Three-phase transformer detection method and system
By extracting the sound signal and soundprint characteristics of the three-phase transformer windings, calculating the similarity, and using the codec model to judge the transformer status, the problem of difficult to control the frequency of regular inspections and poor detection stability is solved, and automatic and objective high accuracy is achieved Sexual testing.
Patent Information
- Application Number
- CN202110967601.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-23
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-08-23
AI Technical Summary
In the prior art, the regular inspection frequency of three-phase transformers is difficult to control, resulting in high costs or delayed failure detection, and the detection effect depends on the inspection workers' technology and equipment, and the stability is poor.
By obtaining the sound signal of each winding of the three-phase transformer within the same time period, extracting the soundprint characteristics, calculating the similarity between the non-repetitive soundprint characteristics, determining whether an abnormality occurred in the transformer, and using the codec model to convert the soundprint characteristics into features Vector, calculate the similarity ratio to determine the transformer state.
It realizes automatic patrol, the detection frequency can be customized, and the detection method is objective, which improves the stability and accuracy of the detection effect and avoids the limitation of the detection results due to sample limitations.
Smart Images

Figure CN113805105B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of sound processing technology, and in particular to a three-phase transformer detection method and system. Background Art
[0002] With the development of power transmission and transformation systems, transformers, as one of the important equipment in power transmission and transformation systems, are used more and more, and their importance is getting higher and higher. Among them, three-phase transformers have a wide range of applications and are often used in substations, industrial and mining enterprises, post and telecommunications, schools, hospitals and national defense, etc. They are also suitable for some low-voltage distribution lines with low power supply voltage and large fluctuations. The three-phase transformer is a transformer composed of three windings. It is mainly divided into three-phase group transformers and three-phase core transformers to adapt to different input voltages and output voltages. Three-phase transformer detection refers to identifying and detecting the operating status of the three-phase transformer in a certain way, and warning and judging the abnormalities that are about to occur or have occurred, so as to avoid the impact of the fault.
[0003] In the related technologies, the main detection method is regular inspection, that is, the inspection workers use technical means to regularly detect and inspect the transformer to identify potential faults. Because the frequency of regular inspections is difficult to control, too high a frequency will lead to high costs, and too low a frequency will lead to delayed fault detection. Furthermore, the detection effect of regular inspections depends on the skills and experience of the inspection workers, and on the inspection equipment used, so the detection effect is not stable enough. Summary of the invention
[0004] Based on this, it is necessary to provide a three-phase transformer detection method and system that can improve the stability of monitoring effect in response to the above technical problems.
[0005] A three-phase transformer detection method, the method comprising:
[0006] Acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three-way voiceprint features;
[0007] Obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two groups of voiceprint features in the three groups of voiceprint features;
[0008] According to the similarity corresponding to each group of voiceprint features, it is determined whether the three-phase transformer is abnormal.
[0009] In one embodiment, obtaining the sound signal generated by each winding in the three-phase transformer in the same time period includes:
[0010] Collecting the original sound signal generated by each winding in the three-phase transformer, filtering out the noise signal in each original sound signal, and intercepting the sound signal generated by each winding in the same time period according to each original sound signal after filtering out the noise signal; or,
[0011] The original sound signal generated by each winding in the three-phase transformer in the same time period is collected, and each original sound signal is subjected to noise reduction processing to obtain the sound signal generated by each winding in the same time period.
[0012] In one embodiment, obtaining the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features includes:
[0013] Obtain a feature vector of each voiceprint feature in each group of voiceprint features;
[0014] According to the feature vector of each voiceprint feature in each group of voiceprint features, the similarity between two voiceprint features in each group of voiceprint features is calculated and used as the similarity corresponding to each group of voiceprint features.
[0015] In one embodiment, obtaining a feature vector of each voiceprint feature in each group of voiceprint features includes:
[0016] Input each voiceprint feature into the codec model, obtain the feature vector output by the encoding end in the codec model, and use it as the feature vector of each voiceprint feature.
[0017] In one embodiment, the training process of the codec model includes:
[0018] The sample voiceprint features are used as the input of the initial model, the sample voiceprint features are used as the expected output of the initial model, and the parameters in the initial model are adjusted until the loss function of the initial model converges to obtain the encoding and decoding model; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
[0019] In one embodiment, judging whether a three-phase transformer is abnormal according to the similarity corresponding to each group of voiceprint features includes:
[0020] Taking any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features respectively, and calculating the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features;
[0021] If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
[0022] A three-phase transformer detection system, the system comprising: a processor and three sensors; each of the three sensors corresponds to each winding in the three-phase transformer, each sensor is arranged at a relative preset position of the corresponding winding, and the three sensors are respectively connected to the processor;
[0023] Three sensors are used to collect sound signals generated by corresponding windings in the same time period;
[0024] The processor is used to extract three voiceprint features from three sound signals, obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, and judge whether an abnormality occurs in the three-phase transformer based on the similarity corresponding to each group of voiceprint features, wherein each group of voiceprint features is composed of two voiceprint features in the three voiceprint features.
[0025] In one embodiment, the relative predetermined position of each sensor and the corresponding winding is the same.
[0026] In one embodiment, the system further includes: a development board; the development board is connected to the three sensors and to the processor;
[0027] The development board is used to set the sampling rates of the three sensors, receive the three sound signals collected by the three sensors, and forward them to the processor.
[0028] In one embodiment, the processor is a cloud server; data is transmitted between the cloud server and the development board via a message queue telemetry transmission protocol.
