A method, medium and system for detecting vibration of transformer core and clamp

Abstract

The present invention provides a method, medium and system for detecting vibration of a transformer core and a clamp, which belongs to the technical field of transformer cores and clamps. The method comprises the following steps: obtaining a vibration signal group and a sound signal group; calculating a split vibration signal set and a plurality of time-domain sound signals; screening the time-domain sound signals to obtain an ambient noise time-domain sound signal; extracting noise features of the ambient noise time-domain sound signal; extracting split vibration features of each split vibration signal; traversing the input split vibration features to obtain a correlation between each noise feature and each split vibration feature; deleting the split vibration signals corresponding to the split vibration features in each split vibration signal set whose correlation is greater than a preset correlation threshold value to generate a clean vibration signal; establishing a motion equation of the transformer core and the clamp subjected to the electromagnetic force in the transformer, and calculating the vibration of each vibration collection point of the core and the clamp; fusing the vibrations of the vibration collection points to obtain and output a fused vibration signal.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer cores and clamps, and in particular relates to a method, medium and system for detecting vibration of transformer cores and clamps. Background Art

[0002] Transformers are crucial equipment in power systems, and their health is directly linked to the safe and stable operation of the entire power grid. The vibration of the transformer core and clamps, core components of the transformer, is a crucial indicator of the transformer's operating condition. Excessive core and clamp vibration can lead to serious faults such as winding displacement and insulation damage. Therefore, accurately detecting and monitoring transformer core and clamp vibration has long been a key concern in the industry.

[0003] There are two main existing methods for detecting transformer core and clamp vibration: one is to directly measure the core and clamp vibration using vibration sensors; the other is to calculate the theoretical vibration of the core and clamp when subjected to electromagnetic forces. While the former can directly measure the actual vibration of the core and clamp, it is prone to losing some detailed vibration information due to the sampling rate limitations of the vibration sensor. The latter can more accurately calculate the theoretical vibration of each point on the core and clamp, but it cannot reflect the vibration characteristics under actual operating conditions. Both methods have their advantages and disadvantages, and a comprehensive detection method that fully utilizes measured vibration data and integrates theoretical calculations is still lacking.

[0004] Furthermore, the complex operating environment of transformer cores and clamps makes them susceptible to interference from the transformer's internal noise, which poses a challenge to accurately detecting vibrations in these cores. Existing vibration monitoring methods often ignore the impact of environmental noise and directly analyze the raw vibration signals collected by vibration sensors. These methods are susceptible to interference from various noise sources, resulting in inaccurate vibration monitoring results. Summary of the Invention

[0005] In view of this, the present invention provides a transformer core and clamp vibration detection method, medium and system, which can solve the technical problem that most existing transformer core and clamp vibration monitoring methods ignore the influence of environmental noise, directly use the original vibration signal collected by the vibration sensor for analysis, and are easily interfered by various noise sources, thereby obtaining inaccurate vibration monitoring results.

[0006] The present invention is achieved in that:

[0007] A first aspect of the present invention provides a method for detecting vibration of a transformer core and a clamp, comprising the following steps:

[0008] S10, obtaining a group of vibration signals collected by vibration sensors installed at multiple vibration collection points on the transformer core and clamps, and a group of sound signals collected by multiple sound pressure sensors installed on the surface of the transformer;

[0009] S20. Perform Fourier decomposition on each vibration signal in the vibration signal group to obtain multiple decomposition vibration signals to form a decomposition vibration signal set; perform time domain decomposition on each sound signal in the sound signal group to obtain multiple time domain sound signals;

[0010] S30, screening each time-domain sound signal using a pre-trained sound signal screening model to obtain an ambient noise time-domain sound signal;

[0011] S40, extracting the time domain features of the ambient noise time domain sound signal, recorded as noise features; extracting the features of each split vibration signal, recorded as split vibration features;

[0012] S50. For each noise feature, using a pre-trained noise-vibration correlation model, traverse and input the split vibration features to obtain a correlation between each noise feature and each split vibration feature;

[0013] S60, deleting the split vibration signals corresponding to the split vibration features having a correlation degree greater than a preset correlation degree threshold in each split vibration signal set, and restoring each split vibration signal set to a vibration signal as a clean vibration signal;

[0014] S70, establishing a motion equation of the transformer core and the clamps subjected to the electromagnetic force within the transformer, and calculating the vibration of each vibration collection point of the core and the clamps;

[0015] S80: Using a pre-trained vibration fusion model, the vibrations of the vibration collection points corresponding to each vibration signal set are fused to obtain and output a fused vibration signal.

[0016] On the basis of the above technical solution, the transformer core and clamp vibration detection method of the present invention can also be improved as follows:

[0017] Among them, the specific steps of S10 include: randomly setting multiple vibration collection points on the transformer core and clamps, and installing a vibration sensor at each collection point to collect vibration signals; at the same time, randomly setting multiple sound pressure sensors on the surface of the transformer to collect sound signals; and forming vibration signal groups and sound signal groups respectively.

