Transformer state evaluation method and system based on acousto-optic heterogeneous data, and medium
By collecting and processing the acousto-optical heterogeneous data of the transformer, and using triangular filter groups and convolutional neural networks for state evaluation, the problem that the identification algorithm in transformer state detection in the prior art is difficult to deal with different state connections, achieving more efficient and accurate status assessment.
Patent Information
- Application Number
- CN202411611419.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2025-02-18
AI Technical Summary
In the state detection of transformer, the acousto-optical heterogeneous data recognition algorithm is difficult to effectively process the relationship between different states, and lacks a method to process the imbalance of the number and difficulty of audio and image.
The transformer state evaluation method based on acousto-optical heterogeneous data is adopted. By collecting the audio data and image data of the transformer in different state dimensions, converting it into a time spectrum diagram, and using a triangular filter group to compress the frequency dimension, and then compress it after compression. The feature input is preset to the convolutional neural network for state evaluation.
It improves the accuracy and efficiency of transformer status evaluation, adapts to different models and specifications of transformers, can effectively process acousto-optical heterogeneous data characteristics, and provides more accurate status evaluation results.
Smart Images
Figure CN119475036A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power transformer state detection, and in particular to a transformer state evaluation method, system and medium based on acoustic and optical heterogeneous data. Background Art
[0002] Power transformers play an important role in voltage conversion and power transmission. The transformers used in power systems are large in number, with a variety of capacity levels and specifications, and they run for a long time, which leads to a corresponding increase in their accident rate. Once a transformer fails, it may cause huge economic losses to the power grid and endanger the personal safety of operation and maintenance personnel. The vibration of the transformer in operation mainly includes winding vibration, core vibration and cooling system vibration. The mechanical waves generated by the vibration radiate outward through the solid structural parts, insulating oil and air of the transformer to form sound wave signals. The sound wave signal contains a lot of information about the working status of the transformer. 20Hz to 20kHz is the sound frequency range that can be heard by the human ear. Experienced substation staff can directly judge whether the status of the running transformer is normal by listening to the sound of the running transformer. For the running transformer, the sound signal contains rich equipment status information. When the transformer is in abnormal working conditions such as overload, loose core, DC bias magnetization, ferromagnetic resonance, etc., the characteristics of the sound signal emitted by the transformer will also change accordingly.
[0003] At present, in the related technology, there are acoustic and optical heterogeneous data recognition algorithms. For example, in the article on the acoustic and optical heterogeneous data recognition model for transformer core loosening fault based on improved MFCC (Mel Frequency Cepstral Coefficients) and 3D-CNN (3D Convolutional Neural Networks), the recognition ability of the neural network algorithm is used to improve the accuracy of acoustic and optical heterogeneous data recognition. However, the different states of the transformer are simply divided into different categories without considering the connection between different states. There is also a lack of processing methods for the imbalance in the number and difficulty of audio and image. Summary of the invention
[0004] Based on this, it is necessary to propose a transformer condition assessment method, system and medium based on acoustic and optical heterogeneous data to address the above problems.
[0005] A transformer condition assessment method based on acoustic and optical heterogeneous data, the method comprising:
[0006] Collect audio and image data of the transformer in different state dimensions.
[0007] A time-frequency spectrum diagram is determined according to the audio data and the image data, wherein the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features of the audio data and the image data.
[0008] The time-frequency spectrum is compressed in the frequency dimension by using a triangular filter bank, so as to complete the compression of the acoustic-optical heterogeneous data features.
[0009] The compressed acoustic and optical heterogeneous data features are input into a preset convolutional neural network to obtain a transformer state assessment result.
[0010] The collecting of audio data and image data of the transformer in different state dimensions specifically includes:
[0011] The audio data and image data of the transformer are collected in two dimensions: current and voltage state and fault state. The current and voltage state dimension includes normal test, open circuit excitation test and short circuit current test; the fault state dimension includes normal test, core loosening test and coil loosening test.
