CN106934237A - Radar cross-section redaction measures of effectiveness creditability measurement implementation method - Google Patents
Radar cross-section redaction measures of effectiveness creditability measurement implementation method Download PDFInfo
- Publication number
- CN106934237A CN106934237A CN201710138743.9A CN201710138743A CN106934237A CN 106934237 A CN106934237 A CN 106934237A CN 201710138743 A CN201710138743 A CN 201710138743A CN 106934237 A CN106934237 A CN 106934237A
- Authority
- CN
- China
- Prior art keywords
- sample
- interval
- credibility
- classification
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of Radar cross-section redaction measures of effectiveness creditability measurement implementation method, for contacting between pattern classification and measures of effectiveness, regards measures of effectiveness as performance ratings classification, and from classification credibility, research assessment is credible.Will indetermination theory be incorporated into the method, realize test data Uncertainty Management.According to disaggregated model Performance Evaluating Indexes, by being calculated assessment models creditability measurement method.Various single sample classification confidence calculations methods are merged, obtains, using extensive list Samples Estimates confidence calculations method, further by credible propagation model, obtaining synthetic effectiveness evaluation creditability measurement result.The present invention can process Effectiveness Evaluation Model of ECM credibility comprehensive analysis, give complete analysis process and corresponding computing formula;The demand of assessment result analysis can be met in the assessment of most of Weapons Systems Effectiveness.
Description
Technical field
The present invention relates to a kind of technology of information processing detection field, specifically a kind of Radar cross-section redaction measures of effectiveness
Creditability measurement implementation method.
Background technology
Assessment credibility is a qualitativing concept for ambiguity, and it represents the satisfaction of the accuracy and authenticity of assessment result
Degree.The data produced under different emulation granularities are produced with real system data ratio in simulation system credibility research
Compared with, according to data distribution, variation tendency, size of data etc., the similitude of judgement and analogue system, and then judge analogue system
It is credible.In order to improve the credibility of assessment, scholars start with from appraisal procedure, by scientific, the conjunction of discussing appraisal procedure
Rationality, and then the credibility of assessment result is illustrated, therefore constantly there is new Assessment theory to occur.For example, dividing from earliest level
Analysis method Fuzzy AHP finally, exactly in order to reduce the influence of the subjective factor to assessment result of people as far as possible, from
And improve the credibility of assessment result.However, up to now, it is directly also little to the credible research of assessment result.
The content of the invention
The present invention for the insurmountable technical problem of prior art including abstract model creditability measurement method and can
The defects such as the specific quantization of letter property, propose a kind of Radar cross-section redaction measures of effectiveness creditability measurement implementation method,
The present invention is achieved by the following technical solutions:
The present invention is directed to contacting between pattern classification and measures of effectiveness, regards measures of effectiveness as performance ratings classification, from
Classification credibility is set out, and research assessment is credible.Will indetermination theory be incorporated into the method, realize test data do not know
Property treatment.According to disaggregated model Performance Evaluating Indexes, by being calculated assessment models creditability measurement method.Merge various lists
Sample classification confidence calculations method, obtains using extensive list Samples Estimates confidence calculations method, further by credible
Spread through sex intercourse model, obtains synthetic effectiveness evaluation creditability measurement result.
The present invention by the credible factor of analyzing influence Radar cross-section redaction measures of effectiveness, obtain impact evaluation result because
It is plain mainly to have assessment data, assessment models and sample to be tested itself etc..The credibility for assessing data is influenceed by data source, by
The uncertain decision of measures of effectiveness experiment activity itself;The credibility of assessment models is influenceed by model accuracy rate;Single sample
This assessment confidence level is influenceed by sample distribution situation, and the distribution of higher dimensional space where sample and neighbour's sample is determined.By dividing
More than analysis several factors, and solve the problems, such as that its credibility quantifies respectively, solution Radar cross-section redaction measures of effectiveness is credible to be quantified
Subjective problem, improves the credible precision of Radar cross-section redaction measures of effectiveness, and letter is operated in actual application process
It is single, it is practical.
The present invention specifically includes following steps:
Step A:Uncertain data in object to be assessed is distinguished according to fuzzy data, random data and interval censored data
Carry out processing the uncertainty for obtaining assessing data<X,Crd(X)>, specifically include:
1. defined from interval value, information of the interval variable in uncertain space is unknown, i.e., it is in interval
In the range of each sample point probability it is equal.It can be considered that interval variable is equally distributed in interval range;Note area
Between variable be [a, b], then being converted to the distribution function after being uniformly distributed is:
2. after the probability density distribution for obtaining interval variable, obtain parameter and take any small interval probability in interval,
Further it is able to interval censored data discretization simply, asks for the probability that parameter takes a certain point value, i.e.,:Fuzzy variable A is by conversion
Operator can be translated into interval variable, and such that it is able to be solved using the method for processing interval value, fuzzy variable is uncertain to be surveyed
Degree problem.
