CN115688017A - FRCMDE-based transformer core fault voiceprint diagnosis method and device - Google Patents

Abstract

The invention belongs to the technical field of power equipment fault diagnosis, and discloses a transformer core fault voiceprint diagnosis method and device based on FRCMDE (frequency-dependent modulation and noise suppression), wherein the method comprises the steps of carrying out successive variation modal decomposition on transformer voiceprint data to obtain a plurality of intrinsic modal components, calculating the kurtosis and energy ratio of each intrinsic modal component, selecting the intrinsic modal components containing useful information, and reconstructing the intrinsic modal components; solving the FRCMDE value of the reconstructed signal in an analysis scale, and removing redundant and invalid parts in the FRCMDE value under different scales by using a Fisher ratio to construct an optimal feature subset; and constructing an improved PODSDOA-LSSVM fault diagnosis model to identify FRCMDE characteristics and output a diagnosis result. The invention can help electric power workers to master the running state of the transformer in time, know latent faults in advance and avoid loss caused by equipment faults.

Description

FRCMDE-based Voiceprint Diagnosis Method and Device for Transformer Core Fault

technical field

The invention relates to the technical field of power equipment faults, in particular to a FRCMDE-based voiceprint diagnosis method and device for transformer iron core faults.

Background technique

As the core equipment in the power system, transformers are increasingly exposed to problems such as state perception and fault warning under complex changing conditions. Once a fault occurs, it will cause serious power supply and distribution problems and major economic losses. Therefore, there is an urgent need for an effective fault state diagnosis method to detect and diagnose the operating state of the transformer in time to ensure its safe and reliable operation in the complex environment of the new power system.

The common faults of transformers mainly include mechanical faults, insulation faults and overheating faults. Among them, the frequency of electrical faults caused by mechanical faults is the highest, and iron core faults account for a large proportion. According to statistical research, iron core faults are in the Most of them are caused by mechanical structural problems at the beginning, and as the fault degree deepens, it will threaten and damage the clamping force and insulation of the iron core, and even lead to serious electrical faults such as multi-point grounding of the iron core. The sound signal during transformer operation contains a large amount of operating state information, and the power transformer fault diagnosis method based on voiceprint analysis has received sufficient attention. This method has the characteristics of comprehensive state perception, efficient information processing, convenient and flexible application, etc. It can effectively improve the fault identification level of power transformers, reduce the failure probability of power transformers, and effectively prevent and reduce major accidents caused by transformer failures.

Contents of the invention

In order to solve the technical problems mentioned in the above-mentioned background technology, the present invention proposes a transformer core fault voiceprint diagnosis method based on FRCMDE, which decomposes the transformer voiceprint data through Successive Variational Mode Decomposition (SVMD) to obtain several Intrinsic mode components (IMF), calculate the kurtosis and energy ratio of each eigenmode component, select the intrinsic mode component (IMF) containing useful information and reconstruct it; find the reconstructed signal in Analyze the fractional-order fine compound multi-scale distribution entropy (FRCMDE) within the scale, use Fisher ratio to eliminate redundant and invalid parts in the fractional-order fine compound multi-scale distribution entropy at different scales, and construct the optimal feature subset; use PODSBOA Optimize the parameters of fractional-order fine composite multi-scale scatter entropy and hyperparameters of LSSVM and build an improved PODSBOA-LSSVM fault diagnosis model to identify the optimal feature subset and output the diagnosis results. The method can be used for non-contact measurement, simple equipment installation, fast measurement speed, easy signal measurement, and will not interfere with the normal operation of the equipment.

In order to achieve the above object, a technical solution adopted by the present invention is: a method for diagnosing voiceprints of transformer iron core faults based on FRCMDE, comprising the following steps:

S1. Decompose the voiceprint data of the transformer successively to obtain several eigenmode components, calculate the kurtosis and energy proportion of each eigenmode component, select the eigenmode components containing useful information, and to refactor;

S2. Calculate the fractional-order fine compound multi-scale distribution entropy of the reconstructed signal within the analysis scale;

S3. Use the Fisher ratio to eliminate redundant and invalid parts in the fractional fine composite multi-scale distribution entropy at different scales, and construct a feature subset of the optimal scale;

S4. Use the PODSBOA algorithm to optimize the super-parameters of LSSVM. Based on the obtained optimal feature subset, construct a PODSBOA-LSSVM fault diagnosis model, diagnose the unknown transformer voiceprint data, and output the diagnosis result.