[0012] The time-frequency spectrum diagram is determined according to the audio data and the image data, and the time-frequency spectrum diagram includes the acoustic-optical heterogeneous data features of the audio data and the image data, specifically including:
[0013] The audio data and the image data are subjected to pre-emphasis, frame division and windowing processing.
[0014] The audio data and image data of each frame are respectively subjected to fast Fourier transform and the power spectrum density is calculated.
[0015] The power spectrum density of the audio data and the image data of each frame is spliced to obtain a time-frequency spectrum diagram, wherein the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features.
[0016] The method of compressing the time-frequency spectrum in the frequency dimension by using a triangular filter bank to compress the features of the acoustic-optical heterogeneous data specifically includes:
[0017] The center frequency of the filter is determined according to the value of each filter in the triangular filter bank at different frequency components.
[0018] The time-frequency spectrum is compressed in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features.
[0019] The method of determining the center frequency of the filter according to the value of each filter in the triangular filter bank at different frequency components specifically includes:
[0020] according to Determine the center frequency of the filter;
[0021] Among them, H m(k) is the value of the m-th filter on the k-th frequency component, f(m) is the center frequency of the m-th filter, k is the number of frequency components, and m is the number of filters.
[0022] The step of compressing the time-frequency spectrum in the frequency dimension by using the center frequency to complete the compression of the acoustic-optical heterogeneous data features specifically includes:
[0023] By S' = g c (f,S) compresses the time-frequency spectrum in the frequency dimension to complete the compression of the acoustic-optical heterogeneous data features, wherein S is the time-frequency spectrum, f is the center frequency, and S' is the compressed time-frequency spectrum.
[0024] Wherein, after compressing the time-frequency spectrum in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features, the method further specifically includes:
[0025] The error loss function is determined based on the error between the predicted state classification and the actual state classification.
[0026] The center frequency is updated according to the error loss function.
[0027] Wherein, updating the center frequency according to the error loss function specifically includes:
[0028] according to Update the center frequency, where f is the center frequency, f′ is the updated center frequency, η is the learning rate of gradient descent, and g n is a neural network, l is the error loss function, g c is the compression function of the center frequency.
[0029] The compressed acoustic and optical heterogeneous data features are input into a preset convolutional neural network to obtain the transformer state assessment result, which specifically includes:
[0030] Extract the compressed acoustic-optical heterogeneous data features.
[0031] The compressed acoustic-optical heterogeneous data features are sequentially input into the convolution layer, batch normalization layer, ReLU function, and maximum pooling layer in the preset convolutional neural network for preliminary feature extraction to obtain primary features.
[0032] The primary features are input into the cascade network of the residual blocks in the preset convolutional neural network to extract deep features and obtain deep features.
[0033] The deep features are input into the global average pooling layer in the preset convolutional neural network, and the deep features are classified from two dimensions of the voltage and current state and the fault state of the transformer to obtain a transformer state assessment result.
[0034] A transformer condition assessment system based on acoustic and optical heterogeneous data, the system comprising:
[0035] The acquisition module is used to collect audio data and image data of the transformer in different state dimensions.
[0036] The time-spectrogram determination module is used to determine the time-spectrogram according to the audio data and the image data, wherein the time-spectrogram includes the acoustic-optical heterogeneous data features of the audio data and the image data.
[0037] The time-spectrogram compression module is used to compress the time-spectrogram in the frequency dimension by using a triangular filter bank to complete the compression of the acoustic-optical heterogeneous data features.
[0038] The transformer state assessment module is used to input the compressed acoustic and optical heterogeneous data features into a preset convolutional neural network to obtain a transformer state assessment result.
[0039] A 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 method described above.
