Bushing partial discharge diagnosis method and system based on multispectrogram fusion, and storage medium

Abstract

The invention discloses a bushing partial discharge diagnosis method and system based on multispectral fusion, and a storage medium. The method comprises the steps of obtaining partial discharge data of a transformer bushing to be diagnosed; identifying the partial discharge data of the to-be-diagnosed transformer bushing by using a trained partial discharge type identification model to obtain the discharge type of the to-be-diagnosed transformer bushing; based on the partial discharge type of the to-be-diagnosed transformer bushing, selecting a trained corresponding type severity identification model; inputting partial discharge data of a to-be-diagnosed transformer bushing into the trained corresponding type severity identification model, and training to obtain a discharge severity diagnosis result; according to the invention, accurate diagnosis of the partial discharge type and accurate diagnosis of the severity of the corresponding type can be realized, the accuracy of equipment insulation defect diagnosis prediction and operation and maintenance and the working efficiency are improved, and the safety and reliability of bushing operation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer bushing partial discharge diagnosis, and in particular to a bushing partial discharge diagnosis method, system and storage medium based on multi-spectrum fusion. Background Art

[0002] In transformer bushings, the generation of partial discharge is usually related to the following factors: manufacturing defects inside the bushing, mechanical damage during installation, insulation aging during long-term operation, and the impact of the external environment on the insulation material. Partial discharge not only accelerates the degradation of the insulation material, but may also cause the overall reliability of the insulation system to decrease. Therefore, it is crucial to quickly diagnose the type of partial discharge and improve the efficiency of bushing maintenance to ensure the safe and stable operation of the transformer.

[0003] At present, the research and application of bushing partial discharge diagnosis technology mainly focus on improving the detection sensitivity and the accuracy of the recognition algorithm. Since partial discharge signals are often accompanied by various physical phenomena such as sound, light, and electricity, how to effectively extract partial discharge feature information from these complex signals is the focus of research. At the same time, the bushing partial discharge detection system is often faced with interference from a strong electromagnetic environment, which puts higher requirements on the detection equipment. In order to improve the accuracy of partial discharge diagnosis, researchers at home and abroad have proposed a variety of partial discharge diagnosis and recognition technologies, such as wavelet transform, support vector machine, artificial neural network, fuzzy logic algorithm, and K nearest neighbor algorithm. These technologies are committed to extracting different types of partial discharge features from partial discharge signals, thereby improving the accuracy of partial discharge type diagnosis. However, most of the current partial discharge diagnosis methods are aimed at two-dimensional spectra or manually extracted partial discharge features. The former automatically extracts the discharge features contained in the spectra through deep learning algorithms, which is not very targeted and requires a very complex network to learn the deep features of partial discharge; the latter artificially defines the partial discharge type through traditional algorithms such as threshold method and decision tree method, which is not flexible and difficult to adapt to the actual operating environment. Therefore, it is necessary to develop a bushing partial discharge diagnosis method that can efficiently extract the deep characteristics of partial discharge.

[0004] In addition, the existing partial discharge diagnosis methods are based on a single spectrum and can only reflect very limited partial discharge characteristics. In addition, since the training samples of the diagnostic model are often obtained by conducting typical partial discharge model tests in the laboratory, there are certain differences between them and the local discharge signals of the bushing measured on site. As a result, when the local characteristics of the discharge deviate, the accuracy of the diagnosis is low. Summary of the invention

[0005] The purpose of the present invention is to provide a bushing partial discharge diagnosis method, system and storage medium based on multi-spectrum fusion, which can realize accurate diagnosis of partial discharge type and accurate diagnosis of corresponding type severity, improve the accuracy and work efficiency of equipment insulation defect diagnosis prediction, operation and maintenance, and greatly improve the safety and reliability of bushing operation.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for diagnosing partial discharge of a bushing based on multi-spectrum fusion, comprising: Acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; Using the trained partial discharge type recognition model to identify the partial discharge data of the transformer bushing to be diagnosed, the discharge type of the transformer bushing to be diagnosed is obtained; Based on the partial discharge type of the transformer bushing to be diagnosed, a trained corresponding type severity recognition model is selected, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model, and a surface discharge severity recognition model; The partial discharge data of the transformer bushing to be diagnosed is input into the trained corresponding type severity recognition model, and the diagnosis result of the discharge severity is obtained through training.