[0018] Among them, the specific steps of S20 include: for each vibration signal in the vibration signal group, using Fourier transform to split it into multiple split vibration signals of different frequency bands to form a split vibration signal set; for each sound signal in the sound signal group, using time domain splitting to split it into multiple time domain sound signals.

[0019] Among them, the specific steps of S30 include: for each time domain sound signal, using a pre-trained sound signal screening model to screen and extract the environmental noise time domain sound signal therefrom; the training process of the sound signal screening model includes: collecting a large number of various sound signals including transformer operating environmental noise as training samples; manually labeling each training sample to mark which are environmental noise components; and using a deep learning algorithm to train to obtain a sound signal screening model.

[0020] The training process of the sound signal screening model is as follows:

[0021] 1) Collect a large number of various sound signals including transformer operating environment noise as training samples.

[0022] 2) Manually label each training sample to identify which ones are environmental noise components.

[0023] 3) Using deep learning algorithms, such as convolutional neural networks or recurrent neural networks, the sound signal screening model is trained using labeled training samples as input. This model can effectively identify and extract environmental noise components from the input time-domain sound signal.

[0024] Among them, the specific steps of S50 include: for each noise feature, using a pre-trained noise-vibration correlation model, traversing and inputting each split vibration feature, and calculating the correlation between each noise feature and each split vibration feature; the training process of the noise-vibration correlation model includes: collecting a large amount of sample data containing environmental noise and mechanical vibration; extracting features of the noise signal and vibration signal in each sample data to form corresponding noise features and vibration features; and using a supervised learning algorithm to train to obtain the noise-vibration correlation model.

[0025] The training process of the noise-vibration correlation model is as follows:

[0026] 1) Collect a large amount of sample data containing environmental noise and mechanical vibration.

[0027] 2) Extract features of the noise signal and vibration signal in each sample data to form corresponding noise features and vibration features.

[0028] 3) Using a supervised learning algorithm, such as a support vector machine or neural network, with noise and vibration features as input and the correlation between noise and vibration as output, a noise-vibration correlation model is developed. This model can effectively assess the correlation between any noise and vibration features.

[0029] The specific steps of S60 include: for each split vibration signal set, deleting vibration signals with a correlation greater than a preset threshold value of 0.7, and restoring the remaining vibration signals to complete vibration signals as clean vibration signals.

[0030] The specific steps of S70 include: establishing a motion equation for the vibration of the transformer core and the clamps caused by the electromagnetic force, and calculating the theoretical vibration condition of each vibration collection point by solving the motion equation.

[0031] Among them, the specific steps of S80 include: using a pre-trained vibration fusion model to fuse the clean vibration signal obtained in step S60 with the theoretical vibration calculated in step S70 to obtain a final fused vibration signal; the training process of the vibration fusion model includes: collecting a large number of measured vibration signals and theoretically calculated vibration signals during the operation of the transformer as training samples; manually inspecting and labeling these samples to ensure that there is a certain difference between the measured vibration signal and the theoretically calculated vibration signal; and using a machine learning algorithm to train to obtain a vibration fusion model.

[0032] The training process of the vibration fusion model is as follows:

[0033] 1) Collect a large number of measured vibration signals and theoretically calculated vibration signals during transformer operation as training samples.

[0034] 2) Manually inspect and label these samples to ensure that there is a certain difference between the measured vibration signals and the theoretically calculated vibration signals.

[0035] 3) Use machine learning algorithms, such as multi-layer perceptrons or convolutional neural networks, with measured vibration signals and theoretically calculated vibration signals as input, output the fused vibration signals, and perform model training.

[0036] 4) The trained vibration fusion model can effectively fuse the measured vibration signal with the theoretically calculated vibration signal to obtain a more accurate and reliable final vibration signal.

[0037] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for detecting vibration of a transformer core and a clamp.

[0038] A third aspect of the present invention provides a transformer core and clamp vibration detection system, which includes the above-mentioned computer-readable storage medium.

[0039] Compared with the prior art, the transformer core and clamp vibration detection method, medium and system provided by the present invention have the following beneficial effects:

[0040] 1. Fusion of measured vibration signals and theoretical vibration calculation results. The present invention not only utilizes the measured vibration signals collected by the vibration sensor but also combines them with the core and fixture vibrations calculated through theoretical calculations, fusing the two to produce the final vibration monitoring results. This approach leverages the advantages of the measured signals while integrating the precision of theoretical calculations, overcoming the sampling limitations of a single vibration sensor and improving the accuracy of vibration detection.

