CN116449256A - A transformer state fault diagnosis system and method based on voiceprint sensing - Google Patents

Abstract

The invention discloses a transformer state fault diagnosis system and method based on voiceprint sensing, which can obtain transformer oil chromatographic diagnosis data, voiceprint diagnosis data, working condition information record data, total hydrocarbon content data after reduced load operation, reduced The total amount of acetylene data after load operation; construct the mass function under various fault conditions, and substitute various data into the mass function under each fault condition. Real-time monitoring of operating status; through AI learning, for the identification and judgment of electrical parameters such as three-phase unbalance and load characteristics during transformer operation, as well as abnormal status information of partial discharge and no-load, combined with vibration and other parameters of the operating environment, in the The front-end performs information fusion processing and early warning in a timely manner, which improves the real-time response to transformer faults, reduces the risk of transformer faults, and improves the efficiency and control capabilities of inspection operations.

Description

A transformer state fault diagnosis system and method based on voiceprint sensing

technical field

The invention relates to the technical field of transformer monitoring, in particular to a transformer fault diagnosis method.

Background technique

Transformer is an important equipment for transmission and distribution of electric energy in the power system. The stability and reliability of its operation directly determine the reliability of power supply for users. The current power transformer operation monitoring work is basically realized by manual inspection, which has problems such as heavy workload, low efficiency, and slow response.

In the past, a small number of experienced inspection personnel could judge whether there was any abnormality by the sound of the equipment when it was running. However, the development of the power grid is changing with each passing day. With the increasing number of power equipment, the traditional "manual auscultation" method is far from meeting the actual needs of intelligent control. And it is easily affected by external environmental interference, such as background noise, voices of personnel, etc.; the acoustic and vibration signals generated by power equipment such as transformers and circuit breakers during operation contain a large amount of status information, and have identification features like human fingerprints. When a device is defective or malfunctions, its voiceprint will change accordingly. Accurate identification of voiceprint information helps O&M personnel diagnose equipment defects and pinpoint the cause of the failure.

For example, the CN110864801A Transformer Noise and Vibration Online Monitoring and Fault Analysis System is pre-fixed at various angles and positions of the main transformer equipment through an external patch-type acoustic sensor, and collects and samples the sound of the main transformer in real time, amplifies the sound signal, filters it and After converted into digital electrical signals, the data is sent to the adjacent HSBC processing unit through wireless or Bluetooth communication modules, and then the HSBC unit sends the data to the online monitoring master server and monitoring background system through switches and optical fiber networks. Not only can the possibility and cause analysis of various fault types be analyzed, but also comprehensive analysis and judgment can be made on the fault, and the location of the fault point can be preliminarily judged according to the installation position of the sensor and the respective sampling data.

However, the accuracy of judging transformer faults by a single sensor is low, and a comprehensive analysis that can combine multiple existing parameters is needed to draw conclusions.

Contents of the invention

The purpose of the present invention is to provide a transformer fault diagnosis method to solve the problems raised in the above-mentioned background technology.

To achieve the above object, the present invention provides the following technical solutions:

A transformer state fault diagnosis method based on voiceprint sensing, characterized in that: comprising:

Obtain transformer oil chromatography diagnostic data, voiceprint diagnostic data, working condition information record data (mainly transformer load status), total hydrocarbon content data after load-down operation, and total acetylene data after load-down operation;

The mass function under various fault conditions is constructed, and various data are substituted into the mass function under each fault condition, and the one with the largest function value can be diagnosed as a transformer fault.

Preferably, the faults include open circuit faults, insulation faults, circuit faults, magnetic circuit faults and no faults;

Open circuit fault mass function:

Insulation fault mass function:

Circuit fault mass function:

Magnetic circuit fault mass function:

No fault mass function:

Wherein, k=1-K, K is a normalization constant;

Open circuit fault and insulation fault

Circuit fault, magnetic circuit fault and no fault

Among them, m ₁ (e ₁ ) is the probability of each fault when the oil chromatography diagnosis is e ₁ , m ₂ (e ₂ ) is the probability of each fault when the voiceprint diagnosis is e ₂ , m ₃ (e ₃ ) is the The condition information is recorded as the probability of each failure when e ₃ is recorded, m ₄ (e ₄ ) is the probability of each failure when the total hydrocarbon content is e ₄ after the load-down operation, m ₅ (e ₅ ) is the total acetylene content after the load-down operation The probability of each fault occurs when the quantity is e ₅ , and A is the type of fault.