[0007] Optionally, the step of training the partial discharge type recognition model includes: Get the spectrum sample library; The spectrogram sample library is divided into training set and test set according to the set ratio, and different discharge types are used as labels; Based on the training set, the partial discharge type recognition model is trained to determine the optimal parameters of the partial discharge type recognition model, thereby obtaining a preliminary partial discharge type recognition model; The preliminary partial discharge type recognition model is tested using the test set to obtain the test results; The average diagnostic accuracy of the discharge type is calculated according to the test results, and based on the average diagnostic accuracy of the discharge type, a trained partial discharge type recognition model is obtained.

[0008] Optionally, the step of acquiring the spectrum sample library includes: Acquire typical partial discharge data sets of different pressure levels of transformer bushings; the typical partial discharge data sets include relevant data sets of four types of discharges: spike, suspension, bubble and surface; the relevant data sets include phase data, amplitude data and time data; The data in the typical partial discharge data set are divided into three severity levels: mild, medium and severe according to the degree of pressure, and a set of three-dimensional φ-Δt-Δu spectra of different severity levels of the four discharge types are obtained by drawing. The three-dimensional φ-Δt-Δu spectrogram sets of the four discharge types with different severity levels are normalized to obtain a spectrogram sample library.

[0009] Optionally, the step of obtaining a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels by drawing includes: Based on a typical type of partial discharge data set with different pressure levels on transformer bushings, the phase φ of each discharge signal in the typical type of partial discharge data set, the time difference Δt' between adjacent discharge signals, and the difference Δu' between the absolute values ​​of the amplitudes of adjacent discharge signals are calculated; The time difference Δt' between adjacent discharge signals and the difference Δu' between the absolute amplitudes of adjacent discharge signals are normalized to obtain the time difference information and amplitude difference information of each discharge signal in the spectrum. The normalization formula is expressed as follows:

[0010]

[0011] Among them, Δt i Indicates the time difference information of the i-th discharge signal in the spectrum; Δu i Represents the amplitude difference information of the i-th discharge signal in the spectrum; Δt' i represents the time difference between the i-th discharge signal and the i-1-th discharge signal; Δu' i represents the difference in absolute value of the amplitude between the i-th discharge signal and the i-1-th discharge signal; Based on the phase φ of each discharge signal and the time difference information and amplitude difference information in the spectrum, a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels are drawn.

[0012] Optionally, the training steps of the spike discharge severity recognition model, the suspension discharge severity recognition model, the bubble discharge severity recognition model and the surface discharge severity recognition model include: Based on the four types of discharge, the training set and test set are divided into 4 sub-training sets with different discharge types and 4 sub-test sets with different discharge types, and different severity levels are used as labels; The sub-training sets of four different discharge types are respectively input into four corresponding type severity recognition models for training, and the optimal parameters of the corresponding type severity recognition models are determined, thereby obtaining four preliminary corresponding type severity recognition models, namely, preliminary thorn discharge severity recognition model, preliminary suspension discharge severity recognition model, preliminary bubble discharge severity recognition model, and preliminary surface discharge severity recognition model. The subtest sets of four different discharge types are input into four preliminary corresponding type severity recognition models in pairs for testing, and different types of severity test results are obtained; According to the test results of different types of severity, the average diagnostic accuracy of different discharge type severities is calculated, and based on the average diagnostic accuracy of different discharge type severities, a trained spike discharge severity recognition model, a trained suspended discharge severity recognition model, a trained bubble discharge severity recognition model and a trained surface discharge severity recognition model are obtained.

[0013] Optionally, the network structures of the local discharge type recognition model, spike discharge severity recognition model, suspended discharge severity recognition model, bubble discharge severity recognition model and surface discharge severity recognition model all adopt the LeNet-5 three-dimensional convolutional neural network model including: a sequentially connected input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a fully connected layer and an output layer.

[0014] Optionally, the calculation formula for the average diagnostic accuracy of the discharge type is expressed as follows:

[0015] in, represents the average diagnostic accuracy of the discharge type; R n It is expressed as the probability that the test set samples with discharge type n are identified as discharge type n by the preliminary partial discharge type recognition model.

