CN104077485A - Model correctness evaluation method based on goodness of fit - Google Patents

Abstract

The invention discloses a model correctness evaluation method based on goodness of fit. The model correctness evaluation method is characterized by comprising the following steps of 1, establishing the hierarchical structure of a model, 2, calculating the individual goodness of fit of each parameter of a child node of the bottommost layer of the hierarchical structure, 3, calculating the weight of each parameter of each child node relative to the child node, 4, calculating the goodness of fit of each child node, 5, calculating the weight of each parameter of each child node relative to a target node, and 6, calculating the goodness of fit of the target node. The model correctness evaluation method based on goodness of fit is capable of evaluating the correctness degree of the model as long as the input and output parameters of the model are known and in case that the output parameters are measureable; the method does not need to find out complex principle and the inherent working mode of the model, and therefore, the method has remarkable advantages.

Description

The evaluation method of the Correctness of model based on the goodness of fit

Technical field

The present invention relates to a kind of evaluation method of Correctness of model, be specifically related to a kind of evaluation method of the Correctness of model based on the goodness of fit, belong to model evaluation method and technology field.

Background technology

The correctness of model directly affects the performance quality of Related product, so the correctness evaluation of model is significant.

Taking skin antenna power electric coupling model as example:

Skin antenna refer to by the radio-frequency enabled part of integrated micro-strip antenna array be embedded in Weapons platform construction, by utilizing the Active Arrays of the Highgrade integration that advanced composite moulding process manufactures, it both can be used as the mechanics bearing function part of Weapons platform construction, also can be used as the electromagnetic function solenoid part of transmitting/receiving wireless.

With traditional antenna contrast, skin antenna has the advantages that the height of structure/circuit merges, and makes antenna become part or all of Weapons platform construction, has reduced antenna weight and space hold rate.Skin antenna can be widely applied in the fields such as opportunity of combat of new generation, unmanned plane, early warning dirigible, intelligent battlebus, stealthy battleship.

The unit for electrical property parameters that affects skin antenna performance mainly contains gain, beam angle, standing-wave ratio (SWR) and secondary lobe, and mechanical property parameters mainly contains stress, strain and displacement.

Skin antenna power electric coupling model representation skin antenna antenna electric performance and mechanical property situation over time under different load.The correctness of this model directly affects the performance quality of skin antenna, so the correctness evaluation of skin antenna power electric coupling model is significant.

But, the method of existing evaluation model correctness is all to adopt specific method for specific model, and must know principle of work and the inherent working method thereof of model complexity, evaluation method is numerous, understanding is difficult, do not have a kind of unification to meeting all evaluation methods of adoptable correctness of model of certain condition.

Summary of the invention

For solving the deficiencies in the prior art, the object of the present invention is to provide the method for the correctness of the known and model that output parameter can be surveyed of a kind of application evaluation input/output argument wide, practical, based on the goodness of fit.

In order to realize above-mentioned target, the present invention adopts following technical scheme:

An evaluation method for Correctness of model based on the goodness of fit, is characterized in that, comprises the following steps:

(1), set up the Recurison order hierarchy structure of model:

Degree of membership relation according to each factor is set up Recurison order hierarchy structure, is divided into top level goal node, middle layer child node, three levels of the sub-child node of bottom, and upper layer node has dominating role to lower level node, and same layer factor is separate;

(2), calculate the individual event goodness of fit of the each parameter of the Recurison order hierarchy child node of the structure bottom:

(1), in the case of the size of model and input parameter known, calculate or emulation obtains the calculated value of the output parameter of model;

(2), record by experiment the actual test value of the output parameter that under identical input, this model is corresponding;

(3), the relative error between calculated value and the actual test value of calculating output parameter;

(4), relative error is mapped as to the goodness of fit by type gauss of distribution function less than normal;

(3), calculate the weight of the each parameter of sub-child node with respect to child node;

(4), calculate the goodness of fit of each child node:

Obtain the goodness of fit of each child node according to the weight calculation of the relative child node of the goodness of fit of the each parameter of sub-child node and each parameter;

(5), calculate the weight of the each parameter of child node with respect to destination node;

(6), calculate the goodness of fit of destination node:

Obtain the goodness of fit of destination node according to the weight calculation of the relative destination node of the goodness of fit of the each parameter of child node and each parameter.