A transformer state fault diagnosis system based on voiceprint sensing, comprising a housing and a channel sound source localization system component installed on the housing, an anti-interference detection component, a voiceprint feature extraction component, and a faulty voiceprint judging component;

The channel sound source localization system component is used to collect the voiceprint data of the transformer, and send the collected voiceprint data to the anti-interference detection component;

The anti-interference detection component performs filtering processing on the received voiceprint data, and then sends the processed voiceprint data to the voiceprint feature extraction component;

The voiceprint feature extraction component is used to extract the fundamental frequency and harmonics corresponding to multiple targets from the processed voiceprint data;

The fault voiceprint judging element is used to compare the fundamental frequency and harmonics of each target with the preset fundamental frequency and harmonics of the target to calculate the similarity, judge the fault type of the transformer, and output the result.

Preferably, the housing is a trumpet-shaped structure with a hollow interior and an open end.

Preferably, the channel sound source localization system components are arranged at the open end of the casing.

Preferably, the component of the channel sound source localization system is a 64-channel sound source localization system, which can cover audible sound and ultrasonic frequency bands.

Preferably, the fault voiceprint judging element constructs a voiceprint detection model based on MobileNetV3, the voiceprint detection model includes a voiceprint feature extraction module, a MobileNetV3 module and a loop module; the voiceprint feature extraction module extracts Features, the MobileNetV3 module classifies the extracted features, and connects them to the loop module, and performs modeling through the information of the front and back frames to improve the accuracy of the model.

Preferably, the fault voiceprint judging element introduces an attention mechanism to construct a voiceprint detection model based on MobileNetV3, performs pooling processing on each channel for the obtained feature matrix, and then obtains an output vector through two fully connected layers , where the number of nodes in the first fully connected layer is 1/4 of the channel of the input feature matrix, and the channel of the second fully connected layer is consistent with the channel of the feature matrix.

Preferably, the loss function is used in the process of voiceprint detection model training authenticating;

Among them, m and n represent the number of samples and the total number of categories, respectively, is the cross-entropy loss corresponding to samples, p _i is the probability that the sample belongs to category i, from which the voiceprint characteristics of transformer faults can be obtained.

Preferably, a vibration sensor is also included, and the vibration sensor is used to collect the vibration signal of the transformer when it is working, and extract the corresponding frequency spectrum features through the processor, and use the obtained frequency spectrum features when the transformer is in various faults as a comparison feature ;

The processor analyzes and compares the spectral features and comparative features to diagnose the working status or fault type of the transformer.

Compared with prior art, the beneficial effect of the present invention is:

On the premise of not affecting the normal operation of the transformer, the real-time monitoring of the operation status of the transformer is realized through a non-contact method;

Through AI learning, for the identification and judgment of electrical parameters such as three-phase unbalance and load characteristics during the operation of the transformer, as well as the abnormal state information of partial discharge and no-load, combined with parameters such as the vibration of the operating environment, information fusion processing is carried out in a timely manner at the front end And early warning, which improves the real-time response to transformer faults, reduces the risk of transformer faults, and improves the efficiency and control capabilities of inspection operations;

With the help of 64-channel sound source localization system components, covering audible sound and ultrasonic frequency bands, quickly locate the location of abnormal sound sources, analyze defect types, and greatly improve the efficiency of equipment inspection;

The voiceprint feature extraction component obtains the characteristic parameter distribution of the dispersion degree of the unconnected voltage level and cooling mode by analyzing the transformer voiceprint feature quantity, which is basically similar. Approximate Gaussian normal distribution characteristics, etc., by weighting each characteristic parameter, calculate the threshold parameters of transformer health status attention and abnormality, calculate the health index parameters of the transformer for each set of detection data received, and compare the parameters with the threshold , to determine the current health status of the transformer;

On the basis of collecting a large number of transformer voiceprint samples, the accuracy of fault diagnosis is improved with the help of algorithms.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings that are required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort.

Fig. 1 is a structural schematic diagram of a voiceprint diagnostic device of the present invention;

Fig. 2 is a structural schematic diagram of the first direction of the housing of the present invention;

Fig. 3 is a schematic diagram of the second direction structure of the casing of the present invention;

Fig. 4 is a structural schematic diagram of the components of the channel sound source localization system of the present invention;

Fig. 5 is a schematic diagram of the vibration source and propagation path of the transformer of the present invention.

1. Channel sound source localization system components; 2. Housing; 3. Anti-interference detection components; 4. Voiceprint feature extraction components; 5. Fault voiceprint judgment components.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

As shown in Figure 1-5:

The opening of the housing 2 is provided with a matching channel sound source localization system component 1 used in combination and connected. The inner end of the channel sound source localization system component 1 is connected with an anti-jamming detection component 3 by a line, and the voiceprint feature extraction component 4 is separated from one side A fault voiceprint judging element 5 is installed at the position, and the fault voiceprint judging element 5 is correspondingly embedded in a local groove point on the surface of the housing 2 .

