CN116720103B

CN116720103B - Large-scale circuit fault diagnosis method and system based on cloud theory and kernel density

Info

Publication number: CN116720103B
Application number: CN202310611976.1A
Authority: CN
Inventors: 刘美容; 段涛; 刘慧�
Original assignee: Ningbo Lidou Intelligent Technology Co ltd
Current assignee: Ningbo Lidou Intelligent Technology Co ltd
Priority date: 2023-05-26
Filing date: 2023-05-26
Publication date: 2025-08-26
Anticipated expiration: 2043-05-26
Also published as: CN116720103A

Abstract

The invention discloses a large-scale circuit fault diagnosis method and system of a Bayesian model based on Epanechnikov kernel density estimation of cloud theory, which belong to the field of large-scale analog circuit fault diagnosis, wherein fault information of a modularized integrated circuit after circuit tearing is collected, digital characteristics of clouds under each fault are generated by a reverse generator, a plurality of cloud drops under each fault mode are obtained by a forward cloud generator, membership values of each characteristic and corresponding fault state under a sample set are calculated by using a standard cloud model, cloud drops under each fault characteristic are determined according to a membership matrix, a fault sample set is expanded, and posterior probability density values under each fault category are calculated by using the Bayesian model based on Epanechnikov kernel density estimation, so that a fault diagnosis result is obtained. The method combines the cloud theory and the Bayesian kernel density, effectively solves the problems of single state information, less fault sample set and the like after the network is torn, integrates data information and priori knowledge, and greatly improves the fault accuracy.

Description

Cloud theory and nuclear density-based large-scale circuit fault diagnosis method and system

Technical Field

The invention belongs to the technical field of large-scale analog circuit fault diagnosis, and particularly relates to a cloud theory-based Bayesian model large-scale circuit fault diagnosis method and system for Epanechnikov kernel density estimation.

Background

Since the development of integrated circuits has been carried out in small, medium and large scale stages, and in the very large scale stage, industrial production has urgent requirements for large scale circuit testing, and the diagnosis of large scale circuit faults has not been fully mature in theory and method.

The accurate module-level fault positioning and clear identification detection result is a topic which needs to be solved by electronic engineering urgently, is a key step of practical application in theory, has huge number of devices contained in topology and continuity of each device parameter in an effective range for a large-scale analog circuit, so that the working condition of the circuit is more complicated, the state information of a large-scale circuit network is single after tearing, a simulation sample is difficult to obtain, the number of fault sample sets formed by tearing node voltages in each fault mode is small, accurate diagnosis of all faults is difficult to carry out, and the traditional fault analysis method is difficult to adapt to the current requirements.

Disclosure of Invention

Aiming at the defects or improvement demands of the prior art, the invention provides a Bayesian model large-scale circuit fault diagnosis method and system based on the Epanechnikov kernel density estimation of cloud theory, so as to improve the diagnosis accuracy and diagnosis efficiency of circuit diagnosis.

To achieve the above object, according to one aspect of the present invention, there is provided a method for diagnosing a large-scale circuit fault of a bayesian model of an Epanechnikov kernel density estimation based on cloud theory, including:

Establishing a large-scale circuit schematic diagram, selecting the voltage of a modularized tearing node as a diagnosis signal, carrying out coding classification on fault types according to different faults of components, and collecting the voltage of the tearing node under different faults to form an initial sample set;

Generating digital characteristic expectations, entropy and super entropy of cloud under each fault by using an inverse generator of a cloud model according to an initial sample set formed by voltages of tearing nodes under different faults;

according to the digital characteristic expectation, entropy and super entropy of the cloud under each fault, a plurality of cloud drops under each fault are obtained by utilizing a forward cloud generator so as to restore a plurality of measurable node voltage values;

Calculating each voltage value under the initial sample set and the membership value of the corresponding fault state by using a standard cloud model to form a membership matrix, and determining cloud drops under each fault voltage value according to the membership matrix of the fault so as to expand the initial sample set;

Dividing the expanded initial sample set into a plurality of folds for cross verification, and calculating posterior probability density values under each fault category in the test set by using a Bayesian model of Epanechnikov kernel density estimation;

And classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result.

