Intelligent transformer control method, device, storage medium, system and transformer

Abstract

The present invention is applicable to the field of transformer control technology, and provides an intelligent transformer control method, device, storage medium, system and transformer, the method comprising: pre-processing multi-channel sound wave signals to obtain a plurality of sample signals, the multi-channel sound wave signals carrying vibration waveform information under the working state of the transformer; based on the learning separation model, separating and processing the sample signals to obtain normal sound waves and impurity sound waves, extracting features of the impurity sound waves to obtain impurity sound wave features; performing similarity comparison between the impurity sound wave features and the abnormal sound wave features in the abnormal feature library, comparing the comparison results of the similarity comparison with the fault possibility threshold, and obtaining different abnormal levels of the transformer; based on different abnormal levels, executing corresponding transformer working state adjustment schemes. The method provided by the present application can detect the working state of the transformer in real time without stopping, and has high detection accuracy and high speed.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer control, and in particular relates to an intelligent transformer control method, device, storage medium, system and transformer. Background Art

[0002] A transformer is an electrical device used to convert electrical energy. It has a high power transfer capacity. It is mainly used to convert electrical energy from one voltage level to another to meet the needs of different power systems or equipment.

[0003] Transformers are usually composed of an iron core, windings, and a casing. The iron core is made of highly permeable materials to concentrate the magnetic field and reduce magnetic losses. The windings are coils made of conductive materials, including low-voltage and high-voltage side windings. The casing is used to protect the internal components of the transformer and provide heat dissipation and insulation.

[0004] In order to ensure the normal operation and safety of high-power transformers, regular maintenance and inspection are required. This includes monitoring temperature and humidity, checking insulation performance, cleaning windings and cooling systems, etc. The existing monitoring and inspection of transformers, in order to ensure the safety of equipment and inspection personnel, the mainstream method is to use switching equipment to connect the load to the backup equipment, and then perform performance tests on the equipment in the shutdown state. Therefore, it is difficult for existing detection methods to obtain the cause and degree of the fault at the first time of equipment failure or abnormality, and it is difficult to monitor or discover potential faults of the equipment in real time. Often, it is only possible to switch to the backup transformer equipment in an emergency after the equipment has obvious abnormalities. This brings risks to the stable operation of the power system and increases the cost of daily operation and maintenance. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide an intelligent transformer control method, which aims to solve the problem that the existing transformer status detection method is difficult to obtain the cause and degree of the fault in the first time of equipment failure or abnormality, and it is difficult to monitor or discover potential faults of the equipment in real time. It is often only possible to switch to the backup transformer equipment in an emergency after the equipment has a more obvious abnormality, which brings risks to the stable operation of the power system and increases the cost of daily operation and maintenance.

[0006] The embodiment of the present application is implemented as follows: a smart transformer control method is provided, the method comprising: pre-processing a plurality of sound wave signals to obtain a plurality of sample signals, the plurality of sound wave signals carrying vibration waveform information of the transformer under working conditions; separating and processing the sample signals based on a learning separation model to obtain normal sound waves and impurity sound waves, the normal sound waves being sound wave waveforms of the transformer under normal working conditions, and the impurity sound waves being sound wave waveforms generated due to abnormal working conditions of the transformer; extracting features of the impurity sound waves to obtain impurity sound wave features; performing a similarity comparison between the impurity sound wave features and abnormal sound wave features in an abnormal feature library, the abnormal sound wave features comprising electromagnetic abnormal features and mechanical abnormal features, the electromagnetic abnormal features being used to characterize the sound wave features of electromagnetic abnormalities caused by electromagnetic abnormal phenomena of the transformer, and the mechanical abnormal features being used to characterize the sound wave features of mechanical abnormalities caused by mechanical failures of the transformer; comparing the comparison result of the similarity comparison with a fault possibility threshold to obtain different abnormal levels of the transformer; obtaining a preset control scheme, and executing corresponding transformer working state adjustment schemes based on different abnormal levels.

[0007] Preferably, the method for preprocessing multiple sound wave signals to obtain a plurality of sample signals is as follows: based on a plurality of vibration sensors, multiple sound wave signals are obtained, wherein the vibration sensors are used to obtain time domain waveform data of vibration conditions at different positions of the transformer; the multiple sound wave signals are sampled, expanded and aligned, and each sound wave signal is set as a sample signal to obtain a plurality of sample signals.

[0008] Preferably, based on the learning separation model, the method for separating and processing the sample signal is: obtaining the normal spectrum historical data of the transformer to be analyzed within the historical normal working time period, and constructing the normal historical data matrix of the transformer based on the normal spectrum historical data:

[0009] M=t0f(β0)+t1f(β1)+t2f(β2)+…+t n f(β n )

[0010] Where:

[0011]

[0012] Among them, t n represents a time period, N represents dividing 24 hours into N equal-length time periods, f(β n ) represents the matrix composed of different audio data corresponding to different loads of the transformer in a period of time, Indicates different load levels, The time series representation of the normal sound wave of the transformer under different loads, s(qm ) represents the time series representation of the normal sound wave waveform, M represents the normal historical data matrix of the transformer; obtain a number of abnormal spectrum data from the abnormal sound wave library, and compare the abnormal spectrum data with s(q m ) are mixed separately to obtain several mixed spectrum data matrices; a learning separation model is constructed based on the Deep Attractor Network network model; the learning separation model network is trained based on the training samples composed of the normal historical data matrix, the mixed spectrum data matrix and the abnormal spectrum data to obtain the learning separation model; the acquisition time of the sample signal to be processed and the load level of the transformer within the acquisition time are obtained; the sample signal to be processed and its acquisition time and the corresponding load level are input as input to the learning separation model, and the normal sound wave and the impurity sound wave in the sample signal are obtained as output.