Described operator is:Wherein:For under interval
Limit,It is the interval upper limit, AL(α) and Au(α) is the α cut sets of fuzzy variable A, the operator energy maximality quality guarantee
Hold the uncertainty of fuzzy variable.
Preferably, in the case of the degree of membership of known fuzzy variable, it is also possible to directly to the uncertainty of fuzzy variable
It is described.When ambiguity uncertain data set A is expressed as:Wherein:μA(Xi) represent data Xi
Degree of membership;Data XiUncertainty measure be:Fuzzy data XiCredibility be:
Preferably, in the case of the probability space of known stochastic variable, stochastic variable directly can be converted into interval change
Amount.
In addition, the comentropy situation of change according to information system solves the problems, such as the filling of incomplete data, before this hair
The bright specific intension that also discuss comentropy, it can be seen that comentropy is to a certain extent probabilistic measure, letter
Breath entropy represents the degree for removing uncertainty, when the probability on domain X point of stochastic uncertainty data in certain process of the test
Cloth is:P={ Px| x ∈ X }, then the uncertainty measure for defining test data X is:
Step B:Assessment models credibility Analysis, i.e., carry out measures of effectiveness using SVM classifier and neural network model etc.
Calculate, after measured data is trained for training sample, classification prediction is carried out to sample data to be assessed, and obtain according to model
Each correct possibility of class sample classification and necessity, and then obtain the credibility of disaggregated model.
Described SVM classifier, specifically:It is provided with sample set ((x1, y1), (x2, y2) ..., (xN, yN), wherein xi∈Rd
Input is represented, y ∈ { ± l } represent target output (wherein i=1 ..., N).If optimal hyperlane is ωTxi+ b=0, then weights to
Amount ω and biasing b must are fulfilled for constraint:yi(ωTxi+b)≥1-εi, wherein:εi£ becomes most for lax, intermediate scheme and ideal line
The departure degree of implementations.The target of SVM models is to find one to make the super flat of training data average error error in classification minimum
Face, so as to optimization problem can be derived:Wherein:C is positive parameter (the punishment system that need to be specified
Number), represent punishment degree of the SVM to mistake point sample.According to method of Lagrange multipliers, the solution of optimal separating hyper plane can be converted
It is optimization problem:Its constraints is:ai>=0, i=
1 ..., N, wherein:It is Lagrange multiplier.K(xi,xj) it is the kernel function for meeting Mercer theorems, conventional is linear
Core (Linear), polynomial kernel (Polynomial), Sigmoid cores and gaussian radial basis function core (RBF) etc..
Described neural network model, specifically refers to:If the training sample set of BP neural network isWherein xi
=[xi1,xi2,…,xiM] it is an input for m dimensions, oiIt is to be input into as xiWhen neutral net output.It is learning rate to make η > 0
Parameter.L-th output of neuron can be derived as in hidden layer:Output layer is defeated
Going out to be derived as:Under incremental processing mode, i-th sample is to neural network learning error function
For:From l-th neuron of hidden layer to the connection weight output layer, updating incremental computations formula is:Wherein:f2' it is the derivative of excitation function.Equally, we can obtain:M-th neuron of input layer is to l-th neuron of hidden layer
Connection weight increments of change, typically using formula:
Equally, we can obtain:Each neuron connection weight weight values for being tried to achieve according to more than and threshold value
Increment changes, and can constantly update the neuron weighted value and threshold value of next round network training, and its formula is: And m=1,2 ..., M.Each
After iteration, study will be re-started, and calculate each global error, judge network error whether in allowed band.Such as
Fruit error meets requirement or reaches setting study number of times, then stop calculating.Otherwise, algorithm continues.
Described SVM classifier, it is new without class label information that its classification accuracy refers to that grader is correctly predicted
The ability of the class label of data sample,Wherein:F is discriminant function, as g (xi)=yiWhen, F (g
(xi),yi)=1, is otherwise 0;A is that the form of sample is with (x comprising the n data set of samplei,yi) exist, xiRepresent sample
Property set, yiRepresent class label.
The classification performance of described neural network model is obtained by drawing ROC curve and AUC curve evaluations.
Using real rate and false positive rate as reference axis, the point on curve represents positive sample and negative sample to described ROC curve
This classification accuracy, the probability that model can be made full use of to obtain sample to be tested prediction, while algorithm can be showed intuitively
To the difference of sample predictions accuracy rate in the case of different distributions.