In some alternative embodiments, the selecting the voltage of the modular tear node as the diagnostic signal comprises:

and carrying out modularized tearing on the large-scale circuit, tearing the limited accessible node into modularization on the premise of the limited accessible node, and selecting the voltage of the torn node in each fault mode as a diagnosis signal to obtain a fault characteristic vector.

In some alternative embodiments, the initial sample set of voltages of tearing nodes at different faults uses an inverse generator of a cloud model to generate digital feature expectations, entropy and super entropy of the cloud at each fault, comprising:

Note the initial sample set x _ij＝[x_i1,x_i2,…,x_in of circuit faults (i=1, 2,..n; j=1, 2,..m.), a j-th class fault that characterizes the i-th fault, n is the dimension of the feature vector in the fault mode, m is the total number of fault modes, then Obtaining the mean value of the initial sample set of the j-th type faultFrom the following components A first order central moment K of the initial sample set of the j-th class of faults is obtained, from the following componentsObtaining the variance S of an initial sample set of the j-th type fault;

since the certainty is unknown, the digital characteristic expectations E _x, the entropy E _n and the super entropy H _e of the cloud model obtained by the inverse cloud generator CG ^-1 without certainty are as follows:

in some alternative embodiments, the obtaining a plurality of cloud droplets per fault using a forward cloud generator includes:

On the basis of the cloud { E _x,E_n,H_e } of each type of faults, setting up the number Num of cloud drops, utilizing a forward cloud model CG (E _x,E_n,H_e, num), firstly, for the j type of faults, generating a model with E _n as an expectation, A normal random number E' _ni that is the variance;

using the generated random number, with E _x as the expectation, Generating a normal random number x _k for variance and calculating the x _k membershipWherein x _k of u _k with certainty represents a cloud droplet in the domain;

Repeating the steps until the set cloud drop number Num is generated, and obtaining a cloud sample set drop (x _k,u_k) of the j-th type fault.

In some alternative embodiments, the calculating the membership value of each voltage value and the corresponding fault state in the initial sample set by using the standard cloud model forms a membership matrix, including:

Describing membership degrees of fault feature vectors corresponding to fault types through a cloud model, substituting standard cloud parameters corresponding to the j-th type fault of the i-th feature vector of an initial fault sample x _ij into the standard cloud model formed under each fault Obtaining the corresponding membership of x _ij, wherein u _ij is the membership of the j-th fault of the i-th feature vector of the initial fault sample, E _nij' is the normal random number generated by the entropy E _nij and the super entropy H _eij ² of the j-th fault of the i-th feature vector, and E _xij is the digital feature expectation of the j-th fault of the i-th feature vector;

After the membership degree of various faults of the initial fault sample under all feature vectors is obtained, membership degree moment is obtained: m represents the total number of failure modes;

the fuzzy vector is formed by fuzzy comprehensive evaluation uncertainty index Obtaining the membership degree of uncertainty, wherein n is a fuzzy comprehensive evaluation vector (u _i1,u_i2,…,u_in), i=1, 2, & n, beta is the maximum membership degree in (u _i1,u_i2,…,u_in), and gamma is the second maximum membership degree in (u _i1,u_i2,…,u_in);

calculating the uncertain membership degree of each fault under the membership degree to form an uncertain membership degree fuzzy characteristic vector And (3) representing the uncertain membership degree of the ith eigenvector to obtain a membership degree matrix of m rows and n+1 columns:

In some alternative embodiments, the dividing the expanded initial sample set into a plurality of fold cross-validations, calculating posterior probability density values for each fault class in the test set using a bayesian model of an Epanechnikov kernel density estimation, includes:

Cross-verifying the expanded initial sample set into a training set and a testing set;

In the Bayesian model, a likelihood function of the Epanechnikov kernel density estimation can be obtained through a plurality of independent samples distributed in the test set, posterior distribution of the likelihood function of the Epanechnikov kernel density estimation is obtained according to the Bayesian theorem, and then posterior probability density values under each fault category are obtained.

