CN115062733B - Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm - Google Patents

Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm Download PDF

Info

Publication number
CN115062733B
CN115062733B CN202210998691.3A CN202210998691A CN115062733B CN 115062733 B CN115062733 B CN 115062733B CN 202210998691 A CN202210998691 A CN 202210998691A CN 115062733 B CN115062733 B CN 115062733B
Authority
CN
China
Prior art keywords
transformer
algorithm
butterfly
population
fault diagnosis
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.)
Active
Application number
CN202210998691.3A
Other languages
Chinese (zh)
Other versions
CN115062733A (en
Inventor
许志浩
高家通
丁贵立
康兵
王宗耀
唐健耀
施嘉兵
袁净帅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanchang Zuochen Technology Co ltd
Jiangxi Paiyuan Technology Co ltd
Original Assignee
Nanchang Zuochen Technology Co ltd
Jiangxi Paiyuan Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanchang Zuochen Technology Co ltd, Jiangxi Paiyuan Technology Co ltd filed Critical Nanchang Zuochen Technology Co ltd
Priority to CN202210998691.3A priority Critical patent/CN115062733B/en
Publication of CN115062733A publication Critical patent/CN115062733A/en
Application granted granted Critical
Publication of CN115062733B publication Critical patent/CN115062733B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1209—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing using acoustic measurements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention belongs to the technical field of power equipment fault diagnosis, and discloses a transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm, wherein noise-containing transformer voiceprint data is subjected to complete set empirical mode decomposition, and environmental noise is filtered by using a spectral subtraction method to obtain pure transformer voiceprint data; decomposing the denoised transformer voiceprint data by using complete set empirical mode decomposition, extracting combined characteristics of composite multi-scale entropy, fuzzy entropy and kurtosis entropy, and reducing dimensions of the combined characteristics by using t-SNE; and finally, constructing an improved PODSBOA-LSSVM fault diagnosis model based on the combined characteristics, diagnosing the voiceprint of the unknown transformer and outputting a diagnosis result. The invention can help electric power staff to master the running state of the transformer in time, know latent faults in advance and avoid loss caused by equipment faults.