[0029] The above-mentioned three-phase transformer detection method and system obtains the sound signal generated by each winding in the three-phase transformer in the same time period, and extracts three-way voiceprint features. The similarity corresponding to each group of voiceprint features in any two non-repetitive groups of voiceprint features is obtained, and according to the similarity corresponding to each group of voiceprint features, it is judged whether the three-phase transformer is abnormal. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using samples to train the model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection results. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 1 is a flow chart of a three-phase transformer detection method in one embodiment;
[0031] Figure 2 is a flow chart of a three-phase transformer detection method in another embodiment;
[0032] Figure 3 A schematic flow chart of a three-phase transformer detection method in another embodiment;
[0033] Figure 4 is a schematic diagram showing a normal distribution of similarity ratios in one embodiment;
[0034] Figure 5 is a line graph showing the value of the similarity ratio over time in one embodiment;
[0035] Figure 6 A schematic diagram of the structure of a three-phase transformer detection system in one embodiment;
[0036] Figure 7 A schematic diagram of the relative preset positions of the sensor and the windings in the three-phase transformer in one embodiment;
[0037] Figure 8 It is a structural schematic diagram of a three-phase transformer detection system in another embodiment;
[0038] Fig. 9 is a structural block diagram of a three-phase transformer detection device in one embodiment;
[0039] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0041] It is understood that the terms "first", "second", etc. used in this application can be used in this article to describe various professional terms, but unless otherwise specified, these professional terms are not limited by these terms. These terms are only used to distinguish one professional term from another professional term. For example, without departing from the scope of this application, the third preset threshold and the fourth preset threshold can be the same or different.
[0042] With the development of power transmission and transformation systems, transformers, as one of the important equipment in power transmission and transformation systems, are used more and more, and their importance is getting higher and higher. Among them, three-phase transformers have a wide range of applications and are often used in substations, industrial and mining enterprises, post and telecommunications, schools, hospitals and national defense, etc. They are also suitable for some low-voltage distribution lines with low power supply voltage and large fluctuations. The three-phase transformer is a transformer composed of three windings. It is mainly divided into three-phase group transformers and three-phase core transformers to adapt to different input voltages and output voltages. Three-phase transformer detection refers to identifying and detecting the operating status of the three-phase transformer in a certain way, and warning and judging the abnormalities that are about to occur or have occurred, so as to avoid the impact of the fault.
[0043] In the related technologies, the main detection method is regular inspection, that is, the inspection workers use technical means to regularly detect and inspect the transformer to identify potential faults. Because the frequency of regular inspections is difficult to control, too high a frequency will lead to high costs, and too low a frequency will lead to delayed fault detection. Furthermore, the detection effect of regular inspections depends on the skills and experience of the inspection workers, and on the inspection equipment used, so the detection effect is not stable enough.
[0044] In addition to the above-mentioned regular inspections, online real-time detection can also be used in related technologies. Online real-time detection refers to obtaining the operating data of the transformer through various forms of devices, and using artificial intelligence, data analysis and other technical means to monitor and warn the status of the transformer in real time after data cleaning, preprocessing and other operations. Online real-time detection can be vibration detection. Vibration detection refers to obtaining the vibration signal of the transformer through a vibration sensor, processing the vibration signal, and then predicting the collected data through artificial intelligence or data analysis. Vibration detection requires the use of vibration sensors when collecting data, such as patch-type vibration sensors. Since the vibration sensor needs to be adsorbed on the surface of the transformer, it brings inconvenience to the normal operation and maintenance of the transformer. In addition, patch-type vibration sensors have certain deficiencies in data accuracy.
[0045] When a transformer component is abnormal, different sounds will be emitted, such as "wow wow" sound when overloaded, "hissing" sound when the bushing surface or the corners of the conductor are discharged, etc. Some experienced maintenance workers use the sound of the transformer operation as a basis for judging the transformer status, so the related abnormality detection research also lists the transformer working sound as a key research object. Therefore, in the relevant technology, voiceprint detection can also be used as a means of online real-time detection. Among them, the technical scheme of voiceprint detection is similar to vibration detection, which is mainly to obtain sound signals, then perform data preprocessing (such as noise reduction or filtering, etc.), and then judge the data through artificial intelligence or data analysis technology. Since the number of fault sound sample libraries is usually limited, the reference basis for subsequent judgment by artificial intelligence or data analysis technology is limited, and the judgment result is not accurate enough. In addition, the above-mentioned voiceprint detection method is sensitive to sound, and in addition to the fault sound of the three-phase transformer, there may be other environmental noises in the actual implementation process, such as bird calls or thunder, and the above environmental noises may lead to fault judgment.
[0046] In view of the problems existing in the above-mentioned related technologies, an embodiment of the present invention provides a three-phase transformer detection method. The method can be applied to the abnormality detection scenario of the three-phase transformer. The execution subject of the method can be a local server or a cloud server with data processing function, and the embodiment of the present invention does not specifically limit this. It should be noted that the number of "multiple" and the like mentioned in each embodiment of the present application refers to the number of "at least two", for example, "multiple" means "at least two". See Figure 1 , the method comprising:
[0047] 101. Acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three voiceprint features;
[0048] 102. Obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two groups of voiceprint features in the three groups of voiceprint features;
[0049] 103. According to the similarity corresponding to each group of voiceprint features, determine whether the three-phase transformer is abnormal.