Sample is that positive positive sample is referred to as really (True by model prediction after disaggregated model is classified
Positive, TP), it is that negative positive sample is referred to as false negative (False Negative, FN) by model prediction, by model prediction for just
Negative sample be referred to as it is false just (False Positive, FP), be that negative negative sample referred to as really bears (True by model prediction
Negative,TN).Real rate (TPR)=TP/ (TP+FN), false positive rate (FPR)=FP/ (FP+TN).
Described AUC curves are integrated on ROC curve and obtain, and its value is closer to 1 property that represent grader
Can be better.
The credibility of described disaggregated model, i.e. model are to the confidence level of A class sample classifications:Wherein:The possibility Pos (A) of event A is A class samples
Originally the probability of A classes is divided into,TAThe accuracy rate of A classes is represented, | TSA| represent A class samples correct
The number of division, | SA| represent the number of A class samples;The necessity Nec (A) of event A, that is, the sample for being not belonging to A classes is divided
To the impossibility of A classes,|FSA| it is not that A classes sample is divided into the individual of A classes by mistake to represent
Count, the credibility of correspondence event A is:
Step C:Test sample sample to Unknown Label is divided according to the forecast confidence size of single sample classification, specifically
Including:
Due to confidence level f (the x)=exp (- 1/d (x)) of SVM classifier, wherein:D (x) is sample to be tested to svm classifier
The distance in face, and the confidence calculations formula of existing nearest neighbor classifier is:T=km(x)/k, wherein:K is that sample to be tested is near
The number of adjoint point, kmX () is the number that sample to be tested judges neighbour's training sample point that generic is included by model;Therefore
The forecast confidence for using single sample classification of nearest neighbor classifier is f1(x)=1-d1(x)/d2(x) and f2(x)=(d2(x)-
d1(x))/(d2(x)+d1(x)), wherein:d1X () is sample to be tested x and nearest training sample xmBetween distance, d2(x)
For in sample to be tested x and training sample with xmThe minimum distance of other training samples for belonging to a different category;And d2(x)
Should be greater than d1X (), then the span of confidence level is 0~1;The classification results confidence level of sample to be tested x is:And the d (y)=0 when in sample to be tested neighborhood without foreign peoples's sample, wherein:km(x)
It is the field sample of the sample to be tested number similar with its, k is the number of sample to be tested neighborhood sample, and d (x) is sample to be tested x
With the Euclidean distance of all similar samples in neighborhood sample and, d (y) is inhomogeneity sample in sample to be tested x and neighborhood sample
Euclidean distance and.
Step D:According to credible series connection and situation in parallel, the credibility of different links in evaluation process is carried out comprehensive
Close, calculate estimating for comprehensive credibility, i.e., in the case of series connection, credibility measure propagation model is:Cr (E)=Cr (Y) * Cr (f
(X);In the case of parallel connection, credibility measure propagation model is:Cr (E)=min { Cr (X), Cr (f (X)) };By propagating mode
Type, is able to for the credibility of different links in evaluation process to carry out synthesis, obtains total assessment credibility measure.
Step E:Credible comprehensive analysis is simultaneously calculated the total classification confidence level of sample to be tested, specifically includes:
1. credibility of the computation interval variable in the segment of certain point:If xi=(xi1,xi2,…,xin) it is n-dimensional space
A certain sample, as sample components xi1Value cannot be obtained by the way that experiment is accurate, but be able to provide the estimation of its interval range,
That is xi1When=[a, b] is interval variable, the interval upper limit and interval limit are respectively adopted in measures of effectiveness carry out efficiency respectively and comment
Estimate, there will be two kinds of situations for correspondence, i.e., it is consistent with the assessment classification results that interval limit is obtained using the interval upper limit, it is now interval
Data do not influence the performance ratings of evaluation object to divide;Or the measures of effectiveness obtained during using the interval upper limit and interval limit
Classification results are inconsistent, now the possibility of two kinds of performance ratings discussed further:When interval upper limit b is taken, by assessing mould
Type is y to the assessment classification results of sample1, when the lower limit a in interval is taken, by assessment models to the assessment classification results of sample
It is y2, now need to calculate the sample respectively and belong to y1With y2Credibility:In known xi1Probability density in interval [a, b]
In the case of function, when by judging xi1When taking the c points in interval, sample xi=(xi1,xi2,…,xin) be on interface;
Now, interval value [a, b] is divided into [a, c] and [c, b], the probability distribution value of [a, c] and [c, b] is calculated respectively, now probability
Distribution is considered as the credibility of assessment classification results in the case of only consideration test data.