Description

Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm
Technical Field
The invention relates to the technical field of power equipment faults, in particular to a transformer voiceprint fault diagnosis method based on empirical mode decomposition and a butterfly algorithm.
Background
After years of operation, the failure probability of the transformer is increased continuously, and the risk of various failures such as insulation aging, component looseness and the like exists. As the power equipment used for electric energy conversion in the system, the number of transformers is large, the running time is long, so that the number of the transformers with faults is larger, the transformers with faults are also larger, the economic loss is caused by only replacing the equipment, and the indirect economic loss caused by the power interruption of the replaced equipment is larger. Transformers are important devices in power systems and it is important to ensure their safe operation. Therefore, the method and the device can detect and diagnose the running state of the transformer, eliminate the hidden danger of the transformer in time and have important significance for the development of a power system.
The main reasons of the transformer faults are that insulation is damaged, insulation aging, maintenance errors and production defects can cause the insulation of the transformer to be damaged by suspension potential, bubble residue and the like, the local field intensity of the insulation is overhigh, and a discharge phenomenon is generated. The discharge causes a further increase in insulation breakdown, eventually leading to breakdown. Loose parts in the transformer or small metal parts falling into the transformer may create floating potentials and also cause discharges. Some mechanical faults may damage the transformer insulation and even result in fault currents and loss of active power. Therefore, it is necessary to perform detection of a failure before causing a larger loss.
Disclosure of Invention
In order to solve the technical problems mentioned in the background art, the invention provides a transformer voiceprint fault diagnosis method based on empirical mode decomposition and a butterfly algorithm, which comprises the steps of carrying out complete set empirical mode decomposition (CEEMDAN) decomposition on voiceprint data of a noise-containing transformer to obtain a plurality of intrinsic mode components (IMF), calculating the kurtosis of each intrinsic mode component to find out the intrinsic mode component (IMF) containing noise and reconstruct the intrinsic mode component (IMF), and filtering environmental noise of a reconstructed signal by using spectral subtraction to obtain pure transformer voiceprints; extracting composite multi-scale entropy, fuzzy entropy and kurtosis entropy of each intrinsic mode component (IMF) as joint features, and performing dimensionality reduction on the joint features by using t-SNE; and finally, constructing an improved PODSOA-LSSVM fault diagnosis model, identifying the combined characteristics and outputting a diagnosis result. The method can realize non-contact measurement, is simple in equipment installation, fast in measurement speed, easy in signal measurement and free from interference on normal operation of equipment.
In order to achieve the above purpose, the invention adopts a technical scheme that: a transformer voiceprint fault diagnosis method based on complete set empirical mode decomposition and butterfly algorithm comprises the following steps:
s1, filtering environmental noise by applying CEEMDAN decomposition and spectral subtraction: carrying out complete set empirical mode decomposition (CEEMDAN) on the voiceprint data of the noise-containing transformer and filtering environmental noise by using a spectral subtraction method to obtain pure voiceprint data of the transformer;
s2, extracting the combined features and performing dimensionality reduction by using t-SNE: decomposing the denoised transformer voiceprint data by completely collecting empirical mode decomposition (CEEMDAN) to obtain a plurality of intrinsic mode components (IMF), extracting composite multi-scale entropy, fuzzy entropy and kurtosis entropy of each intrinsic mode component (IMF) to construct a joint feature, and reducing the dimension of the joint feature by using a t-SNE algorithm;
s3, constructing an improved PODSDOA-LSSVM fault diagnosis model for diagnosis: and (3) performing parameter optimization on a Least Square Support Vector Machine (LSSVM) by using an improved butterfly algorithm (PODSOA) on the basis of the obtained data set of the combined characteristics, constructing a PODSOA-LSSVM fault diagnosis model, diagnosing unknown transformer voiceprint data and outputting a diagnosis result.
Further, in the step S1, complete set empirical mode decomposition (CEEMDAN) is performed on the noisy transformer voiceprint data to obtain a plurality of intrinsic mode components (IMF), the kurtosis of each intrinsic mode component (IMF) is calculated to find out the noisy intrinsic mode component and reconstruct the noisy intrinsic mode component, and the reconstructed signal is used to filter the environmental noise by using spectral subtraction to obtain pure transformer voiceprint data.
Further, the specific process of step S1 is as follows:
s101: carrying out complete set empirical mode decomposition (CEEMDAN) on the voiceprint data of the noise-containing transformer to obtain a plurality of intrinsic mode components (IMF);
s102: calculating the kurtosis of each intrinsic mode component (IMF), wherein the calculation formula is as follows:
Figure 223995DEST_PATH_IMAGE001
in the formula:y b is as followsiOf the intrinsic mode componentbA data value for each location;μ i is a firstiA mean of the individual eigenmode components;σ i is as followsiStandard deviation of the individual eigenmode components;u i is a firstiThe number of data of each eigenmode component; selecting a proper kurtosis value to find out intrinsic mode components containing noise for signal reconstruction;
s103: extracting the environmental noise from the reconstructed intrinsic mode component containing the noise by adopting a spectral subtraction method, and performing normalization processing;
s104: and the extracted environmental noise is subtracted from the noise-containing transformer voiceprint data to obtain pure transformer voiceprint data, and the signal-to-noise ratio is higher.
Further, the specific process of step S2 is as follows:
s201: performing complete set empirical mode decomposition (CEEMDAN) on the de-noised transformer voiceprint data;
s202: calculating the correlation coefficient of each intrinsic modal component and the denoised transformer voiceprint datar i The calculation formula is as follows:
Figure 937873DEST_PATH_IMAGE002
in the formula (I), the compound is shown in the specification,x i is as followsiThe number of the intrinsic mode components is increased,
Figure 779928DEST_PATH_IMAGE003
is composed ofx i Is determined by the average value of (a) of (b),y i for the second time after noise eliminationiThe voice print data of each transformer is stored in a memory,
Figure 34191DEST_PATH_IMAGE004
is composed ofy i Average value of (a); selecting eigenmode components with correlation coefficients larger than a certain threshold value;
s203: calculating the composite multi-scale entropy, fuzzy entropy and kurtosis entropy of the selected eigenmode components to form a joint feature;
s204: and reducing the dimension of the combined feature by using a t-SNE algorithm.
Further, the specific process of step S3 is as follows:
s301: dividing a data set of joint features extracted from the transformer voiceprint data into a training set and a testing set;
s302: improved butterfly algorithm (PODSBO) utilizes a training set to penalize a Least Squares Support Vector Machine (LSSVM)cAnd radial basis inner product function parametersgOptimizing to obtain optimal parameters;
s303: training a Least Squares Support Vector Machine (LSSVM) with optimal parameters, and testing by using a test set;
s304: and building 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.
The improved butterfly algorithm (PODSBO) in the step S3 is based on population optimization strategies of optimizing an initial population, improving a disadvantaged population and the like, dynamic search parameters which can be adjusted in a self-adaptive mode are introduced, variable weight position updating factor strategies are introduced to carry out combined optimization on the butterfly algorithm, the advantages of the two optimization strategies are integrated, and the butterfly algorithm based on the population optimization and the dynamic parameter strategies is formed. The steps of the improved butterfly algorithm are as follows:
step A1: initializing butterfly algorithm search parameters: set the 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 generaten*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 following formula based on the current number of iterationsβPower index ofaDynamic search switching probabilitypAnd location update operatorw 1 、w 2 ；
Sensory modalities that are updated with an iterative process
Figure 645301DEST_PATH_IMAGE005
Power exponent updated with iterative process
Figure 202185DEST_PATH_IMAGE006
Dynamic switching probability updated with iterative process
Figure 268885DEST_PATH_IMAGE007
In the formula:tfor the current iteration numberCounting;N 1 is the maximum number of iterations.
Step A7: iterative optimization: if dynamic search switching probabilityp>rand，randGlobal updating is carried out on the position of the individual for random numbers between 0~1; if dynamic search switching probabilityp<randLocally updating the position of the individual; updating the global optimum;
step A8: and (3) border crossing checking: 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.
The invention has the beneficial effects that: firstly, filtering environmental noise by adopting CEEMDAN and spectral subtraction method, and removing the interference of the environmental noise; and secondly, performing combined optimization on the butterfly algorithm based on population optimization strategies such as initial population optimization, inferior population improvement and the like, introducing dynamic search parameters which can be adjusted in a self-adaptive manner, introducing variable weight position update factor strategies, integrating the advantages of the two optimization strategies, forming the butterfly algorithm based on the population optimization and the dynamic parameter strategies, constructing an improved PODSOA-LSSVM diagnostic model, diagnosing a fault noise signal of the transformer, and effectively improving the accuracy of fault diagnosis of the transformer.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a flow chart of extracting joint features used in the present invention.