[0016] Optionally, the calculation formula for the average diagnostic accuracy of different discharge type severities is expressed as follows:

[0017] in, represents the average diagnostic accuracy of the severity of n discharge types; R nq It is expressed as the probability that a test set sample with discharge type n and severity q is identified as having severity q by the preliminary corresponding type and severity recognition model.

[0018] In a second aspect, the present invention provides a bushing partial discharge diagnosis system based on multi-spectrum fusion, comprising: A data acquisition module, used to acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; A first identification module is used to identify the partial discharge data of the transformer bushing to be diagnosed by using a trained partial discharge type identification model to obtain the discharge type of the transformer bushing to be diagnosed; A selection module is used to select a trained corresponding type severity recognition model based on the partial discharge type of the transformer bushing to be diagnosed, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model and a surface discharge severity recognition model; The second recognition module is used to input the partial discharge data of the transformer bushing to be diagnosed into the trained corresponding type severity recognition model, and train to obtain the diagnosis result of the discharge severity.

[0019] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed, implements the method for diagnosing partial discharge of bushings based on multi-spectrum fusion as described in the first aspect of claim 1.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a bushing partial discharge diagnosis method, system and storage medium based on multi-spectrum fusion. The method can accurately identify the partial discharge type through a trained partial discharge type recognition model, and select a trained corresponding type severity recognition model according to the identified discharge type to accurately identify the severity of the corresponding discharge type, thereby obtaining an accurate diagnosis result of the bushing partial discharge type and severity; improving the accuracy and work efficiency of equipment insulation defect diagnosis prediction, operation and maintenance, and greatly improving the safety and reliability of bushing operation.

[0021] The present invention provides a casing partial discharge diagnosis method based on multi-spectrum fusion. In the training process of the partial discharge type recognition model and four corresponding type severity recognition models, the three-dimensional φ-Δt-Δu spectrum of the fusion feature of the partial discharge is used to train the model, and the shallow discharge feature is preliminarily extracted, so that the model has a high accuracy rate during training and the diagnosis accuracy is improved. The partial discharge type recognition model and the four corresponding type severity recognition models of the present invention all use the LeNet-5 three-dimensional convolutional neural network model to recognize the spectra, which can explore the deeper connections between time, phase and amplitude, obtain features that are more representative of the discharge type, and further improve the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic flow chart of a bushing partial discharge diagnosis method in one embodiment of the present invention; FIG2 (a) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum of a spike discharge in an embodiment of the present invention; FIG2( b ) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum of a suspended discharge in an embodiment of the present invention; FIG2( c ) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum of bubble discharge in one embodiment of the present invention; FIG2( d ) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum of surface discharge in an embodiment of the present invention; FIG3 (a) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum diagram of a spike discharge with a slight severity in one embodiment of the present invention; FIG3( b ) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum of a spike discharge with a medium severity in one embodiment of the present invention; FIG. 3( c ) is a schematic diagram of a three-dimensional φ-Δt-Δu spectrum diagram of a spike discharge with a severe severity in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention. Example 1

[0024] like Figure 1 As shown, the embodiment of the present invention introduces a method for diagnosing partial discharge of a bushing based on multi-spectrum fusion, comprising the following steps: S1: Acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; S2: using the trained partial discharge type recognition model to identify the partial discharge data of the transformer bushing to be diagnosed to obtain the discharge type of the transformer bushing to be diagnosed; S3: based on the partial discharge type of the transformer bushing to be diagnosed, selecting a trained corresponding type severity recognition model, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model, and a surface discharge severity recognition model; S4: inputting the partial discharge data of the transformer bushing to be diagnosed into the trained corresponding type severity recognition model, and training to obtain the diagnosis result of the discharge severity.

[0025] The present embodiment provides a method for diagnosing local discharge of bushings based on multi-spectrum fusion, which can accurately identify the local discharge type through a trained local discharge type recognition model, and select a trained corresponding type severity recognition model according to the identified discharge type to accurately identify the severity of the corresponding discharge type, thereby obtaining an accurate diagnosis result of the local discharge type and severity of the bushing.