2. when same sample multiple component has uncertain parameters, by index in evaluation index system according to " benefit
Type " and " cost-effectivenes " are classified, and the uncertain parameters that will belong to " benefit " and " cost-effectivenes " take the upper limit in interval respectively
And lower limit, judge the efficiency classification of evaluation object under most vantage;Then the lower and upper limit in interval are taken respectively again, is judged
The efficiency classification of evaluation object in the case of least favorable, the efficiency classification in the case of two kinds is identical, then do not consider experiment temporarily
The influence of data uncertainty, the change of efficiency classification and its confidence level when otherwise needing to discuss that each parameter value is different respectively.
When obvious index parameter includes a small amount of interval value variable, in interval value upper and lower bound, the different sample of efficiency classification results is
Sample near classification boundaries, therefore single or a small amount of index change, cause the change of sample and interface distance, finally quilt
It is divided into different classifications.No matter this kind of sample is divided into which kind of, and the credibility of its result is not high, cannot further verify area
Between value parameter concrete numerical value when, the confidence level size of different efficiency classification can be respectively provided, for policymaker's reference.
3. total classification confidence level Cr (the x)=Cr of sample to be tested is calculatedd(x)*Crm(f)*Crs(x), wherein:Crd(x) table
Show the confidence level of test data, CrmF () represents the credibility of model, CrsX () represents the confidence level of the assessment classification of single sample.
Technique effect
Compared with prior art, the present invention can process Effectiveness Evaluation Model of ECM credibility comprehensive analysis, give
Whole analysis process and corresponding computing formula;The objective metric method that the present invention can not be limited by expertise, is given
Operation result has stronger robustness;In the assessment of most of Weapons Systems Effectiveness, the measures of effectiveness in the present invention is credible
Analysis method can meet the demand of assessment result analysis.
Brief description of the drawings
Fig. 1 credibility measure propagation model schematic diagrames;
Comprehensive credibility Analysis flow chart in Fig. 2 present invention.
Specific embodiment
As shown in Fig. 2 the present embodiment is comprised the following steps:
Step 1, gathers training sample and sample to be tested:
It is the 140 groups of data such as table 1 for collecting by taking Radar cross-section redaction measures of effectiveness credibility Analysis as an example in the present embodiment
It is shown;
Table 1, thunder anti-jamming effectiveness valuated data
Step 2:Assessment models credibility Analysis
Step 2.1:By taking SVMs (SVM) assessment models as an example, training sample is 75 groups, and sample to be tested is 65 groups,
Wherein:15, first kind sample, 12, Equations of The Second Kind sample, the 3rd 18, class sample, the 4th 20, class sample.Using SVM models
It is predicted, there are 10 sample predictions results incorrect.Predict the outcome as shown in the table.
Error Rate is:0.153846
Wrongly Predicted Data Index is:4, Wrongly Predicted Label is:2
Wrongly Predicted Data Index is:13, Wrongly Predicted Label is:2
Wrongly Predicted Data Index is:18, Wrongly Predicted Label is:1
Wrongly Predicted Data Index is:22, Wrongly Predicted Label is:4
WronglyPredicted Data Index is:30, Wrongly Predicted Label is:2
Wrongly Predicted Data Index is:39, Wrongly Predicted Label is:4
Wrongly Predicted Data Index is:40, Wrongly Predicted Label is:3
WronglyPredicted Data Index is:47, Wrongly predicted Label is:2
Wrongly Predicted Data Index is:52, WronglyPredicted Label is:3
Wrongly Predicted Data Index is:62, Wrongly Predicted Label is:3
Table 2, using SVM assessment algorithms to the assessment result of sample to be tested.
The first kind is divided into 2 of Equations of The Second Kind, Equations of The Second Kind is divided into the first kind and the 4th class is 1, by the 3rd class point
For the first kind, Equations of The Second Kind and the 4th class are 1, the 4th class is divided into respectively 1 and 2 of Equations of The Second Kind and the 3rd class.
Step 2.2:Each correct possibility of class sample classification is obtained according to model and necessity is respectively:
Step 2.3:It is according to the credibility that model obtains every class sample classification result:
Step 3:Sample to be tested Confidence Analysis
Step 3.1:Table 3 is 3 groups of achievement datas of the equipment to be measured of experiment collection, and which part data cannot be obtained accurately
Data, but be able to obtain value scope (as shown in table 3).
Table 3, sample to be tested information
Step 3.2:T3In average false track improvement factor and T2In anti-interference covering of the fan availability two indices experiment
Data not exact value but interval range.In the case of further cannot obtaining exact value by experiment again, the present invention will be begged for
By the influence that interval censored data works measures of effectiveness.By analysis above, the present invention will first consider the interval upper limit and lower limit
In the case of, it is estimated respectively as two groups of data.Judge two groups of data under performance ratings it is whether consistent.By analysis,
T3In when taking interval limit, performance ratings are 3, and when taking the interval upper limit, performance ratings are 4.T2In, take the interval upper limit and lower limit effect
Energy level results are identical.Therefore, the present invention is needed to T3Data are further analyzed, by the method for interval search, present invention hair
Now when value is 8.81, performance ratings are 3, and when value is 8.82, performance ratings are 4, therefore the present invention approximate thinks
8.815 is the separation of efficiency.