Fig. 3 is a diagnostic flowchart of the improved PODSBOA-LSSVM diagnostic model in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the transformer voiceprint fault diagnosis method based on the complete set of empirical mode decomposition and butterfly algorithm of the embodiment includes the following steps:
s1, filtering environmental noise by applying CEEMDAN decomposition and spectral subtraction: carrying out complete set empirical mode decomposition (CEEMDAN) on the voiceprint data of the noise-containing transformer and filtering environmental noise by using a spectral subtraction method to obtain pure voiceprint data of the transformer;
s2, extracting the combined features and performing dimensionality reduction by using t-SNE: decomposing the denoised transformer voiceprint data through complete set empirical mode decomposition (CEEMDAN) to obtain a plurality of intrinsic mode components (IMF), extracting composite multi-scale entropy, fuzzy entropy and kurtosis entropy of each intrinsic mode component (IMF) to construct a combined feature, and reducing the dimension of the combined feature by using a t-SNE algorithm;
s3, constructing an improved PODSDOA-LSSVM fault diagnosis model for diagnosis: based on the obtained data set of the combined characteristics, an improved butterfly algorithm (PODSDOA) is used for performing parameter optimization on a Least Square Support Vector Machine (LSSVM), an improved PODSDOA-LSSVM fault diagnosis model is constructed, unknown transformer voiceprint data are diagnosed, and a diagnosis result is output.
In step S1 of this embodiment, complete set empirical mode decomposition (CEEMDAN) is performed on the noise-containing transformer voiceprint data to obtain a plurality of intrinsic mode components (IMF), the kurtosis of each intrinsic mode component (IMF) is calculated to find out the noise-containing intrinsic mode component and reconstruct the noise-containing intrinsic mode component, and the reconstructed signal is used to filter the environmental noise by using a spectral subtraction method, so as to obtain pure transformer voiceprint data. The specific process is as follows:
s101: carrying out complete set empirical mode decomposition (CEEMDAN) on the voiceprint data of the noise-containing transformer to obtain a plurality of intrinsic mode components (IMF);
s102: calculating the kurtosis of each intrinsic mode component (IMF), wherein the calculation formula is as follows:
Figure 326840DEST_PATH_IMAGE008
in the formula:y b is as followsiPersonal notebookCharacterizing modal componentsbA data value for each location;μ i is as followsiA mean of the individual eigenmode components;σ i is as followsiStandard deviation of the individual eigenmode components;u i is as followsiThe data number of each eigenmode component; selecting a proper kurtosis value to find out intrinsic mode components containing noise for signal reconstruction;
s103: extracting the environmental noise from the reconstructed intrinsic mode component containing the noise by adopting a spectral subtraction method, and performing normalization processing;
s104: and the extracted environmental noise is subtracted from the noise-containing transformer voiceprint data to obtain pure transformer voiceprint data, and the signal-to-noise ratio is higher.
In this embodiment, the CEEMDAN method parameters are set as follows: the average number of noise additions was taken to be 70, the amplitude of the additional noise was taken to be 0.2 of the standard deviation of the original signal, and the maximum number of iterations was taken to be 350. The spectral subtraction parameters were set as follows: estimating that the segment without environmental noise at the front segment takes 0.15s, the window length takes 200, the frame shift takes 80, and the over-subtraction factor takes 4; the gain compensation factor takes 0.001.
As shown in fig. 2, the specific process of step S2 in this embodiment is as follows:
s201: performing complete set empirical mode decomposition (CEEMDAN) on the denoised transformer voiceprint data;
s202: calculating the correlation coefficient of each intrinsic modal component and the denoised transformer voiceprint datar i The calculation formula is as follows:
Figure 58036DEST_PATH_IMAGE009
in the formula,x i is as followsiThe number of the intrinsic mode components is increased,
Figure 848137DEST_PATH_IMAGE010
is composed ofx i Is determined by the average value of (a) of (b),y i for the second time after noise eliminationiThe voice print data of each transformer is stored in a memory,
Figure 664783DEST_PATH_IMAGE011
is composed ofy i Average value of (d); selecting the eigenmode components with the correlation coefficient larger than a certain threshold value;
s203: calculating the composite multi-scale entropy, fuzzy entropy and kurtosis entropy of the selected eigenmode components to form a joint characteristic:
composite multi-scale entropy:
Figure 464112DEST_PATH_IMAGE012
in the formula:
Figure 721918DEST_PATH_IMAGE013