[0026] In this embodiment, the training step of the partial discharge type recognition model in step S2 specifically includes: S21: Obtain spectrum sample library; S22: Divide the spectrum sample library into a training set and a test set according to a set ratio, and use different discharge types as labels; S23: Based on the training set, the partial discharge type recognition model is trained to determine the optimal parameters of the partial discharge type recognition model, thereby obtaining a preliminary partial discharge type recognition model; S24: testing the preliminary partial discharge type recognition model using the test set to obtain a test result; S25: Calculate the average diagnostic accuracy of the discharge type according to the test results, and obtain a trained partial discharge type recognition model based on the average diagnostic accuracy of the discharge type.

[0027] Specifically, the step of acquiring the spectrum sample library in step S21 includes: S211: Through the bushing partial discharge experiment in the laboratory, the pulse current method is used to obtain the typical type of partial discharge data set of the transformer bushing with different pressurization degrees; the typical type of partial discharge data set includes: related data sets of four types of discharge: spike, suspension, bubble and surface; the related data set includes phase data, amplitude data and time data; S212: dividing the data in the typical type partial discharge data set into three severity levels: mild, medium, and severe according to the pressure level, and obtaining a set of three-dimensional φ-Δt-Δu spectra of different severity levels of the four types of discharge by drawing; Specifically, the steps of obtaining a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels include: Based on the typical type partial discharge data set of different transformer bushing pressurization levels, the phase φ of each discharge signal in the typical type partial discharge data set, the time difference Δt' between adjacent discharge signals, and the difference Δu' in the absolute value of the amplitude between adjacent discharge signals are calculated according to the following table;

[0028] After discarding 5% of the data with higher values ​​of Δt't and Δu', normalization is performed according to the following formula to obtain the time difference information and amplitude difference information of each discharge signal in the spectrum, thereby drawing the three-dimensional φ-Δt-Δu spectra of the four types of discharges, as shown in Figures 2(a)-2(d); The normalized formula is as follows:

[0029]

[0030] Among them, Δt i Indicates the time difference information of the i-th discharge signal in the spectrum; Δu i Represents the amplitude difference information of the i-th discharge signal in the spectrum; Δt' i represents the time difference between the i-th discharge signal and the i-1-th discharge signal; Δu' i represents the difference in absolute value of the amplitude between the i-th discharge signal and the i-1-th discharge signal; Based on the above method of drawing a single spectrum, a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels were obtained by combined drawing.

[0031] Specifically, in this embodiment, 1200 three-dimensional φ-Δt-Δu spectra are drawn for each discharge type, wherein 400 severity levels of each discharge type are drawn; as shown in FIG3 (a)-(c), taking spike discharge as an example, three-dimensional φ-Δt-Δu spectra of severity levels of “mild”, “moderate” and “severe” are drawn; S213 normalizes the three-dimensional φ-Δt-Δu spectrogram sets of the four discharge types with different severity levels to obtain a spectrogram sample library.

[0032] Specifically, the data is accumulated in a discrete space of 256x256x256. The number of discharge points falling at each position is counted, and the entire 256x256x256 three-dimensional φ-Δt-Δu spectrogram is normalized to form a spectrogram sample library.

[0033] Specifically, in step S22, the spectrum sample library is divided into a training set and a test set according to a set ratio, and different discharge types are used as labels. As shown in the following table, the spectrum sample library is divided into a training set and a test set according to a ratio of 3:1;

[0034] Specifically, in step S25, the average diagnostic accuracy of the discharge type is calculated according to the test results, and based on the average diagnostic accuracy of the discharge type, a trained partial discharge type recognition model is obtained, which specifically includes: The calculation formula for the average diagnostic accuracy of discharge type is calculated according to the following formula,

[0035] in, represents the average diagnostic accuracy of the discharge type; R n It is expressed as the probability that the test set samples with discharge type n are identified as discharge type n by the preliminary partial discharge type recognition model.

[0036] When the average accuracy of the discharge type diagnosis is higher than 98%, the partial discharge type recognition model is considered to have been trained, otherwise it is retrained.