[0013] Preferably, the feature extraction of the impurity sound wave to obtain the feature of the impurity sound wave is as follows: dividing the sound wave signal into a number of overlapping frames, and applying a Hamming window to each frame for windowing; applying a fast Fourier transform to each frame to obtain the spectrum of each frame; applying a number of Mel filters to filter the spectrum of each frame to obtain the energy of each filter channel; taking the logarithm of the energy value of each Mel filter group to obtain the logarithmic energy of each frame; applying a discrete cosine transform to the logarithmic energy vector of each frame to obtain a number of cepstral coefficients; setting the number of cepstral coefficients as the impurity sound wave feature corresponding to the impurity sound wave;

[0014] The method for obtaining abnormal sound wave features in the abnormal feature library is as follows: obtaining a number of abnormal spectrum data from the abnormal sound wave library, performing feature extraction on the abnormal spectrum data, and obtaining a single type of abnormal sound wave features; selecting any two of the single type of abnormal sound wave features, setting them as features F1 and F2, and subjecting the features F1 and F2 to spatial pyramid pooling block processing to obtain their respective multi-scale feature representations P1 and P2:

[0015] P1=SpatialPyramidPooling(F1)

[0016] P2=SpatialPyramidPooling(F2)

[0017] Among them, SpatialPyramidPooling represents the spatial pyramid pooling block processing; the multi-scale feature representation is fused to obtain the fused feature F0:

[0018] F0=Concatenate(P1,P2)

[0019] Among them, Concatenate represents a fusion operation; the abnormal sound wave feature includes the fusion feature and the single type of abnormal sound wave feature.

[0020] Preferably, the method for constructing the abnormal feature library is: obtaining a mechanical fault sound database and an electromagnetic fault sound database of the transformer, the mechanical fault sound database containing a number of typical mechanical fault audio data, and the electromagnetic fault sound database containing a number of typical electromagnetic fault audio data; performing feature extraction on the typical mechanical fault audio data and the typical electromagnetic fault audio data, respectively, to obtain the mechanical abnormality feature and the electromagnetic abnormality feature.

[0021] Preferably, the method for performing similarity comparison between the impurity acoustic wave feature and the abnormal acoustic wave feature in the abnormal feature library is:

[0022] The two acoustic wave features to be compared are regarded as vectors A and B in space respectively; the similarity coefficient Similarity(A,B) of the two features to be compared is obtained based on the following formula:

[0023]

[0024] Among them, A and B represent any two vectors whose similarities are to be compared.

[0025] Another object of the embodiment of the present application is to provide an intelligent transformer control device, the intelligent transformer control device comprising:

[0026] The sample signal acquisition module is used to pre-process the multi-channel sound wave signals to obtain a plurality of sample signals, wherein the multi-channel sound wave signals carry the vibration waveform information of the transformer under the working state; the sample signal separation module is used to separate and process the sample signals based on the learning separation model to obtain normal sound waves and impurity sound waves, wherein the normal sound wave is the sound wave waveform of the transformer under the normal working state, and the impurity sound wave is the sound wave waveform caused by the abnormal working of the transformer; the impurity sound wave feature acquisition module is used to extract the features of the impurity sound wave to obtain the impurity sound wave features; the similarity comparison module is used to compare the impurity sound wave features with the The abnormal sound wave features in the abnormal feature library are compared for similarity, and the abnormal sound wave features include electromagnetic abnormal features and mechanical abnormal features. The electromagnetic abnormal features are used to characterize the sound wave features of electromagnetic abnormalities caused by electromagnetic abnormal phenomena of the transformer, and the mechanical abnormal features are used to characterize the sound wave features of mechanical abnormalities caused by mechanical failures of the transformer; an abnormality level acquisition module is used to compare the comparison result of the similarity comparison with the fault possibility threshold to obtain different abnormal levels of the transformer; a working state adjustment module is used to obtain a preset control scheme and execute the corresponding transformer working state adjustment scheme based on different abnormal levels.

[0027] Another object of an embodiment of the present application is to provide a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the processor executes the steps of the intelligent transformer control method as described above.

[0028] Another object of an embodiment of the present application is to provide an intelligent transformer control system, which, when running, executes the steps of the intelligent transformer control method as described above.

[0029] Another object of an embodiment of the present application is to provide an intelligent transformer, which, when in operation, executes the steps of the intelligent transformer control method as described above to monitor its operating status.