The evaluation method of the aforesaid Correctness of model based on the goodness of fit, is characterized in that, in step (three), calculates the each parameter of sub-child node and comprises following sub-step with respect to the weight of child node:

(1), Judgement Matricies:

In the Recurison order hierarchy structure of model, the each element of child node is as first element of judgment matrix, the element that is under the jurisdiction of its each sub-child node is sequentially arranged in the first row and the 1st row thereafter, for the criterion of judgment matrix, the importance degree of each element is pressed to 1-9 assignment;

(2), Consistency Check in Judgement Matrix;

(3), calculate weight vectors:

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, obtain the relative weighting of each index.

The evaluation method of the aforesaid Correctness of model based on the goodness of fit, is characterized in that, the step of Consistency Check in Judgement Matrix is:

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix;

B, by obtain coincident indicator C.I.;

C, table look-up determine corresponding mean random coincident indicator R.I.;

D, calculating consistance ratio C.R. also judge:

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix.

The evaluation method of the aforesaid Correctness of model based on the goodness of fit, is characterized in that, in step (five), calculates the each parameter of child node and comprises following sub-step with respect to the weight of destination node:

(1), Judgement Matricies:

In the Recurison order hierarchy structure of model, destination node element is as first element of judgment matrix, the element that is under the jurisdiction of its each child node is sequentially arranged in the first row and the 1st row thereafter, for the criterion of judgment matrix, the importance degree of each element is pressed to 1-9 assignment;

(2), Consistency Check in Judgement Matrix:

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix;

B, by obtain coincident indicator C.I.;

C, table look-up determine corresponding mean random coincident indicator R.I.;

D, calculating consistance ratio C.R. also judge:

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix;

(3), calculate weight vectors:

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, obtain the relative weighting of each index.

The evaluation method of the aforesaid Correctness of model based on the goodness of fit, it is characterized in that, in step (), by the goodness of fit of each output parameter of known models as the sub-child node of bottom, as middle layer child node, use total goodness of fit of the model that can embody Correctness of model as top level goal node by the goodness of fit of all kinds of integrated performance indexs under output parameter.

Usefulness of the present invention is: the input/output argument that only need know model, and in the detectable situation of output parameter, the Correctness of model evaluation method based on the goodness of fit that adopts the present invention to propose can be evaluated the correctness degree of model, the method does not need to know principle and the inherent working method thereof of model complexity, and therefore method of the present invention has significant advantage.

Brief description of the drawings

Fig. 1 is the process flow diagram that calculates the goodness of fit;

Fig. 2 is the process flow diagram of evaluating the correctness of skin antenna electromechanical Coupling Model;

Fig. 3 is the appraisement system figure of skin antenna electromechanical Coupling Model;

Fig. 4 is the evaluation model figure of skin antenna electromechanical Coupling Model;

Fig. 5 is the curve map that calculates the type gauss of distribution function less than normal of the single index goodness of fit;

Fig. 6 is the hierarchy structure chart of weight calculation.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is done to concrete introduction.

With reference to Fig. 2, evaluation method of the present invention is applicable to the model that input/output argument is known and output parameter can be surveyed, and its principle is as follows:

Thinking of the present invention is in the case of the input of known models, calculating or emulation obtain the calculated value of corresponding output parameter, experiment records the actual test value of the output parameter that under identical input, this model is corresponding, by the relative error between calculated value and the actual test value of calculating output parameter, and this relative error is mapped as to certain score by type gauss of distribution function less than normal, the number of this score just embodies the size of the goodness of fit, it is the correctness degree that the size of the goodness of fit has embodied model, so can verify the correctness of model by the goodness of fit.

Definition to the goodness of fit: considering in each index key factor situation the tolerance of calculated value and actual test value degree of closeness.

The calculation process of the goodness of fit as shown in Figure 1, first, is measured correlation parameter by exemplar/model machine, obtains the actual test value of parameter; Then, record the calculated value of corresponding parameter by known models (under identical input); Finally, calculate the relative error between parameter theory calculated value and actual test value, and this relative error is obtained to corresponding scoring by certain mapping function, after comprehensively, obtain the comprehensive evaluation of calculated value and actual test value degree of agreement.

Understand evaluation method of the present invention for convenience of those skilled in the art, existing known taking input/output argument and skin antenna power electric coupling model that output parameter can be surveyed describes in detail as example.

