Turn-to-turn short-circuit fault diagnosis method and system for reactor, and medium

Abstract

A turn-to-turn short-circuit fault diagnosis method and system for a reactor, and a medium. The method comprises: using operating data of a reactor to be tested to plot a target Bowditch curve of said reactor (101); calculating feature parameters of the target Bowditch curve (102); inputting the target Bowditch curve into a preset embedded target diagnosis network model, so as to extract a feature map of the target Bowditch curve (103); using the feature parameters and the feature map of the target Bowditch curve to generate feature information (104); and comparing whether the feature information is consistent with preset feature information, and on the basis of a comparison result, determining a fault state of said reactor (105). The method solves the technical problem of existing detection methods having poor sensitivity and being unable to detect incipient turn-to-turn short-circuit conditions of reactors in a timely manner, and realizes the real-time detection and determination of turn-to-turn short-circuit faults in dry-type air-core reactors.

Description

A method, system, and medium for diagnosing inter-turn short-circuit faults in reactors.

[0001] This application claims priority to Chinese Patent Application No. 202410808842.3, filed on June 21, 2024, entitled "A Method, System and Medium for Diagnosing Inter-turn Short Circuit Faults in Reactors", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of reactor technology, and in particular to a method, system and medium for diagnosing inter-turn short-circuit faults in reactors. Background Technology

[0003] With continuous technological breakthroughs and development, higher demands are being placed on the safe and stable operation of power system equipment to meet the electricity needs of economic and social development and accelerate the construction of smart grids. Dry-type air-core reactors are widely used in power systems due to their linear inductance, high mechanical strength, and low noise. However, inter-turn short-circuit faults are the main type of fault in dry-type air-core reactors. These faults can cause rapid localized heating of the reactor, leading to damage to the reactor winding insulation, and in severe cases, fire or even burnout, posing serious safety hazards and economic losses to the power system.

[0004] Therefore, methods such as spatial magnetic field detection, electrical parameter detection, and frequency response are commonly used to detect inter-turn short circuit faults in reactors. However, these detection methods have poor sensitivity and cannot detect the initial inter-turn short circuit condition of the reactor in a timely manner.

[0005] Summary of the Invention

[0006] This invention provides a method, system, and medium for diagnosing inter-turn short-circuit faults in reactors, which solves the technical problem that existing detection methods have poor sensitivity and cannot detect the initial inter-turn short-circuit condition of reactors in a timely manner.

[0007] The first aspect of this invention provides a method for diagnosing inter-turn short-circuit faults in a reactor, comprising:

[0008] Using the operating data of the reactor under test, the target Bowditch curve of the reactor under test is plotted.

[0009] Calculate the characteristic parameters of the target Bowditch curve;

[0010] The target Bowditch curve is input into a preset embedded target diagnostic network model to extract the feature map of the target Bowditch curve;

[0011] Feature information is generated using the feature parameters and feature map of the target Bowditch curve;

[0012] The fault state of the reactor under test is determined based on the comparison result by comparing whether the feature information is consistent with the preset feature information.

[0013] Optionally, the step of plotting the Bowditch curve of the reactor under test using its operating data includes:

[0014] Collect operating data of the reactor under test; wherein, the operating data includes current data and voltage data;

[0015] Extract the power frequency components corresponding to the voltage data and the current data respectively, and reconstruct the voltage signal and the current signal using the power frequency components;

[0016] The initial Bowditch curve of the reactor under test is plotted using the reconstructed voltage signal and the reconstructed current signal.

[0017] The initial Bowditch curve is preprocessed to generate the target Bowditch curve.

[0018] Optionally, the step of performing image preprocessing on the initial Bowditch curve to generate the target Bowditch curve includes:

[0019] The initial Bowditch curve is denoised using Gaussian denoising method.

[0020] A local sharpening method was used to enhance the image of the denoised Bowditch curve.

[0021] The enhanced Bowditch curve is subjected to grayscale conversion, image rotation, image stretching, and image shrinking in sequence to generate the target Bowditch curve.

[0022] Optionally, it also includes:

[0023] Acquire normal operating data of reactors in a healthy state and fault operating data of reactors under multiple different fault states;

[0024] A sample set is constructed using the normal operation data and the fault operation data.

[0025] Using the running data from the sample set, plot the initial sample Bowditch curve corresponding to each of the running data sets;

[0026] The initial sample Bowditch curves are classified according to the running status of each running data in the sample set, and status labels are added to the initial sample Bowditch curves of each running status category according to the classification results to generate updated sample Bowditch curves.