Further preferably, the specific process of step S2 is as follows:

的不重叠区域，并求每个区域的平均值，以此得到粗粒化序列； S201: Let the signal sequence X =[ x ₁ , x ₂ ,… x _n ] reconstructed by successive variational mode decomposition, n is the length of the signal sequence X x _n is the nth data, and the entropy is distributed in the compound multi-scale In the algorithm, the initial point is [1, τ ] and the signal is continuously divided into lengths

The non-overlapping areas of , and calculate the average value of each area, so as to obtain the coarse-grained sequence;

S202: Calculate the average value of the probability of the corresponding distribution pattern of the coarse-grained sequence:

Use the standard normal distribution function to map the coarse-grained sequence to the mapping sequence Y = [ y ₁ , y ₂ ,..., y _n ] in the range of [0,1], where y _n is the nth mapping signal; that is

In the formula, μ , σ are the expectation and standard deviation of signal sequence X, t is time, x _i is the i-th signal after SVMD decomposition and reconstruction, and y _i is the i-th mapped signal;

Through the linear transformation algorithm, the mapping sequence Y is mapped to an integer in the range [1, c ] to obtain a linear transformation signal, namely

为第i个线性变换信号； In the formula, round is a rounding function; c is the number of categories;

is the ith linear transformation signal;

进行相空间重构，则嵌入向量为： right

For phase space reconstruction, the embedding vector is:

映射到一个散布模式r _v0, _v1⋯vm-1，其中，

, 由于散布模式r _v0, _v1⋯vm-1内含有m个元 素, 每个元素可取[1,c]的任意整数，所有可能的散布模式的数量为c ^m ； In the formula, m is the embedding dimension, d is the time delay, each

Maps to a scatter pattern r _{v 0} , _{v 1⋯ vm -1} , where,

, since the scatter pattern r _{v 0} , _{v 1⋯ vm -1} contains m elements, each element can be any integer of [1,c], the number of all possible scatter patterns is c ^m ;

Calculate the probability P of each scatter mode r _{v 0} , _{v 1⋯ vm -1} ^{under cm} , namely:

对应的散布模式的个数； In the formula, Num ( r _{v 0} , _{v 1⋯ vm -1} ) is

The number of corresponding scatter patterns;

Calculate the average value of the probability of the corresponding scatter pattern of coarse-grained sequences of different scales, namely:

为尺度τ下的第k个粗粒化序列对应的散步模式的概率,

(r _v0v1⋯vm-1)不同尺度下粗粒化序列对应的散步模式的平均概率； In the formula,

is the probability of the scatter mode corresponding to the kth coarse-grained sequence under the scale τ ,

( r _{v 0 v 1⋯ vm -1} ) the average probability of the spread pattern corresponding to the coarse-grained sequence at different scales;

S203: Calculating the fractional fine compound multi-scale distribution entropy at different scales, namely:

即

(r _v0v1⋯vm-1)，

为gamma函数，Ψ为digamma函数，α为分数阶因子。 In the formula,

Right now

( r _{v 0 v 1⋯ vm -1} ),

is the gamma function, Ψ is the digamma function, and α is the fractional order factor.

Further preferably, the specific process of step S3 is as follows:

S301: Calculate the Fisher ratio for the feature vector set formed by the fractional-order fine compound multi-scale dispersal entropy of the transformer voiceprint data, namely:

表示第k维特征向量的类间离 散度，

表示第k维特征向量的类内离散度； In the formula, F _{( k )} represents the Fisher ratio of the feature vector of the k -th dimension,

Indicates the inter-class dispersion of the k -th dimensional feature vector,

Indicates the intra-class dispersion of the k- dimensional feature vector;

S302: Sort the Fisher ratio, and select a feature subset with an optimal scale.

Further preferably, the specific process of step S4 is as follows:

S401: Divide the optimal scale feature subset into a training set and a test set;

S402: The improved butterfly algorithm uses the training set pair and the penalty factor c of the least squares support vector machine and the parameter g of the radial basis inner product function to optimize to obtain the optimal parameters;

S403: Train the least squares support vector machine with optimal parameters, and use the test set to test;

S404: Construct a PODSBOA-LSSVM fault diagnosis model according to the training test results, use the PODSBOA-LSSVM diagnosis model to diagnose the unknown transformer voiceprint data, and output a diagnosis result.