Step 3.3:It is close by being uniformly distributed Interval Valued Probability when the possibility that the parameter takes each point value in interval is identical
Degree function, then the present invention is it is concluded that be:When average false track improvement factor value is [8.79,8.815], efficiency etc.
Level is 3, and its probability size is 0.42;When average false track improvement factor value is [8.815,8.85], performance ratings are
4, its probability size is 0.58.
Step 3.4:Sample in table 4 is assessed by SVM, the result is that the no correct present invention is not aware that, but
It is the credibility for being calculated model to the accuracy rate of sample classification according to model of the invention, the credibility of model is to a certain degree
On represent model to three groups of credibilities of sample efficiency division result above.But the description of above-mentioned steps can be seen that only
The credibility for just knowing that model is inadequate, accordingly, it would be desirable to the further confidence level of the single sample of analysis.
Table 4, T1The nearest samples and relevant information of sample
Step 3.5:Calculating the confidence level of single sample needs to calculate Euclidean distance of the sample with neighborhood sample and neighbour respectively
The relation of the classification of domain sample and itself classification.Table 3 gives T1The nearest sample and range information of sample.Obtained by table 4, T1
The classification confidence of sample is:
Step 3.6:The present invention is to T2And T3Two groups of data are analyzed.With T2Interval limit value as the accurate of index
Sample during value is designated as T21, T is designated as using interval higher limit as the sample during exact value of index22, T is calculated respectively21And T22's
Neighborhood sample and its distance, are computed T21And T22Neighbouring sample it is identical, range information is as shown in table 5.
Table 5, T2The nearest samples and relevant information of sample
T is obtained by table 521And T22The classification confidence of sample is respectively:
As can be seen that T2Sample takes classification of the interval upper and lower bound on sample influences very little, in the sample, substantially may be used
With the influence without considering interval value (uncertain data).
For T3The situation of sample can be analyzed using same method, try to achieve its classification confidence.Here no longer go to live in the household of one's in-laws on getting married
State.
From the credibility Analysis flow that Fig. 2 is given can be seen that comprehensive analysis Samples Estimates result it is credible when, root
According to the difference of uncertainty measure data type in experiment, it is translated into interval value or provides credibility measure, both
Method can process the data uncertainty in experiment.The processing mode for being converted to interval value is actually by uncertain number
According to the thought for being divided into multiple precise informations.After assessment models credibility and single sample credibility is calculated, by credibility
Propagation model, carries out synthesis, you can to obtain sample by the confidence level of the credible of test data, the credible and single sample of model
The credible measured value of this assessment result.
Above-mentioned propagation model refers specifically to the propagation model described in step D, as:Cr (E)=Cr (Y) * Cr (f
(X)。
Step 4, comprehensive credible calculating:
Step 4.1:By above-mentioned steps, credibility of the model to each class sample classification is obtained, while giving list
The confidence level of sample.Therefore, by the method for step D, can obtain total assessment credibility is:Cr(T1)=Cr (A2)*f
(T1)=0.88*0.68=0.60, Cr (T2)=Cr (A2)*f(T2)=0.89*0.84=0.75.
Step 4.2:So far, the present embodiment has been calculated sample T1And T2Representative equipment Efficacy assessment result for etc.
The credibility of level 2 and grade 3 is respectively 60% and 75%.In the present embodiment and do not need expert participate in marking, to manpower requirements
It is low.The result has clearly showed the credibility of the Radar cross-section redaction measures of effectiveness result that measures of effectiveness personnel provide, to comment
Estimate the foundation solid using offer of conclusion, and the result is not limited by using the present invention staff's know-how, confidence level
It is higher.
Above-mentioned specific implementation can by those skilled in the art on the premise of without departing substantially from the principle of the invention and objective with difference
Mode local directed complete set is carried out to it, protection scope of the present invention is defined and not by above-mentioned specific implementation institute by claims
Limit, each implementation in the range of it is by the constraint of the present invention.
Claims (9)
1. a kind of Radar cross-section redaction measures of effectiveness creditability measurement implementation method, it is characterised in that comprise the following steps:
Step A:Uncertain data in object to be assessed is carried out respectively according to fuzzy data, random data and interval censored data
Treatment obtains assessing uncertain the < X, Cr of datad(X)〉;
Step B:Assessment models credibility Analysis, i.e., carry out measures of effectiveness calculating using SVM classifier and neural network model, real
After survey data are trained for training sample, classification prediction is carried out to sample data to be assessed, and each class is obtained according to model
The correct possibility of sample classification and necessity, and then obtain the credibility of disaggregated model;
Step C:Test sample sample to Unknown Label is divided according to the forecast confidence size of single sample classification;
Step D:According to credible series connection and situation in parallel, the credibility of different links in evaluation process is carried out into synthesis, counted
Calculate estimating for comprehensive credibility;
Step E:Credible comprehensive analysis is simultaneously calculated the total classification confidence level of sample to be tested.