[0041] 2. Effectively suppresses environmental noise interference. In addition to collecting vibration signals, the present invention also utilizes a sound pressure sensor to capture the sound signals of the transformer's operating environment. Through signal processing and pattern recognition techniques, the influence of environmental noise is effectively removed from the vibration signals. This not only improves the reliability of vibration detection but also provides more accurate basic data for transformer condition monitoring and fault diagnosis.

[0042] 3. Achieve multi-source information fusion. This solution integrates multiple sources of information, including vibration signals and sound signals collected on-site from the transformer, as well as theoretically calculated vibration conditions. Through advanced signal processing and machine learning algorithms, it achieves comprehensive analysis and accurate detection of transformer core and clamp vibration. This multi-source information fusion approach is more reliable and effective than traditional methods that rely solely on vibration sensors or theoretical calculations.

[0043] In summary, the solution of the present invention solves the technical problem that most existing transformer core and clamp vibration monitoring methods ignore the influence of environmental noise and directly use the original vibration signal collected by the vibration sensor for analysis, which is easily interfered by various noise sources, thereby obtaining inaccurate vibration monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0047] like Figure 1 FIG. 1 is a flow chart of a method for detecting vibration of a transformer core and a clamp provided by the present invention. The method comprises the following steps:

[0048] S10, obtaining a group of vibration signals collected by vibration sensors installed at multiple vibration collection points on the transformer core and clamps, and a group of sound signals collected by multiple sound pressure sensors installed on the surface of the transformer;

[0049] S20. Perform Fourier decomposition on each vibration signal in the vibration signal group to obtain multiple decomposition vibration signals to form a decomposition vibration signal set; perform time domain decomposition on each sound signal in the sound signal group to obtain multiple time domain sound signals;

[0050] S30, screening each time-domain sound signal using a pre-trained sound signal screening model to obtain an ambient noise time-domain sound signal;

[0051] S40, extracting the time domain features of the ambient noise time domain sound signal, recorded as noise features; extracting the features of each split vibration signal, recorded as split vibration features;

[0052] S50. For each noise feature, use a pre-trained noise-vibration correlation model to traverse the input split vibration features to obtain the correlation between each noise feature and each split vibration feature;

[0053] S60, deleting the split vibration signals corresponding to the split vibration features having a correlation degree greater than a preset correlation degree threshold in each split vibration signal set, and restoring each split vibration signal set to a vibration signal as a clean vibration signal;

[0054] S70, establishing a motion equation of the transformer core and the clamps subjected to the electromagnetic force within the transformer, and calculating the vibration of each vibration collection point of the core and the clamps;

[0055] S80: Using a pre-trained vibration fusion model, the vibrations of the vibration collection points corresponding to each vibration signal set are fused to obtain and output a fused vibration signal.

[0056] The specific implementation of the above steps is described below:

[0057] The specific implementation of step S10 is as follows: first, multiple vibration collection points are set up on the transformer core and clamps, and a vibration sensor is installed at each collection point to collect vibration signals. Simultaneously, multiple sound pressure sensors are installed on the transformer surface to collect sound signals. These sensors collect a group of vibration and sound signals during transformer operation.

[0058] In step S20, each vibration signal in the vibration signal group is split into multiple vibration signals of different frequency bands using Fourier transform to form a set of split vibration signals. Each sound signal in the sound signal group is split into multiple time-domain sound signals using time-domain splitting.

[0059] In step S30, a pre-trained sound signal screening model is used to screen each time domain sound signal to extract the ambient noise time domain sound signal. The training process of the sound signal screening model is as follows:

[0060] 1) Collect a large number of various sound signals including transformer operating environment noise as training samples.

[0061] 2) Manually label each training sample to identify which ones are environmental noise components.

[0062] 3) Using deep learning algorithms, such as convolutional neural networks or recurrent neural networks, the sound signal screening model is trained using labeled training samples as input. This model can effectively identify and extract environmental noise components from the input time-domain sound signal.

[0063] In step S40, the time domain features of the ambient noise time domain sound signal are first extracted, including signal amplitude, frequency, phase, and other indicators, which are recorded as noise features. At the same time, the features of each split vibration signal are extracted, such as amplitude, frequency, energy, etc., which are recorded as split vibration features.

[0064] In step S50, for each noise feature, a pre-trained noise-vibration correlation model is used to traverse and input each split vibration feature, and the correlation between each noise feature and each split vibration feature is calculated. The training process of the noise-vibration correlation model is as follows:

[0065] 1) Collect a large amount of sample data containing environmental noise and mechanical vibration.

[0066] 2) Extract features of the noise signal and vibration signal in each sample data to form corresponding noise features and vibration features.

[0067] 3) Using a supervised learning algorithm, such as a support vector machine or neural network, with noise and vibration features as input and the correlation between noise and vibration as output, a noise-vibration correlation model is developed. This model can effectively assess the correlation between any noise and vibration features.