Further preferably, the steps of the improved butterfly algorithm are as follows:

Step A1: Initialize the search parameters of the butterfly algorithm: set the number of butterfly populations to N , set the maximum number of iterations of the algorithm to N ₁ , the population boundary conditions [ L _b , U _b ], and the optimization problem dimension dim ;

Step A2: Generate the initial butterfly population according to the boundary conditions: in the boundary range, use random numbers to generate an initial butterfly population of N *dim size, and expand the initial population size to 2 N*dim through spatial symmetry;

Step A3: Calculation of fitness: According to the fitness criterion function, calculate the individual fitness of the expanded population of butterflies;

Step A4: Population recovery: select N individuals with the best fitness through the elite retention strategy and record them as the recovery population, find and record the best individual in the current recovery population;

Step A5: Inferior population update: select the two butterfly individuals with the worst fitness, and perform cross processing and mutation operations on them;

Step A6: Dynamic update of algorithm parameters: update sensory modality β , power exponent a , dynamic search switching probability p and position update operators w ₁ and w ₂ according to the current iteration number according to the following formula;

Step A7: Iterative optimization: if the dynamic search switching probability p>rand , rand is a random number between 0 and 1, the individual's position is updated globally; if the dynamic search switching probability p<rand , the individual's position is locally updated update; update the global optimum;

Step A8: Boundary check: check whether the updated individual is out of the boundary, and perform boundary correction on the position of the new individual that exceeds the boundary;

Step A9: Determine whether the current iteration end condition of the algorithm is satisfied: if the end condition is not met, the algorithm proceeds to step A5 and continues to execute; otherwise, the current optimal result is output, and the algorithm ends.

The present invention also provides a FRCMDE-based transformer core fault voiceprint diagnosis device, which includes a non-volatile computer storage medium, the computer storage medium stores computer-executable instructions, and the computer-executable instructions can execute the above-mentioned FRCMDE-based transformer Iron core fault soundprint diagnosis method.

The FRCMDE-based transformer core fault voiceprint diagnosis device provided by the present invention includes electronic equipment, and the electronic equipment includes: one or more processors and memory, and also includes an input device and an output device, a processor, a memory, an input device and The output device is connected through a bus or other means, the memory is a non-volatile computer-readable storage medium, and the processor executes various functional applications of the server by running non-volatile software programs, instructions and modules stored in the memory. Data processing, that is, to implement the FRCMDE-based transformer core fault voiceprint diagnosis method, the input device receives input digital or character information, and generates keys related to user settings and function control of the FRCMDE-based transformer core fault voiceprint diagnosis device signal input.

The beneficial effects of the present invention are as follows: firstly, the superiority of SVMD in signal processing and the effectiveness of FRCMDE in fault feature extraction are combined to effectively filter out the noise interference components in the signal, and learn from the advantages of fractional calculus, It is introduced into the fine compound multi-scale distribution entropy to improve the sensitivity of entropy features to transformer sound fault characteristics; secondly, the Fisher ratio is used to construct the optimal feature subset to reduce the feature dimension; the PODSBOA algorithm is used to optimize the hyperparameters of LSSVM, An improved PODSBOA-LSSVM diagnostic model is constructed to diagnose transformer fault noise signals, which effectively improves the accuracy of transformer fault diagnosis.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Fig. 2 is a schematic structural diagram of an electronic device.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

As shown in Figure 1, the FRCMDE-based transformer core fault voiceprint diagnosis method of the present embodiment includes the following steps:

S1. Perform successive variational mode decomposition (SVMD) on the transformer voiceprint data to obtain several intrinsic mode components (IMF), calculate the kurtosis and energy ratio of each intrinsic mode component and select Intrinsic Mode Components (IMF) containing useful information and reconstructed;

S2. Calculate the fractional-order fine composite multi-scale dispersion entropy (FRCMDE value) of the reconstructed signal within the analysis scale;

S3. Use the Fisher ratio to eliminate redundant and invalid parts in the fractional fine composite multi-scale distribution entropy at different scales, and construct a feature subset of the optimal scale;

S4. Use the PODSBOA algorithm to optimize the super-parameters of LSSVM. Based on the obtained optimal feature subset, construct a PODSBOA-LSSVM fault diagnosis model, diagnose the unknown transformer voiceprint data, and output the diagnosis result.