In some alternative embodiments, the classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain the diagnosis result includes:

Classifying fault data by using an Epanechnikov kernel density estimation Bayesian model, regarding voltage data points of faults to be detected as a vector, and estimating probability density functions of each fault type by using an Epanechnikov kernel density estimation method;

And projecting the voltage data points of the faults to be detected onto the probability density function of each fault class, calculating the probability density value under each class, and classifying the voltage data points of the faults to be detected into the class with the highest probability density value.

In some optional embodiments, the projecting the voltage data points of the fault to be detected onto the probability density function of each fault class, calculating the probability density value under each class, and classifying the voltage data points of the fault to be detected into the class with the highest probability density value includes:

Taking the fault voltage data point to be detected as a vector, wherein the vector comprises a plurality of features, subtracting the posterior mean value of the class from the features in the vector, dividing each element obtained by the posterior square difference of the class, substituting the obtained element into an Epanechnikov kernel function as an independent variable, and taking the average value of the obtained new vector to obtain the probability density value of the fault sample to be detected under the class;

calculating the reciprocal of the square root of the determinant of the posterior variance of the class, multiplying the reciprocal by the obtained probability density value to normalize the probability value;

and calculating the posterior probability density value of the fault voltage data point to be detected under each fault category according to the Bayesian theorem, and classifying the fault voltage data point to be detected into the category with the highest probability density value.

According to another aspect of the present invention, there is provided a system for diagnosing a large-scale circuit fault of a bayesian model of an Epanechnikov kernel density estimation based on a cloud theory, including:

The circuit acquisition module to be diagnosed is used for establishing a large-scale circuit schematic diagram, selecting the voltage of the modularized tearing node as a diagnosis signal, carrying out coding classification on fault types according to different faults of components and parts, and collecting the voltage of the tearing node under different faults to form an initial sample set;

The cloud theory calculation module is used for generating digital characteristic expectations, entropies and super entropies of clouds under each fault by using an initial sample set formed by voltages of tearing nodes under different faults and using a reverse generator of a cloud model, and obtaining a plurality of cloud drops under each fault by using the forward cloud generator according to the digital characteristic expectations, the entropies and the super entropies of the clouds under each fault so as to restore a plurality of measurable node voltage values;

The membership degree expansion sample module is used for calculating each voltage value under the initial sample set and the membership degree value corresponding to the fault state to form a membership degree matrix by utilizing the standard cloud model, and determining cloud drops under each fault voltage value according to the membership degree matrix of the fault so as to expand the initial sample set;

The fault diagnosis module is used for dividing the expanded initial sample set into a plurality of fold cross-validation, calculating posterior probability density values under each fault category in the test set by using a Bayesian model of Epanechnikov kernel density estimation, and classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result.

According to another aspect of the present invention there is provided a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of any of the methods described above.

In general, the above technical solutions conceived by the present invention, compared with the prior art, enable the following beneficial effects to be obtained:

The Bayesian model of the Epanechnikov kernel density estimation based on the cloud theory is used for large-scale circuit fault diagnosis, cloud drops generated by combining a forward cloud model and a reverse cloud model in the cloud theory are used for expanding a fault sample set, the generation of a sample based on the cloud fuses the randomness and ambiguity uncertainty theory of the cloud model, characteristic data deviating from normal distribution is filtered, the model has certain anti-noise capability, the Bayesian model of the Epanechnikov kernel function does not need to make any assumption on the distribution of target variables, and the model based on the circuit fault diagnosis obtained on the basis can be better suitable for different data sets and problems, so that the accuracy and the diagnosis efficiency of circuit fault diagnosis are effectively improved.