[0050] In the above step 101, the three-phase transformer refers to three independent windings, which are connected in different ways (such as star or triangle) to input a three-phase AC power supply, and the same is true for its output. The three-phase transformer can adapt to different input voltages through three windings, and can output different voltages. In order to obtain the sound signal generated by each winding in the three-phase transformer, sensors can be arranged around the three-phase transformer in advance. Among them, each sensor is used to collect the sound signal generated by a winding in the three-phase transformer, that is, the three sensors correspond to the three windings in the three-phase transformer one by one. The sensor can be a sound collection equipment such as a microphone, and the arrangement position of the sensor can be related to the internal layout of the three-phase transformer, such as being arranged on a bracket that is convenient for placement inside the three-phase transformer. It should be noted that in order to collect sound signals with higher quality as much as possible, the sensor can be arranged in front of the winding in the three-phase transformer.
[0051] Through the above step 101, the sound signals generated by each of the three windings in a certain time period can be obtained, and a total of three voiceprint features can be extracted. If the three voiceprint features are combined, three non-repetitive groups of voiceprint features can actually be obtained. For example, if the three voiceprint features are used as the first voiceprint feature, the second voiceprint feature and the third voiceprint feature, respectively, the first voiceprint feature and the second voiceprint feature can form a group of voiceprint features. Since the second voiceprint feature and the third voiceprint feature form another group of voiceprint features, the "any two groups of voiceprint features" in the above step 102 can be obtained.
[0052] It should be noted that the reason why two groups of voiceprint features are given in step 102 is that in the actual real-time process, the three voiceprint features can be combined to form three groups of voiceprint features, namely, the first group of voiceprint features is formed by the first voiceprint features and the second voiceprint features, the second group of voiceprint features is formed by the first voiceprint features and the third voiceprint features, and the third group of voiceprint features is formed by the second voiceprint features and the third voiceprint features. In step 102, only two groups of voiceprint features need to be selected from the three groups of voiceprint features, and the embodiment of the present invention does not specifically limit the selection method. The "non-repetitive" in the above step 102 means that any two groups of voiceprint features are not completely composed of the same two voiceprint features. For example, if one group of voiceprint features in the any two groups of voiceprint features is composed of the first voiceprint features and the second voiceprint features, then the other group of voiceprint features cannot also be composed of the first voiceprint features and the second voiceprint features, but there are differences. For example, the other group of voiceprint features can be composed of the first voiceprint features and the third voiceprint features, or the second voiceprint features and the third voiceprint features.
[0053] It should also be noted that the similarity corresponding to each group of voiceprint features is actually the similarity between the two voiceprint features in each group of voiceprint features. When obtaining the similarity between the two voiceprint features, since the voiceprint feature is mainly the sound wave spectrum, a cepstrum analysis can be performed based on the sound waveform matching principle to score the matching degree between the two sound wave spectra, and the scoring result is used as the similarity between the two voiceprint features.
[0054] It is understandable that when the three-phase transformer is working normally, the timbre of the sound is usually fixed, and there will be differences only when it is abnormal. On this basis, under the premise of normal operation, the similarity corresponding to each group of voiceprint features should be fixed or stable within a preset range. Therefore, in the above step 103, the similarity value interval corresponding to each group of voiceprint features during normal operation can be pre-determined based on the statistical results.
[0055] It can be seen from the above embodiments that three voiceprint features can constitute three groups of voiceprint features, so that three similarity value intervals can be predetermined. Specifically, the first voiceprint features and the second voiceprint features constitute a first group of voiceprint features corresponding to a first similarity value interval, the first voiceprint features and the third voiceprint features constitute a second group of voiceprint features corresponding to a second similarity value interval, and the second voiceprint features and the third voiceprint features constitute a third group of voiceprint features corresponding to a third similarity value interval.
[0056] If in the above step 102, any two groups of voiceprint features refer to the first channel voiceprint features and the second channel voiceprint features constituting the first group of voiceprint features, and the first channel voiceprint features and the third channel voiceprint features constituting the second group of voiceprint features, then in step 103, it can be determined whether the similarity corresponding to the first group of voiceprint features is within the first similarity value interval, and whether the similarity corresponding to the second group of voiceprint features is within the first similarity value interval.
[0057] If the similarity corresponding to the first group of voiceprint features is not within the first similarity value interval, and / or the similarity corresponding to the second group of voiceprint features is not within the second similarity value interval, it means that there is an abnormality in the winding of the three-phase transformer, and it can be determined that the three-phase transformer is abnormal. At this time, an alarm can be issued through the early warning platform and the early warning information can be displayed. Among them, the displayed early warning information can include the specific location information of the abnormal three-phase transformer.
[0058] If the similarity corresponding to the first group of voiceprint features is within the first similarity value interval, and the similarity corresponding to the second group of voiceprint features is within the second similarity value interval, it means that there is no abnormality in the winding of the three-phase transformer, and it can be determined that the three-phase transformer is working normally.
[0059] The method provided by the embodiment of the present invention obtains the sound signal generated by each winding in the three-phase transformer in the same time period, and extracts three-way voiceprint features. The similarity corresponding to each group of voiceprint features in any two non-repetitive groups of voiceprint features is obtained, and according to the similarity corresponding to each group of voiceprint features, it is judged whether the three-phase transformer is abnormal. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using samples to train the model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results.
[0060] In combination with the content of the above embodiments, in one embodiment, the embodiment of the present invention does not specifically limit the method of obtaining the sound signal generated by each winding in the three-phase transformer in the same time period, including but not limited to the following two methods:
[0061] (1) The first method: collect the original sound signal generated by each winding in the three-phase transformer, filter out the noise signal in each original sound signal, and intercept the sound signal generated by each winding in the same time period based on each original sound signal after filtering out the noise signal.