[0030] An intelligent transformer control method provided in an embodiment of the present application has the advantages that: by learning the separation model, it is beneficial to improve the resolution accuracy of impurity sound waves and normal working sound waves, which is convenient for feature comparison in subsequent steps; by performing a similarity comparison between the impurity sound wave features and the abnormal sound wave features in the abnormal feature library, it is possible to improve the separation and recognition of different types of faults, and it is beneficial to expand the feature sound wave library that characterizes the faults, thereby improving the accuracy of judging different fault types. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flow chart of an intelligent transformer control method provided in an embodiment of the present application;

[0032] Figure 2 A module diagram of an intelligent transformer control device provided in an embodiment of the present application;

[0033] Figure 3 FIG. 4 is a block diagram of the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not intended to limit the present invention.

[0035] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script without departing from the scope of this application.

[0036] like Figure 1 As shown, in one embodiment, a smart transformer control method is proposed. This embodiment mainly uses this method to be applied to the smart transformer device body, and can also be applied to the cloud or local monitoring device used to monitor or control the transformer device. It can also be applied to integrated cloud hosts, cloud computers, local control computers, cloud servers and other devices. This application does not limit this. A smart transformer control method, the method may include the following steps:

[0037] S10: Preprocessing the multi-channel sound wave signals to obtain a number of sample signals, wherein the multi-channel sound wave signals carry vibration waveform information of the transformer under working conditions.

[0038] In this embodiment, the sound wave signals collected by the system may include vibration waveforms and sounds caused by vibrations, etc.

[0039] S20: Based on the learning separation model, the sample signal is separated and processed to obtain a normal sound wave and an impurity sound wave, wherein the normal sound wave is a sound wave waveform of the transformer under normal working conditions, and the impurity sound wave is a sound wave waveform generated due to abnormal operation of the transformer.

[0040] In this embodiment, the learning separation model can be a data processing model built based on a self-learning neural network architecture, which is used to separate the impurity sound wave from the mixed sound wave, and then facilitate the feature extraction of the impurity sound wave in the subsequent steps. Since the transformer also generates noise during normal operation, if the feature extraction of the mixed signal is performed directly, it will be affected by the noise generated by the normal operation of the transformer equipment, resulting in a large number of features that cannot represent the sound characteristics of the fault sound source itself, and the extraction accuracy is low.

[0041] S30: extracting features of the impurity sound wave to obtain impurity sound wave features.

[0042] S40: performing a similarity comparison between the impurity acoustic wave feature and the abnormal acoustic wave feature in the abnormal feature library, wherein the abnormal acoustic wave feature includes an electromagnetic abnormal feature and a mechanical abnormal feature. The electromagnetic abnormal feature is used to characterize the acoustic wave feature of the electromagnetic abnormality caused by the electromagnetic abnormality of the transformer, and the mechanical abnormality feature is used to characterize the acoustic wave feature of the mechanical abnormality caused by the mechanical failure of the transformer.

[0043] S50: Compare the comparison result of the similarity comparison with the fault possibility threshold to obtain different abnormality levels of the transformer.

[0044] In this embodiment, the impurity acoustic wave characteristics are compared with the acoustic wave characteristics of each type. When the comparison credibility is higher than a threshold, it can be considered that the transformer has this type of abnormality. Each abnormality has its corresponding preset fault level.

[0045] S60: Obtain a preset control plan, and execute a corresponding transformer working state adjustment plan based on different abnormal levels.

[0046] In the embodiment of the present application, the vibration of the transformer under working condition or the information of the sound waves generated by the vibration is first collected to obtain a large number of sample signals. Then the normal working sound waves in the sample are filtered out to obtain the impurity sound waves, and feature extraction is performed. Finally, the features of the abnormal sound waves are compared with the features of the typical abnormalities, and a threshold can be designed, that is, if the comparison similarity is higher than the threshold, it can be considered that the abnormality has occurred. The transformer is processed accordingly based on the severity of the abnormality.

[0047] Those skilled in the art will know that normal vibration occurs when the transformer is working. This is generally because the silicon wafer, iron core and winding of the transformer are subjected to periodically changing electromagnetic forces under the action of the alternating electromagnetic field. When the alternating magnetic flux passes through the iron core, a regular magnetostrictive effect will be generated, causing the iron core and the internal structure to undergo slight deformations, which will cause vibrations of the iron core and the windings. For example, the iron core material will undergo slight dimensional changes during the magnetization and demagnetization process, a phenomenon called magnetostriction. Although this dimensional change is very small, it can still cause significant vibration and noise due to the high frequency of change. The above noises can all be called regular normal sound waves generated by the transformer during normal operation, and will not affect the operation of the transformer. In some cases, the transformer becomes abnormal and then emits abnormal sounds. In this application, these abnormal sounds are classified into two categories: sounds caused by mechanical failures and sounds caused by electromagnetic anomalies.

[0048] The present application obtains several fault characteristics based on the characteristics of abnormal sounds in different situations. Among them, the fault characteristics can include several levels, preferably divided into four levels, the first level can represent a slight abnormality, the second level can represent an abnormality that requires warning and attention, the third level can represent an abnormality that requires adjustment of the transformer, and the fourth level can represent a serious or emergency abnormality.