[0040] The embodiments of the present invention have the following beneficial effects:
[0041] The present invention first collects audio data and image data of the transformer in different state dimensions. By converting the audio data and image data into a time-frequency spectrum, the frequency and intensity characteristics at different time points, namely, the acoustic-optical heterogeneous data characteristics, can be extracted. Furthermore, by compressing the time-frequency spectrum in the frequency dimension, the redundancy of the data can be reduced and the processing efficiency can be improved. Among them, the triangular filter group can effectively remove high-frequency noise and redundant information while maintaining the main characteristics of the signal. Finally, by inputting the compressed acoustic-optical heterogeneous data characteristics into a preset convolutional neural network, the transformer state assessment result is obtained, which improves the accuracy and efficiency of the assessment, and can also adapt to transformers of different models and specifications, and has a wide range of applicability. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0043] in:
[0044] Figure 1 It is a flow chart of an embodiment of a transformer state assessment method based on acoustic and optical heterogeneous data;
[0045] Figure 2 A schematic flow chart of another embodiment of a transformer condition assessment method based on acoustic and optical heterogeneous data;
[0046] Figure 3 It is a structural schematic diagram of an embodiment of a transformer condition assessment system based on acoustic and optical heterogeneous data;
[0047] Figure 4 A schematic structural diagram of an embodiment of the medium provided by the present invention. DETAILED DESCRIPTION
[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0049] like Figure 1 As shown, Figure 1 The figure is a flow chart of an embodiment of a transformer condition assessment method based on acoustic and optical heterogeneous data. A transformer condition assessment method based on acoustic and optical heterogeneous data, the method comprising:
[0050] S101: Collect audio data and image data of the transformer in different state dimensions.
[0051] Exemplarily, audio data and image data of the transformer are collected in two dimensions: current and voltage state and fault state. The current and voltage state dimension includes normal state, open circuit excitation state and short circuit current state; the fault state dimension includes normal state, loose core and loose coil.
[0052] Specifically, the state classification under the two dimensions of current and voltage state and fault state is shown in Table 1:
[0053] Voltage and current\fault No trouble Loose core Loose coil No abnormality Status 1 Status 2 State three Open circuit excitation State 4 Status Five Status 6 Short circuit current Status seven Status eight Status Nine
[0054] Table 1: Status classification table under two dimensions: current and voltage status and fault status
[0055] The first row in Table 1 is the simulation experiments of three transformer fault states, including: no-fault experiment, core loosening experiment and coil loosening experiment; the first column is the simulation experiments of three transformer voltage and current states, including no-abnormal experiment, open-circuit excitation experiment and short-circuit current experiment.
[0056] Combining the simulation experiments of voltage and current states with the simulation experiments of fault states, 9 types of transformer states can be obtained, as shown in Table 1. The specific contents of the simulation experiments are as follows:
[0057] Fault-free experiment: There are no faults in the transformer core, coil and other structures.
[0058] Core loosening experiment: Loosen the horizontal fastening screws of the transformer core, and use the number of turns required to completely loosen the screws as the benchmark to simulate a looseness of 50%.
[0059] Coil loosening experiment: Loosen the fastening screws in the vertical direction of the transformer coil winding, and use the number of turns required to completely loosen the screws as the benchmark to simulate looseness levels of 50% and 100%.
[0060] No abnormal experiment: Transformer voltage and current are rated values.
[0061] Open circuit excitation experiment: Open the high voltage side of the transformer and apply voltage to the low voltage side at a gradient of 40%, 70%, 90%, 100%, 105%, 110%, and 115% based on the rated voltage.
[0062] Short-circuit current experiment: The low-voltage side of the transformer is short-circuited, and current is applied on the high-voltage side at a gradient of 40%, 70%, 90%, 100%, and 110% based on the rated current.
[0063] Due to the differences between the transformer on-site environment and the simulation experiment environment, in order to make the audio and images obtained from the simulation experiment closer to the on-site audio and images, according to the characteristics of the on-site audio and images and the characteristics of the audio and images of the simulation experiment, the audio and images obtained from the simulation experiment were subjected to sample expansion processing as shown in Table 2. These sample expansion processing, on the one hand, make the audio and images of the simulation experiment as close as possible to the on-site audio and images, and on the other hand, increase the number of samples, which can improve the generalization effect of the neural network.