[0062] Among them, the embodiment of the present invention mainly needs to analyze in real time whether the three-phase transformer is abnormal, and the basis of the real-time analysis is also the original sound signal collected within the historical time period. Thus, the sound signal required for analysis is screened out. Specifically, if an original sound signal is collected within the historical time period from 10:00 am on August 13, 2021 to 10:10 am on August 13, 2021, and there may be several fragments of noise signals in the original sound signal. Therefore, in the first method, these noise signal fragments can be directly screened out from the original sound signal.
[0063] Corresponding to the three windings in the three-phase transformer, the three windings will generate three original sound signals, so each original sound signal can filter out the noise signal fragments in the above manner. It should be noted that, since the sound signal generated by each winding in the same time period is required for further processing, after the above-mentioned process of filtering out the noise signal fragments, there may be inconsistencies in the time period, so each original sound signal after filtering out the noise signal can be time-aligned and intercepted to obtain the sound signal generated by each winding in the same time period.
[0064] Specifically, if a certain original sound signal previously corresponds to the time period from 10:00 a.m. on August 13, 2021 to 10:10 a.m. on August 13, 2021, and after filtering out the noise signal fragment, it becomes the time period from 10:02 a.m. on August 13, 2021 to 10:08 a.m. on August 13, 2021. The other original sound signal also previously corresponded to the time period from 10:00 a.m. on August 13, 2021 to 10:10 a.m. on August 13, 2021, but after filtering out the noise signal fragment, it becomes the time period from 10:03 a.m. on August 13, 2021 to 10:09 a.m. on August 13, 2021. Based on the above description, the two original sound signals after filtering out the noise signal can be time-aligned and intercepted, so that each of them corresponds to the time period from 10:03 am on August 13, 2021 to 10:08 am on August 13, 2021 after interception.
[0065] It should be noted that the above-mentioned screening out of the noise signal in the original sound signal requires identification, which can be done through manual identification or model identification in actual implementation, and the embodiment of the present invention does not specifically limit this.
[0066] (2) The second method: collect the original sound signal generated by each winding in the three-phase transformer in the same time period, perform noise reduction processing on each original sound signal, and obtain the sound signal generated by each winding in the same time period.
[0067] Different from the first method mentioned above, in the second method, it is not necessary to filter out the noise signal segments, but to perform noise reduction processing, such as using an adaptive filter, etc., which is not specifically limited in the embodiment of the present invention.
[0068] The method provided in the embodiment of the present invention removes noise from the original sound signal by screening the noise signal or performing noise reduction processing, and uses the original sound signal after noise removal as the basis for subsequent abnormality analysis, thereby avoiding the interference of environmental noise on abnormality detection, and further improving the accuracy of the detection result.
[0069] In combination with the contents of the above embodiments, in one embodiment, see Figure 2 The embodiment of the present invention does not specifically limit the method of obtaining the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, including but not limited to:
[0070] 201. Obtain a feature vector of each voiceprint feature in each group of voiceprint features;
[0071] 202. Calculate the similarity between two voiceprint features in each voiceprint feature group according to the feature vector of each voiceprint feature in each voiceprint feature group, and use it as the similarity corresponding to each voiceprint feature group.
[0072] In the above step 201, the feature vector may specifically be short-time spectrum features such as Mel-frequency cepstral coefficients and perceptual linear prediction features, or may be features such as D-vector or Deep feature, which is not specifically limited in the embodiment of the present invention.
[0073] In the above step 202, when calculating the similarity between the two voiceprint features, the Euclidean distance or cosine similarity may be used for calculation, which is not specifically limited in the embodiment of the present invention.
[0074] The method provided by the embodiment of the present invention obtains the feature vector of each voiceprint feature in each group of voiceprint features. According to the feature vector of each voiceprint feature in each group of voiceprint features, the similarity between two voiceprint features in each group of voiceprint features is calculated, and used as the similarity corresponding to each group of voiceprint features. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using samples to train the model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, that is, abnormality detection is achieved through the sound similarity between the windings, so that the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results.
[0075] In combination with the contents of the above embodiments, in one embodiment, the embodiment of the present invention does not specifically limit the method of obtaining the feature vector of each voiceprint feature in each group of voiceprint features, including but not limited to: inputting each voiceprint feature into the codec model, obtaining the feature vector output by the encoding end in the codec model, and using it as the feature vector of each voiceprint feature.
[0076] Among them, the encoding and decoding model can be an Encoder-Decoder model applied to the input and output sequences. The structure of the encoding segment and the decoding end can be selected according to the needs, such as a convolutional neural network model or a long short-term memory neural network model, etc. The embodiment of the present invention does not specifically limit the selected model structure. In addition, it should be noted that, similar to image decoding, some local information in the image will have a significant impact on the decoding result. For example, if face recognition in a photo is required, and the position of the face in the photo is usually located in the middle of the photo, the local information in the middle of the photo will have a significant impact on the decoding result, thereby increasing the weight corresponding to the local information in the middle of the photo. That is, an attention mechanism can be used, and a multi-scale attention mechanism can be used. Among them, multi-scale means that there is more than one local information that needs to be paid attention to.