Drawings

FIG. 1 is a schematic flow chart of a circuit fault diagnosis method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a video amplifying circuit according to an embodiment of the present invention;

FIG. 3 is a tearing diagram of a video amplifying circuit according to an embodiment of the present invention;

FIG. 4 is a fault cloud sample generation process provided by an embodiment of the present invention;

FIG. 5 is a graph comparison of cloud models under normal and fault conditions provided by an embodiment of the present invention;

FIG. 6 is a confusion matrix of fault diagnosis results provided by an embodiment of the present invention;

Fig. 7 is a schematic structural diagram of a circuit fault diagnosis device according to an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.

Referring to fig. 1, a schematic flow chart of a method provided by an embodiment of the present invention is shown, and the method shown in fig. 1 includes the following steps:

(1) Establishing a simulation model of the video amplifying circuit, wherein a simulation circuit diagram is shown in fig. 2, carrying out modularized division on the video amplifying circuit under the condition that node tearing is met, and collecting voltage state information of torn nodes to obtain fault characteristic vectors, wherein tolerance of resistance in the circuit is 5%, considering that a semiconductor device and a passive device are easier to generate faults, and according to practical experience, the circuit is provided with 9 fault modes including normal states, as shown in table 1;

TABLE 1 failure states and categories

Fault labels	Fault state	Fault coding
			F0	Normal state	0
F1	V ₁ open circuit	1
			F2	R ₁ short circuit	2
F3	Q ₄ BE extremely short circuit	3
			F4	Q ₂ CE extremely short circuit	4
F5	R ₉ short circuit	5
			F6	V ₂ open circuit	6
F7	V ₃ open circuit	7
			F8	Q ₁₀ CE extremely short circuit	8

In step (1), the large-scale circuit is subjected to modularized tearing, and one network needs to meet the following conditions that 1, the torn node is an accessible node and comprises a public node, 2, no topological relation and no coupling between parameters exist among all sub-networks after tearing, 3, all sub-networks are minimum sub-networks relative to the torn node, all sub-networks are only used for enabling the node to be torn, namely, the sub-network scale is minimum, the fewer the number of faults is combined, and the higher the diagnosis rate is.

(2) Sending an initial sample set formed by the voltage of the tearing node under each fault mode to a reverse cloud generator to obtain cloud { E _x,E_n,H_e };

(3) A plurality of cloud drops under each fault mode are obtained by utilizing a forward cloud generator algorithm to restore a plurality of measurable node voltage values;

(4) Calculating membership values of each voltage characteristic and corresponding fault state under the initial sample set by using a standard cloud model, and determining cloud drops under each fault voltage characteristic according to a membership matrix of an original fault to expand the initial sample set;

(5) Dividing the expanded initial sample set into a plurality of folds for cross verification, and calculating posterior probability density values under each fault category in the test set by using a Bayesian model of Epanechnikov kernel density estimation;

(6) And classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result.

In this embodiment, the above step (2) may be implemented by:

The cloud sample generation needs a reverse cloud and forward cloud model algorithm at the same time, and for each circuit fault, the cloud can be used for describing the uncertain conversion between the qualitative concept and the fixed value of the fault mode so as to reflect the uncertainty of things or human knowledge concepts in the natural world, namely ambiguity and randomness, which not only give explanation from random theory and fuzzy set theory, but also reflect the relevance between the ambiguity and randomness, so that the mapping between quantification and qualitative is formed, as shown in fig. 4, and the generation process of the fault cloud sample is realized. Elements in a sample set in a fault mode can be represented by cloud drops, the sampled voltage is converted from a fixed value to a qualitative concept by an inverse cloud generator, and the qualitative concept is represented by a digital feature { E _x,E_n,H_e }, wherein the digital feature is expected to be E _x, entropy E _n and super-entropy H _e.