In step S1 of this embodiment, the transformer voiceprint data is subjected to successive variational mode decomposition (SuccessiveVariational Mode Decomposition, SVMD) to obtain several intrinsic mode components (IMF), and the kurtosis sum of each intrinsic mode component is calculated Energy proportioning picks out Intrinsic Mode Components (IMFs) that contain useful information and reconstructs them. The specific process is:

S101: Performing Sequential Variational Mode Decomposition (SVMD) on the transformer voiceprint data to obtain several Intrinsic Mode Components (IMF);

S102: Calculate the kurtosis of each intrinsic mode component (IMF), the calculation formula of which is:

In the formula: y _b is the data value of the bth position of the i -th eigenmode component; μ _i is the mean value of the i -th eigenmode component; σ _i is the standard value of the i -th eigenmode component difference; u _i is the number of data of the ith eigenmode component;

S103: Calculate the energy proportion of each intrinsic mode component (IMF), the calculation formula of which is:

In the formula: e _i is the energy of the i -th eigenmode component (IMF), and M represents the number of eigenmode components (IMF).

S104: Select an eigenmode component containing useful information by using a kurtosis criterion and an energy ratio criterion to perform signal reconstruction.

The specific process of step S2 in the present embodiment is as follows:

的不重叠区域，并求每个区域的平均值，以此得到粗粒化序列。即： S201: Let the signal sequence X after SVMD decomposition and reconstruction =[ x ₁ , x ₂ ,… x _n ], n is the length of the signal sequence X x _n is the nth data, in the RCMDE algorithm (composite multi-scale distribution entropy) In , the initial point is to continuously divide the signal into lengths according to [1, τ ]

The non-overlapping regions of , and calculate the average value of each region, so as to obtain the coarse-grained sequence. Right now:

为信号序列X粗粒化后的序列，k为同一尺度下的序列号，j为信号序列被 划分的等长段数，也是粗粒化序列的数据长度； In the formula,

is the coarse-grained sequence of the signal sequence X, k is the sequence number at the same scale, j is the number of equal-length segments into which the signal sequence is divided, and is also the data length of the coarse-grained sequence;

S202: Calculate the average value of the probability of the corresponding distribution pattern of the coarse-grained sequence:

(1) Use the standard normal distribution function to map the coarse-grained sequence x to the mapping sequence Y in the range of [0,1] = [ y ₁ , y ₂ ,⋯, y _n ], y _n is the nth mapping signal ;Right now

In the formula, μ , σ are the expectation and standard deviation of signal sequence X, t is time, x _i is the i-th signal after SVMD decomposition and reconstruction, and y _i is the i-th mapped signal;

(2) Through the linear transformation algorithm, the mapping sequence Y is mapped to integers in the range [1, c] to obtain the linear transformation signal, namely

为第i个线性变换信号。 In the formula, round is a rounding function; c is the number of categories;

is the ith linear transformation signal.

进行相空间重构，则嵌入向量为： (3) yes

For phase space reconstruction, the embedding vector is:

映射到一个散布模式r _v0, _v1⋯vm-1，其中，

, 由于散布模式r _v0, _v1⋯vm-1内含有m个元素, 每个元素可取[1,c]的任意整数，因此所有可能的散布模式的数量为c ^m 。 In the formula, m is the embedding dimension, d is the time delay, each

Maps to a scatter pattern r _{v 0} , _{v 1⋯ vm -1} , where,

, since the scatter pattern r _{v 0} , _{v 1⋯ vm -1} contains m elements, and each element can be any integer of [1,c], so the number of all possible scatter patterns is c ^m .

(4) Calculate the probability P of each scatter pattern r _{v 0} , _{v 1⋯ vm -1} ^{under cm} , namely

对应的散布模式的个数。 In the formula, Num ( r _{v 0} , _{v 1⋯ vm -1} ) is

The number of corresponding scatter patterns.