[0077] Similarly, based on the above principles, when analyzing the voiceprint features, some local information also needs to be focused on. It can be seen from the actual practice that when the three-phase transformer is working abnormally, the sound emitted is usually low-frequency, that is, it is in the audio range with lower frequency. Therefore, in the embodiment of the present invention, when the decoding end introduces the multi-scale attention mechanism, the decoding end can introduce several lower-frequency sound information in the voiceprint features during decoding, and decode them accordingly. It should be noted that the embodiment of the present invention mainly uses the encoding end of the encoding and decoding model, but a single encoding end cannot perform supervised training. Therefore, in the embodiment of the present invention, when obtaining the feature vector of the voiceprint feature, in order to train the encoding end, it is necessary to introduce a coding and decoding model with an encoding end and a decoding end at the same time. In addition, the final output of the encoding end in the encoding and decoding model is a hidden state in the format of a feature vector.
[0078] The method provided by the embodiment of the present invention obtains the feature vector output by the encoding end of the encoding and decoding model by inputting each voiceprint feature into the encoding and decoding model, and uses it as the feature vector of each voiceprint feature. Since the inspection can be automatically realized, the detection frequency can be self-determined, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using the sample training model and then using the model for voiceprint analysis, but using the sound correlation when the internal winding of the three-phase transformer is working, that is, the abnormality detection is realized by the sound similarity between the windings, so that the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, since it is not simply using the sample training model and then using the model for voiceprint analysis, but using the encoding and decoding model to convert the voiceprint feature into a feature vector, and thus calculating the similarity ratio between the feature vectors of the voiceprint feature, and then detecting the three-phase transformer based on the ratio statistical distribution when the three-phase transformer is working normally, it can avoid the detection result being limited due to sample limitation, and thus improve the accuracy of the detection result.
[0079] In combination with the contents of the above embodiments, in one embodiment, the embodiment of the present invention does not specifically limit the training method of the codec model, including but not limited to: taking the sample voiceprint features as the input of the initial model, taking the sample voiceprint features as the expected output of the initial model, adjusting the parameters in the initial model until the loss function of the initial model converges to obtain the codec model; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
[0080] It should be noted that in the embodiment of the present invention, when the codec model used is trained, the voiceprint feature is input, and since only the encoding end is used, a supervised training method is required. During the training process, in order to have a basis for measuring whether the model has completed the training, the corresponding loss function is designed based on this. In the embodiment of the present invention, the model of the codec model before training is recorded as the initial model, and the sample voiceprint feature can be used as the input of the initial model, and at the same time, the sample voiceprint feature is used as the output of the initial model (specifically the expected output), that is, after the encoding end encodes the sample voiceprint feature, the decoding end can restore the output result of the encoding end to the original input sample voiceprint feature, and this degree can be reflected by the loss function. Through the above method, known input and output are realized, and supervised training of the initial model is completed based on this. Therefore, during the training process, the higher the similarity between the input sample voiceprint feature (that is, the expected output) and the actual output, the better the model effect. The convergence of the loss function mentioned in the above process can refer to the similarity between the expected output and the actual output being greater than a preset threshold, and the embodiment of the present invention does not specifically limit this.
[0081] The method provided by the embodiment of the present invention uses the sample voiceprint feature as the input of the initial model, uses the sample voiceprint feature as the expected output of the initial model, adjusts the parameters in the initial model until the loss function of the initial model converges, and obtains the encoding and decoding model. Since the inspection can be automatically realized, the detection frequency can be self-determined, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using the sample training model and then using the model for voiceprint analysis, but using the sound correlation when the internal winding of the three-phase transformer is working, that is, the abnormality detection is realized by the sound similarity between the windings, so that the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, since it is not simply using the sample training model and then using the model for voiceprint analysis, but using the encoding and decoding model to convert the voiceprint feature into a feature vector, and thus calculate the similarity ratio between the feature vectors of the voiceprint feature, and then based on the statistical distribution of the ratio when the three-phase transformer is working normally, the three-phase transformer is detected, thereby avoiding the detection result being limited due to sample limitation, and thus improving the accuracy of the detection result.
[0082] In combination with the contents of the above embodiments, in one embodiment, see Figure 3 The embodiment of the present invention does not specifically limit the method of judging whether an abnormality occurs in the three-phase transformer according to the similarity corresponding to each group of voiceprint features, including but not limited to:
[0083] 301. Taking any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features, respectively, and calculating the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features;
[0084] 302. If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
[0085] It should be noted that the first group of voiceprint features is the first channel voiceprint features and the second channel voiceprint features, and the second group of voiceprint features is the first channel voiceprint features and the third channel voiceprint features. The above embodiment uses the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features as the basis for measuring whether the three-phase transformer is abnormal because it is found through statistics that the ratio of the two is normally distributed. For details, please refer to Figure 4 .
[0086] exist Figure 4 In the figure, the horizontal axis represents the specific value of the ratio, and the vertical axis represents the frequency of each ratio. Since the three-phase transformer is in normal working state most of the time, Figure 4It can be clearly seen that the ratio is located in the middle area most of the time, that is, the probability that the ratio is in the value interval corresponding to the middle interval is the highest. Therefore, based on the above monitoring summary, the value interval of the ratio when the three-phase transformer is in normal working state most of the time can be obtained, that is, the preset interval mentioned in the above step 302. Figure 4 , the preset interval can be 0.5 to 0.65.