[0027] Image preprocessing is performed on the updated sample Bowditch curves to generate the target sample Bowditch curves;

[0028] The Bowditch curves of each target sample are divided into a training set and a test set.

[0029] Optionally, it also includes:

[0030] Construct an embedded initial diagnostic network model; wherein the embedded initial diagnostic network model includes an input layer, a first convolutional layer, a first inverse residual layer and multiple second inverse residual layers, second convolutional layers, average pooling layers, a third convolutional layer and an output layer;

[0031] The target sample Bowditch curve of the training set is input into the input layer;

[0032] The image size of the target sample Bowditch curve is normalized through the input layer and then input into the first convolutional layer;

[0033] The first convolutional layer performs convolution processing on the size-normalized Bowditch curve of the target sample to generate a first feature map, which is then sequentially input into the first inverse residual layer.

[0034] The first feature map is up-dimensionalized by the first inverse residual layer to generate a second feature map, which is then sequentially input into multiple second inverse residual layers.

[0035] The second feature map is convolved by each of the second inverted residual layers to generate a third feature map, which is then input into the second convolutional layer.

[0036] The third feature map is convolved by the second convolutional layer to generate a fourth feature map, which is then input into the average pooling layer.

[0037] The fourth feature map is reduced in dimensionality by the average pooling layer to generate a fifth feature map, which is then input into the third convolutional layer.

[0038] The fifth feature map is convolved by the third convolutional layer to generate multiple sixth feature maps, which are then input into the output layer.

[0039] The output layer classifies each of the sixth feature maps to generate multiple target feature maps.

[0040] Optionally, it also includes:

[0041] Determine whether the running state category corresponding to each target feature map is consistent with the preset running state category;

[0042] If the running status categories are consistent, the target sample Bowditch curve of the test set is input into the embedded initial diagnostic network model for testing;

[0043] When the test results are consistent with the training results, the embedded initial diagnostic network model is determined as the embedded target diagnostic network model.

[0044] If the running state categories are inconsistent, the model parameters of the embedded initial diagnostic network model are optimized, and the process jumps to the step of inputting the target sample Bowditch curve of the training sample into the input layer until the running state category corresponding to the target feature map is consistent with the preset running state category, and then the next step is executed.

[0045] Optionally, the step of calculating the characteristic parameters of the target Bowditch curve includes:

[0046] Extract the characteristic parameters of the target Bowditch curve of the reactor under test; wherein, the characteristic parameters include the tilt angle, major axis and minor axis;

[0047] The formula for calculating the tilt angle is as follows:

[0048] The formula for extracting the major axis is:

[0049] The formula for extracting the minor axis is:

[0050] In the formula, a is the major axis, b is the minor axis, θ is the inclination angle, and U c I represents the voltage amplitude. c φ is the current amplitude. u The initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

[0051] Optionally, the step of comparing whether the feature information is consistent with preset feature information and determining the fault state of the reactor under test based on the comparison result includes:

[0052] Compare whether the feature parameters and feature map corresponding to the target Bowditch curve are consistent with the feature parameters and feature map of the Bowditch curve under the preset healthy state;

[0053] If the characteristic parameters and characteristic diagrams are consistent, then it is determined that the reactor under test is not faulty.

[0054] If the feature parameters and feature map are inconsistent, then determine whether the feature parameters and feature map are consistent with the feature parameters and feature map of the Bowditch curve under the preset fault state, respectively.

[0055] If the characteristic parameters and the characteristic map are consistent, then the current fault state of the reactor under test is determined according to the fault state corresponding to the consistency of the characteristic parameters and the characteristic map.

[0056] A second aspect of the present invention provides a reactor inter-turn short-circuit fault diagnosis system, comprising:

[0057] The plotting module is used to plot the target Bowditch curve of the reactor under test using the operating data of the reactor under test.

[0058] The calculation module is used to calculate the characteristic parameters of the target Bowditch curve;

[0059] The feature map module is used to input the target Bowditch curve into a preset embedded target diagnostic network model and extract the feature map of the target Bowditch curve.

[0060] The feature information module is used to generate feature information using the feature parameters and feature map of the target Bowditch curve;

[0061] The comparison module is used to compare whether the feature information is consistent with the preset feature information, and to determine the fault state of the reactor under test based on the comparison result.

[0062] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the reactor inter-turn short-circuit fault diagnosis method as described in any of the preceding claims.