One, set up the Recurison order hierarchy structure of model

Degree of membership relation according to each factor is set up Recurison order hierarchy structure, is divided into top level goal node, middle layer child node, three levels of the sub-child node of bottom.Simple understand be exactly, the factors of same layer be subordinated to the factor of last layer or on upper strata because of have impact, the factor of one deck or be subject to the effect of lower floor's factor under simultaneously arranging again.

Concrete, the goodness of fit of each output parameter of use skin antenna electromechanical Coupling Model is as the sub-child node of bottom, and the mechanical performance index that for example affects antenna power electric coupling performance comprises stress, strain, displacement, at stress goodness of fit x for sub-child node ₁₁, strain goodness of fit x ₁₂, displacement goodness of fit x ₁₃represent respectively the individual event goodness of fit of stress, strain and displacement, the electrical characteristics index that affects antenna power electric coupling performance comprises gain, beam angle, standing-wave ratio (SWR), secondary lobe, at gain goodness of fit x for sub-child node ₂₁, beam angle goodness of fit x ₂₂, standing-wave ratio (SWR) goodness of fit x ₂₃with secondary lobe goodness of fit x ₂₄represent respectively the individual event goodness of fit of gain, beam angle, standing-wave ratio (SWR), secondary lobe; By the goodness of fit of all kinds of integrated performance indexs under output parameter as middle layer child node, such as the integrated performance index that affects destination layer skin antenna power electric coupling performance comprises mechanical property and electrical performance indexes, the therefore goodness of fit X with mechanical characteristic in child node ₁goodness of fit X with electrical characteristics ₂represent respectively the goodness of fit of child node; With total goodness of fit D of model that can embody Correctness of model as top level goal node.

The Recurison order hierarchy structure of so just having set up skin antenna power electric coupling model, in this Recurison order hierarchy structure, upper layer node has dominating role to lower level node, and same layer factor is separate.As shown in Figure 4.Wherein:

$D=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{W}_i{X}_i,X_1=\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_{1i}{x}_{1i},X_2=\underset{j=1}{\overset{4}{\mathrm{\&Sigma;}}}{w}_{2j}{x}_{2j}$

W _irepresent the weight of child node i with respect to top mode,

W _1irepresent the weight of sub-child node i with respect to child node 1,

W _2jrepresent the weight of sub-child node j with respect to child node 2.

So after learning the goodness of fit and the weight (importance degree) of every individual sub-child node with respect to middle layer child node of each sub-child node of bottom, by identical computing formula, just can obtain the goodness of fit of each child node of middle layer, after the goodness of fit of the each child node in known middle layer and the weight of each child node with respect to destination layer node thereof, just the total goodness of fit of destination layer node can be obtained, the correctness degree of known skin antenna power electric coupling model can be verified by the stool and urine of total goodness of fit.

The appraisement system of skin antenna power electric coupling model as Fig. 3, evaluation model as shown in Figure 4.

Two, calculate the individual event goodness of fit of the each parameter of the Recurison order hierarchy child node of the structure bottom

1, in the case of the size of model and input parameter known, calculate or emulation obtains the calculated value of the output parameter of model.

3 parameters (stress, strain and displacement) of the sub-child node comprising for child node 1,4 parameters (gain, beam angle, standing-wave ratio (SWR) and secondary lobe) of the sub-child node that child node 2 comprises, by calculating or emulation obtains the calculated value y of the output parameter of model ₁₁, y ₁₂, y ₁₃, y ₂₁, y ₂₂, y ₂₃, y ₂₄.

2, record by experiment the actual test value of the output parameter that under identical input, this model is corresponding.The actual test value z of output parameter ₁₁, z ₁₂, z ₁₃, z ₂₁, z ₂₂, z ₂₃, z ₂₄represent.

Table 1 force characteristic parameter is with the changing value of load

Table 2 unit for electrical property parameters is with the changing value of load

3, the relative error between calculated value and the actual test value of calculating output parameter.

$\mathrm{\&Delta;}{E}_{1m}=\frac{|y_{1m}-z_{1m}|}{y_{1m}}\&times;100\%$ m＝1,2,3；

$\mathrm{\&Delta;}{E}_{2n}=\frac{|y_{2n}-z_{2n}|}{y_{2n}}\&times;100\%$ n＝1,2,3,4。

The each relative error of parameter of table 3

4, relative error is mapped as to the goodness of fit by type gauss of distribution function less than normal.