，τis a scale factor;kis a serial number at the same scale,
Figure 948500DEST_PATH_IMAGE014
is to time series data
Figure 49180DEST_PATH_IMAGE015
The coarse-grained treatment is carried out,x(i) As elements of time series datamIs dimension, generally 1 or 2;r for similar tolerances, the length to be determined is generally taken to beNTime series of
Figure 123971DEST_PATH_IMAGE016
Standard deviation of (2)
Figure 829759DEST_PATH_IMAGE017
10% -25% of the total weight of the composition,
Figure 899346DEST_PATH_IMAGE018
in order to compound the multi-scale entropy,
Figure 425005DEST_PATH_IMAGE019
representative sample entropy
Figure 362874DEST_PATH_IMAGE020
，
Figure 657589DEST_PATH_IMAGE021
Is that the two sequences are within a similar tolerancerThe probability of matching in the m-dimension is lowered,
Figure 491553DEST_PATH_IMAGE022
is that the two sequences are within a similar tolerancerThe probability of matching m +1 dimensions is lower.
Fuzzy entropy:
Figure 504509DEST_PATH_IMAGE023
in the formula (I), the compound is shown in the specification,
Figure 121435DEST_PATH_IMAGE024
is of length ofNThe time series of (a) and (b),x(i) In the form of a time-series data element,min order to be a dimension, the number of the channels,r in order to be of a similar tolerance,
Figure 270656DEST_PATH_IMAGE025
is composed of
Figure 278451DEST_PATH_IMAGE026
Wherein
Figure 575440DEST_PATH_IMAGE027
Figure 792795DEST_PATH_IMAGE028
Figure 999785DEST_PATH_IMAGE030
Figure 175552DEST_PATH_IMAGE031
Figure 163099DEST_PATH_IMAGE032
，
Figure 184145DEST_PATH_IMAGE033
Is the mean value of m continuous data in the time sequence data X;
Figure 104696DEST_PATH_IMAGE034
is a sequence of times
Figure 451364DEST_PATH_IMAGE035
The m-dimensional vector after the phase space reconstruction is performed,
Figure 651840DEST_PATH_IMAGE036
is composed of
Figure 742155DEST_PATH_IMAGE037
And
Figure 658159DEST_PATH_IMAGE038
the maximum value of the absolute value of the difference of the corresponding element,
Figure 644569DEST_PATH_IMAGE039
is that
Figure 403447DEST_PATH_IMAGE040
And
Figure 766295DEST_PATH_IMAGE041
the similarity of (c).
Kurtosis entropy:
Figure 130280DEST_PATH_IMAGE042
in the formula:
Figure 225275DEST_PATH_IMAGE043
denotes to the firstiThe kurtosis of each intrinsic mode component accounts for the sum of the kurtosis of all the intrinsic mode components;
Figure 409132DEST_PATH_IMAGE044
is a firstiThe kurtosis of the eigenmode components;N 2 representing the number of eigenmode components.
Combined characteristics:
Figure 637988DEST_PATH_IMAGE045
；
s204: and reducing the dimension of the combined feature by using a t-SNE algorithm.
As shown in fig. 3, in the present embodiment, the specific process of step S3 is as follows:
s301: dividing a data set of combined characteristics extracted from the transformer voiceprint data into a training set and a testing set;
s302: improved butterfly algorithm utilizes penalty factor of training set to Least Square Support Vector Machine (LSSVM)cAnd radial basis inner product function parametersgOptimizing to obtain optimal parameters;
s303: training a Least Squares Support Vector Machine (LSSVM) with optimal parameters, and testing by using a test set;
s304: and constructing a PODSDOA-LSSVM fault diagnosis model according to the training test structure, diagnosing unknown transformer voiceprint data by using the PODSDOA-LSSVM fault diagnosis model, and outputting a diagnosis result.
In this embodiment, the improved butterfly algorithm in step S3 is a butterfly algorithm that is combined and optimized based on population optimization strategies such as optimizing an initial population, improving a disadvantaged population, and the like, introducing a dynamic search parameter that can be adaptively adjusted, and introducing a variable weight location update factor strategy, and combines the advantages of the two optimization strategies to form a butterfly algorithm based on population optimization and a dynamic parameter strategy. The steps of the improved butterfly algorithm are as follows:
step A1: initializing 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 generaten*dimThe initial butterfly population of the size is expanded to 2 by space symmetryn*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: 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 ；
Sensory modalities that are updated with iterative processes
Figure 593830DEST_PATH_IMAGE046
Power exponent updated with iterative process
Figure 453202DEST_PATH_IMAGE047
Dynamic switching probability updated with iterative process
Figure 858775DEST_PATH_IMAGE048
In the formula:tthe current iteration number is;N 1 is the maximum number of iterations.
Step A7: iterative optimization: if dynamic search switching probabilityp>rand，randGlobal updating is carried out on the position of the individual for random numbers between 0~1; if dynamic search switching probabilityp<randLocally updating the position of the individual; updating the global optimum;
step A8: and (3) border crossing checking: 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.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent flow transformations made by using the contents of the specification and drawings, or applied directly or indirectly to other related technical fields, are included in the scope of the present invention.