[0049] In the embodiment of the present application, based on different abnormal levels, the method for executing the corresponding preset control scheme can be that the first-level abnormal state can refer to the abnormal sound such as vibration caused by general electromagnetic abnormality caused by temporary sudden change of transformer load, temporary three-phase imbalance, etc., which does not need to be adjusted, and can only be recorded in the monitoring database without processing; the second-level abnormal level can be the noise generated by the eddy current induced in the iron core by the AC magnetic field, or the vibration of the local area of ​​the iron core induced by the local heating and additional electromagnetic force or local additional electromagnetic force caused by the eddy current, etc. Such abnormalities can only provide an alarm when they occur for a short time. When these occur frequently, they can be identified as faults, thereby increasing the level of abnormal faults. The third level can refer to faults that require intervention and adjustment of the transformer working state due to overvoltage or frequency fluctuations, excessive load, and the occurrence of high-order harmonics. For example, after the abnormality occurs, the transformer load is reduced, the filtering level is increased, and other operations are performed to adjust the transformer working state. The fourth level can refer to noise caused by local short circuit or discharge of metal parts, serious loosening of core fasteners, and serious resonance of equipment. In this case, the transformer can be shut down for maintenance. Through different control schemes, the status of the transformer can be detected in real time, potential risks can be discovered in time, and problems can be avoided from escalating.

[0050] Those skilled in the art will know that for large transformers, such as transformers of 500KV or above, the cost is often in the billions. If it is difficult to detect minor faults in time, these faults will gradually deteriorate over time and eventually cause serious property losses. The separation control method provided in the present application first separates the noise from the normal working noise, then performs feature comparison on the noise separately, performs feature identification, and then takes preset means to control the transformer. The advantage is that by learning the separation model, it is helpful to improve the resolution accuracy of impurity sound waves and normal working sound waves, which is convenient for feature comparison in subsequent steps; by comparing the impurity sound wave features with the abnormal sound wave features in the abnormal feature library for similarity, it is possible to improve the separation recognition of different types of faults, and it is helpful to expand the feature sound wave library that characterizes faults, thereby improving the accuracy of judging different types of faults.

[0051] In a preferred embodiment, a method for preprocessing multiple sound wave signals to obtain a plurality of sample signals is as follows: based on a plurality of vibration sensors, multiple sound wave signals are obtained, wherein the vibration sensors are used to obtain time domain waveform data of vibration conditions at different positions of the transformer; the multiple sound wave signals are sampled, expanded and aligned, and each sound wave signal is set as a sample signal to obtain a plurality of sample signals.

[0052] In the embodiment of the present application, a large number of vibration sensors can be arranged at the same or different positions of the transformer, and these sensors can collect vibration information or sound information emitted by vibration sound waves. This information is sampled based on time to obtain several time periods, and then the amplitude can be expanded first to reduce the difficulty of signal processing. Finally, the sampling time duration can be aligned to unify the data processing template and reduce the difficulty of data processing. Finally, several sample signals to be processed are obtained.

[0053] In a preferred embodiment, based on the learning separation model, the method for separating and processing the sample signal is: obtaining the normal spectrum historical data of the transformer to be analyzed within the historical normal working time period, and constructing the normal historical data matrix of the transformer based on the normal spectrum historical data:

[0054] M=t0f(β0)+t1f(β1)+t2f(β2)+…+t n f(β n )

[0055] Where:

[0056]

[0057] Among them, t n represents a time period, N represents dividing 24 hours into N equal-length time periods, f(β n ) represents the matrix composed of different audio data corresponding to different loads of the transformer in a period of time, Indicates different load levels, The time series representation of the normal sound wave of the transformer under different loads, s(q m ) represents the time series representation of the normal sound wave waveform, M represents the normal historical data matrix of the transformer; obtain a number of abnormal spectrum data from the abnormal sound wave library, and compare the abnormal spectrum data with s(q m ) are mixed separately to obtain several mixed spectrum data matrices; a learning separation model is constructed based on the Deep Attractor Network network model; the learning separation model network is trained based on the training samples composed of the normal historical data matrix, the mixed spectrum data matrix and the abnormal spectrum data to obtain the learning separation model; the acquisition time of the sample signal to be processed and the load level of the transformer within the acquisition time are obtained; the sample signal to be processed and its acquisition time and the corresponding load level are input as input to the learning separation model, and the normal sound wave and the impurity sound wave in the sample signal are obtained as output.

[0058] In this embodiment, since the working conditions and ambient temperature of the transformer may be different in different time periods, and the ambient temperature and the load level of the transformer may affect the specific waveform of the normal noise generated by the transformer during normal operation, causing it to change. In order to accurately separate the impurity noise from the normal noise, the present application constructs a normal noise data matrix corresponding to different time periods and different loads based on the load and time information in the transformer historical data, so as to characterize the normal sound wave conditions of the transformer under different time periods and different loads. When filtering noise, the historical data of normal noise in the same time period and under the same load as the sample to be processed is used as the sample of the self-learning model, and the normal sound wave model in the same time period and under the same load as the sample data is compared and filtered, thereby improving the separation accuracy of the sample to be processed.