(5) Calculate the average value of the corresponding distribution pattern probabilities of coarse-grained sequences of different scales, namely:

为尺度τ下的第τ个粗粒化序列对应的散步模式的概率,

(r _v0v1⋯vm-1)不同尺度下粗粒化序列对应的散步模式的平均概率； In the formula,

is the probability of the scatter mode corresponding to the τth coarse-grained sequence under the scale τ ,

( r _{v 0 v 1⋯ vm -1} ) the average probability of the spread pattern corresponding to the coarse-grained sequence at different scales;

S203: Calculate the fractional order fine composite multi-scale dispersion entropy (FRCMDE value) at different scales, namely

即

(r _v0v1⋯vm-1)，

为gamma函数，Ψ为digamma函数，α为分数阶因子。 In the formula,

Right now

( r _{v 0 v 1⋯ vm -1} ),

is the gamma function, Ψ is the digamma function, and α is the fractional order factor.

In this embodiment, m is usually 2 or 3, c is between [4,8], and delay d is generally 1.

In the present embodiment, the specific process of step S3 is as follows:

S301: Calculate the Fisher ratio for the eigenvector set formed by the fractional-order fine composite multi-scale spread entropy of the transformer voiceprint data, namely

表示第k维特征向量的类间离 散度，

表示第k维特征向量的类内离散度； In the formula, F _{( k )} represents the Fisher ratio of the feature vector of the k -th dimension,

Indicates the inter-class dispersion of the k -th dimensional feature vector,

Indicates the intra-class dispersion of the k-th dimension feature vector;

S302: Sort the Fisher ratio, and select a feature subset with an optimal scale.

In this embodiment, the specific process of step S4 is as follows:

S401: Divide the optimal scale feature subset into a training set and a test set;

S402: The improved butterfly algorithm (PODSBOA) is optimized by using the training set pair and the penalty factor c of the least squares support vector machine (LSSVM) and the parameter g of the radial basis inner product function to obtain the optimal parameters;

S403: Train a least squares support vector machine (LSSVM) with optimal parameters, and use a test set for testing;

S404: Construct a PODSBOA-LSSVM fault diagnosis model according to the training test results, use the PODSBOA-LSSVM diagnosis model to diagnose the unknown transformer voiceprint data, and output a diagnosis result.

In this embodiment, the improved butterfly algorithm described in step S4 is based on population optimization strategies such as optimizing the initial population, improving disadvantaged populations, introducing dynamic search parameters that can be adaptively adjusted, and introducing a variable weight position update factor strategy to combine the butterfly algorithm. Optimization, combining the advantages of the two optimization strategies, a butterfly algorithm based on population optimization and dynamic parameter strategy is formed. The steps of the improved butterfly algorithm are as follows:

Step A1: Initialize the search parameters of the butterfly algorithm: set the number of butterfly populations to N , set the maximum number of iterations of the algorithm to N ₁ , the population boundary conditions [ L _b , U _b ], and the optimization problem dimension dim ;

Step A2: Generate the initial butterfly population according to the boundary conditions: in the boundary range, use random numbers to generate an initial butterfly population of N *dim size, and expand the initial population size to 2 N*dim through spatial symmetry;

Step A3: Calculation of fitness: According to the fitness criterion function, calculate the individual fitness of the expanded population of butterflies;

Step A4: Population recovery: select N individuals with the best fitness through the elite retention strategy and record them as the recovery population, find and record the best individual in the current recovery population;

Step A5: Inferior population update: select the two butterfly individuals with the worst fitness, and perform cross processing and mutation operations on them;

Step A6: Dynamic update of algorithm parameters: update sensory modality β , power exponent a , dynamic search switching probability p and position update operators w ₁ and w ₂ according to the current iteration number according to the following formula;

Step A7: Iterative optimization: if the dynamic search switching probability p>rand , rand is a random number between 0 and 1, the individual's position is updated globally; if the dynamic search switching probability p<rand , the individual's position is locally updated update; update the global optimum;

Step A8: Boundary check: check whether the updated individual is out of the boundary, and perform boundary correction on the position of the new individual that exceeds the boundary;

Step A9: Determine whether the current iteration end condition of the algorithm is satisfied: if the end condition is not met, the algorithm proceeds to step A5 and continues to execute; otherwise, the current optimal result is output, and the algorithm ends.

The embodiment of the present invention also provides a FRCMDE-based transformer core fault voiceprint diagnosis device, including a non-volatile computer storage medium, the computer storage medium stores computer-executable instructions, and the computer-executable instructions can execute any of the above-mentioned embodiments FRCMDE-based voiceprint diagnosis method for transformer core faults.