[0063] As can be seen from the above technical solutions, the present invention has the following advantages:

[0064] This invention collects real-time operating data of the reactor under test port and uses this operating data to plot a Bowditch curve. By extracting the feature information of the Bowditch curve through a target diagnostic network model and comparing this feature information with the feature information of a preset Bowditch curve, the current fault state of the reactor under test can be obtained, thereby realizing real-time detection and judgment of inter-turn short-circuit faults in dry-type air-core reactors. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 is a flowchart of the steps of a reactor inter-turn short-circuit fault diagnosis method provided in Embodiment 1 of the present invention;

[0067] Figure 2 is a flowchart of the steps of a reactor inter-turn short-circuit fault diagnosis method provided in Embodiment 2 of the present invention;

[0068] Figure 3 is a schematic diagram of an image preprocessing method provided in Embodiment 2 of the present invention;

[0069] Figure 4 is a schematic diagram of the inverted residual structure of a MobileNetV2 intelligent diagnostic network provided in Embodiment 2 of the present invention;

[0070] Figure 5 is a diagnostic schematic diagram of a fault diagnosis system provided in Embodiment 2 of the present invention;

[0071] Figure 6 is a structural block diagram of a reactor inter-turn short-circuit fault diagnosis system provided in Embodiment 3 of the present invention. Detailed Implementation

[0072] This invention provides a method, system, and medium for diagnosing inter-turn short-circuit faults in reactors, addressing the technical problem that existing detection methods have poor sensitivity and cannot detect the initial inter-turn short-circuit condition of reactors in a timely manner.

[0073] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0074] Example 1

[0075] Please refer to Figure 1, which is a flowchart of the steps of a reactor inter-turn short circuit fault diagnosis method provided in Embodiment 1 of the present invention.

[0076] The present invention provides a method for diagnosing inter-turn short-circuit faults in reactors, comprising the following steps:

[0077] Step 101: Using the operating data of the reactor under test, plot the target Bowditch curve of the reactor under test.

[0078] It should be noted that the reactor under test refers to a dry-type air-core reactor that requires fault diagnosis.

[0079] In practice, high-precision voltage or current transformers are used to collect port voltage and port current data of the dry-type air-core reactor that needs to be diagnosed for faults. The power frequency components of the voltage and current data are extracted, the voltage and current signals are reconstructed, and an initial Bowditch curve (i.e., Bowditch curve) is plotted based on the reconstructed voltage and current signals. The target Bowditch curve can be obtained by performing image preprocessing on the initial Bowditch curve.

[0080] It should be noted that image preprocessing includes image denoising and image enhancement operations.

[0081] Step 102: Calculate the characteristic parameters of the target Bowditch curve.

[0082] It should be noted that the characteristic parameters include the tilt angle, major axis, and minor axis.

[0083] In practice, the tilt angle, major axis, and minor axis of the Bowditch curve can be obtained by using the calculation equations for the tilt angle, major axis, and minor axis.

[0084] Step 103: Input the target Bowditch curve into the preset embedded target diagnostic network model and extract the feature map of the target Bowditch curve.

[0085] It should be noted that the embedded target diagnostic network model refers to the MobileNetV2 intelligent diagnostic network model obtained through training. Specifically, this invention is mainly applied to an intelligent diagnostic system for winding faults in dry-type air-core reactors. The MobileNetV2 intelligent diagnostic network model is embedded within this intelligent fault diagnostic system to facilitate subsequent fault diagnosis between turns of the dry-type air-core reactor windings.

[0086] In practical implementation, the feature map of the target Bowditch curve is extracted through the embedded MobileNetV2 intelligent diagnostic network model.

[0087] Step 104: Use the feature parameters and feature map of the target Bowditch curve to generate feature information.

[0088] It should be noted that the feature information refers to the feature map and feature parameters of the Bowditch curve.

[0089] Step 105: Compare whether the feature information is consistent with the preset feature information, and determine the fault state of the reactor under test based on the comparison result.

[0090] It should be noted that the preset feature information refers to the feature parameters and feature maps of the Bowditch curve under healthy conditions or multiple fault conditions.

[0091] In practice, the characteristic parameters of the target Bowditch curve are compared with those of the Bowditch curves under healthy conditions or multiple fault conditions, or the characteristic graph of the target Bowditch curve is compared with that of the Bowditch curves under healthy conditions or multiple fault conditions. If the characteristic parameters are consistent with those of a certain condition, or the characteristic graph is consistent with those of a certain condition, then that condition is determined to be the condition of the reactor under test. If that condition is healthy, then the condition of the reactor under test is also healthy. Similarly, if that condition is an outer encapsulation fault - minor fault, then the condition of the reactor under test is also an outer encapsulation fault - minor fault.