The algorithm of the one-dimensional reverse cloud generator is realized as follows:

Inputting a quantitative value of an initial fault sample set;

outputting an expected value E _x, entropy E _n and super entropy H _e of the qualitative concept A represented by cloud drops;

note the initial sample set x _ij＝[x_i1,x_i2,…,x_in of circuit faults (i=1, 2,., n; j=1, 2, m), calculating a sample mean Absolute central moment of first-order sampleSample variance

In the domain space, a cloud is the point that best represents the qualitative concept, which is expected to be the central value in the domain space, which is expected

Entropy is determined by randomness and ambiguity of qualitative concepts, and represents a measurable granularity of a qualitative concept, E _n is a measure of randomness of the qualitative concept, reflects the discrete degree of the cloud, reflects the margin of the qualitative concept, reflects the range of values of the cloud which can be accepted by the qualitative concept in the domain space, is a measure of ambiguity of the qualitative concept, and generally, the larger the entropy is, the larger the range of values of the cloud which can be accepted by the qualitative concept is, the more the qualitative concept is ambiguous, and reflects the relativity between the randomness and the ambiguity, and the entropy can be obtained by the average value of samples

Super entropy is a measure of uncertainty of entropy, revealing the cohesiveness of uncertainty of all points of language values in the domain space and the association of ambiguity and randomness, indirectly reflecting the thickness of cloud, and is obtained by sample variance and super entropy

In this embodiment, the above step (3) may be implemented by:

The forward cloud generator is a mapping from qualitative concepts to quantitative representations thereof, and generates cloud droplets according to the digital features { E _x,E_n,H_e } of the cloud obtained above, each cloud droplet being a specific implementation of the concept.

The algorithm of the one-dimensional forward cloud generator is realized as follows:

Three digital characteristic values E _x,E_n,H_e representing the qualitative concept A and a cloud drop number Num are input;

Outputting quantitative values of Num cloud drops, wherein each cloud drop represents the certainty degree of the concept A;

the generation is expected to take E _n as a matter of course, A normal random number E' _ni that is the variance;

using the generated random number, with E _x as the expectation, Generating a normal random number x _k for variance and calculating the membership of x _k Wherein x _k of u _k with certainty represents a cloud droplet in the domain;

Repeating the steps until a set number of cloud drops Num is generated, and obtaining a cloud sample set drop (x _k,u_k).

In this example, the number of droplets may be num=10000.

As shown in fig. 5, the cloud model and the sample distribution situation with 10000 cloud droplets are generated by the fault labels F0, F3, F5 and F7 respectively.

From comparison, the expectation, the certainty (membership), and the randomness (discrete degree) are seen from the high to low limit of the cloud model analysis. The range of the cloud drops accepted by each fault in the domain space and the aggregation degree of the cloud drops are greatly different, and the normal state and the fault state are greatly different.

In this embodiment, the above step (4) may be implemented by:

in order to avoid excessively long feature vectors when a fault sample set is expanded, the embodiment of the invention selects a cloud membership matrix under the feature nodes to obtain cloud droplets with larger occurrence probability and certainty as the feature vectors so as to facilitate the next step of fault mode identification.

In order to generate a cloud sample set without increasing data dimension, describing membership degrees of each fault feature vector corresponding to each fault type through a cloud model, calculating the membership degree of the j-th fault of the i-th feature vector by utilizing a standard cloud model formed under each fault, and substituting standard cloud parameters corresponding to the j-th fault of the i-th feature of an initial fault sample x _ij into the cloud modelAnd obtaining a corresponding membership degree, wherein u _ij is the membership degree of the j-th class fault of the i-th feature vector of the initial fault sample, and E _nij' is a normal random number generated by the entropy E _nij and the super entropy H _eij ² of the j-th class fault of the i-th feature vector.

Meanwhile, the uncertainty causes of data errors and external interference in the fault diagnosis process are considered, a fuzzy vector is formed by fuzzy comprehensive evaluation uncertainty indexes, and the membership calculation formula of the uncertainty is as follows: n is the fuzzy comprehensive evaluation vector (u _i1,u_i2,…,u_in), i=1, 2,..n the number of membership degrees, β is the maximum membership degree of (u _i1,u_i2,…,u_in), and γ is the second largest membership degree of (u _i1,u_i2,…,u_in). Calculating the uncertain membership degree of each fault under the membership degree to form an uncertain membership degree fuzzy characteristic vector as shown in the formula In the middle ofAn uncertain membership representing the ith feature. By integrating the above formula, the membership matrix of m rows and n+1 columns can be obtained again.