[0059] In this embodiment, preferably, t n It can be set to 60 seconds. The self-learning of the separation model can refer to obtaining the sound waveform of the normal working state of the historical data of the transformer to be tested for learning, and using big data to learn the sound waveform of other transformer equipment has low accuracy. The present application finds that even for transformers of the same model and batch produced, affected by differences in their installation accuracy, installation environment, installation site, and complex electromagnetic conditions of the internal components of the transformer equipment, the normal working noise generated by each device under its working condition is more prominently different. If the sound of its normal operation is not collected separately for each transformer, this will significantly affect the separation accuracy of the noise in the subsequent steps, as well as the accuracy of fault identification. Therefore, the present application collects the historical working data of the transformer to be monitored itself as a data source, that is, it has self-learning ability, and the longer the normal working time, the more accurate the fault analysis.

[0060] In this embodiment, the learning separation model network uses the Deep Attractor Network deep neural network model, which uses a deep neural network combined with the concept of attractor dynamics to achieve sound separation by learning the attractor in the audio signal. This network architecture has significant advantages for the processing of noise characteristics of transformer equipment. The network includes an embedding network and an attractor generation network, and the goal of using mixed audio data as input is to learn to generate embedding vectors related to each sound source. In the embedding space, clustering algorithms (such as k-means) can be used to cluster the embedding vectors to generate attractor vectors for each source. The attractor generation network is trained by minimizing the Euclidean distance from the embedding vector to the corresponding attractor. And then it is achieved by using the loss function between the label of the attractor and the embedding vector. Finally, the embedding network and the attractor generation network are jointly trained to optimize the entire model to complete the training. Through this method, abnormal sounds can be separated from the mix.

[0061] In a preferred embodiment, the method for extracting the features of the impurity sound wave to obtain the features of the impurity sound wave can be: dividing the sound wave signal into several overlapping frames, and applying a Hamming window to each frame for windowing; applying a fast Fourier transform to each frame to obtain the spectrum of each frame; applying several Mel filters to filter the spectrum of each frame to obtain the energy of each filter channel; taking the logarithm of the energy value of each Mel filter group to obtain the logarithmic energy of each frame; applying a discrete cosine transform to the logarithmic energy vector of each frame to obtain several cepstral coefficients; and setting the several cepstral coefficients to the impurity sound wave features corresponding to the impurity sound wave.

[0062] In the embodiments of the present application, a method for extracting the characteristics of impurity sound waves is provided. It can be understood that the method can also be used for extracting the characteristics of all sound waves in the present application.

[0063] In a preferred embodiment, the method for obtaining abnormal sound wave features in the abnormal feature library is as follows: obtaining a number of abnormal spectrum data from the abnormal sound wave library, performing feature extraction on the abnormal spectrum data, and obtaining a single type of abnormal sound wave features; selecting any two of the single type of abnormal sound wave features, setting them as features F1 and F2, and subjecting the features F1 and F2 to spatial pyramid pooling block processing to obtain their respective multi-scale feature representations P1 and P2:

[0064] P1=SpatialPyramidPooling(F1)

[0065] P2=SpatialPyramidPooling(F2)

[0066] Among them, SpatialPyramidPooling represents the spatial pyramid pooling block processing; the multi-scale feature representation is fused to obtain the fused feature F0:

[0067] F0=Concatenate(P1,P2)

[0068] Among them, Concatenate represents a fusion operation; the abnormal sound wave feature includes the fusion feature and the single type of abnormal sound wave feature.

[0069] In this embodiment, considering that there may be more than one type of fault when a fault occurs, a large number of mixed features are constructed after the sound features of a single fault are extracted. That is, the pyramid pooling method can be used in the feature extraction model to pool the features, and then different types of features are fused. When comparing the features, if the abnormal features of the sample correspond to the fused features, it can be known that the device has multiple faults.

[0070] In a preferred embodiment, assuming that the sampling rate of the audio signal is 16kHz, we set the frame length to 25 milliseconds (i.e., 400 sampling points) and the frame shift to 10 milliseconds (i.e., 160 sampling points). Select to use 40 filters (Mel filter banks) and retain 13 MFCC coefficients. First, the audio signal can be divided into overlapping frames, each frame length is 400 sampling points, and the frame shift is 160 sampling points. Each frame is windowed using a Hamming window. Then a fast Fourier transform is applied to each frame to obtain the spectrum of each frame. 40 Mel filters are applied to the spectrum of each frame to calculate the energy of each filter channel. The logarithm of the energy value of each Mel filter bank is taken to obtain the logarithmic energy of each frame. Finally, discrete cosine transform is applied to the logarithmic energy vector of each frame to obtain the cepstral coefficients. For example, the 13 cepstral coefficients obtained are [5.2619, 3.0218, 1.4183, 0.6929, 0.3918, 0.2445, 0.1633, 0.1137, 0.0818, 0.0601, 0.0445, 0.0329, 0.0243], which can be used as sound wave features for comparison and analysis retrieval in subsequent steps.