[0087] in addition, Figure 5 A line graph of the ratio over time is given when the three-phase transformer is abnormal. Figure 5 In the figure, the horizontal axis indicates the advancement of time, and the vertical axis indicates the specific value of the ratio. Figure 5 The upper value line in the figure shows the value of the ratio over time when the three-phase transformer is working abnormally. Figure 5 The value curve at the bottom of the figure shows the value of the ratio over time when the three-phase transformer is working normally. The reason why there are two curves is mainly due to the different calculation objects when calculating the ratio. It can be seen that when the three-phase transformer is working normally, the ratio is usually stable, but when it is working abnormally, it can be referred to Figure 5 The ratio of the value curve in the upper middle part is obviously larger.
[0088] The method provided by the embodiment of the present invention calculates the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features by taking any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features respectively. If the ratio is within the preset interval, it is determined that there is no abnormality in the three-phase transformer. If the ratio is not within the preset interval, it is determined that there is an abnormality in the three-phase transformer. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using the sample training model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, that is, the abnormality detection is realized by the ratio of the sound similarity between the windings, so that the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results.
[0089] It should be understood that although Figure 1 , Figure 2 and Figure 3The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 , Figure 2 and Figure 3 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.
[0090] It should be noted that the technical solutions described above can be implemented as independent embodiments in the actual implementation process, or they can be combined with each other and implemented as a combined embodiment. In addition, when describing the contents of the above embodiments of the present invention, different embodiments are described in a corresponding order based on the idea of convenient description, such as the order of data flow, instead of limiting the execution order between different embodiments. Accordingly, in the actual implementation process, if it is necessary to implement multiple embodiments provided by the present invention, it is not necessary to follow the execution order provided when the embodiments are described by the present invention, but the execution order between different embodiments can be arranged according to the needs.
[0091] In combination with the contents of the above embodiments, in one embodiment, a three-phase transformer detection system is provided. Figure 6 The system comprises:
[0092] A processor 601 and three sensors, the three sensors are respectively denoted as 602, 603 and 604; each of the three sensors corresponds to each winding in the three-phase transformer, each sensor is arranged at a relative preset position of the corresponding winding, and the three sensors are respectively connected to the processor 601;
[0093] Three sensors are used to collect sound signals generated by corresponding windings in the same time period;
[0094] The processor is used to extract three voiceprint features from three sound signals, obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, and judge whether an abnormality occurs in the three-phase transformer based on the similarity corresponding to each group of voiceprint features, wherein each group of voiceprint features is composed of two voiceprint features in the three voiceprint features.
[0095] Among them, the relative preset position refers to the placement position of the sensor, which is not an absolute position, but a relative position to the winding. The reason why the setting position is emphasized here is to minimize the impact of environmental factors on abnormal detection. For other specific implementation processes, please refer to the content of the above method embodiment, which will not be repeated here.
[0096] The system provided by the embodiment of the present invention collects the sound signals generated by each winding in the three-phase transformer in the same time period through three sensors. The processor extracts three voiceprint features from the three sound signals, obtains the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, and judges whether the three-phase transformer is abnormal based on the similarity corresponding to each group of voiceprint features. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using samples to train the model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection results. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results.
[0097] In combination with the content of the above embodiment, in one embodiment, the relative preset position of each sensor and the corresponding winding is the same. It should be noted that in the embodiment of the present invention, the reason why the relative preset position of each sensor and the corresponding winding is required to be the same is because the similarity between the voiceprint features needs to be calculated later, which requires minimizing the acquisition error when collecting the sound signal, and the size of the acquisition error may be affected by the relative preset position between the sensor and the corresponding winding.
[0098] For example, Figure 6In the example, if the relative preset positions of two sensors and the corresponding windings are the same, and the relative preset positions of the third sensor and the corresponding windings are different from the relative preset positions of the first two sensors and the corresponding windings, for example, the first two sensors are facing their respective corresponding windings, and the third sensor is not facing its corresponding winding, then when the third sensor collects the sound signal of the corresponding winding, it will affect the collected sound signal because it is not facing the sound source position, that is, the collection of the sound signal will cause collection errors. Therefore, this error will be further transmitted to the calculation result of the similarity between the voiceprint features, and finally to the ratio between the similarities, thereby affecting the accuracy of abnormality detection. Therefore, it is necessary that the relative preset positions of each sensor and the corresponding winding are the same, so that the three sensors are in the same collection environment as much as possible when collecting sound signals.
[0099] It should be noted that the same relative preset position can be quantified by coordinates. Specifically, the three sensors can be installed at the same height from the ground (that is, c is the same). In addition, they can be facing the center of the winding (that is, b is the same, for the point in the winding facing the sensor, the distance from the point to the edge of the winding), and the distances from the centers of the corresponding windings are the same (that is, a is the same). For details, please refer to Figure 7 .
[0100] The system provided by the embodiment of the present invention can avoid acquisition errors caused by inconsistent positions by setting the relative preset positions of each sensor and the corresponding winding to be the same, thereby improving the accuracy of the detection results.
[0101] In combination with the contents of the above embodiments, in one embodiment, see Figure 8 , the system further includes: a development board 605; the development board 605 is connected to the three sensors and to the processor 601;
[0102] The development board 605 is used to set the sampling rates of the three sensors, and to receive the three sound signals collected by the three sensors and forward them to the processor 601 .
[0103] It should be noted that the sampling rate here is related to the inspection frequency. The higher the sampling rate, the higher the inspection frequency; otherwise, the lower the inspection frequency.
[0104] The system provided by the embodiment of the present invention collects the sound signals generated by each winding in the three-phase transformer in the same time period through three sensors. The processor extracts three voiceprint features from the three sound signals, obtains the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, and judges whether the three-phase transformer is abnormal based on the similarity corresponding to each group of voiceprint features. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable.