[0092] Example 2

[0093] Please refer to Figures 2 to 5. Figure 2 is a flowchart of the steps of a reactor inter-turn short circuit fault diagnosis method provided in Embodiment 2 of the present invention.

[0094] The present invention provides a method for diagnosing inter-turn short-circuit faults in reactors, comprising the following steps:

[0095] Step 201: Collect the operating data of the reactor under test; the operating data includes current data and voltage data.

[0096] It should be noted that high-precision voltage transformers (PTs) or current transformers (CTs) are used to collect port voltage data of dry-type air-core reactors that require fault diagnosis. and port current data

[0097] Step 202: Extract the power frequency components corresponding to the voltage and current data respectively, and reconstruct the voltage and current signals using the power frequency components.

[0098] It should be noted that the voltage and current signals are reconstructed by extracting the power frequency components, and the reconstructed signal expression is as follows:

[0099] In the formula, u is the voltage signal, i is the current signal, and U c To extract The amplitude, I c To extract The amplitude, f is the frequency, which is 50Hz, t is the time, and φ is the amplitude.u The initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

[0100] Step 203: Using the reconstructed voltage signal and the reconstructed current signal, plot the initial Bowditch curve of the reactor under test.

[0101] It should be noted that the initial Bowditch curve of the reactor under test is plotted based on the reconstructed voltage and current signals. The expression for plotting the Bowditch curve is as follows:

[0102] In the formula, u is the voltage signal, i is the current signal, and U c To extract The amplitude, I c To extract The amplitude, φ u The initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

[0103] Step 204: Perform image preprocessing on the initial Bowditch curve to generate the target Bowditch curve.

[0104] Optionally, step 204 includes the following steps S11-S13:

[0105] S11. Based on Gaussian denoising method, perform image denoising processing on the initial Bowditch curve;

[0106] S12. Use a local sharpening method to enhance the image of the denoised Bowditch curve;

[0107] S13. Perform grayscale conversion, image rotation, image stretching and image shrinking on the enhanced Bowditch curve in sequence to generate the target Bowditch curve.

[0108] It should be noted that, as shown in Figure 3, Gaussian denoising is employed. A Gaussian function is used as a template and convolved with the original curve image to achieve noise reduction, making the image smoother, reducing image noise, and lowering detail levels. Considering that the Bowditch curve image features are mainly concentrated in the central region of the image, local sharpening is used to enhance the central region, highlighting its contours and details. After image denoising and enhancement, all color images are converted to grayscale images. Simultaneously, rotation, stretching, and shrinking operations can be performed on all images to further enhance them, thus obtaining the target Bowditch curve.

[0109] Optionally, this method further includes the following steps S21-S26:

[0110] S21. Obtain normal operating data of reactors under healthy conditions and fault operating data of reactors under multiple different fault conditions;

[0111] S22. Construct a sample set using normal operation data and fault operation data;

[0112] S23. Using the running data from the sample set, plot the initial sample Bowditch curve corresponding to each running data.

[0113] S24. Classify the initial sample Bowditch curves according to the running status of each running data in the sample set, and add status labels to the initial sample Bowditch curves of each running status category according to the classification results to generate updated sample Bowditch curves.

[0114] S25. Perform image preprocessing on the updated sample Bowditch curve to generate the target sample Bowditch curve;

[0115] S26. Divide the Bowditch curves of each target sample into a training set and a test set.

[0116] It should be noted that the normal operating data refers to the reactor port voltage and current data under healthy conditions.

[0117] The fault operation data consists of reactor port voltage and current data under various fault states. Specifically, the fault states include outer encapsulation fault, middle encapsulation fault, and inner encapsulation fault, and each fault state corresponds to a fault severity category (i.e., operating state category), including minor fault, moderate fault, and severe fault. Corresponding fault state labels are generated for different fault operating states, with label formats such as: outer encapsulation fault_minor fault.

[0118] In practice, voltage transformers (PTs) and current transformers (CTs) are used to simultaneously collect reactor port voltage / current data under healthy conditions. and It is also necessary to collect a large amount of voltage and current data under fault conditions of dry-type air-core reactors, including historical data of actual reactors, model simulation data, and publicly available data.