[0068] In step S60, for each split vibration signal set, vibration signals with a correlation greater than a preset threshold are deleted. This correlation threshold can be set to 0.7. Specifically, when the correlation between a noise feature and a vibration feature exceeds 0.7, the vibration signal is considered to be significantly affected by ambient noise and needs to be removed. After removing these noise-affected vibration signals, the remaining vibration signals are restored to complete vibration signals and treated as clean vibration signals.

[0069] In step S70, a motion equation is established for the vibrations generated by the transformer core and clamps due to electromagnetic forces. This motion equation describes the vibration characteristics of each vibration collection point on the core and clamps, including amplitude, frequency, and phase. By solving this motion equation, the theoretical vibration condition at each vibration collection point can be calculated.

[0070] In step S80, a pre-trained vibration fusion model is used to fuse the clean vibration signal obtained in step S60 with the theoretical vibration calculated in step S70 to obtain the final fused vibration signal. The training process of the vibration fusion model is as follows:

[0071] 1) Collect a large number of measured vibration signals and theoretically calculated vibration signals during transformer operation as training samples.

[0072] 2) Manually inspect and label these samples to ensure that there is a certain difference between the measured vibration signals and the theoretically calculated vibration signals.

[0073] 3) Use machine learning algorithms, such as multi-layer perceptrons or convolutional neural networks, with measured vibration signals and theoretically calculated vibration signals as input, output the fused vibration signals, and perform model training.

[0074] 4) The trained vibration fusion model can effectively fuse the measured vibration signal with the theoretically calculated vibration signal to obtain a more accurate and reliable final vibration signal.

[0075] In order to better describe the specific embodiments of the present invention, the following is a detailed explanation with reference to the formula:

[0076] Step S10: Vibration and sound signal collection

[0077] N vibration collection points are set on the transformer core and clamps, and a vibration sensor is installed at each collection point to collect vibration signals. At the same time, M sound pressure sensors are set on the surface of the transformer to collect sound signals. During the operation of the transformer, the vibration sensor and the sound pressure sensor respectively collect the vibration signal group V = {v1, v2, ..., v N} and the sound signal group S={s1,s2,...,s M}.

[0078] Step S20: Signal Splitting

[0079] For each vibration signal v in the vibration signal set V i , Fourier transform is used to split it into L vibration signals of different frequency bands {v i1 ,v i2 ,...,v iL}, forming a split vibration signal set V d ={v d1 ,v d2 ,...,v dN}, where v di ={v i1 ,v i2 ,...,v iL}.

[0080] For each sound signal s in the sound signal group S j , using the time domain splitting method to split it into K time domain sound signals {s j1 ,s j2 ,...,s jK}, forming a time domain sound signal set S t ={s t1 ,s t2 ,...,s tM}, where s tj ={s j1 ,s j2 ,...,s jK}.

[0081] Step S30: Extracting the time domain sound signal of ambient noise

[0082] For the time domain sound signal set S t Each time domain sound signal s in tj , using a pre-trained sound signal screening model Screening and extracting the time domain sound signal of environmental noise

[0083] The sound signal screening model The training process is as follows:

[0084] 1) Collect a large number of various sound signals S including transformer operating environment noise train ={s train1 ,s train2 ,...,s trainP} as training samples, where each s trainp It is a time domain sound signal sequence.

[0085] 2) For each training sample s trainp Perform manual annotation to obtain the annotation vector y trainp ={y train1 ,y train2 ,...,y trainK}, where y trainl =1 means s trainl is the ambient noise, otherwise it is 0.

[0086] 3) Use deep learning algorithms, such as convolutional neural networks (CNN) or recurrent neural networks (RNN), to build a sound signal screening model The input of the model is the time domain sound signal s, and the output is the probability of the noise component in the signal

[0087] 4) Take training sample S train and its label vector {y trainp} as supervision information, train the sound signal screening model

[0088] By applying the trained model It can be effectively obtained from the time domain sound signal set S t Extract the time domain sound signal of the ambient noise

[0089] Step S40: Feature extraction

[0090] 1) Extracting time-domain sound signals from ambient noise The time domain characteristics of the signal, including signal amplitude A, frequency f and phase φ, form the noise feature set F noise ={f noise1 ,f noise2 ,...,f noiseM}, where f noisej ={A j ,f j ,φ j}.

[0091] 2) Extract and split vibration signal set V d Each split vibration signal v di The characteristics of the vibration, including amplitude A, frequency f and energy E, form the split vibration feature set Fvib ={f vib1 ,f vib2 ,...,f vibN}, where f vibi ={A i1 ,A i2 ,...,A iL ;f i1 ,f i2 ,...,f iL ;E i1 ,E i2 ,...,E iL}.