[0037] In this embodiment, the training steps of the four corresponding types of severity recognition models in step S3: spike discharge severity recognition model, suspended discharge severity recognition model, bubble discharge severity recognition model and surface discharge severity recognition model are as follows: (1) Spike discharge severity identification model S311: Based on the four types of discharges, the training set and the test set are divided into 4 sub-training sets of different discharge types and 4 sub-test sets of different discharge types, and different severity levels are used as labels, thereby obtaining the sub-training set and sub-test set of spike discharges, as shown in the following table:

[0038] S312: inputting the spike discharge sub-training set into the spike discharge severity recognition model for training, determining the optimal parameters of the spike discharge severity recognition model, and obtaining a preliminary spike discharge severity recognition model; S313: inputting the spike discharge subtest set into the preliminary spike discharge severity recognition model for testing, and obtaining a severity test result of the discharge type being spike discharge; S314: The root discharge type is a spike discharge severity test result, calculating the average diagnostic accuracy of the spike discharge severity, and obtaining a trained spike discharge severity recognition model based on the average diagnostic accuracy of the spike discharge severity.

[0039] (2) Suspension discharge severity identification model S321: Based on the four types of discharge, the training set and the test set are divided into 4 sub-training sets of different discharge types and 4 sub-test sets of different discharge types, and different severity levels are used as labels, thereby obtaining the sub-training set and sub-test set of suspended discharge, as shown in the following table:

[0040] S322: inputting the suspension discharge training set into the suspension discharge severity recognition model for training, determining the optimal parameters of the suspension discharge severity recognition model, and obtaining a preliminary suspension discharge severity recognition model; S323: inputting the sub-test set of suspended discharge into the preliminary suspended discharge severity recognition model for testing, and obtaining a test result of the severity of the discharge type being suspended discharge; S324: the root discharge type is a suspension discharge severity test result, calculating the average diagnostic accuracy of the suspension discharge severity, and obtaining a trained suspension discharge severity recognition model based on the average diagnostic accuracy of the suspension discharge severity.

[0041] (3) Bubble discharge severity identification model S331: Based on the four discharge types, the training set and the test set are divided into 4 sub-training sets of different discharge types and 4 sub-test sets of different discharge types, and different severity levels are used as labels, thereby obtaining the sub-training set and sub-test set of bubble discharge, as shown in the following table:

[0042] S332: inputting the sub-training set of bubble discharge into the bubble discharge severity recognition model for training, determining the optimal parameters of the bubble discharge severity recognition model, and obtaining a preliminary bubble discharge severity recognition model; S333: inputting the sub-test set of bubble discharge into the preliminary bubble discharge severity recognition model for testing, and obtaining a severity test result of the discharge type being bubble discharge; S334: based on the severity test result of the root discharge type being bubble discharge, calculate the average diagnostic accuracy of the bubble discharge severity, and obtain a trained bubble discharge severity recognition model based on the average diagnostic accuracy of the bubble discharge severity.

[0043] (4) Surface discharge severity identification model S341: Based on the four types of discharge, the training set and the test set are divided into 4 sub-training sets of different discharge types and 4 sub-test sets of different discharge types, and different severity levels are used as labels, thereby obtaining the sub-training set and sub-test set of surface discharge, as shown in the following table:

[0044] S342: inputting the surface discharge sub-training set into the surface discharge severity recognition model for training, determining the optimal parameters of the surface discharge severity recognition model, and obtaining a preliminary surface discharge severity recognition model; S343: inputting the sub-test set of surface discharge into the preliminary surface discharge severity recognition model for testing, and obtaining a severity test result of the discharge type being surface discharge; S344: based on the test result of severity of the root discharge type being creeping discharge, the average diagnostic accuracy of the creeping discharge severity is calculated, and based on the average diagnostic accuracy of the creeping discharge severity, a trained creeping discharge severity recognition model is obtained.

[0045] Specifically, the calculation formula for the average diagnostic accuracy of the severity of discharge type n is expressed as follows:

[0046] in, represents the average diagnostic accuracy of the severity of n discharge types; R nq It is expressed as the probability that the test set sample with discharge type n and severity q is identified as having severity q by the preliminary corresponding type and severity recognition model; When the average diagnostic accuracy of the severity of discharge type n is higher than 98%, the discharge severity recognition model of discharge type n is considered to have been trained, otherwise it is retrained.