Consider the calculated value of each parameter and the difference value of actual measured value the smaller the better (the error more correctness degree of minimodel is just higher), so the relative error that we adopt type gauss of distribution function less than normal that step 3 is obtained is mapped as the goodness of fit, calculate the inclined to one side of the single index goodness of fit

Small-sized Gaussian distribution curve figure as shown in Figure 5.

Computing formula is as follows:

$x_{1m}=\left\{\begin{array}{c}100,\mathrm{\&Delta;}{E}_{1m}\&le;\underset{\&OverBar;}{x}\\ {100e}^{-{\left(\frac{\mathrm{\&Delta;}{E}_{1m}-\underset{\&OverBar;}{x}}{\mathrm{\&delta;}}\right)}^2},\underset{\&OverBar;}{x}<\mathrm{\&Delta;}{E}_{1m}<\stackrel{\&OverBar;}{x}\\ 0,\mathrm{\&Delta;}{E}_{1m}\&GreaterEqual;x\end{array}\right.$

m＝1,2,3；

$x_{2n}=\left\{\begin{array}{c}100,\mathrm{\&Delta;}{E}_{2n}\&le;\underset{\&OverBar;}{x}\\ {100e}^{-{\left(\frac{\mathrm{\&Delta;}{E}_{2n}-\underset{\&OverBar;}{x}}{\mathrm{\&delta;}}\right)}^2},\underset{\&OverBar;}{x}<\mathrm{\&Delta;}{E}_{2n}<\stackrel{\&OverBar;}{x}\\ 0,\mathrm{\&Delta;}{E}_{2n}\&GreaterEqual;\stackrel{\&OverBar;}{x}\end{array}\right.$

n＝1,2,3,4，

X _1mrepresent by relative error △ E _1mthe goodness of fit of the sub-child node m that mapping obtains;

X _2nrepresent by relative error △ E _2nthe goodness of fit of the sub-child node n that mapping obtains;

represent relative error error value bound.

In above-mentioned calculating formula, generally get so just obtain the goodness of fit x of 3 parameters of the included sub-child node of child node 1 ₁₁, x ₁₂, x ₁₃, the goodness of fit x of 4 parameters of the included sub-child node of child node 2 ₂₁, x ₂₂, x ₂₃, x ₂₄.

The each parameter individual event of table 4 goodness of fit

The individual event goodness of fit that is the each parameter of sub-child node is:

x ₁₁＝90、x ₁₂＝88、x ₁₃＝89，

x ₂₁＝93、x ₂₂＝92、x ₂₃＝87、x ₂₄＝91。

Three, calculate the weight of the each parameter of sub-child node with respect to child node

Utilize analytical hierarchy process to calculate the weight of sub-child node with respect to each child node, in Recurison order hierarchy structure, for convenient quantificational expression below, generally represent different levels with A, B, C... from top to bottom, same level from left to right uses 1,2,3,4 ... represent different factors, as shown in Figure 6.

Calculating sub-child node specifically comprises the steps: with respect to the weight of each child node

1, Judgement Matricies

In the Recurison order hierarchy structure of model, each has the element of downward membership as first element (being positioned at the upper left corner) of judgment matrix, each element that is under the jurisdiction of it is sequentially arranged in the first row and first row thereafter, be the each element of child node first element as judgment matrix, the element that is under the jurisdiction of its each sub-child node is sequentially arranged in the first row and the 1st row thereafter.For the criterion of judgment matrix, relatively how much important which is important, between two for two elements, and the importance degree of each element is pressed to 1-9 assignment, and importance scale value sees the following form 5.

Table 5 importance scale implication table

Fill in judgment matrix according to the importance scale implication table shown in table 5, the judgment matrix filling is as shown in table 6, table 7:

Table 6 judgment matrix B ₁

B 1	C 1	C 2	C 3
				C 1	1	1/5	1/3
C 2	5	1	3
				C 3	3	1/3	1

Table 7 judgment matrix B ₂

B 2	C 4	C 5	C 6	C 7
					C 4	1	1/7	1/3	1/5
C 5	7	1	5	3
					C 6	3	1/5	1	1/3
C 7	5	1/3	3	1

2, Consistency Check in Judgement Matrix

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix.