[0092] Step S50: Calculation of noise and vibration correlation

[0093] For the noise feature set F noise Each noise feature f in noisej , using the pre-trained noise vibration correlation model , traverse the input split vibration feature set F vib Each split vibration feature f in vibi , calculate the correlation c between them ji :

[0094]

[0095] The noise vibration correlation model The training process is as follows:

[0096] 1) Collect a large amount of sample data containing environmental noise and mechanical vibration

[0097] Each sample includes the noise feature f trainq , vibration characteristics And the correlation between them c trainq .

[0098] 2) Use supervised learning algorithms, such as support vector machines (SVM) or neural networks (NN), to establish a noise-vibration correlation model The input of the model is the noise feature f noise and vibration characteristics f vib , the output is the correlation between them

[0099] 3) Take training sample D train As supervision information, the noise vibration correlation model is trained

[0100] By applying the trained model The noise feature set F can be effectively calculatednoise Each noise feature in the split vibration feature set F vib The correlation between each vibration feature in .

[0101] Step S60: Clean vibration signal extraction

[0102] For each split vibration signal set v di , delete the split vibration signals whose correlation is greater than the preset threshold δ=0.7, and restore the undeleted vibration signals to the complete vibration signals Forming a clean vibration signal set The purpose of this is to delete vibration signals that are severely interfered by environmental noise and retain relatively clean vibration signals.

[0103] Step S70: Theoretical vibration calculation

[0104] The motion equation for the vibration of the transformer core and the clamps caused by electromagnetic force is established as:

[0105]

[0106] Where m is the mass of the core and the clamp, c is the damping coefficient, k is the stiffness coefficient, F(t) is the electromagnetic force acting on the core and the clamp, and x(t) is the displacement of the core and the clamp at each vibration collection point.

[0107] By solving the motion equation, the theoretical vibration conditions of each vibration collection point can be calculated, including amplitude A, frequency f and phase φ, etc., to form a theoretical vibration signal set X = {x1, x2, ..., x N}, where x i ={A i ,f i ,φ i}.

[0108] Step S80: Vibration signal fusion

[0109] Use pre-trained vibration fusion model The clean vibration signal set obtained in step S60 The final fused vibration signal Y = {y1, y2, ..., y N}.

[0110] The vibration fusion model The training process is as follows:

[0111] 1) Collect a large number of measured vibration signals during transformer operation And theoretical calculation vibration signal X train As training samples, form a sample set

[0112] 2) Manually inspect and label these samples to ensure the actual vibration signal The vibration signal x is calculated theoretically trainq There are certain differences between

[0113] 3) Use machine learning algorithms, such as multi-layer perceptron (MLP) or convolutional neural network (CNN), to build a vibration fusion model The input of the model is a clean vibration signal And the theoretical vibration signal x, the output is the fusion vibration signal

[0114] 4) Take training sample D train As supervision information, the vibration fusion model is trained

[0115] By applying the trained vibration fusion model Can effectively convert clean vibration signals The final fused vibration signal Y is obtained by fusing it with the theoretical vibration signal X. This not only fully utilizes the advantages of the measured vibration signal, but also combines the vibration characteristics of the theoretical calculation, and can obtain more accurate and reliable vibration monitoring results.

[0116] The following is an explanation of the meaning of the relevant variables and formulas:

[0117] N: The number of vibration collection points set on the transformer core and clamps.

[0118] M: The number of sound pressure sensors installed on the transformer surface.

[0119] V={v1,v2,...,v N}: Vibration signal group, a set of vibration signals collected by N vibration sensors.

[0120] S={s1,s2,...,s M}: Sound signal group, a collection of sound signals collected by M sound pressure sensors.

[0121] L: Each vibration signal v i The number of frequency bands to split into.

[0122] V d ={v d1 ,v d2 ,...,v dN}: Split the vibration signal set, where v di ={v i1 ,v i2 ,...,viL} represents the L split vibration signals obtained after the i-th vibration signal is split.

[0123] K: Each sound signal s j The number of time domain signals to be split.

[0124] S t ={s t1 ,s t2 ,...,s tM}: Time domain sound signal set, where s tj ={s j1 ,s j2 ,...,s jK} represents the K time domain sound signals obtained after splitting the j-th sound signal.

[0125] A pre-trained sound signal screening model is used to extract environmental noise components from time-domain sound signals.

[0126] S train ={s train1 ,s train2 ,...,s trainP}: Used to train the sound signal screening model training sample set, where each s trainp It is a time domain sound signal sequence.

[0127] y trainp ={y train1 ,y train2 ,...,y trainK}: For training sample s trainp The manual annotation results, where y trainl =1 means s trainl is the ambient noise, otherwise it is 0.

[0128] From the time domain sound signal set S t The time domain sound signal of the ambient noise is extracted.