[0105] In combination with the content of the above embodiments, in one embodiment, the processor 601 is a cloud server; data is transmitted between the cloud server and the development board via a message queue telemetry transmission protocol.
[0106] Among them, the message queue telemetry transmission protocol is a "lightweight" communication protocol based on the publish / subscribe model, which is built on the TCP / IP protocol. The biggest advantage of the message queue telemetry transmission protocol is that it can provide real-time and reliable message services for remote devices with very little code and limited bandwidth. As an instant messaging protocol with low overhead and low bandwidth usage, it has a wide range of applications in the Internet of Things, small devices, mobile applications, etc. The message queue telemetry transmission protocol is actually a message publish / subscribe transmission protocol based on client-server.
[0107] The system provided by the embodiment of the present invention can automatically implement patrol inspection, the detection frequency can be self-determined and the detection method is objective, so the detection effect is relatively stable. In addition, the use of the message queue telemetry transmission protocol can reduce bandwidth occupancy overhead.
[0108] In combination with the contents of the above embodiments, in one embodiment, Fig. 9 As shown, a three-phase transformer detection device is provided, including: a first acquisition module 901, a second acquisition module 902 and a judgment module 903, wherein:
[0109] The first acquisition module 901 is used to acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three-way voiceprint features;
[0110] The second acquisition module 902 is used to obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two groups of voiceprint features in the three groups of voiceprint features;
[0111] The judgment module 903 is used to judge whether an abnormality occurs in the three-phase transformer according to the similarity corresponding to each group of voiceprint features.
[0112] In one embodiment, the first acquisition module 901 is used to collect the original sound signal generated by each winding in the three-phase transformer, filter out the noise signal in each original sound signal, and intercept the sound signal generated by each winding in the same time period based on each original sound signal after filtering out the noise signal; or, collect the original sound signal generated by each winding in the three-phase transformer in the same time period, perform noise reduction processing on each original sound signal, and obtain the sound signal generated by each winding in the same time period.
[0113] In one embodiment, the second acquisition module 902 includes:
[0114] An acquisition unit, used for acquiring a feature vector of each voiceprint feature in each group of voiceprint features;
[0115] The calculation unit is used to calculate the similarity between two voiceprint features in each group of voiceprint features according to the feature vector of each voiceprint feature in each group of voiceprint features, and use it as the similarity corresponding to each group of voiceprint features.
[0116] In one embodiment, the acquisition unit is used to input each voiceprint feature into a codec model, acquire a feature vector output by an encoding end in the codec model, and use the feature vector as the feature vector of each voiceprint feature.
[0117] In one embodiment, the second acquisition module 902 further includes:
[0118] The training unit is used to take the sample voiceprint features as the input of the initial model, take the sample voiceprint features as the expected output of the initial model, adjust the parameters in the initial model until the loss function of the initial model converges, and obtain the encoding and decoding model; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
[0119] In one embodiment, the judgment module 903 is used to take any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features, respectively, and calculate the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features; if the ratio is within the preset interval, it is determined that there is no abnormality in the three-phase transformer; if the ratio is not within the preset interval, it is determined that there is an abnormality in the three-phase transformer.
[0120] The device provided by the embodiment of the present invention obtains the sound signal generated by each winding in the three-phase transformer in the same time period, and extracts three-way voiceprint features. The similarity corresponding to each group of voiceprint features in any two non-repetitive groups of voiceprint features is obtained, and according to the similarity corresponding to each group of voiceprint features, it is judged whether the three-phase transformer is abnormal. Since the inspection can be automatically realized, the detection frequency can be customized, and the detection method is objective, the detection effect is relatively stable. In addition, since it is not simply using samples to train the model and then using the model for voiceprint analysis, but using the sound correlation of the internal windings of the three-phase transformer when working, the detection method is not independent of the transformer body, but can be combined with the mechanism characteristics of the three-phase transformer, thereby improving the accuracy of the detection result. Finally, instead of simply using samples to train the model and then using the model for voiceprint analysis, the voiceprint features are converted into feature vectors using a codec model, and the similarity ratio between the feature vectors of the voiceprint features is calculated. The three-phase transformer is then detected based on the statistical distribution of the ratio when the three-phase transformer is working normally. This can avoid the detection results being limited due to sample limitations, and thus improve the accuracy of the detection results.
[0121] The specific definition of the three-phase transformer detection device can be found in the definition of the three-phase transformer detection method above, which will not be repeated here. Each module in the above-mentioned three-phase transformer detection device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0122] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store preset thresholds. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a three-phase transformer detection method is implemented.
[0123] Those skilled in the art will understand that Fig.10The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0124] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:
[0125] Acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three-way voiceprint features;
[0126] Obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two groups of voiceprint features in the three groups of voiceprint features;
[0127] According to the similarity corresponding to each group of voiceprint features, it is determined whether the three-phase transformer is abnormal.
[0128] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0129] Collecting the original sound signal generated by each winding in the three-phase transformer, filtering out the noise signal in each original sound signal, and intercepting the sound signal generated by each winding in the same time period according to each original sound signal after filtering out the noise signal; or,
[0130] The original sound signal generated by each winding in the three-phase transformer in the same time period is collected, and each original sound signal is subjected to noise reduction processing to obtain the sound signal generated by each winding in the same time period.
[0131] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0132] Obtain a feature vector of each voiceprint feature in each group of voiceprint features;
[0133] According to the feature vector of each voiceprint feature in each group of voiceprint features, the similarity between two voiceprint features in each group of voiceprint features is calculated and used as the similarity corresponding to each group of voiceprint features.