Claims (5)

1. A transformer voiceprint fault diagnosis method based on complete set empirical mode decomposition and butterfly algorithm is characterized by comprising the following steps: the method comprises the following steps:
s1, filtering environmental noise by applying CEEMDAN decomposition and spectral subtraction: completely collecting the voiceprint data of the noise-containing transformer, carrying out empirical mode decomposition, and filtering environmental noise by using a spectral subtraction method to obtain pure voiceprint data of the transformer;
s2, extracting the combined features and reducing the dimension by using t-SNE: decomposing the denoised transformer voiceprint data through complete set empirical mode decomposition to obtain a plurality of intrinsic mode components, extracting combined features of composite multi-scale entropy, fuzzy entropy and kurtosis entropy of each intrinsic mode component, and reducing the dimension of the combined features by using a t-SNE algorithm;
s3, constructing an improved PODSDOA-LSSVM fault diagnosis model for diagnosis: based on the obtained data set of the combined characteristics, performing parameter optimization on a least square support vector machine by using an improved butterfly algorithm, constructing a PODSDOA-LSSVM fault diagnosis model, diagnosing voiceprint data of an unknown transformer and outputting a diagnosis result; the specific process of step S3 is as follows:
s301: dividing a data set of joint features extracted from the transformer voiceprint data into a training set and a testing set;
s302: improved butterfly algorithm utilizes punishment factor of training set to least square support vector machinecAnd radial basis inner product function parametersgOptimizing to obtain optimal parameters;
s303: training a least square support vector machine with optimal parameters, and testing by using a test set;
s304: building 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;
the improved butterfly algorithm comprises the following steps:
step A1: initializing butterfly algorithm search parameters: set the 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 generaten*dimThe initial butterfly population of the size is expanded to 2 by space symmetryn*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 probabilityp(ii) a And updating the location update operatorw 1 、w 2 ；
Sensory modalities that are updated with iterative processes
Figure 440762DEST_PATH_IMAGE001
Power exponent update with iterative process
Figure 844062DEST_PATH_IMAGE002
Dynamic switching probability updated with iterative process
Figure 430901DEST_PATH_IMAGE003
In the formula:tthe current iteration number is;N 1 is the maximum iteration number;
step A7: iterative optimization: if dynamic search switching probabilityp>rand，randGlobal updating is carried out on the position of the individual for random numbers between 0~1; if dynamic search switching probabilityp<randLocally updating the position of the individual; updating the global optimum;
step A8: and (3) border crossing checking: 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.
2. The transformer voiceprint fault diagnosis method based on the complete set of empirical mode decomposition and butterfly algorithm according to claim 1, which is characterized in that: in the step S1, complete set empirical mode decomposition is performed on the noisy transformer voiceprint data to obtain a plurality of intrinsic mode components, the kurtosis of each intrinsic mode component is calculated to find out the noisy intrinsic mode component and reconstruct the noisy intrinsic mode component, and the reconstructed signal is used for filtering environmental noise by using a spectral subtraction method to obtain pure transformer voiceprint data.
3. The transformer voiceprint fault diagnosis method based on the complete set of empirical mode decomposition and butterfly algorithm as claimed in claim 2, wherein: the specific process of step S1 is as follows:
s101: completely collecting the voiceprint data of the noise-containing transformer by empirical mode decomposition to obtain a plurality of intrinsic mode components;
s102: calculating the kurtosis of each intrinsic mode component, wherein the calculation formula is as follows:
Figure 654072DEST_PATH_IMAGE004
in the formula:y b is as followsiOf the intrinsic mode componentbA data value for each location;μ i is as followsiA mean of the individual eigenmode components;σ i is as followsiStandard deviation of the individual eigenmode components;u i is as followsiThe number of data of each eigenmode component; selecting a proper kurtosis value to find out intrinsic mode components containing noise for signal reconstruction;
s103: extracting the environmental noise from the reconstructed intrinsic mode component containing the noise by adopting a spectral subtraction method, and performing normalization processing;
s104: and the extracted environmental noise is subtracted from the noise-containing transformer voiceprint data to obtain pure transformer voiceprint data, and the signal-to-noise ratio is higher.
4. The transformer voiceprint fault diagnosis method based on the complete set of empirical mode decomposition and butterfly algorithm according to claim 1, which is characterized in that: the specific process of step S2 is as follows:
s201: carrying out complete set empirical mode decomposition on the de-noised transformer voiceprint data;
s202: calculating the correlation coefficient of each intrinsic modal component and the denoised transformer voiceprint datar i The calculation formula is as follows:
Figure 735291DEST_PATH_IMAGE005
in the formula (I), the compound is shown in the specification,x i is as followsiThe number of the intrinsic mode components is increased,
Figure 993097DEST_PATH_IMAGE006
is composed ofx i Is determined by the average value of (a) of (b),y i for the second time after noise eliminationiThe voice print data of each transformer is stored in a memory,
Figure 16417DEST_PATH_IMAGE007
is composed ofy i Average value of (d); selecting eigenmode components with correlation coefficients larger than a certain threshold value;
s203: calculating the composite multi-scale entropy, fuzzy entropy and kurtosis entropy of the selected eigenmode components to form a combined feature;
s204: and reducing the dimension of the combined feature by using a t-SNE algorithm.
5. The transformer voiceprint fault diagnosis method based on the complete set of empirical mode decomposition and butterfly algorithm according to claim 1, which is characterized in that: the improved butterfly algorithm in the step S3 is a butterfly algorithm based on population optimization and dynamic parameter strategies, which is formed by performing combined optimization on the butterfly algorithm based on a population optimization strategy for optimizing an initial population and improving a disadvantaged population, introducing dynamic search parameters which can be adaptively adjusted and introducing variable weight position update factor strategies.
CN202210998691.3A 2022-08-19 2022-08-19 Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm Active CN115062733B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210998691.3A CN115062733B (en) 2022-08-19 2022-08-19 Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210998691.3A CN115062733B (en) 2022-08-19 2022-08-19 Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm

Publications (2)

Publication Number Publication Date
CN115062733A CN115062733A (en) 2022-09-16
CN115062733B true CN115062733B (en) 2022-11-11

Family

ID=83207292

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210998691.3A Active CN115062733B (en) 2022-08-19 2022-08-19 Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm

Country Status (1)

Country Link
CN (1) CN115062733B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115602191A (en) * 2022-12-12 2023-01-13 杭州兆华电子股份有限公司（Cn） Noise elimination method of transformer voiceprint detection system
CN115688017A (en) * 2022-12-29 2023-02-03 南昌工程学院 FRCMDE-based transformer core fault voiceprint diagnosis method and device
CN116125235B (en) * 2023-04-14 2023-07-18 南昌工程学院 GIS partial discharge fault diagnosis method based on ultrasonic signals

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8620842B1 (en) * 2013-03-15 2013-12-31 Gordon Villy Cormack Systems and methods for classifying electronic information using advanced active learning techniques
US9465994B1 (en) * 2015-02-23 2016-10-11 Amazon Technologies, Inc. Predicting performance and success of large-scale vision algorithms
CN113532829A (en) * 2021-07-15 2021-10-22 东北石油大学 Reciprocating compressor fault diagnosis method based on improved RCMDE
CN114818780A (en) * 2022-03-28 2022-07-29 南京航空航天大学 Vehicle bearing fault diagnosis method based on CEEMDAN and APSO-SVM
CN114899457A (en) * 2022-05-23 2022-08-12 淮阴工学院 Fault detection method for proton exchange membrane fuel cell system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113552891A (en) * 2021-08-27 2021-10-26 金陵科技学院 Robot multi-target path planning based on improved butterfly optimization algorithm