The FRCMDE-based transformer core fault voiceprint diagnosis device provided in this embodiment includes an electronic device. FIG. 2 is a schematic structural diagram of the electronic device provided by the embodiment of the present invention. The electronic device includes: one or more processors 100 and For the memory 200, a processor 100 is taken as an example in FIG. 2 . The electronic device may further include: an input device 300 and an output device 400 . The processor 100, the memory 200, the input device 300, and the output device 400 may be connected via a bus or in other ways. In FIG. 2, connection via a bus is taken as an example. The memory 200 is the above-mentioned non-volatile computer-readable storage medium. The processor 100 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 200, that is, implements the FRCMDE-based voiceprint diagnosis method for transformer core faults. The input device 300 can receive input digital or character information, and generate key signal input related to user settings and function control of the FRCMDE-based transformer iron core fault voiceprint diagnosis device. The output device 400 may include a display device such as a display screen.

The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place , or can also be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative effort.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (7)

1. A transformer core fault voiceprint diagnosis method based on FRCMDE is characterized by comprising the following steps:

s1, carrying out successive variational modal decomposition on transformer voiceprint data to obtain a plurality of intrinsic modal components, calculating the kurtosis and energy ratio of each intrinsic modal component, selecting the intrinsic modal components containing useful information, and reconstructing the intrinsic modal components;

s2, solving a fractional order fine composite multi-scale dispersion entropy of the reconstructed signal in an analysis scale;

s3, removing redundant and invalid parts in fractional order fine composite multi-scale scattered entropies under different scales by using a Fisher ratio, and constructing a feature subset of an optimal scale;

and S4, optimizing the hyperparameters of the LSSVM by using a PODSOA algorithm, constructing a PODSOA-LSSVM fault diagnosis model on the basis of the obtained optimal feature subset, diagnosing the voiceprint data of the unknown transformer and outputting a diagnosis result.

2. The FRCMDE-based transformer core fault voiceprint diagnostic method of claim 1, wherein the specific process of step S2 is as follows:

s201: setting successive variation modal decomposition reconstructed signal sequenceX=[x ₁ ,x ₂ ,⋯x _n ]N is the length X of the signal sequence X _n For the nth data, in the composite multi-scale entropy-spread algorithm, the initial points are set to be the values in [1,τ]continuously dividing the signal into lengths ofτAnd averaging each region to obtain a coarse grained sequence;

s202: calculating the average of the corresponding scatter pattern probabilities for the coarse grained sequence:

mapping the coarse grained sequence to [0,1 ] using a standard normal distribution function]Mapping sequences within a rangeY＝[y ₁ , y ₂ ,⋯,y _n ]，y _n Mapping the signal for the nth; namely, it is

In the formula (I), the compound is shown in the specification,μ，σfor the expected and standard deviation of the signal sequence X,tin the form of a time, the time,x _i the reconstructed ith signal is decomposed for SVMD,y _i mapping the signal for the ith;

mapping sequence is converted by linear transformation algorithmYThe mapping is to be made to a1,c]among integers within the range, a linear transformation signal is obtained, i.e.

In the formula (I), the compound is shown in the specification,roundis a rounding function;cis a classCounting;

is as followsiA linear transform signal;

for is to

And performing phase space reconstruction, wherein the embedded vector is as follows:

in the formula (I), the compound is shown in the specification,min order to embed the dimension number, the number of the embedded dimension,dfor time delay, each

Mapping to a scatter patternr _v0 , _{v vm1⋯-1} Wherein, in the process,

due to the scattering patternr _v0 , _{v vm1⋯-1} Contains thereinmAn element, each element being taken as [1, c ]]Of all possible scattering patterns isc ^m ；

Computingc ^m Each scatter patternr _v0 , _{v vm1⋯-1} Probability of (2)PNamely:

in the formula (I), the compound is shown in the specification,Num(r _v0 , _{v vm1⋯-1} ) Is composed of

The number of corresponding scattering patterns;

calculating the average value of the probability of the corresponding scattering mode of the coarse graining sequences with different scales, namely:

in the formula (I), the compound is shown in the specification,

is a scaleτFirst ofkThe probability of the corresponding walking pattern of a coarse grained sequence,

(r _{v v vm01⋯-1} ) Average probability of the corresponding walking mode of the coarse graining sequence under different scales;

s203: calculating fractional order fine composite multi-scale dispersion entropy at different scales, namely:

in the formula (I), the compound is shown in the specification,

namely, it is

(r _{v v vm01⋯-1} )，

Is composed ofgammaThe function of the function is that of the function,Ψis composed ofdigammaThe function of the function(s) is,αis a fractional order factor.