The housing 2 has a horn-shaped/cylindrical/ear-like structure, which is beneficial to sound collection. The channel sound source localization system component 1 is a 64-channel sound source localization system, which can cover audible sound and ultrasonic frequency bands, and quickly locate abnormal sound sources. Analyze the type of defect and greatly improve the efficiency of equipment inspection;

The anti-interference detection element 3 obtains the transformer noise as a stable signal by monitoring the characteristics of the transformer acoustic signal, and the fault voiceprint judgment element 5 calculates the health index parameters of the transformer for each set of detection data received;

The working principle of anti-interference detection element 3: the transformer noise is a stable signal obtained by monitoring the characteristics of the transformer acoustic signal, mainly concentrated in the 50Hz even-number harmonic frequency component within the range of 700Hz, the cooling fan noise is a stable signal, and the spectrum is mainly located in the 2kHz range, except The energy distribution of other frequency components other than the rotation frequency is relatively uniform. However, there are many disturbances in the outside world, such as insects, frogs, birds, vehicles, etc. They are generally sudden, intermittent and concentrated in a certain range. With the help of packet spectrum method, short time window method and filter method The external environment sound interference is reduced.

The working principle of the feature extraction element 4 is as follows: different transformers, loads, aging of cooling devices, winding deformation, different harmonic content, DC bias, and other working conditions can obtain multi-dimensional transformer voiceprint feature quantities and frequency Ratio feature, wavelet packet energy, and dispersion feature, and its central tendency, dispersion degree, skewness and kurtosis are obtained.

The working principle of fault voiceprint judgment element 5: use feature combination to judge transformer fault type, assign weight to each feature parameter, calculate transformer health status attention, abnormal threshold parameters, and calculate transformer health index parameters for each set of detection data received , and compare the parameter with the threshold to determine the current health status of the transformer.

The voiceprint detection model based on MobileNetV3 is adopted, and the model is mainly divided into voiceprint feature extraction module, MobileNetV3 module and loop module. The voiceprint feature extraction module extracts features through the time domain image, while the MobileNetV3 module classifies the features and connects them to the loop module to model through the information of the front and rear frames to improve the accuracy of the model. Use the loss function during training For verification, m and n respectively represent the number of samples and the total number of categories, /> is the cross-entropy loss corresponding to samples, p _i is the probability that the sample is category i. In this way, the voiceprint characteristics of transformer faults can be obtained, and with the real-time acquisition of electrical parameters such as three-phase unbalance and load characteristics during transformer operation, as well as abnormal status information of partial discharge and no-load, the real-time response to transformer faults can be improved.

Compared with the previous model, MobileNetv3 adds the attention mechanism (SE), pools the obtained feature matrix for each channel, and then obtains the output vector through two fully connected layers, of which the first one is fully connected Layer, the number of fully connected layer nodes is equal to 1/4 of the input feature matrix channel, and the channel of the second fully connected layer is consistent with the channel of our feature matrix. After average pooling + two fully connected layers, the output feature vector can be understood as analyzing a weight relationship for each channel of the feature matrix before SE. It thinks that the more important channel will be given a greater weight. For the less important channel dimension, it corresponds to a relatively small weight. Therefore, compared with the previous model, its classification accuracy rate has increased by 3.2%, and the calculation delay has been reduced by 20%.

It also includes a vibration sensor, which is temporarily adsorbed on the external surface of the transformer oil tank to be detected through a permanent magnet during use; the vibration sensor is used to collect the vibration signal of the transformer during operation, and extract the corresponding frequency spectrum characteristics through the processor;

As shown in Figure 5, the vibration of the transformer oil tank has multiple sources, and the vibration frequencies of different sources are different. The vibration spectrum characteristics of the transformer oil tank caused by various situations are pre-stored in the processor;

The processor compares the spectral features of the transformer obtained on site with various pre-stored spectral features, and the spectral feature with the highest similarity is the working state or vibration source of the on-site transformer. Determining the vibration source of the on-site transformer can provide a reference for subsequent transformer faults.