[0064]
[0065] Table 2: Comparison of audio and image characteristics and post-processing methods between on-site and simulated experiments
[0066] After post-processing, the audio and images of the simulation experiment together with the on-site audio and images constitute the audio data and image data. There are 9,000 audio and image data in total, 1,000 of which are from the on-site audio and images, and 8,000 are from the simulation experiment. Each audio and image is 5 seconds long and is divided into training set, validation set and test set in a ratio of 8:1:1. The training set is used for the subsequent training of the preset convolutional neural network model, the validation set is used to verify the model structure and hyperparameter effects, and the test set is used to test and evaluate the results.
[0067] S102: Determine a time-frequency spectrum diagram according to the audio data and the image data, where the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features of the audio data and the image data.
[0068] Exemplarily, the audio data and image data are pre-emphasized, framed and windowed; each frame of the audio data and image data is subjected to fast Fourier transform and the power spectral density is calculated; the power spectral density of each frame of the audio data and image data is spliced to obtain a time-frequency spectrum diagram, which includes acoustic and optical heterogeneous data features.
[0069] S103: compressing the time-frequency spectrum in the frequency dimension using a triangular filter bank to complete the compression of the acoustic-optical heterogeneous data features.
[0070] Exemplarily, the center frequency of the filter is determined according to the value of each filter in the triangular filter bank at different frequency components. The center frequency of the filter is shown in the following formula:
[0071]
[0072] Among them, H m (k) is the value of the m-th filter on the k-th frequency component, f(m) is the center frequency of the m-th filter, k is the number of frequency components, and m is the number of filters.
[0073] Furthermore, the time-spectrum graph is compressed in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features. The time-spectrum graph is compressed in the frequency dimension according to the following formula:
[0074] S'=g c (f,S);
[0075] Where S is the time-frequency spectrum, f is the center frequency, and S' is the compressed time-frequency spectrum. If the center frequency f is regarded as a variable, the process of compressing the time-frequency spectrum S to obtain S' can be regarded as a function of compressing the center frequency, that is, g c .
[0076] S104: Input the compressed acoustic and optical heterogeneous data features into a preset convolutional neural network to obtain transformer state assessment results.
[0077] Exemplarily, the features of the compressed acoustic and optical heterogeneous data are extracted; the compressed acoustic and optical heterogeneous data features are sequentially input into the convolution layer, batch normalization layer, ReLU function, and maximum pooling layer in the preset convolutional neural network for preliminary feature extraction to obtain primary features; the primary features are input into the cascade network of the residual block in the preset convolutional neural network for deep feature extraction to obtain deep features; the deep features are input into the global average pooling layer in the preset convolutional neural network, and the deep features are classified from two dimensions: the voltage and current state and the fault state of the transformer, to obtain the transformer state assessment result.
[0078] It can be seen from the above description that the present invention first collects audio data and image data of the transformer in different state dimensions, and by converting the audio data and image data into a time-frequency spectrum, the frequency and intensity characteristics at different time points, that is, the acoustic-optical heterogeneous data characteristics, can be extracted. Furthermore, by compressing the time-frequency spectrum in the frequency dimension, the redundancy of the data can be reduced and the processing efficiency can be improved. Among them, the triangular filter group can effectively remove high-frequency noise and redundant information while maintaining the main characteristics of the signal. Finally, by inputting the compressed acoustic-optical heterogeneous data characteristics into a preset convolutional neural network, the transformer state assessment result is obtained, which improves the accuracy and efficiency of the assessment, and can also adapt to transformers of different models and specifications, and has a wide range of applicability.
[0079] like Figure 2 As shown, Figure 2 The figure is a flow chart of another embodiment of a transformer condition assessment method based on acoustic and optical heterogeneous data. A transformer condition assessment method based on acoustic and optical heterogeneous data, the method comprising:
[0080] S201: Collect audio data and image data of the transformer in two dimensions: current and voltage state and fault state. The current and voltage state dimension includes normal test, open circuit excitation test and short circuit current test; the fault state dimension includes normal test, core loosening test and coil loosening test.
[0081] S201: Perform pre-emphasis, frame division and windowing processing on audio data and image data.
