)}---\left(19\right)$

The multiple attributive decision making method that is based on the cost function method that mainly uses at present, in L network, the computing formula of obtaining maximum cost function value is as follows:

$f({\mathrm{<figure-callout\; id="1"\; label="r\; p"\; filenames="121120174138.png"\; state="\{\{state\}\}">r</figure-callout>}}_p,{\mathrm{<figure-callout\; id="2"\; label="r\; p"\; filenames="121120174138.png"\; state="\{\{state\}\}">r</figure-callout>}}_p,...,r_{\mathrm{pN}})=\arg \underset{p\&Element;L}{\max }\underset{q=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_q{r}_{\mathrm{pq}}---\left(20\right)$

W wherein _qANP attribute weight vector W _anpIn the weighted value of q attribute.

Calculate each cost function value to its sequencing selection maximum as objective network, complete primary network and select, in the emulation mapping, the objective network of ordinate represents respectively network c with sequence number 1 to 6 ₃₁To c ₃₆Inscribe network selection result figure such as the accompanying drawing 3 of ANP algorithm during part, find to select the number of times of WiWAX-1 and WiMAX-2 many, the weights W that visible ANP algorithm is tried to achieve _netReally can reflect the overall performance of each objective network.

In sum, Analytic Network Process method ANP is applied to solve in the heterogeneous network vertical handover procedure, determine dependence between index simultaneously the network plan characteristic to be taken into account influencing factor by qualitative analysis, estimated preferably network from subjectivity and objectivity, decision-making more tallies with the actual situation.The weight that comprises 6 network objectives in the weight that Analytic Network Process method ANP draws can roughly be analyzed the good and bad situation of these networks when selected from weight.Situation for the real-time voice business, the average delay of the network sequencing selection result of utilization Analytic Network Process method ANP, shake are all less than other algorithms, the mean handoff number of whole network is minimum, illustrates that the ANP algorithm can select suitable network according to scene preferably.

Below carry out each parameter average result table after network is selected for each algorithm:

Claims (1)

1. the heterogeneous network multiple attributive decision making method of an analytic hierarchy process (AHP) Network Based is characterized in that the method comprises the following steps:

A. the foundation of the network structure model of network selective system: the network layer inside of whole heterogeneous network selective system is the formed network configuration of interactional element, is divided into function group C with affecting attribute factor that network selects and all factors of objective network ₁, cost group C ₂With scheme group C ₃, C ₁Be divided into available bandwidth c ₁₁With total bandwidth c ₁₂, C ₂Be divided into packet delay c ₂₁, packet jitter c ₂₂, packet loss c ₂₃With price c ₂₄, C ₃The objective network that comprises has UMTS-1, UMTS-2, and WLAN-1, WLAN-2, WiWAX-1 and WiMAX-2 correspond respectively to element c ₃₁, c ₃₂, c ₃₃, c ₃₄, c ₃₅And c ₃₆, C _iRepresent i element set, c _ikK the element that represents i element set, K _iThe element number that represents i element set, i=1,2,3, k=1 ..., K _i, K ₁=2, K ₂=4, K ₃=6, three groups of 12 elements altogether;

B. the foundation of the judgment matrix of element between group in the group: to above three groups, according to interdependence and the feedback relationship of hierarchical structure, utilize analytic hierarchy process (AHP) first with element set C ₁Interior element c ₁₁And c ₁₂Respectively with each element c _ikCompare in twos during as criterion, obtain the judgment matrix of 12 2*2 dimensions, each value reflection element c in matrix ₁₁And c ₁₂For element c _ikSignificance level, set up respectively element set C with analytic hierarchy process (AHP) equally ₂Interior element and C ₃Interior element is respectively with each element c _ikThe judgment matrix that compares in twos during for criterion obtains respectively the judgment matrix of 12 4*4 dimensions and the judgment matrix of 12 6*6 dimensions, supposes above judgment matrix B=(b _mn) expression, b _mnThe relative significance level of relative n the element of m element under certain criterion that represents with digital 1-9 or its inverse;

C. the foundation of the judgment matrix of element set: take objective network as main criterion, equally according to analytic hierarchy process (AHP), respectively with C ₁, C ₂, C ₃Compare in twos as three element set of inferior criterion structure, obtain the judgment matrix of 3 3*3 dimensions, suppose that judgment matrix is expressed as E=(e _mn), e _mnThat relative n the element set of m element set that represent with digital 1-9 or its inverse is at element set C _jRelative significance level under criterion;

D. the judgment matrix of the calculating of the characteristic vector of judgment matrix and characteristic value: step b is B=(b _mn), use the summation method calculate its characteristic vector as G=(g (1) ..., g (m) ..., g (K _i)) ^T, wherein

M, n=1,2 ..., K _i, i, j=1,2,3, its maximum characteristic root is

B is judgment matrix, and G is the judgment matrix characteristic vector, (BG) _mM vector value for gained vector (BG); The Consistency Ratio of judgment matrix B represents with CR, and CR=CI/RI, RI are the mean random coincident indicator, and CI is coincident indicator and CI=(λ _max-K _i)/(K _i-1), if CR〉0.1, need to adjust again to comparing matrix, return to step b; For the judgment matrix E between the element set that draws in step c, method is the same obtains its characteristic vector and eigenvalue of maximum, and does consistency check, if CR〉0.1 item return to step c and adjust matrix;

E. the formation of weighting hypermatrix not: suppose element set C _iIn K _iIndividual element is based on element set C _jIn k element c _jkThe characteristic vector of the judgment matrix B that sets up

Obtain in steps d, i, j=1,2,3, k=1 ..., K _j, it is expressed as $G_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}={(g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(1\right),...g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(m\right)...,g_{\mathrm{ij}}^{\left(c_{\mathrm{jk}}\right)}\left(K_i\right))}^T,$

The expression vector

In m vector value, successively from c _j1Arrive

Calculate respectively its characteristic vector

The vectorial permutation that draws is filled into submatrix W _ijIn, namely In final network layer element set and all W that between group, element influences each other and consists of _ijMatrix forms not weighting hypermatrix W=(W together _ij), i, j=1,2,3;

F. the formation of weighting matrix: with element set C _jAs inferior criterion to C ₁, C ₂, C ₃Three element set are relatively set up judgment matrix in twos

Its characteristic vector is expressed as by asking with same method in steps d

a _1j, a _2j, a _3jDifference representation feature vector value, j=1,2,3, vectorial permutation is filled in weighting matrix A, $A=(H^{\left(C_1\right)},H^{\left(C_2\right)},H^{\left(C_3\right)})=\left(a_{\mathrm{ij}}\right),$ i，j=1,2,3；

G. the formation of weighting hypermatrix: weighting matrix W not is weighted processes to guarantee the row normalizing, obtain weighting hypermatrix M=(M _ij), its submatrix M _ij=a _ij* W _ij, weighting hypermatrix M is carried out power operation S=M 2k+1 time ^2k+1, until limit hypermatrix S=MS occurs, also calculated characteristics is vectorial again otherwise need to adjust judgment matrix, limit of utilization hypermatrix S calculates the ANP weights W of network attribute _anpWeight vectors W with network objectives _net

H. network is selected: network parameter is distinguished normalization according to benefit type and the cost type of attribute, then calculate the cost function value of each network with the cost function method, to its sequence, replace the network of valency functional value maximum as objective network.

Images