2. implementation method according to claim 1, it is characterized in that, described step A includes:
1. based on interval variable it is uniformly distributed in interval range, note interval variable is [a, b], then be converted to after being uniformly distributed
Distribution function be:
2. after the probability density distribution for obtaining interval variable, obtain parameter and take any small interval probability in interval, enter one
Step is able to interval censored data discretization simply, asks for the probability that parameter takes a certain point value, i.e.,:Fuzzy variable A passes through operator
Interval variable is translated into, such that it is able to solve the problems, such as fuzzy variable uncertainty measure using the method for processing interval value.
3. implementation method according to claim 2, it is characterized in that, described operator is:
Wherein:It is interval limit,It is the interval upper limit,
AL(α) and Au(α) is the α cut sets of fuzzy variable A, and the uncertainty of fuzzy variable is held in the operator energy maximality quality guarantee.
4. implementation method according to claim 2, it is characterized in that, in the case of the degree of membership of known fuzzy variable, directly
Connect and the uncertainty of fuzzy variable is described:When ambiguity uncertain data set A is expressed as:Wherein:μA(Xi) represent data XiDegree of membership;Data XiUncertainty measure be:Fuzzy data XiCredibility be:
5. implementation method according to claim 2, it is characterized in that, in the case of the probability space of known stochastic variable, directly
Connect and stochastic variable is converted into interval variable.
6. implementation method according to claim 1, it is characterized in that, the credibility of described disaggregated model, i.e., model is to A
The confidence level of class sample classification is:Wherein:Event A
Possibility Pos (A) be probability that A class samples are divided into A classes,TAThe accuracy rate of A classes is represented, |
TSA| the number that A class samples are correctly divided is represented, | SA| represent the number of A class samples;The necessity Nec (A) of event A, i.e., not
The sample for belonging to A classes is divided into the impossibility of A classes,FSARepresentative is not A class samples
It is divided into the number of A classes by mistake, the credibility of correspondence event A is:
7. implementation method according to claim 1, it is characterized in that, the forecast confidence of described single sample classification is f1(x)
=1-d1(x)/d2(x) and f2(x)=(d2(x)-d1(x))/(d2(x)+d1(x)), wherein:d1(x) be sample to be tested x with it is nearest
Training sample xmBetween distance, d2(x) be in sample to be tested x and training sample with xmOther training samples for belonging to a different category
This minimum distance;And d2X () should be greater than d1X (), then the span of confidence level is 0~1;The classification knot of sample to be tested x
Fruit confidence level is:And when in sample to be tested neighborhood without foreign peoples's sample d (y)=
0, wherein:kmX () is the field sample of the sample to be tested number similar with its, k is the number of sample to be tested neighborhood sample, d (x)
For sample to be tested x and the Euclidean distance of all similar samples in neighborhood sample and, d (y) be sample to be tested x with neighborhood sample in not
The Euclidean distance of similar sample and.
8. implementation method according to claim 1, it is characterized in that, described step D:In the case of series connection, credibility is surveyed
Spending propagation model is:Cr (E)=Cr (Y) * Cr (f (X);In the case of parallel connection, credibility measure propagation model is:Cr (E)=
min{Cr(X),Cr(f(X))};By propagation model, it is able to for the credibility of different links in evaluation process to carry out synthesis, obtains
To total assessment credibility measure.