[0082] S202: Perform fast Fourier transform on the audio data and image data of each frame respectively and calculate the power spectrum density.
[0083] S203: splicing the power spectrum density of the audio data and the image data of each frame to obtain a time-frequency spectrum diagram, which includes acoustic and optical heterogeneous data features.
[0084] It should be noted that steps S201-S203 are Figure 1 This has been discussed in detail in the implementation scenario shown and will not be repeated here.
[0085] S204: Determine the center frequency of the filter according to the value of each filter in the triangular filter bank at different frequency components.
[0086] Exemplarily, the center frequency of the filter is determined according to the value of each filter in the triangular filter bank at different frequency components. The center frequency of the filter is shown in the following formula:
[0087]
[0088] Among them, H m (k) is the value of the m-th filter on the k-th frequency component, f(m) is the center frequency of the m-th filter, k is the number of frequency components, and m is the number of filters.
[0089] S205: compressing the time-frequency spectrum in the frequency dimension through the center frequency to complete the compression of the acoustic-optical heterogeneous data features.
[0090] Exemplarily, the time-frequency spectrum is compressed in the frequency dimension by the following formula to complete the compression of the acoustic-optical heterogeneous data features:
[0091] S'=g c (f,S);
[0092] Among them, S is the time-frequency spectrum, f is the center frequency, and S' is the compressed time-frequency spectrum.
[0093] Furthermore, the error loss function is determined according to the error between the predicted state classification and the actual state classification. The error loss function is shown as follows:
[0094] y p =g n (S') = g n (g c (f,S))
[0095] l = g l (y p ,y);
[0096] Among them, g n is a neural network, y p is the predicted state classification, S is the time-frequency spectrum, f is the center frequency, S' is the compressed time-frequency spectrum, l is the error loss function, y is the actual state classification, g 1 is the initial compression function of the center frequency.
[0097] Furthermore, the center frequency is updated according to the error loss function, and the updated center frequency is shown as follows:
[0098]
[0099] Among them, f is the center frequency, f′ is the updated center frequency, η is the learning rate of gradient descent, and g n is a neural network, l is the error loss function, g c is the compression function of the center frequency.
[0100] S206: Extracting features of the compressed acoustic-optical heterogeneous data.
[0101] S207: Input the compressed acoustic and optical heterogeneous data features into the convolution layer, batch normalization layer, ReLU function, and maximum pooling layer in the preset convolutional neural network in sequence for preliminary feature extraction to obtain primary features.
[0102] Exemplarily, the convolutional neural network includes a convolutional layer, a batch normalization layer, a ReLU function, a maximum pooling layer, a residual block, and a global average pooling layer. The compressed acoustic and optical heterogeneous data features are sequentially input into the convolutional layer, the batch normalization layer, the ReLU function, and the maximum pooling layer for preliminary feature extraction to obtain primary features.
[0103] S208: Input the primary features into the cascade network of the residual block in the preset convolutional neural network to extract deep features and obtain deep features.
[0104] Exemplarily, the primary features are input into a cascade network of six residual blocks to perform deep feature extraction to obtain deep features.
[0105] S209: Input the deep features into the global average pooling layer in the preset convolutional neural network, classify the deep features from two dimensions: the voltage and current state and the fault state of the transformer, and obtain the transformer state assessment result.
[0106] Exemplarily, the deep features are input into the global average pooling layer, and the global average pooling layer classifies the deep features from two dimensions: voltage and current state and fault state, to obtain the transformer state assessment result. Considering the connection between the voltage and current state and the fault state on the transformer state, the sound of the transformer in different states is collected and the acoustic and optical heterogeneous data features are extracted, which is conducive to comparing the acoustic and optical heterogeneous data features in different states of the transformer and obtaining the corresponding transformer state assessment result.
[0107] It can be seen from the above description that the present invention obtains the corresponding transformer state assessment result by collecting and classifying the acoustic and optical heterogeneous data features of audio data and image data from different state dimensions.