B, by obtain coincident indicator C.I..

C, table look-up determine corresponding mean random coincident indicator R.I..

Table 8 mean random coincident indicator R.I. table

D, calculating consistance ratio C.R. also judge

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix.

Table 9 is to judgment matrix B ₁consistency check result

Table 10 is to judgment matrix B ₂consistency check result

Can find out that by the result of calculation of table 9, table 10 each judgment matrix all meets coherence request, i.e. CR<0.1.

3, calculate weight vectors

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, can obtain the relative weighting of each index.

Each sub-child node is respectively with respect to child node weighted value:

w ₁₁＝0.105、w ₁₂＝0.637、w ₁₃＝0.258，

w ₂₁＝0.550、w ₂₂＝0.064、w ₂₃＝0.118、w ₂₄＝0.268。

Four, calculate the goodness of fit of each child node

The goodness of fit x of the each parameter of sub-child node obtaining according to step 2 ₁₁, x ₁₂, x ₁₃, x ₂₁, x ₂₂, x ₂₃, x ₂₄, and the each sub-child node obtaining according to step 3 is with respect to child node weight w ₁₁, w ₁₂, w ₁₃, w ₂₁, w ₂₂, w ₂₃, w ₂₄, can calculate goodness of fit X1 and the X2 of each child node:

$X_1=\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}{w}_{1i}{x}_{1i}=90.815,X_2=\underset{j=1}{\overset{4}{\mathrm{\&Sigma;}}}{w}_{2j}{x}_{2j}=89.302.$

Five, calculate the weight of the each parameter of child node with respect to destination node

Computing method are identical with step 3, specific as follows:

1, Judgement Matricies

In the Recurison order hierarchy structure of model, each has the element of downward membership as first element of judgment matrix, each element that is under the jurisdiction of it is sequentially arranged in the first row and first row thereafter, it is destination node element first element as judgment matrix, the element that is under the jurisdiction of its each child node is sequentially arranged in the first row and the 1st row thereafter, for the criterion of judgment matrix, the importance degree of each element is pressed to 1-9 assignment.Each child node is as shown in table 11 below with respect to the judgment matrix A of destination node:

Table 11 judgment matrix A

A	B 1	B 2
			B 1	1	3
B 2	1/3	1

2, Consistency Check in Judgement Matrix

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix;

B, by obtain coincident indicator C.I.;

C, table look-up determine corresponding mean random coincident indicator R.I.;

D, calculating consistance ratio C.R. also judge:

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix.

Child node is as shown in table 12 below with respect to the each index relative weighting of destination node and consistency check result thereof.

The consistency check result of table 12 to judgment matrix A

3, calculate weight vectors

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, obtain the relative weighting W of each index ₁and W ₂.

Each child node is respectively with respect to destination node weighted value:

W ₁＝0.492、W ₂＝0.508。

Six, calculate the goodness of fit of destination node

The goodness of fit X of the each parameter of child node obtaining according to step 4 ₁and X ₂, and the weights W of the relative destination node of each parameter ₁and W ₂, calculate the goodness of fit D of destination node:

$D=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{W}_1{X}_1=90.046.$

Can evaluate the correctness of skin antenna power electric coupling model by total goodness of fit D: if total goodness of fit D value is larger, the correctness degree of this model is higher; Otherwise the correctness degree of this model is lower.

Evaluate: because technology concrete in engineering and manufacturability difficulty have determined the location that Correctness of model is evaluated, if technology and manufacturability difficulty are all higher, the index that model is met to actual requirement that so can be corresponding reduces a bit, such as thinking that in this case this model just meets indices requirement in the time that total goodness of fit reaches 85, can practical operation application in the middle of engineering; If technology and manufacturability difficulty are all lower, and relate to the important step in project, it is stricter that the index that model can be met the demands specifies, could meet indices requirement such as working as in this case when total goodness of fit of model reaches 95.So the evaluation method of the skin antenna electromechanical Coupling Model based on the goodness of fit in this paper, sets up evaluation model, has calculated net result, this result is a quantitative criterion, and the concrete central related request of project that must rely on carrys out rational evaluation.