[0047] In this embodiment, the network structures of the partial discharge type recognition model, the spike discharge severity recognition model, the suspended discharge severity recognition model, the bubble discharge severity recognition model and the surface discharge severity recognition model all adopt the LeNet-5 three-dimensional convolutional neural network model including: a sequentially connected input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a fully connected layer and an output layer; its detailed structural composition and parameters are shown in the following table:

[0048] In summary, the method for diagnosing local discharge of bushings based on multi-spectrum fusion proposed in this embodiment can realize accurate identification of local discharge types and accurate identification of the severity of the corresponding discharge types, thereby obtaining accurate diagnosis results of the local discharge type and severity of the bushings. On the technical level, the present invention will greatly improve the accuracy and work efficiency of on-site local discharge testing and bushing operation and maintenance, avoid equipment with defects from being put into operation, and greatly promote the progress of electrical equipment testing and detection technology. From the perspective of economic benefits, if the core technology of the present invention is industrialized and marketed, it will greatly improve the accuracy and efficiency of bushing local discharge diagnosis, reduce the probability of insulation failure in bushings, and have significant direct and indirect economic benefits. Example 2 This embodiment provides a bushing partial discharge diagnosis system based on multi-spectrum fusion, including:

[0049] A data acquisition module, used to acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; A first identification module is used to identify the partial discharge data of the transformer bushing to be diagnosed by using a trained partial discharge type identification model to obtain the discharge type of the transformer bushing to be diagnosed; A selection module is used to select a trained corresponding type severity recognition model based on the partial discharge type of the transformer bushing to be diagnosed, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model and a surface discharge severity recognition model; The second recognition module is used to input the partial discharge data of the transformer bushing to be diagnosed into the trained corresponding type severity recognition model, and train to obtain the diagnosis result of the discharge severity. Example 3

[0050] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed, the method for diagnosing partial discharge of bushings based on multi-spectrum fusion as described in Embodiment 1 of Claim 1 is implemented.

[0051] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may 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 codes.

[0052] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0053] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0054] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method, system and storage medium for diagnosing partial discharge of bushing based on multi-spectrum fusion, characterized in that: include: Acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; Using the trained partial discharge type recognition model to identify the partial discharge data of the transformer bushing to be diagnosed, the discharge type of the transformer bushing to be diagnosed is obtained; Based on the partial discharge type of the transformer bushing to be diagnosed, a trained corresponding type severity recognition model is selected, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model, and a surface discharge severity recognition model; The partial discharge data of the transformer bushing to be diagnosed is input into the trained corresponding type severity recognition model, and the diagnosis result of the discharge severity is obtained through training.

2. The method for diagnosing partial discharge of bushing based on multi-spectrum fusion according to claim 1, characterized in that: The training step of the partial discharge type recognition model includes: Get the spectrum sample library; The spectrogram sample library is divided into a training set and a test set according to the set ratio, and different discharge types are used as labels; Based on the training set, the partial discharge type recognition model is trained to determine the optimal parameters of the partial discharge type recognition model, thereby obtaining a preliminary partial discharge type recognition model; The preliminary partial discharge type recognition model is tested using the test set to obtain the test results; The average diagnostic accuracy of the discharge type is calculated according to the test results, and based on the average diagnostic accuracy of the discharge type, a trained partial discharge type recognition model is obtained.

3. The method for diagnosing partial discharge of bushing based on multi-spectrum fusion according to claim 2, characterized in that: The step of acquiring the spectrum sample library comprises: Acquire typical partial discharge data sets of different pressure levels of transformer bushings; the typical partial discharge data sets include relevant data sets of four types of discharges: spike, suspension, bubble and surface; the relevant data sets include phase data, amplitude data and time data; The data in the typical partial discharge data set are divided into three severity levels: mild, medium and severe according to the degree of pressure, and a set of three-dimensional φ-Δt-Δu spectra of different severity levels of the four discharge types are obtained by drawing. The three-dimensional φ-Δt-Δu spectrogram sets of the four discharge types with different severity levels are normalized to obtain a spectrogram sample library.