9. implementation method according to claim 1, it is characterized in that, described step E is specifically included:
1. credibility of the computation interval variable in the segment of certain point:Work as xi=(xi1,xi2,…,xin) it is a certain of n-dimensional space
Sample, as sample components xi1Value cannot be obtained by the way that experiment is accurate, but be able to provide the estimation of its interval range, i.e. xi1
When=[a, b] is interval variable, the interval upper limit and interval limit are respectively adopted in measures of effectiveness carries out measures of effectiveness respectively, right
Should there will be two kinds of situations, i.e.,:
A. consistent with the assessment classification results that interval limit is obtained using the interval upper limit, now interval censored data does not influence evaluated right
The performance ratings of elephant are divided;Or
The measures of effectiveness classification results obtained when b. using the interval upper limit and interval limit are inconsistent, then further:
B1. to the assessment classification results of sample it is y by assessment models when interval upper limit b is taken1,
B2. to the assessment classification results of sample it is y by assessment models when the lower limit a in interval is taken2, the sample is now calculated respectively
Originally y is belonged to1With y2Credibility:In known xi1In the case of probability density function in interval [a, b], when by judging xi1
When taking the c points in interval, sample xi=(xi1,xi2,…,xin) be on interface, then by interval value [a, b] be divided into [a, c] and
[c, b], calculates the probability distribution value of [a, c] and [c, b] respectively, and now probability distribution only considers assessment point in the case of test data
The credibility of class result;
2. when same sample multiple components have uncertain parameters, by index in evaluation index system according to " profit evaluation model " and
" cost-effectivenes " is classified, will belong to " benefit " and " cost-effectivenes " uncertain parameters take respectively interval the upper limit and under
Limit, judges the efficiency classification of evaluation object under most vantage;Then the lower and upper limit in interval are taken respectively again, is judged least
The efficiency classification of evaluation object in the case of profit, the efficiency classification in the case of two kinds is identical, then temporarily do not consider test data
Probabilistic influence, the change of efficiency classification and its confidence level when otherwise needing to discuss that each parameter value is different respectively;
3. total classification confidence level Cr (the x)=Cr of sample to be tested is calculatedd(x)*Crm(f)*Crs(x), wherein:CrdX () represents experiment
The confidence level of data, CrmF () represents the credibility of model, CrsX () represents the confidence level of the assessment classification of single sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710138743.9A CN106934237A (en) | 2017-03-09 | 2017-03-09 | Radar cross-section redaction measures of effectiveness creditability measurement implementation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710138743.9A CN106934237A (en) | 2017-03-09 | 2017-03-09 | Radar cross-section redaction measures of effectiveness creditability measurement implementation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106934237A true CN106934237A (en) | 2017-07-07 |
Family
ID=59433826
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710138743.9A Pending CN106934237A (en) | 2017-03-09 | 2017-03-09 | Radar cross-section redaction measures of effectiveness creditability measurement implementation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106934237A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299555A (en) * | 2018-09-30 | 2019-02-01 | 上海机电工程研究所 | Infrared Imaging Seeker anti-jamming performance evaluation method and system |
CN109684689A (en) * | 2018-12-12 | 2019-04-26 | 云南电网有限责任公司电力科学研究院 | A kind of insulating gas comprehensive performance evaluation method of multi-parameter fusion |
CN109738871A (en) * | 2018-12-14 | 2019-05-10 | 中国人民解放军63893部队 | Data processing method is assessed in the radar EW test ruling of rule-based engine |
CN110059743A (en) * | 2019-04-15 | 2019-07-26 | 北京致远慧图科技有限公司 | Determine method, equipment and the storage medium of forecasting reliability measurement |
CN110288005A (en) * | 2019-05-30 | 2019-09-27 | 南京理工大学 | Millimeter wave detector New Jamming Effects Evaluation Method based on SVM |
CN110348752A (en) * | 2019-07-16 | 2019-10-18 | 中国人民解放军火箭军工程大学 | A kind of large scale industry system structure security assessment method considering environmental disturbances |
CN110618405A (en) * | 2019-10-16 | 2019-12-27 | 中国人民解放军海军大连舰艇学院 | Radar active interference efficiency measuring and calculating method based on interference mechanism and decision-making capability |
CN111157963A (en) * | 2020-01-31 | 2020-05-15 | 中国人民解放军32802部队 | Online evaluation method for interference effect of active phased array radar |
CN112945018A (en) * | 2021-01-29 | 2021-06-11 | 南京理工大学 | Laser non-lethal weapon injury evaluation method |
CN112990125A (en) * | 2021-04-26 | 2021-06-18 | 中国人民解放军国防科技大学 | Method for judging whether radiation source radar belongs to target platform |
CN114018317A (en) * | 2021-09-30 | 2022-02-08 | 北京航天华腾科技有限公司 | Data acquisition device and method for marine environment |
-
2017
- 2017-03-09 CN CN201710138743.9A patent/CN106934237A/en active Pending