[0071] In a preferred embodiment, the method for constructing the abnormal feature library is: obtaining a mechanical fault sound database and an electromagnetic fault sound database of the transformer, the mechanical fault sound database containing a number of typical mechanical fault audio data, and the electromagnetic fault sound database containing a number of typical electromagnetic fault audio data; performing feature extraction on the typical mechanical fault audio data and the typical electromagnetic fault audio data, respectively, to obtain the mechanical abnormality feature and the electromagnetic abnormality feature.

[0072] In the embodiment of the present application, a large amount of various types of transformer fault information can be collected. It should be noted that different severity levels of the same type of fault can be counted as different fault information. For example, the fault characteristics corresponding to a slight three-phase unbalanced fault and a severe three-phase unbalanced fault are different, and they can be set as a variety of different fault characteristics. Through this solution, the comparison accuracy is further improved.

[0073] In a preferred embodiment, the method for performing similarity comparison between the impurity acoustic wave feature and the abnormal acoustic wave feature in the abnormal feature library is:

[0074] The two acoustic wave features to be compared are regarded as vectors A and B in space respectively;

[0075] The similarity coefficient Similarity(A,B) of the two features to be compared is obtained based on the following formula:

[0076]

[0077] Among them, A and B represent any two vectors whose similarities are to be compared.

[0078] In the embodiment of the present application, since the features are multidimensional, the cosine space similarity comparison method is adopted. If the threshold is low, the result acquisition time after the abnormal fault occurs is short, but it may cause false alarms. Therefore, the threshold can be designed based on the actual situation.

[0079] like Figure 2 As shown, in one embodiment, an intelligent transformer control device is provided, which can be integrated into an intelligent transformer, and specifically may include:

[0080] The sample signal acquisition module 510 is used to pre-process the multi-channel sound wave signals to obtain a number of sample signals, wherein the multi-channel sound wave signals carry the vibration waveform information of the transformer under the working state.

[0081] The sample signal separation module 520 is used to separate the sample signal based on the learning separation model to obtain normal sound waves and impurity sound waves. The normal sound wave is the sound wave waveform of the transformer under normal working conditions, and the impurity sound wave is the sound wave waveform generated by abnormal operation of the transformer.

[0082] The impurity sound wave feature acquisition module 530 is used to extract features of the impurity sound wave to obtain impurity sound wave features.

[0083] The similarity comparison module 540 is used to perform a similarity comparison between the impurity acoustic wave features and the abnormal acoustic wave features in the abnormal feature library. The abnormal acoustic wave features include electromagnetic abnormal features and mechanical abnormal features. The electromagnetic abnormal features are used to characterize the acoustic wave features of electromagnetic abnormalities caused by electromagnetic abnormal phenomena of the transformer, and the mechanical abnormal features are used to characterize the acoustic wave features of mechanical abnormalities caused by mechanical failures of the transformer.

[0084] The abnormality level acquisition module 550 is used to compare the comparison result of the similarity comparison with the fault possibility threshold to obtain different abnormality levels of the transformer.

[0085] The working state adjustment module 560 is used to obtain a preset control scheme and execute a corresponding transformer working state adjustment scheme based on different abnormality levels.

[0086] In the embodiment of the present application, the explanation and description of the above-mentioned intelligent transformer control device can refer to the explanation and description of the above-mentioned corresponding method. For the description of the above-mentioned intelligent transformer control method, please refer to the above, which will not be repeated here.

[0087] In the embodiment of the present application, the device helps to improve the resolution accuracy of impurity sound waves and normal working sound waves by learning the separation model, which is convenient for feature comparison in subsequent steps; by comparing the similarity of the impurity sound wave features with the abnormal sound wave features in the abnormal feature library, the separation and recognition of different types of faults can be improved, and it is beneficial to expand the feature sound wave library that characterizes faults, thereby improving the accuracy of judging different types of faults.

[0088] Figure 3 The internal structure diagram of a computer device in one embodiment is shown. The device may be a device for running the above method in the embodiment of the present application. As shown in 3, the computer device includes a processor, a memory, a network interface, an input device and a display screen connected via a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program, which, when executed by the processor, enables the processor to implement the intelligent transformer control method. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to execute the intelligent transformer control method. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0089] Those skilled in the art will understand that Figure 3 The 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.

[0090] In one embodiment, the intelligent transformer control device provided by the present application can be implemented in the form of a computer program. The computer program can be Figure 3 The computer device shown in the figure is run on the computer device. The memory of the computer device can store various program modules constituting the intelligent transformer control device, such as: Figure 2 The modules 510, 520 and 530 shown in the figure, etc. The computer program composed of various program modules enables the processor to execute the steps of the intelligent transformer control method of various embodiments of the present application described in this specification.

[0091] For example, Figure 3 The computer device shown can be Figure 2The module 510 in the intelligent transformer control device shown executes step S10. The computer device can execute step S20 through the module 520. And so on.

[0092] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the processor executes the steps of the intelligent transformer control method as described above.

[0093] In the embodiment of the present application, please refer to the above for the description of the above-mentioned intelligent transformer control method, which will not be repeated here.