3. The FRCMDE-based transformer core fault voiceprint diagnostic method of claim 1, wherein the specific process of step S3 is as follows:

s301: calculating a Fisher ratio for a feature vector set formed by fractional order fine composite multi-scale dispersion entropy of transformer voiceprint data, namely:

in the formula (I), the compound is shown in the specification,F _k() denotes the firstkThe Fisher ratio of the feature vector of the dimension,

is shown askThe inter-class dispersion of the dimensional feature vectors,

is shown askIntra-class dispersion of dimensional feature vectors;

s302: and sorting the Fisher ratios, and selecting the feature subset with the optimal scale.

4. The FRCMDE-based transformer core fault voiceprint diagnostic method of claim 1, wherein the specific process of step S4 is as follows:

s401: dividing the feature subset with the optimal scale into a training set and a testing set;

s402: improved butterfly algorithm utilizes punishment factors of training set pairs and least square support vector machinecAnd radial basis inner product function parametersgOptimizing to obtain optimal parameters;

s403: training a least square support vector machine with optimal parameters, and testing by using a test set;

s404: and constructing a PODSDOA-LSSVM fault diagnosis model according to the training test result, diagnosing unknown transformer voiceprint data by using the PODSDOA-LSSVM fault diagnosis model, and outputting a diagnosis result.

5. The FRCMDE-based transformer core fault voiceprint diagnostic method of claim 4, wherein the modified butterfly algorithm comprises the steps of:

step A1: butterfly pairInitializing butterfly algorithm search parameters: set butterfly population quantity asNSetting the maximum iteration number of the algorithm asN ₁ Population boundary condition [ 2 ]L _b , U _b ]Optimization problem dimensiondim；

Step A2: generating an initial butterfly population according to boundary conditions: in the boundary range, random number is adopted to generate N*dimAn initial butterfly population of a size that is scaled up to 2 by spatially symmetric amplification of the initial populationN*dim；

Step A3: and (3) fitness calculation: calculating the individual fitness of the butterfly of the amplified population according to a fitness criterion function;

step A4: and (3) population recovery: selection by Elite Retention strategyNRecording the individuals with the best fitness as a recovery population, and finding and recording the best individuals of the current recovery population;

step A5: updating the inferior population: selecting two butterfly individuals with the worst fitness, and performing cross processing and mutation operation on the two butterfly individuals;

step A6: and (3) dynamically updating algorithm parameters: updating the sensory modality according to the current iteration number and the following formulaβPower index ofaDynamic search switching probabilitypAnd location update operatorw ₁ 、w ₂ ；

Step A7: iterative optimization: if dynamic search switching probabilityp>rand，randGlobally updating the position of the individual for a random number between 0 and 1; if dynamic search switching probabilityp<randLocally updating the position of the individual; updating the global optimum;

step A8: and (3) border crossing inspection: checking whether the updated individual exceeds the boundary, and performing boundary correction on the position of a new individual exceeding the boundary;

step A9: judging whether the iteration end condition of the algorithm is met or not at present: if the end condition is not met, the algorithm is switched to the step A5 to be continuously executed; otherwise, outputting the current optimal result, and ending the algorithm.

6. An FRCMDE-based transformer core fault voiceprint diagnostic apparatus, comprising a non-volatile computer storage medium having computer-executable instructions stored thereon, the computer-executable instructions being capable of performing the FRCMDE-based transformer core fault voiceprint diagnostic method of any one of claims 1 to 5.

7. The FRCMDE-based transformer core fault voiceprint diagnostic apparatus of claim 6 comprising an electronic device comprising one or more processors and memory, and further comprising an input device and an output device, wherein the processors, memory, input device and output device are connected by a bus or other means, the memory is a non-volatile computer readable storage medium, the processor executes various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, and the input device receives input digital or character information and generates key signal inputs related to user settings and function control of the FRCMDE-based transformer core fault voiceprint diagnostic apparatus.

Images