[0129] F noise ={f noise1 ,f noise2 ,...,f noiseM}: noise feature set, where f noisej ={A j ,f j ,φ j} represents the time domain characteristics of the jth environmental noise time domain sound signal, including amplitude A, frequency f and phase φ.

[0130] F vib ={fvib1 ,f vib2 ,...,f vibN}: Split vibration feature set, where f vibi ={A i1 ,A i2 ,...,A iL ;f i1 ,f i2 ,...,f iL ;E i1 ,E i2 ,...,E iL} represents the characteristics of the i-th split vibration signal, including the amplitude A, frequency f and energy E of each frequency band.

[0131] The pre-trained noise-vibration correlation model is used to calculate the correlation between noise features and vibration features.

[0132]

[0133] Used to train noise vibration correlation model A training sample set, where each sample includes the noise feature f trainq , vibration characteristics And the correlation between them c trainq .

[0134] Noise characteristics f noisej and vibration characteristics f vibi The correlation between them.

[0135] δ=0.7: correlation threshold, used to delete vibration signals that are severely interfered by environmental noise.

[0136] Clean vibration signal set is the complete vibration signal obtained by deleting the vibration signal affected by noise.

[0137] m, c, k: mass, damping coefficient and stiffness coefficient of transformer core and clamps.

[0138] F(t): Electromagnetic force acting on the core and the clamp.

[0139] x(t): Displacement of the core and the clamp at each vibration collection point.

[0140] X={x1,x2,...,x N}: Theoretical vibration signal set, where x i ={A i ,f i ,φ i} represents the theoretical vibration characteristics of the i-th vibration collection point.

[0141] Pre-trained vibration fusion model for clean vibration signals The final fused vibration signal Y is obtained by fusing it with the theoretical vibration signal X.

[0142]

[0143] : Used to train vibration fusion models A training sample set, where each sample includes a measured vibration signal Theoretical vibration signal x trainq and fused vibration signal y trainq .

[0144] Y={y1,y2,...,y N}: The final fused vibration signal.

[0145]

[0146] This is a second-order differential equation that describes the vibration motion of the core and clamp. m is the mass of the core and clamp, c is the damping coefficient, k is the stiffness coefficient, F(t) is the electromagnetic force acting on the core and clamp, and x(t) is the displacement of the core and clamp. By solving this equation, the theoretical vibration characteristics of each vibration sampling point on the core and clamp can be calculated.

[0147]

[0148] This is the noise vibration correlation model trained Calculate the noise characteristic f noisej and vibration characteristics f vibi Correlation between ji formula.

[0149]

[0150] This is the vibration fusion model trained Clean vibration signal The formula for fusing the theoretical vibration signal x to obtain the final fused vibration signal y is obtained.

[0151] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for detecting vibration of a transformer core and a clamp.

[0152] A third aspect of the present invention provides a transformer core and clamp vibration detection system, which includes the above-mentioned computer-readable storage medium.

[0153] Specifically, the principle of the present invention is:

[0154] 1. Multi-source information fusion

[0155] The solution of the present invention fully utilizes multiple sources of information, including vibration signals and sound signals collected on-site from the transformer, as well as theoretically calculated vibration conditions. First, a vibration sensor and a sound pressure sensor collect the measured vibration signals and ambient sound signals from the transformer core and clamps, respectively. Simultaneously, based on a theoretical model of the vibrations generated by electromagnetic forces in the transformer core and clamps, theoretical vibration characteristics are calculated. These three types of information encompass different aspects of the vibration of the transformer core and clamps. By integrating this information, a more comprehensive and accurate representation of the actual vibration state of the core and clamps can be achieved.

[0156] 2. Ambient noise suppression

[0157] Measured vibration signals inevitably contain various noise interferences caused by the transformer's operating environment. Therefore, the present invention utilizes ambient sound signals collected by a sound pressure sensor and trains a sound signal screening model using a machine learning algorithm. This model effectively separates the ambient noise components from the measured vibration signal, resulting in a relatively clean vibration signal. This significantly improves the reliability of vibration detection and avoids misjudgments caused by noise interference.

[0158] 3. Fusion of theoretical vibration and measured vibration

[0159] The solution of the present invention not only utilizes measured vibration signals but also incorporates vibration characteristics derived from theoretical calculations. First, by establishing the motion equations for the transformer core and clamp vibrations, the theoretical vibration characteristics of each vibration collection point on the core and clamp are calculated. Then, using a pre-trained vibration fusion model, the clean measured vibration signals are fused with the theoretical vibration signals to obtain the final, accurate vibration monitoring results. This fusion approach leverages the advantages of measured signals and the accuracy of theoretical calculations, compensating for the shortcomings of a single method and improving the overall accuracy of vibration detection.