[0094] In the embodiment of the present application, the program run by the method stored in the storage medium of the embodiment of the present application, by learning the separation model, is conducive to improving the resolution accuracy of impurity sound waves and normal working sound waves, and is convenient for feature comparison in subsequent steps; by comparing the similarity of the impurity sound wave features with the abnormal sound wave features in the abnormal feature library, the separation and recognition of different types of faults can be improved, and it is conducive to expanding the feature sound wave library that characterizes the faults, thereby improving the accuracy of judging different fault types.

[0095] In one embodiment, an intelligent transformer control system is provided. When the system is running, the system executes the steps of the intelligent transformer control method as described above.

[0096] In the embodiment of the present application, the system may be a computer software or program system, which runs in hardware containing a processing device, and when the system is running, the corresponding method is executed. For the description of the intelligent transformer control method, please refer to the above, which will not be repeated here.

[0097] In the embodiment of the present application, by learning the separation model, it is helpful to improve the resolution accuracy of impurity sound waves and normal working sound waves, which is convenient for feature comparison in subsequent steps; by comparing the similarity of the impurity sound wave features with the abnormal sound wave features in the abnormal feature library, the separation and recognition of different types of faults can be improved, and it is helpful to expand the feature sound wave library that characterizes faults, thereby improving the accuracy of judging different fault types.

[0098] In one embodiment, a smart transformer is provided. When the transformer is running, the steps of the smart transformer control method described above are executed to monitor its operating state. In the embodiment of the present application, the smart transformer is beneficial to improve the resolution accuracy of impurity sound waves and normal working sound waves by learning the separation model, which is convenient for feature comparison in subsequent steps; by comparing the impurity sound wave features with the abnormal sound wave features in the abnormal feature library for similarity, the separation recognition of different types of faults can be improved, and it is beneficial to expand the feature sound wave library that characterizes faults, thereby improving the accuracy of judging different types of faults.

[0099] It should be understood that, although each step in the flow chart of each embodiment of the present application is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0100] Those skilled 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 program can be stored in a non-volatile computer-readable storage medium. When the 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 non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0101] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described 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.

[0102] 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 present application. 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 present application shall be subject to the attached claims.

Claims

1. An intelligent transformer control method, characterized in that: The method comprises: Preprocessing the multi-channel sound wave signals to obtain a plurality of sample signals, wherein the multi-channel sound wave signals carry vibration waveform information of the transformer under working conditions; Based on the learning separation model, the sample signal is separated and processed to obtain a normal sound wave and an impurity sound wave, wherein the normal sound wave is the sound wave waveform of the transformer under normal working condition, and the impurity sound wave is the sound wave waveform generated due to abnormal working of the transformer; Extracting features of the impurity sound wave to obtain impurity sound wave features; Compare the impurity acoustic wave features with the abnormal acoustic wave features in the abnormal feature library for similarity, wherein the abnormal acoustic wave features include electromagnetic abnormal features and mechanical abnormal features, wherein the electromagnetic abnormal features are used to characterize the acoustic wave features of electromagnetic abnormalities caused by electromagnetic abnormal phenomena of the transformer, and the mechanical abnormal features are used to characterize the acoustic wave features of mechanical abnormalities caused by mechanical failures of the transformer; Comparing the comparison result of the similarity comparison with the fault possibility threshold to obtain different abnormality levels of the transformer; Obtain the preset control plan and execute the corresponding transformer working state adjustment plan based on different abnormal levels; Based on the learning separation model, the method for separating and processing the sample signal is as follows: Obtain the historical data of the normal spectrum of the transformer to be analyzed during the historical normal working period. Based on the normal spectrum historical data, the normal historical data matrix of the transformer is constructed: Where: in, Represents a period of time. Indicates dividing 24 hours into N equal-length time periods. It represents the matrix composed of different audio data corresponding to different loads of the transformer in a period of time. Indicates different load levels, The timing representation of the normal acoustic waveform of the transformer under different loads. A timing representation of a normal sound wave shape. Represents the normal historical data matrix of the transformer; Acquire some abnormal spectrum data from the abnormal sound wave library, and compare the abnormal spectrum data with the normal historical data function. Perform mixing respectively to obtain several mixed spectrum data matrices; Build a learning separation model based on the Deep Attractor Network model; Training the learning separation model network based on the training samples consisting of the normal historical data matrix, the mixed spectrum data matrix and the abnormal spectrum data to obtain the learning separation model; Acquire the acquisition time of the sample signal to be processed and the load level of the transformer during the acquisition time; The sample signal to be processed, its acquisition time and the corresponding load level are input into the learning separation model as input quantities, and the normal sound wave and the impurity sound wave in the sample signal are output.

2. The intelligent transformer control method according to claim 1, characterized in that: The method of preprocessing multi-channel sound wave signals to obtain several sample signals is as follows: Based on a plurality of vibration sensors, a multi-channel acoustic wave signal is obtained, wherein the vibration sensors are used to obtain time domain waveform data of vibration conditions at different positions of the transformer; The multiple sound wave signals are sampled, expanded and aligned, and each sound wave signal is set as a sample signal to obtain a plurality of sample signals.