As can be seen here, evaluation method tool of the present invention has the following advantages:

Only need know the input/output argument of model and in the detectable situation of output parameter, do not need to know principle and the inherent working method thereof of model complexity, the method that just can adopt the present invention to propose is evaluated the correctness of model, not only practical, and application is wide.

It should be noted that, above-described embodiment does not limit the present invention in any form, and all employings are equal to replaces or technical scheme that the mode of equivalent transformation obtains, all drops in protection scope of the present invention.

Claims (5)

1. the evaluation method of the Correctness of model based on the goodness of fit, is characterized in that, comprises the following steps:

(1), set up the Recurison order hierarchy structure of model:

Degree of membership relation according to each factor is set up Recurison order hierarchy structure, is divided into top level goal node, middle layer child node, three levels of the sub-child node of bottom, and upper layer node has dominating role to lower level node, and same layer factor is separate;

(2), calculate the individual event goodness of fit of the each parameter of the Recurison order hierarchy child node of the structure bottom:

(1), in the case of the size of model and input parameter known, calculate or emulation obtains the calculated value of the output parameter of model;

(2), record by experiment the actual test value of the output parameter that under identical input, this model is corresponding;

(3), the relative error between calculated value and the actual test value of calculating output parameter;

(4), relative error is mapped as to the goodness of fit by type gauss of distribution function less than normal;

(3), calculate the weight of the each parameter of sub-child node with respect to child node;

(4), calculate the goodness of fit of each child node:

Obtain the goodness of fit of each child node according to the weight calculation of the relative child node of the goodness of fit of the each parameter of sub-child node and each parameter;

(5), calculate the weight of the each parameter of child node with respect to destination node;

(6), calculate the goodness of fit of destination node:

Obtain the goodness of fit of destination node according to the weight calculation of the relative destination node of the goodness of fit of the each parameter of child node and each parameter.

2. the evaluation method of the Correctness of model based on the goodness of fit according to claim 1, is characterized in that, in step (three), calculates the each parameter of sub-child node and comprises following sub-step with respect to the weight of child node:

(1), Judgement Matricies:

In the Recurison order hierarchy structure of model, the each element of child node is as first element of judgment matrix, the element that is under the jurisdiction of its each sub-child node is sequentially arranged in the first row and the 1st row thereafter, for the criterion of judgment matrix, the importance degree of each element is pressed to 1-9 assignment;

(2), Consistency Check in Judgement Matrix;

(3), calculate weight vectors:

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, obtain the relative weighting of each index.

3. the evaluation method of the Correctness of model based on the goodness of fit according to claim 2, is characterized in that, the step of Consistency Check in Judgement Matrix is:

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix;

B, by obtain coincident indicator C.I.;

C, table look-up determine corresponding mean random coincident indicator R.I.;

D, calculating consistance ratio C.R. also judge:

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix.

4. the evaluation method of the Correctness of model based on the goodness of fit according to claim 3, is characterized in that, in step (five), calculates the each parameter of child node and comprises following sub-step with respect to the weight of destination node:

(1), Judgement Matricies:

In the Recurison order hierarchy structure of model, destination node element is as first element of judgment matrix, the element that is under the jurisdiction of its each child node is sequentially arranged in the first row and the 1st row thereafter, for the criterion of judgment matrix, the importance degree of each element is pressed to 1-9 assignment;

(2), Consistency Check in Judgement Matrix:

A, calculated the maximum characteristic root λ of judgment matrix R by | λ I-R|=0 _max, wherein I is unit matrix;

B, by obtain coincident indicator C.I.;

C, table look-up determine corresponding mean random coincident indicator R.I.;

D, calculating consistance ratio C.R. also judge:

$C.R.=\frac{C.I.}{R.I.}$

In the time of C.R.<0.1, the consistance of judgment matrix can receive;

In the time of C.R.>=0.1, judgment matrix does not meet coherence request, need to again revise this judgment matrix;

(3), calculate weight vectors:

To meeting conforming judgment matrix, the normalization of rank of advanced units vector, then ask row and normalization, obtain the relative weighting of each index.

5. according to the evaluation method of the Correctness of model based on the goodness of fit described in claim 1 to 4 any one, it is characterized in that, in step (), by the goodness of fit of each output parameter of known models as the sub-child node of bottom, as middle layer child node, use total goodness of fit of the model that can embody Correctness of model as top level goal node by the goodness of fit of all kinds of integrated performance indexs under output parameter.