4. The method for diagnosing partial discharge of bushings based on multi-spectrum fusion according to claim 3, characterized in that: The step of obtaining a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels by drawing comprises: Based on a typical type of partial discharge data set with different pressure levels on transformer bushings, the phase φ of each discharge signal in the typical type of partial discharge data set, the time difference Δt' between adjacent discharge signals, and the difference Δu' between the absolute values ​​of the amplitudes of adjacent discharge signals are calculated; The time difference Δt' between adjacent discharge signals and the difference Δu' between the absolute amplitudes of adjacent discharge signals are normalized to obtain the time difference information and amplitude difference information of each discharge signal in the spectrum. The normalization formula is expressed as follows: ； ； Among them, Δt i Indicates the time difference information of the i-th discharge signal in the spectrum; Δu i Indicates the amplitude difference information of the i-th discharge signal in the spectrum; Δt' i represents the time difference between the i-th discharge signal and the i-1-th discharge signal; Δu' i represents the difference in absolute value of the amplitude between the i-th discharge signal and the i-1-th discharge signal; Based on the phase φ of each discharge signal and the time difference information and amplitude difference information in the spectrum, a set of three-dimensional φ-Δt-Δu spectra of four discharge types with different severity levels are drawn.

5. The method for diagnosing partial discharge of bushing based on multi-spectrum fusion according to claim 2, characterized in that: The training steps of the spike discharge severity recognition model, the suspension discharge severity recognition model, the bubble discharge severity recognition model and the surface discharge severity recognition model include: Based on the four types of discharge, the training set and test set are divided into 4 sub-training sets with different discharge types and 4 sub-test sets with different discharge types, and different severity levels are used as labels; The sub-training sets of four different discharge types are respectively input into four corresponding type severity recognition models for training, and the optimal parameters of the corresponding type severity recognition models are determined, thereby obtaining four preliminary corresponding type severity recognition models, namely, preliminary thorn discharge severity recognition model, preliminary suspension discharge severity recognition model, preliminary bubble discharge severity recognition model, and preliminary surface discharge severity recognition model. The subtest sets of four different discharge types are input into four preliminary corresponding type severity recognition models for testing, and different types of severity test results are obtained; According to the test results of different types of severity, the average diagnostic accuracy of different discharge type severities is calculated, and based on the average diagnostic accuracy of different discharge type severities, a trained spike discharge severity recognition model, a trained suspended discharge severity recognition model, a trained bubble discharge severity recognition model and a trained surface discharge severity recognition model are obtained.

6. The method for diagnosing partial discharge of bushings based on multi-spectrum fusion according to claim 5, characterized in that: The network structures of the partial discharge type recognition model, the spike discharge severity recognition model, the suspended discharge severity recognition model, the bubble discharge severity recognition model and the surface discharge severity recognition model all adopt the LeNet-5 three-dimensional convolutional neural network model, including: a sequentially connected input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a fully connected layer and an output layer.

7. The method for diagnosing partial discharge of bushings based on multi-spectrum fusion according to claim 6, characterized in that: The calculation formula of the average diagnostic accuracy of the discharge type is expressed as follows: ； in, represents the average diagnostic accuracy of the discharge type; R n It is expressed as the probability that the test set samples with discharge type n are identified as discharge type n by the preliminary partial discharge type recognition model.

8. The method for diagnosing partial discharge of bushings based on multi-spectrum fusion according to claim 7, characterized in that: The calculation formula for the average diagnostic accuracy of the different discharge type severities is expressed as follows: ； in, represents the average diagnostic accuracy of the severity of n discharge types; R nq It is expressed as the probability that a test set sample with discharge type n and severity q is identified as having severity q by the preliminary corresponding type and severity recognition model.

9. A bushing partial discharge diagnosis system based on multi-spectrum fusion, characterized in that: include: A data acquisition module, used to acquire partial discharge data of the transformer bushing to be diagnosed, wherein the partial discharge data includes: phase information, amplitude information and time information of the discharge signal; A first identification module is used to identify the partial discharge data of the transformer bushing to be diagnosed by using a trained partial discharge type identification model to obtain the discharge type of the transformer bushing to be diagnosed; A selection module is used to select a trained corresponding type severity recognition model based on the partial discharge type of the transformer bushing to be diagnosed, wherein the corresponding type severity recognition model is any one of a spike discharge severity recognition model, a suspended discharge severity recognition model, a bubble discharge severity recognition model and a surface discharge severity recognition model; The second recognition module is used to input the partial discharge data of the transformer bushing to be diagnosed into the trained corresponding type severity recognition model, and train to obtain the diagnosis result of the discharge severity.

10. A computer-readable storage medium, characterized in that: The computer program is stored therein, and when the computer program is executed, the bushing partial discharge diagnosis method based on multi-spectrum fusion according to any one of claims 1 to 8 is implemented.

Images