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299555B (en) * | 2018-09-30 | 2020-02-04 | 上海机电工程研究所 | Anti-interference performance evaluation method and system for infrared imaging seeker |
CN109299555A (en) * | 2018-09-30 | 2019-02-01 | 上海机电工程研究所 | Infrared Imaging Seeker anti-jamming performance evaluation method and system |
CN109684689A (en) * | 2018-12-12 | 2019-04-26 | 云南电网有限责任公司电力科学研究院 | A kind of insulating gas comprehensive performance evaluation method of multi-parameter fusion |
CN109738871A (en) * | 2018-12-14 | 2019-05-10 | 中国人民解放军63893部队 | Data processing method is assessed in the radar EW test ruling of rule-based engine |
CN109738871B (en) * | 2018-12-14 | 2022-12-23 | 中国人民解放军63893部队 | Radar countermeasure equipment test judgment evaluation data processing method based on rule engine |
CN110059743A (en) * | 2019-04-15 | 2019-07-26 | 北京致远慧图科技有限公司 | Determine method, equipment and the storage medium of forecasting reliability measurement |
CN110059743B (en) * | 2019-04-15 | 2021-10-29 | 北京致远慧图科技有限公司 | Method, apparatus and storage medium for determining a predicted reliability metric |
CN110288005A (en) * | 2019-05-30 | 2019-09-27 | 南京理工大学 | Millimeter wave detector New Jamming Effects Evaluation Method based on SVM |
CN110288005B (en) * | 2019-05-30 | 2022-09-27 | 南京理工大学 | SVM-based millimeter wave detector interference effect evaluation method |
CN110348752A (en) * | 2019-07-16 | 2019-10-18 | 中国人民解放军火箭军工程大学 | A kind of large scale industry system structure security assessment method considering environmental disturbances |
CN110348752B (en) * | 2019-07-16 | 2023-07-25 | 中国人民解放军火箭军工程大学 | Large industrial system structure safety assessment method considering environmental interference |
CN110618405A (en) * | 2019-10-16 | 2019-12-27 | 中国人民解放军海军大连舰艇学院 | Radar active interference efficiency measuring and calculating method based on interference mechanism and decision-making capability |
CN111157963A (en) * | 2020-01-31 | 2020-05-15 | 中国人民解放军32802部队 | Online evaluation method for interference effect of active phased array radar |
CN112945018A (en) * | 2021-01-29 | 2021-06-11 | 南京理工大学 | Laser non-lethal weapon injury evaluation method |
CN112990125A (en) * | 2021-04-26 | 2021-06-18 | 中国人民解放军国防科技大学 | Method for judging whether radiation source radar belongs to target platform |
CN112990125B (en) * | 2021-04-26 | 2021-07-30 | 中国人民解放军国防科技大学 | Method for judging whether radiation source radar belongs to target platform |
CN114018317A (en) * | 2021-09-30 | 2022-02-08 | 北京航天华腾科技有限公司 | Data acquisition device and method for marine environment |
CN114018317B (en) * | 2021-09-30 | 2022-05-17 | 北京航天华腾科技有限公司 | Data acquisition device and method for marine environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106934237A (en) | Radar cross-section redaction measures of effectiveness creditability measurement implementation method | |
CN108280553B (en) | Mountain torrent disaster risk zoning and prediction method based on GIS-neural network integration | |
CN103974311B (en) | Based on the Condition Monitoring Data throat floater detection method for improving Gaussian process regression model | |
CN105117602B (en) | A kind of metering device running status method for early warning | |
Nayak et al. | Fuzzy computing based rainfall–runoff model for real time flood forecasting | |
CN112506990A (en) | Hydrological data anomaly detection method based on spatiotemporal information | |
CN112348290B (en) | River water quality prediction method, river water quality prediction device, storage medium and storage device | |
Elbagalati et al. | Development of an enhanced decision-making tool for pavement management using a neural network pattern-recognition algorithm | |
CN112257914B (en) | Aviation safety causal prediction method based on random forest | |
Zhao et al. | The application of BP neural networks to analysis the national vulnerability. | |
WO2017071369A1 (en) | Method and device for predicting user unsubscription | |
CN113822499B (en) | Train spare part loss prediction method based on model fusion | |
Kargah-Ostadi | Comparison of machine learning techniques for developing performance prediction models | |
CN114611372A (en) | Industrial equipment health prediction method based on Internet of things edge calculation | |
Alsayadi et al. | Improving the regression of communities and crime using ensemble of machine learning models | |
Santos et al. | An artificial intelligence method for improving upscaling in complex reservoirs | |
CN112434887B (en) | Water supply network risk prediction method combining network kernel density estimation and SVM | |
CN113626929A (en) | Multi-stage multi-topology ship traffic complexity measuring method and system | |
CN112241808A (en) | Road surface technical condition prediction method, device, electronic equipment and storage medium | |
Abbasabadi et al. | A framework for urban building energy use modeling | |
CN107977727B (en) | Method for predicting blocking probability of optical cable network based on social development and climate factors | |
CN111126694A (en) | Time series data prediction method, system, medium and device | |
CN115544886A (en) | Method, system, apparatus and medium for predicting failure time node of high-speed elevator | |
CN115545342A (en) | Risk prediction method and system for enterprise electric charge recovery | |
CN115330526A (en) | Enterprise credit scoring method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170707 |
|
RJ01 | Rejection of invention patent application after publication |