The types of faults diagnosed by the values obtained by each sensor are different, but at the same time, there are often overlaps in the diagnosis of faults between different monitoring systems. Different sensors can be used to monitor the compatibility of different sensors to narrow the range of faults diagnosed. For example, when oil When the medium dissolved gas chromatographic monitoring system finds an overheating fault, it can be analyzed in combination with the monitoring data of the grounding current monitoring system to rule out or confirm whether the multi-point grounding fault of the iron core has caused the insulation to overheat, thereby generating the corresponding gas. In this way, the scope of the fault can be further narrowed, and in some cases the type of fault can be directly determined;

The faults of power transformers are usually divided into external and internal faults. The root causes of internal failures are mainly mechanical looseness of power windings and insulating cores, winding resonance, overheating, degradation of insulating oil, oxidation, moisture, contamination by various chemical solid substances of insulating oil, partial discharge, design and manufacturing defects. Causes of external failures include system failures such as short circuit, lightning strike, system overload and system switching misoperation. Usually, the root causes of internal failures mainly include mechanical looseness of power windings and insulating cores, winding resonance, overheating, degradation of insulating oil, oxidation, moisture, pollution of various chemical solid substances of insulating oil, partial discharge, design and manufacturing defects; traditional methods It cannot accurately detect loose faults of structural components such as windings, iron cores, and clamps, but these faults can be effectively and accurately monitored by vibration.

e ₁ is the obtained oil chromatographic diagnosis data (using David’s triangle method), e ₂ is the voice print diagnosis data obtained through this system, e ₃ is the obtained working condition information record, e ₄ is the obtained total Hydrocarbon content, e ₅ is the total amount of acetylene obtained after reduced load operation; insulation fault, short circuit fault, discharge fault, protection and malfunction fault can be preliminarily judged through e ₁ -e ₅ .

K is a normalization constant, k=1-K; where K is a conflict factor, indicating the conflict between different evidences, the larger K is, the greater the conflict between different evidences, 1/(1-k) is the normalization factor. m ₁ , m ₂ , m ₃ , m ₄ , and m ₅ are the corrected failure probability values of e ₁ , e ₂ , e ₃ , e ₄ , and e ₅ respectively;

The calculation process of data fusion:

By solving the above mass function, the mass function value of each fault is obtained, and the one with the largest function value can be diagnosed as a transformer fault.

By measuring and analyzing the vibration signal on the surface of the transformer through the vibration sensor to judge and predict the fault of the winding and core. Cooperate with gas chromatography analysis method to judge. After detecting the chemical composition of various chemical substances in the dissolved gas in the machine oil and the content of various gases in it through chromatography, it can be well judged and analyzed. For example, when performing chromatographic analysis on the average hydrogen oxide gas content and other component chemical substances contained in dissolved gases in transformer oil, the main causes of gas failures such as iron cores and overheating are in these factors. ₄ and other components such as tetracycloalkene, the average hydrogen oxide concentration content is relatively high, and the average hydrogen oxide concentration content of components such as CO and CO ₂ has little change compared with the previous gases, while the intermittent The fault of multi-point continuous grounding is mainly manifested in the chromatographic analysis, which has been found to include gases such as C ₂ H ₂ .

Embodiment one:

Take the identification of circuit faults, magnetic circuit faults, and no faults as an example:

m ₄ (e ₄ ) is the probability of occurrence of each fault when the total hydrocarbon content is e ₄ after reduced load operation, m ₅ (e ₅ ) is the probability of occurrence of each fault when the total amount of acetylene is e ₅ after reduced load operation;

Circuit fault mass function:

Magnetic circuit fault mass function:

No fault mass function:

After calculating all the formulas, the value of the circuit fault is the largest, and the circuit fault is the fault that occurred;

Through oil chromatography David's triangle diagnosis, voiceprint diagnosis, and working condition information records, insulation faults, short circuit faults, discharge faults, protection and malfunction faults can also be judged.

Embodiment two:

Take the identification of short-circuit faults, insulation faults, and discharge faults as an example:

Short circuit fault:

Insulation fault:

Discharge failure:

After calculating all formulas, the insulation fault value is the largest, and the insulation fault is the fault that occurred.

In the description of this specification, descriptions referring to the terms "one embodiment", "example", "specific example" and the like mean that specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment of the present invention. In an embodiment or example. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The preferred embodiments of the invention disclosed above are only to help illustrate the invention. The preferred embodiments are not exhaustive in all detail, nor are the inventions limited to specific embodiments described. Obviously, many modifications and variations can be made based on the contents of this specification. This description selects and specifically describes these embodiments in order to better explain the principle and practical application of the present invention, so that those skilled in the art can well understand and utilize the present invention. The invention is to be limited only by the claims, along with their full scope and equivalents.