[0108] like Figure 3 As shown, Figure 3 The structure diagram of an embodiment of a transformer condition assessment system based on acoustic and optical heterogeneous data is shown in FIG. A transformer condition assessment system 10 based on acoustic and optical heterogeneous data includes:
[0109] The acquisition module 11 is used to acquire audio data and image data of the transformer in different state dimensions.
[0110] The time-spectrogram determining module 12 is used to determine a time-spectrogram according to the audio data and the image data, wherein the time-spectrogram includes the acoustic-optical heterogeneous data features of the audio data and the image data.
[0111] The time-spectrogram compression module 13 is used to compress the time-spectrogram in the frequency dimension by using a triangular filter bank, so as to complete the compression of the acoustic-optical heterogeneous data features.
[0112] The transformer state assessment module 14 is used to input the compressed acoustic and optical heterogeneous data features into a preset convolutional neural network to obtain a transformer state assessment result.
[0113] Exemplarily, in the acquisition module 11, audio data and image data of the transformer in two dimensions, current and voltage state and fault state, are collected respectively, wherein the current and voltage state dimension includes normal test, open circuit excitation test and short circuit current test; the fault state dimension includes normal test, core loosening test and coil loosening test. In the time-frequency spectrum determination module 12, the audio data and image data are pre-emphasized, framed and windowed; the audio data and image data of each frame are fast Fourier transformed and the power spectrum density is calculated; the power spectrum density of the audio data and image data of each frame is spliced to obtain a time-frequency spectrum, which includes acoustic and optical heterogeneous data features. In the time-frequency spectrum compression module 13, the center frequency of the filter is determined according to the value of each filter in the triangular filter group at different frequency components; the time-frequency spectrum is compressed in the frequency dimension by the center frequency to complete the compression of the acoustic and optical heterogeneous data features. In the transformer state assessment module 14, the compressed acoustic and optical heterogeneous data features are extracted; the compressed acoustic and optical heterogeneous data features are sequentially input into the convolution layer, batch normalization layer, ReLU function, and maximum pooling layer in the preset convolutional neural network for preliminary feature extraction to obtain primary features; the primary features are input into the cascade network of the residual block in the preset convolutional neural network for deep feature extraction to obtain deep features; the deep features are input into the global average pooling layer in the preset convolutional neural network, and the deep features are classified from the two dimensions of the voltage and current state and the fault state of the transformer to obtain the transformer state assessment result.
[0114] like Figure 4 As shown, Figure 4The structure diagram of an embodiment of the medium provided by the present invention is shown in FIG. The medium 20 stores at least one computer program 21, which is executed by a processor to implement the following Figure 1 and Figure 2 In one embodiment, the medium 20 may be a storage chip, a hard disk, a mobile hard disk, a USB flash drive, an optical disk, or other readable and writable storage tools, or a server, etc.
[0115] The above describes specific embodiments of the present specification, and other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily have to be performed in the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0116] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer-readable storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.
[0117] The apparatus, device, non-volatile computer-readable storage medium and method provided in the embodiments of this specification correspond to each other, and therefore, the apparatus, device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the corresponding apparatus, device, and non-volatile computer storage medium will not be repeated here.
[0118] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
[0119] For the convenience of description, the above device is described by being divided into various units according to their functions and described separately. Of course, when implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware. It should be understood by those skilled in the art that this specification embodiment can be provided as a method, system, or computer program product. Therefore, this specification embodiment can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification embodiment can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0120] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0121] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0122] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.
Claims
1. A transformer condition assessment method based on acoustic and optical heterogeneous data, characterized in that: The method comprises: Collect audio data and image data of the transformer in different state dimensions; Determine a time-frequency spectrum diagram according to the audio data and the image data, wherein the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features of the audio data and the image data; Using a triangular filter bank to compress the time-frequency spectrum in the frequency dimension, so as to complete the compression of the acoustic-optical heterogeneous data features; The compressed acoustic and optical heterogeneous data features are input into a preset convolutional neural network to obtain a transformer state assessment result.