Taking the influence of external interference into consideration, collecting 200 cloud drops under each membership degree u+/-0.01, and finally obtaining 11 multiplied by 9 multiplied by 200 fault samples with 11 fault characteristic attributes and 9 fault categories.

In this embodiment, the above step (5) may be implemented by:

the sample set is cross-validated into a training set and a test set.

The epannechnikov kernel density estimation is a non-parametric probability density estimation method that can be used to estimate arbitrary distributions, while the bayesian model is a bayesian-based probability model that can be used to estimate posterior distributions of unknown parameters. In the bayesian model, unknown parameters are regarded as random variables, posterior distribution of the parameters is updated through observation data, and n independent samples X ₁,X₂,…X_n with the same distribution are obtained, so that the likelihood function expression of the Epanechnikov kernel density estimation can be obtained as follows:

Wherein h is a bandwidth parameter, determines the smoothness of a kernel function, and obtains posterior distribution of pi (f|X ₁,…,X_n)∞L(f|X₁,…,X_n) pi (f) according to Bayesian theorem;

In this bayesian model, it is assumed that the variable y is generated from an unknown probability distribution, and then an prior probability distribution is constructed using an Epanechnikov kernel function, which represents the density distribution of y throughout the input space. And combining the observed data x with the prior probability distribution through Bayesian rules to obtain posterior probability distribution. Finally, the posterior probability distribution can be used to predict the target variable value for the new data point. Unlike traditional parametric regression models, the bayesian model based on the Epanechnikov kernel function does not require any assumptions on the distribution of the target variables, and thus can better accommodate different datasets and problems.

In this embodiment, step (6) may be implemented by:

The method comprises classifying fault data by using a Bayesian model of Epanechnikov kernel density estimation, regarding each fault voltage data point as a vector, and estimating a probability density function of each fault category by using an Epanechnikov kernel density estimation method. The data points to be classified are then projected onto the probability density function of each fault class, and their probability density values under each class are calculated. Finally, the data points are classified into the category with the highest probability density value. Specifically, the present invention relates to a method for manufacturing a semiconductor device.

Regarding the fault data point to be tested as a vector, the fault data point to be tested comprises a plurality of characteristics, subtracting the characteristics in the vector from the posterior mean value of the class, dividing each element obtained by the posterior variance of the class as an independent variable, substituting the independent variable into an Epanechnikov kernel function, obtaining a new vector average value, namely a probability density value of a sample to be tested under the class, calculating the reciprocal of the square root of the determinant of the posterior variance of the class, and multiplying the probability density value obtained in the last step to normalize the probability value. And calculating the posterior probability density value of the data point to be classified under each fault category according to the Bayesian theorem, and classifying the data point to be classified into the category with the highest probability density value.

The confusion matrix of the fault classification results is shown in fig. 6, and the diagnosis accuracy is 99.4%.

According to the large-scale circuit fault diagnosis method of the Bayesian model based on the Epanechnikov kernel density estimation of the cloud theory, disclosed by the invention, a fault sample set is expanded by utilizing cloud drops generated by combining a forward cloud model and a reverse cloud model in the cloud theory, the situation that the number of samples after network tearing is small or a simulation sample is difficult to obtain under a specific condition is effectively solved, the Bayesian model based on the Epanechnikov kernel density estimation is used for carrying out pattern recognition, the probability density function of Gaussian distribution to estimate class conditions is solved, and the diagnosis degree of fault diagnosis is improved due to the fact that the probability density function is too fit caused by the small number.

As shown in fig. 7, in another embodiment of the present invention, there is further provided a system for diagnosing a large-scale circuit fault of a bayesian model of an Epanechnikov kernel density estimation based on a cloud theory, including:

The specific implementation of each module may refer to the description of the above method embodiment, and this embodiment will not be repeated.