3. The intelligent transformer control method according to claim 1, characterized in that: The method for extracting the features of the impurity sound wave to obtain the features of the impurity sound wave is: The acoustic wave signal is divided into a number of overlapping frames, and each frame is windowed using a Hamming window; Apply fast Fourier transform to each frame to obtain the spectrum of each frame; Apply several Mel filters to the spectrum of each frame and get the energy of each filter channel; Take the logarithm of the energy value of each Mel filter bank to obtain the logarithmic energy of each frame; Apply discrete cosine transform to the logarithmic energy vector of each frame to obtain several cepstral coefficients; The plurality of cepstrum coefficients are set as impurity sound wave characteristics corresponding to the impurity sound wave; The method for obtaining abnormal sound wave features in the abnormal feature library is: Acquire a number of abnormal spectrum data from an abnormal sound wave library, perform feature extraction on the abnormal spectrum data, and obtain a single type of abnormal sound wave feature; Select any two of the single-type abnormal sound wave features and set them as features and Features , the features and Features After being processed by the spatial pyramid pooling block, their respective multi-scale feature representations are obtained and : in, Represents spatial pyramid pooling block processing; The multi-scale feature representation is fused to obtain the fused feature : in, represents a fusion operation; The abnormal sound wave features include the fusion features and the single-type abnormal sound wave features.

4. The intelligent transformer control method according to claim 1, characterized in that: The method for constructing the abnormal feature library is as follows: Acquire a mechanical fault sound database and an electromagnetic fault sound database of the transformer, wherein the mechanical fault sound database contains a number of typical mechanical fault audio data, and the electromagnetic fault sound database contains a number of typical electromagnetic fault audio data; Feature extraction is performed on the typical mechanical failure audio data and the typical electromagnetic failure audio data respectively to obtain the mechanical abnormality feature and the electromagnetic abnormality feature.

5. The intelligent transformer control method according to claim 1, characterized in that: The method for performing similarity comparison between the impurity acoustic wave feature and the abnormal acoustic wave feature in the abnormal feature library is: The two acoustic wave features to be compared are regarded as vectors A and B in space respectively; The similarity coefficient of the two features to be compared is obtained based on the following formula : Among them, A and B represent any two vectors whose similarities are to be compared.

6. An intelligent transformer control device, characterized in that: The intelligent transformer control device comprises: A sample signal acquisition module is used to pre-process the multi-channel sound wave signals to obtain a number of sample signals, wherein the multi-channel sound wave signals carry the vibration waveform information of the transformer under the working state; A sample signal separation module is used to separate and process the sample signal based on a learning separation model to obtain a normal sound wave and an impurity sound wave, wherein the normal sound wave is a sound wave waveform of the transformer under normal working conditions, and the impurity sound wave is a sound wave waveform generated due to abnormal operation of the transformer; An impurity sound wave feature acquisition module, used to extract features of the impurity sound wave to obtain impurity sound wave features; A similarity comparison module is used to compare the impurity acoustic wave features with the abnormal acoustic wave features in the abnormal feature library for similarity, wherein the abnormal acoustic wave features include electromagnetic abnormal features and mechanical abnormal features, wherein the electromagnetic abnormal features are used to characterize the acoustic wave features of electromagnetic abnormalities caused by electromagnetic abnormal phenomena of the transformer, and the mechanical abnormal features are used to characterize the acoustic wave features of mechanical abnormalities caused by mechanical failures of the transformer; An abnormality level acquisition module, used to compare the comparison result of the similarity comparison with the fault possibility threshold to obtain different abnormality levels of the transformer; A working state adjustment module is used to obtain a preset control scheme and execute the corresponding transformer working state adjustment scheme based on different abnormal levels; Based on the learning separation model, the method for separating and processing the sample signal is: Obtain the historical data of the normal spectrum of the transformer to be analyzed during the historical normal working period. Based on the normal spectrum historical data, the normal historical data matrix of the transformer is constructed: Where: in, Represents a period of time. Indicates dividing 24 hours into N equal-length time periods. It represents the matrix composed of different audio data corresponding to different loads of the transformer in a period of time. Indicates different load levels, The timing representation of the normal acoustic waveform of the transformer under different loads. A timing representation of a normal sound wave shape. Represents the normal historical data matrix of the transformer; Acquire some abnormal spectrum data from the abnormal sound wave library, and compare the abnormal spectrum data with the normal historical data function. Perform mixing respectively to obtain several mixed spectrum data matrices; Build a learning separation model based on the Deep Attractor Network model; Training the learning separation model network based on the training samples consisting of the normal historical data matrix, the mixed spectrum data matrix and the abnormal spectrum data to obtain the learning separation model; Acquire the acquisition time of the sample signal to be processed and the load level of the transformer during the acquisition time; The sample signal to be processed, its acquisition time and the corresponding load level are input into the learning separation model as input quantities, and the normal sound wave and the impurity sound wave in the sample signal are output.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the steps of the intelligent transformer control method according to any one of claims 1 to 5.

8. An intelligent transformer control system, characterized in that: When the system is running, the steps of the intelligent transformer control method according to any one of claims 1 to 5 are executed.

9. An intelligent transformer, characterized in that: When the smart transformer is running, the steps of the smart transformer control method according to any one of claims 1 to 5 are executed to monitor its running state.

Images