2. A transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 1, characterized in that: The collecting of audio data and image data of the transformer in different state dimensions specifically includes: The audio data and image data of the transformer are collected in two dimensions: current and voltage state and fault state. The current and voltage state dimension includes normal test, open circuit excitation test and short circuit current test; the fault state dimension includes normal test, core loosening test and coil loosening test.
3. The transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 1 is characterized in that: The determining of a time-frequency spectrum diagram according to the audio data and the image data, wherein the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features of the audio data and the image data, specifically includes: Performing pre-emphasis, framing and windowing processing on the audio data and image data; Performing fast Fourier transform on the audio data and image data of each frame and calculating power spectrum density; The power spectrum density of the audio data and the image data of each frame is spliced to obtain a time-frequency spectrum diagram, wherein the time-frequency spectrum diagram includes acoustic-optical heterogeneous data features.
4. A transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 3, characterized in that: The compressing the time-frequency spectrum diagram in the frequency dimension by using a triangular filter bank to complete the compression of the acoustic-optical heterogeneous data features specifically includes: Determine the center frequency of the filter according to the value of each filter in the triangular filter bank at different frequency components; The time-frequency spectrum is compressed in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features.
5. A transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 4, characterized in that: The method of determining the center frequency of the filter according to the value of each filter in the triangular filter bank at different frequency components specifically includes: according to Determine the center frequency of the filter; Among them, H m (k) is the value of the m-th filter on the k-th frequency component, f(m) is the center frequency of the m-th filter, k is the number of frequency components, and m is the number of filters.
6. A transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 5, characterized in that: The compressing the time-frequency spectrum in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features specifically includes: By S' = g c (f,S) compresses the time-frequency spectrum in the frequency dimension, so as to complete the compression of the acoustic-optical heterogeneous data features, where S is the time-frequency spectrum, f is the center frequency, S′ is the compressed time-frequency spectrum, and g c is the compression function of the center frequency.
7. The transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 4 is characterized in that: After the time-frequency spectrum is compressed in the frequency dimension by the center frequency to complete the compression of the acoustic-optical heterogeneous data features, the method further specifically includes: Determining an error loss function based on the error between the predicted state classification and the actual state classification; The center frequency is updated according to the error loss function.
8. The transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 7 is characterized in that: The updating of the center frequency according to the error loss function specifically includes: according to Update the center frequency, where f is the center frequency, f′ is the updated center frequency, η is the learning rate of gradient descent, and g n is a neural network, l is the error loss function, g c is the compression function of the center frequency.
9. The transformer condition assessment method based on acoustic and optical heterogeneous data according to claim 4 is characterized in that: The compressed acoustic and optical heterogeneous data features are input into a preset convolutional neural network to obtain the transformer state assessment result, specifically including: Extracting features of the compressed acoustic-optical heterogeneous data; The compressed acoustic-optical heterogeneous data features are sequentially input into the convolution layer, batch normalization layer, ReLU function, and maximum pooling layer in the preset convolutional neural network for preliminary feature extraction to obtain primary features; Inputting the primary features into the cascade network of the residual blocks in the preset convolutional neural network to extract deep features and obtain deep features; The deep features are input into the global average pooling layer in the preset convolutional neural network, and the deep features are classified from two dimensions of the voltage and current state and the fault state of the transformer to obtain a transformer state assessment result.
10. A transformer condition assessment system based on acoustic and optical heterogeneous data, characterized in that: The system comprises: An acquisition module, used to collect audio data and image data of the transformer in different state dimensions; A time-spectrogram determination module, used to determine a time-spectrogram according to the audio data and the image data, wherein the time-spectrogram includes acoustic-optical heterogeneous data features of the audio data and the image data; A time-spectrogram compression module, used to compress the time-spectrogram in the frequency dimension using a triangular filter bank to complete the compression of the acoustic-optical heterogeneous data features; The transformer state assessment module is used to input the compressed acoustic and optical heterogeneous data features into a preset convolutional neural network to obtain a transformer state assessment result.
11. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 9.