It should be noted that each step/component described in the present application may be split into more steps/components, or two or more steps/components or part of operations of the steps/components may be combined into new steps/components, according to the implementation needs, to achieve the object of the present application.

It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.

Claims

1. A cloud theory-based method for diagnosing a large-scale circuit fault of a bayesian model of an Epanechnikov kernel density estimation, comprising the steps of:

Obtaining a plurality of cloud drops under each fault by utilizing digital characteristic expectation, entropy and super entropy of the cloud under each fault by utilizing a forward cloud generator to restore a plurality of measurable node voltage values, describing randomness and ambiguity of the fault mode and the relevance between the two by utilizing the cloud for each circuit fault mode, wherein elements in a fault sample set under the fault mode can be regarded as formed cloud drops;

classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result;

dividing the expanded initial sample set into a plurality of folds for cross-validation, calculating posterior probability density values under each fault category in the test set by using a Bayesian model of Epanechnikov kernel density estimation, wherein the method comprises the following steps:

In a Bayesian model, a likelihood function of Epanechnikov kernel density estimation can be obtained through a plurality of independent samples distributed in the test set, posterior distribution of the likelihood function of the Epanechnikov kernel density estimation is obtained according to the Bayesian theorem, and then posterior probability density values under each fault category are obtained;

The step of classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result comprises the following steps:

Projecting voltage data points of faults to be detected onto probability density functions of each fault class, calculating probability density values under each class, and classifying the voltage data points of the faults to be detected into the class with the highest probability density values;

Projecting the voltage data points of the faults to be detected onto probability density functions of each fault class, calculating probability density values under each class, and classifying the voltage data points of the faults to be detected into the class with the highest probability density values, wherein the method comprises the following steps:

Taking the fault voltage data point to be detected as a vector, wherein the vector comprises a plurality of features, subtracting the posterior mean value of the class from the features in the vector, dividing each element obtained by the posterior square difference of the class, and substituting the posterior square difference into an Epanechnikov kernel function as an independent variable to obtain a new vector average value, and obtaining the probability density value of the fault sample to be detected under the class;

2. The fault diagnosis method according to claim 1, wherein selecting the voltage of the modular tear node as the diagnosis signal comprises:

3. The fault diagnosis method according to claim 1 or 2, wherein the initial sample set of voltages of tearing nodes at different faults uses a reverse generator of a cloud model to generate digital characteristic expectations, entropy and super entropy of the cloud at each fault, comprising:

Note that the initial sample set x _ij＝[x_i1,x_i2,…,x_in of circuit faults (i=1, 2,..n; j=1, 2,..m) represents the j-th fault of the i-th fault feature, n is the dimension of the feature vector in the fault mode, and m is the total number of fault modes, and then the fault is represented by the following formula Obtaining the mean value of the initial sample set of the j-th type faultFrom the following components A first order central moment K of the initial sample set of the j-th class of faults is obtained, from the following componentsObtaining the variance S of an initial sample set of the j-th type fault;

4. a fault diagnosis method according to claim 3, wherein said obtaining a plurality of cloud droplets for each fault using a forward cloud generator comprises:

Using the generated random number, generating a random number expected by E _x, Generating a normal random number x _k for variance and calculating the x _k membershipWherein x _k of u _k with certainty represents a cloud droplet in the domain;

5. The fault diagnosis method according to claim 4, wherein calculating the membership matrix of each voltage value and the membership value of the corresponding fault state in the initial sample set using the standard cloud model comprises:

6. A cloud theory-based, bayesian model based system for diagnosing faults in a large scale of circuit for estimating the density of an Epanechnikov kernel, comprising:

The fault diagnosis module is used for dividing the expanded initial sample set into a plurality of fold cross-validation, calculating posterior probability density values under each fault category in the test set by using a Bayesian model of Epanechnikov kernel density estimation, and classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result;

The method for calculating the posterior probability density value under each fault category in the test set by using the Bayesian model of the Epanechnikov kernel density estimation comprises the following steps:

7. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 5.