[0160] To sum up, the core technical principle of the transformer core and clamp vibration detection method of the present invention is to achieve accurate detection and reliable analysis of the vibration state of the transformer core and clamps through key technical means such as multi-source information fusion, environmental noise suppression, and fusion of theoretical vibration and measured vibration.

[0161] The following is a specific embodiment of the present invention:

[0162] A 110kV transformer at a power company experienced abnormal vibration during operation. To promptly detect and diagnose the problem, the company decided to use the transformer core and clamp vibration detection method proposed in this invention to monitor the transformer's vibration status. The specific implementation steps are as follows:

[0163] 1. Signal Acquisition

[0164] A total of 12 vibration collection points were placed on the transformer core and clamps, each equipped with a triaxial accelerometer to collect vibration signals during transformer operation. Eight microphones were also placed on the transformer surface to collect ambient sound signals from the transformer.

[0165] After 24 hours of continuous monitoring, a total of 12 vibration signals and 8 sound signals were collected, with sampling frequencies of 5kHz and 20kHz respectively.

[0166] 2. Signal Splitting

[0167] For the 12 vibration signals, short-time Fourier transform is used to split each signal into 8 frequency bands to form a split vibration signal set. For the 8 sound signals, Hilbert-Huang transform is used to split each signal into 16 time domain signals to form a time domain sound signal set.

[0168] 3. Ambient noise suppression

[0169] A pre-trained convolutional neural network sound signal screening model is used to process the time-domain sound signal set and extract the ambient noise components. The model takes the time-domain sound signal as input and outputs the probability of the noise component in each time-domain signal. By processing all the time-domain sound signals, a set of ambient noise time-domain sound signals is ultimately obtained.

[0170] 4. Feature Extraction

[0171] From the ambient noise time-domain sound signal set, the time-domain features of each signal, such as amplitude, frequency, and phase, are extracted to form a noise feature set. Simultaneously, from the split vibration signal set, the features of each split vibration signal, such as amplitude, frequency, energy, crest factor, and form factor, are extracted to form a split vibration feature set.

[0172] Taking a split vibration signal as an example, the extracted feature parameters are shown in Table 1.

[0173] Table 1 Characteristic parameters of a split vibration signal

[0174] parameter value amplitude <![CDATA[0.22m / s 2 ]]> frequency 65Hz energy 0.48J Crest Factor 3.8 Form Factor 1.2

[0175] After similar processing, a total of 12 feature sets of split vibration signals and 8 feature sets of noise time domain signals were obtained.

[0176] 5. Noise impact analysis

[0177] Using the pre-trained support vector machine noise-vibration correlation model, the input split vibration feature set and noise feature set are traversed to calculate the correlation between each noise feature and each split vibration feature.

[0178] With a certain noise characteristic f noise ={A=0.18,f=120Hz,φ=1.2rad} as an example, the correlation calculation results with the 12 split vibration features are shown in Table 2.

[0179] Table 2 Correlation between a certain noise feature and the split vibration feature

[0180]

[0181]

[0182] As can be seen from the table, the correlation between this noise feature and the 3rd, 5th and 11th split vibration features is high, exceeding the threshold of 0.7, indicating that these vibration signal components are greatly affected by environmental noise.

[0183] By calculating the correlation between all noise features and split vibration features, a total of 4 split vibration signals were judged to be greatly affected by noise and needed to be deleted from the split vibration signal set.

[0184] 6. Theoretical vibration calculation

[0185] Based on the theoretical model of the vibration of the transformer core and clamps under the action of electromagnetic force, the following motion equation is established:

[0186]

[0187] Where m = 1200 kg is the mass of the core and the clamp, c = 3200 N·s / m is the damping coefficient, k = 7.8 × 10 8 N / m is the stiffness coefficient. By using the Runge-Kutta numerical integration method to solve the equation, the theoretical vibration conditions of the 12 vibration collection points were calculated, including amplitude, frequency, and phase. The results are shown in Table 3.

[0188] Table 3 Theoretical vibration conditions of 12 vibration collection points

[0189]

[0190]

[0191] 7. Vibration signal fusion

[0192] Finally, the pre-trained convolutional neural network vibration fusion model is used to fuse the clean vibration signal set obtained in step 5 and the theoretical vibration conditions calculated in step 6 to obtain the final fused vibration signal.

[0193] The fused vibration signals of the 12 vibration collection points were comprehensively analyzed and the following results were obtained:

[0194] The vibration peak is about 0.22m / s 2 , less than the manufacturer's recommended 0.3m / s 2 warning value.

[0195] The main vibration frequency is concentrated around 50Hz, which is basically consistent with the power supply frequency.

[0196] The vibration phase differences at each collection point are small, indicating that the overall vibration is relatively coordinated.