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8620842B1 (en) * 2013-03-15 2013-12-31 Gordon Villy Cormack Systems and methods for classifying electronic information using advanced active learning techniques
US9465994B1 (en) * 2015-02-23 2016-10-11 Amazon Technologies, Inc. Predicting performance and success of large-scale vision algorithms
CN113532829A (en) * 2021-07-15 2021-10-22 东北石油大学 Reciprocating compressor fault diagnosis method based on improved RCMDE
CN114818780A (en) * 2022-03-28 2022-07-29 南京航空航天大学 Vehicle bearing fault diagnosis method based on CEEMDAN and APSO-SVM
CN114899457A (en) * 2022-05-23 2022-08-12 淮阴工学院 Fault detection method for proton exchange membrane fuel cell system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
BP Neural Network and its Improved Algorithm in the Power System Transformer Fault Diagnosis;Li Nan Zhang等;《 Applied Mechanics and Materials 》;20131231;全文 *
变压器故障声纹检测与诊断方法研究;吴国鑫;《中国硕士学位论文全文数据库（电子期刊）》;20220115;全文 *
基于最小二乘支持向量机的变压器故障诊断研究;高洁等;《万方数据知识服务平台》;20201029;全文 *

Also Published As

Publication number Publication date
CN115062733A (en) 2022-09-16

Similar Documents

Publication Publication Date Title
CN115062733B (en) Transformer voiceprint fault diagnosis method based on empirical mode decomposition and butterfly algorithm
CN108805059B (en) Sparse regularization filtering and self-adaptive sparse decomposition gearbox fault diagnosis method
CN103630808B (en) A kind of partial discharge signal denoising method based on lifting wavelet transform
CN111007566B (en) Curvature-driven diffusion full-convolution network seismic data bad channel reconstruction and denoising method
CN106353649A (en) Method for denoising partial discharge signals on basis of lifting wavelet transformation
CN111914705A (en) Signal generation method and device for improving health state evaluation accuracy of reactor
CN112329914B (en) Fault diagnosis method and device for buried transformer substation and electronic equipment
CN112014692A (en) Partial discharge ultrahigh frequency signal blind source separation denoising method based on principal component analysis
CN113269082A (en) Partial discharge threshold denoising method based on improved variational modal decomposition
CN108801568B (en) Bridge dynamic deflection noise reduction method and system
CN114648048B (en) Electrocardiosignal noise reduction method based on variational self-coding and PixelCNN model
CN116578940A (en) Bearing fault new type identification and diagnosis method based on mixed depth self-coding
Sun et al. Data-driven fault diagnosis method based on second-order time-reassigned multisynchrosqueezing transform and evenly mini-batch training
CN109635428B (en) GIS mechanical fault diagnosis method based on mechanical state signal analysis
CN110333054B (en) Creep micro-fault detection method for body-in-white welding equipment
Tran et al. Denoising induction motor sounds using an autoencoder
CN115859054B (en) MIC-CEEMDAN-based hydroelectric generating set draft tube pressure pulsation signal denoising method
CN115186758A (en) Inverter fault diagnosis method based on complex convolutional neural network
CN112595515A (en) Power shafting bearing fault detection method and system
CN108573698B (en) Voice noise reduction method based on gender fusion information
Qibing et al. A wavelet theory about online wavelets denoising based on Moving Window and Principal Component Analysis (PCA)
CN115659150B (en) Signal processing method, device and equipment
CN116296243B (en) Pneumatic identification method based on large-size nuclear dense blocks
CN115078934A (en) Method and system for detecting arc discharge fault based on wavelet LSTM self-encoder single-class reconstruction
CN115114955A (en) Equipment fault detection technology based on sound and vibration signals

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
GR01 Patent grant
GR01 Patent grant