[0134] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0135] Input each voiceprint feature into the codec model, obtain the feature vector output by the encoding end in the codec model, and use it as the feature vector of each voiceprint feature.
[0136] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0137] The sample voiceprint features are used as the input of the initial model, the sample voiceprint features are used as the expected output of the initial model, and the parameters in the initial model are adjusted until the loss function of the initial model converges to obtain the encoding and decoding model; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
[0138] In one embodiment, when the processor executes the computer program, the following steps are also implemented:
[0139] Taking any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features respectively, and calculating the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features;
[0140] If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
[0141] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
[0142] Acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three-way voiceprint features;
[0143] Obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two groups of voiceprint features in the three groups of voiceprint features;
[0144] According to the similarity corresponding to each group of voiceprint features, it is determined whether the three-phase transformer is abnormal.
[0145] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0146] Collecting the original sound signal generated by each winding in the three-phase transformer, filtering out the noise signal in each original sound signal, and intercepting the sound signal generated by each winding in the same time period according to each original sound signal after filtering out the noise signal; or,
[0147] The original sound signal generated by each winding in the three-phase transformer in the same time period is collected, and each original sound signal is subjected to noise reduction processing to obtain the sound signal generated by each winding in the same time period.
[0148] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0149] Obtain a feature vector of each voiceprint feature in each group of voiceprint features;
[0150] According to the feature vector of each voiceprint feature in each group of voiceprint features, the similarity between two voiceprint features in each group of voiceprint features is calculated and used as the similarity corresponding to each group of voiceprint features.
[0151] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0152] Input each voiceprint feature into the codec model, obtain the feature vector output by the encoding end in the codec model, and use it as the feature vector of each voiceprint feature.
[0153] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0154] The sample voiceprint features are used as the input of the initial model, the sample voiceprint features are used as the expected output of the initial model, and the parameters in the initial model are adjusted until the loss function of the initial model converges to obtain the encoding and decoding model; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
[0155] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:
[0156] Taking any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features respectively, and calculating the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features;
[0157] If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
[0158] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0159] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0160] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. A three-phase transformer detection method, characterized in that: The method comprises: Acquire the sound signal generated by each winding in the three-phase transformer in the same time period, and extract three-way voiceprint features; Obtaining the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where each group of voiceprint features is composed of two voiceprint features in the three voiceprint features, and non-repeating means that the any two groups of voiceprint features are not completely composed of the same two voiceprint features; Taking the arbitrary two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features respectively, and calculating the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features; If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
2. The method according to claim 1, characterized in that The step of obtaining the sound signal generated by each winding in the three-phase transformer in the same time period includes: Collecting the original sound signal generated by each winding in the three-phase transformer, filtering out the noise signal in each original sound signal, and intercepting the sound signal generated by each winding in the same time period according to each original sound signal after filtering out the noise signal; or, The original sound signal generated by each winding in the three-phase transformer in the same time period is collected, and noise reduction processing is performed on each original sound signal to obtain the sound signal generated by each winding in the same time period.
3. The method according to claim 1, characterized in that The obtaining of the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features includes: Obtain a feature vector of each voiceprint feature in each group of voiceprint features; According to the feature vector of each voiceprint feature in each group of voiceprint features, the similarity between two voiceprint features in each group of voiceprint features is calculated and used as the similarity corresponding to each group of voiceprint features.
4. The method according to claim 3, characterized in that The step of obtaining a feature vector of each voiceprint feature in each group of voiceprint features includes: Each voiceprint feature is input into the codec model, and a feature vector output by the encoding end in the codec model is obtained and used as the feature vector of each voiceprint feature.
5. The method according to claim 4, characterized in that The training process of the encoding and decoding model includes: The sample voiceprint features are used as the input of the initial model, the sample voiceprint features are used as the expected output of the initial model, the parameters in the initial model are adjusted until the loss function of the initial model converges, and the encoding and decoding model is obtained; wherein the loss function is used to measure the similarity between the actual output of the initial model and the expected output.
6. A three-phase transformer detection system, characterized in that: The system comprises: a processor and three sensors; each of the three sensors corresponds to each winding in the three-phase transformer, each sensor is arranged at a relative preset position of the corresponding winding, and the three sensors are respectively connected to the processor; The three sensors are respectively used to collect sound signals generated by corresponding windings in the same time period; The processor is used to extract three voiceprint features from three sound signals, obtain the similarity corresponding to each group of voiceprint features in any two non-repeating groups of voiceprint features, where non-repeating means that the any two groups of voiceprint features are not completely composed of the same two groups of voiceprint features, and use the any two groups of voiceprint features as the first group of voiceprint features and the second group of voiceprint features, respectively, and calculate the ratio between the similarity corresponding to the first group of voiceprint features and the similarity corresponding to the second group of voiceprint features; If the ratio is within the preset interval, it is determined that the three-phase transformer is not abnormal; if the ratio is not within the preset interval, it is determined that the three-phase transformer is abnormal.
7. The system according to claim 6, characterized in that The relative preset position of each sensor and the corresponding winding is the same.
8. The system according to claim 7, characterized in that Also includes: Development board; The development board is connected to the three sensors and to the processor; The development board is used to set the sampling rates of the three sensors, and to receive the three sound signals collected by the three sensors and forward them to the processor.
9. The system according to claim 8, characterized in that The processor is a cloud server; data is transmitted between the cloud server and the development board via a message queue telemetry transmission protocol.