Claims (10)

1. A transformer state fault diagnosis system based on voiceprint sensing, characterized in that: it comprises a housing (2) and a channel sound source localization system component (1) installed on the housing (2), an anti-jamming detection component (3), voiceprint feature extraction component (4), fault voiceprint judgment component (5); The channel sound source localization system component (1) is used to collect the voiceprint data of the transformer, and send the collected voiceprint data to the anti-interference detection component (3); The anti-interference detection element (3) performs filtering processing on the received voiceprint data, and then sends the processed voiceprint data to the voiceprint feature extraction element (4); The voiceprint feature extraction component (4) is used to extract the fundamental frequency and harmonics corresponding to a plurality of targets from the processed voiceprint data; The fault voiceprint judging element (5) is used to compare the fundamental frequency and harmonics of each target with the preset fundamental frequency and harmonics of the target to calculate the similarity, judge the fault type of the transformer, and output the result. 2. A transformer status fault diagnosis system based on voiceprint sensing according to claim 1, characterized in that: the housing (2) is a trumpet-shaped structure with a hollow interior and an opening at one end. 3. A transformer state fault diagnosis system based on voiceprint sensing according to claim 2, characterized in that: the channel sound source localization system component (1) is arranged at the open end of the casing (2). 4. A transformer state fault diagnosis system based on voiceprint sensing according to claim 1, characterized in that: said channel sound source localization system component (1) is a 64-channel sound source localization system, which can cover audible Acoustic and ultrasonic frequency bands. 5. A kind of transformer state fault diagnosis system based on voiceprint sensing according to claim 1, characterized in that: the fault voiceprint judging element (5) builds a voiceprint detection model based on MobileNetV3, and the voiceprint The detection model includes a voiceprint feature extraction module, a MobileNetV3 module and a circulation module; the voiceprint feature extraction module extracts features through a time-domain map, and the MobileNetV3 module classifies the extracted features, and accesses them to the circulation module. The information of the frame is modeled to improve the accuracy of the model. 6. a kind of transformer state fault diagnosis system based on voiceprint sensing according to claim 5, is characterized in that: described fault voiceprint judging element (5) introduces attention mechanism and constructs the voiceprint detection model based on MobileNetV3, For the obtained feature matrix, each channel is pooled, and then the output vector is obtained through two fully connected layers. The first fully connected layer has the number of fully connected layer nodes that is 1 of the input feature matrix channel. /4, the channel of the second fully connected layer is consistent with the channel of the feature matrix. 7. A transformer state fault diagnosis system based on voiceprint sensing according to claim 6, characterized in that: loss function is used in the process of voiceprint detection model training authenticating; Among them, m and n represent the number of samples and the total number of categories, respectively, is the cross-entropy loss corresponding to samples, p _i is the probability that the sample belongs to category i, from which the voiceprint characteristics of transformer faults can be obtained. 8. A transformer state fault diagnosis system based on voiceprint sensing according to claim 7, characterized in that: it also includes a vibration sensor, the vibration sensor is used to collect the vibration signal of the transformer during operation, and through The processor extracts the corresponding spectral features, and uses the obtained spectral features of the transformer during various faults as comparison features; The processor analyzes and compares the spectral features and comparative features to diagnose the working status or fault type of the transformer. 9. A transformer state fault diagnosis method based on voiceprint sensing, characterized in that: comprising: Obtain transformer oil chromatography diagnostic data, voiceprint diagnostic data, working condition information record data, total hydrocarbon content data after load-down operation, and total acetylene data after load-down operation; The mass function under various fault conditions is constructed, and various data are substituted into the mass function under each fault condition, and the one with the largest function value can be diagnosed as a transformer fault. 10. A transformer state fault diagnosis method based on voiceprint sensing according to claim 9, characterized in that: said faults include open circuit faults, insulation faults, circuit faults, magnetic circuit faults and no faults; Open circuit fault mass function: Insulation fault mass function: Circuit fault mass function: Magnetic circuit fault mass function: No fault mass function: Wherein, k=1-K, K is a normalization constant; In open circuit faults and insulation faults: In circuit fault, magnetic circuit fault and no fault: Among them, m ₁ (e ₁ ) is the probability of each fault when the oil chromatography diagnosis is e ₁ , m ₂ (e ₂ ) is the probability of each fault when the voiceprint diagnosis is e ₂ , m ₃ (e ₃ ) is the The condition information is recorded as the probability of each failure when e ₃ is recorded, m ₄ (e ₄ ) is the probability of each failure when the total hydrocarbon content is e ₄ after the load-down operation, m ₅ (e ₅ ) is the total acetylene content after the load-down operation The probability of each fault occurs when the quantity is e ₅ , and A is the type of fault.