[0197] By applying the method of the present invention, not only is the measured vibration signal fully utilized, but theoretical calculation results are also integrated. While effectively suppressing the influence of environmental noise, more accurate and reliable vibration monitoring results of the transformer core and clamps are obtained. This provides reliable basic data support for the condition assessment and fault diagnosis of the transformer. In the above-mentioned embodiment, the core and clamp can be regarded as a whole or as two independent entities, and the implementation of the steps of the present invention is not affected. That is, in the steps of the embodiment, the process of each step of the present invention is implemented on a separate core or a separate clamp.

[0198] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for detecting vibration of transformer core and clamp, characterized in that: The following steps are involved: S10, obtaining a group of vibration signals collected by vibration sensors installed at multiple vibration collection points on the transformer core and clamps, and a group of sound signals collected by multiple sound pressure sensors installed on the surface of the transformer; S20. Perform Fourier decomposition on each vibration signal in the vibration signal group to obtain multiple decomposition vibration signals to form a decomposition vibration signal set; perform time domain decomposition on each sound signal in the sound signal group to obtain multiple time domain sound signals; S30. Using a pre-trained sound signal screening model to screen each time-domain sound signal to obtain an ambient noise time-domain sound signal. The specific steps include: using a pre-trained sound signal screening model to screen each time-domain sound signal to extract the ambient noise time-domain sound signal; the training process of the sound signal screening model includes: collecting a large number of various sound signals including transformer operating ambient noise as training samples; manually labeling each training sample to identify which ones are ambient noise components; and using a deep learning algorithm to train a sound signal screening model. S40, extracting the time domain features of the ambient noise time domain sound signal, recorded as noise features; extracting the features of each split vibration signal, recorded as split vibration features; S50. For each noise feature, using a pre-trained noise-vibration correlation model, traversing and inputting the split vibration features, and obtaining a correlation between each noise feature and each split vibration feature. The specific steps include: for each noise feature, using a pre-trained noise-vibration correlation model, traversing and inputting each split vibration feature, and calculating a correlation between each noise feature and each split vibration feature. The training process of the noise-vibration correlation model includes: collecting a large amount of sample data containing environmental noise and mechanical vibration; performing feature extraction on the noise signal and vibration signal in each sample data to form corresponding noise features and vibration features; and using a supervised learning algorithm to train and obtain the noise-vibration correlation model. S60, deleting the split vibration signals corresponding to the split vibration features having a correlation degree greater than a preset correlation degree threshold in each split vibration signal set, and restoring each split vibration signal set to a vibration signal as a clean vibration signal; S70, establishing a motion equation of the transformer core and the clamps subjected to the electromagnetic force within the transformer, and calculating the vibration of each vibration collection point of the core and the clamps; S80: Using a pre-trained vibration fusion model, the vibrations of the vibration collection points corresponding to each vibration signal set are fused to obtain and output a fused vibration signal.

2. A transformer core and clamp vibration detection method according to claim 1, characterized in that: The specific steps of S10 include: randomly setting multiple vibration collection points on the transformer core and the clamp, and installing a vibration sensor at each collection point to collect vibration signals; and randomly setting multiple sound pressure sensors on the surface of the transformer to collect sound signals; forming a vibration signal group and a sound signal group respectively.

3. A transformer core and clamp vibration detection method according to claim 1, characterized in that: The specific steps of S20 include: for each vibration signal in the vibration signal group, using Fourier transform to split it into multiple split vibration signals of different frequency bands to form a split vibration signal set; for each sound signal in the sound signal group, using time domain splitting to split it into multiple time domain sound signals.

4. A transformer core and clamp vibration detection method according to claim 1, characterized in that: The specific steps of S60 include: for each split vibration signal set, deleting vibration signals with a correlation greater than a preset threshold value of 0.7, and restoring the remaining vibration signals to complete vibration signals as clean vibration signals.

5. A transformer core and clamp vibration detection method according to claim 1, characterized in that: The specific steps of S70 include: establishing a motion equation for the vibration of the transformer core and the clamping parts caused by the electromagnetic force, and calculating the theoretical vibration condition of each vibration collection point by solving the motion equation.

6. A transformer core and clamp vibration detection method according to claim 1, characterized in that: The specific steps of S80 include: using a pre-trained vibration fusion model to fuse the clean vibration signal obtained in step S60 with the theoretical vibration calculated in step S70 to obtain a final fused vibration signal; the training process of the vibration fusion model includes: collecting a large number of measured vibration signals and theoretically calculated vibration signals during the operation of the transformer as training samples; manually inspecting and labeling these samples to ensure that there is a certain difference between the measured vibration signals and the theoretically calculated vibration signals; and using a machine learning algorithm to train to obtain the vibration fusion model.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program instructions, which, when executed, are used to execute the transformer core and clamp vibration detection method according to any one of claims 1 to 6.

8. A transformer core and clamp vibration detection system, characterized in that: Contains the computer-readable storage medium of claim 7.

Images