[0119] Plot the initial sample Bowditch curves (i.e., initial sample Bowditch curves) using the voltage / current data of the reactor port under healthy conditions, and plot the initial sample Bowditch curves for each fault condition using the voltage and current data of the reactor port under each fault condition.

[0120] A sample set is constructed using all initial Bowditch curves.

[0121] Add labels to all the Bowditch curve image data plotted above. In this example, they are divided into "healthy winding", "outer layer encapsulation fault", "middle layer encapsulation fault" and "inner layer encapsulation fault". The fault severity label is the fault severity category (i.e. operating status category) under the corresponding fault type, including "minor fault", "moderate fault" and "serious fault", with the label format as: "outer layer encapsulation fault_minor fault".

[0122] As shown in Figure 3, after adding labels, the updated Bowditch curves are obtained. Image denoising and enhancement operations are then performed on all Bowditch curve images. In this example, Gaussian denoising is used for image denoising. A Gaussian function is used as a template and convolved with the original image to achieve denoising, making the image smoother, reducing image noise, and lowering detail levels. Considering that the features of the Bowditch curve image are mainly concentrated in the central region of the entire image, local sharpening is used to enhance the central region of the image, highlighting the contours and details of the central region.

[0123] After completing image denoising and enhancement operations, all color images are converted to grayscale images. At the same time, all images can be enhanced by rotation, stretching, shrinking and other operations to obtain a sample set of the target Bowditch curve. Based on all image data, normal and faulty image datasets are constructed. 70% of the data in the dataset is used as training samples and 30% is used as test samples. The classification accuracy of the test samples is used as the evaluation index of the model's quality. The dataset is divided into training set and test set, laying the groundwork for training the MobileNetV2 intelligent diagnostic network model.

[0124] Optionally, this method further includes the following steps S31-S310:

[0125] S31. Construct an embedded initial diagnostic network model; wherein, the embedded initial diagnostic network model includes an input layer, a first convolutional layer, a first inverse residual layer and multiple second inverse residual layers, a second convolutional layer, an average pooling layer, a third convolutional layer and an output layer;

[0126] S32. Input the target sample Bowditch curve from the training set into the input layer;

[0127] S33. The image size of the target sample Bowditch curve is normalized through the input layer and then input into the first convolutional layer;

[0128] S34. The Bowditch curve of the target sample after size normalization is convolved by the first convolutional layer to generate the first feature map and then input into the first inverse residual layer in sequence.

[0129] S35. The first feature map is up-dimensionalized by the first inverse residual layer to generate a second feature map, which is then sequentially input into multiple second inverse residual layers.

[0130] S36. The second feature map is convolved through each second inverted residual layer to generate the third feature map, which is then input into the second convolutional layer.

[0131] S37. The third feature map is convolved by the second convolutional layer to generate the fourth feature map and then input into the average pooling layer.

[0132] S38. The fourth feature map is reduced in dimensionality by an average pooling layer to generate the fifth feature map, which is then input into the third convolutional layer.

[0133] S39. The fifth feature map is convolved by the third convolutional layer to generate multiple sixth feature maps, which are then input into the output layer.

[0134] S310. Classify each sixth feature map through the output layer to generate multiple target feature maps.

[0135] It should be noted that the embedded initial diagnostic network model is established, namely the MobileNetV2 intelligent diagnostic network model. The MobileNetV2 intelligent diagnostic network is a subclass of VGG-Net within the CNN family. The MobileNetV2 network structure is composed of multiple stacked blocks, the most important part of which is the inverse residual structure, as shown in Figure 4. The inverse residual structure mainly consists of three parts: channel expansion, feature extraction, and channel compression. The complete detailed structure of the MobileNetV2 intelligent diagnostic network model is shown in the following sections and Table 1:

[0136] First layer: Input layer, image size is 224×224, the size of the input image is normalized.

[0137] The second layer is a convolutional layer with dimensions of 224×224×3, 32 output channels, 1 module repetition, and a stride of 2.

[0138] The third layer is the inverted residual layer, with dimensions of 112×112×32, an upsizing factor of 1, 16 output channels, a module repetition count of 1, and a step size of 1.

[0139] Fourth layer: Inverted residual layer, with dimensions of 112×112×16, a dimensionality increase of 6, 24 output channels, module repetition times of 2, and a step size of 2.

[0140] Fifth layer: Inverted residual layer, with dimensions of 56×56×24, dimensionality increase ratio of 6, 32 output channels, module repetition times of 3, and step size of 2.

[0141] The sixth layer is the inverted residual layer, with dimensions of 28×28×32, an upsizing factor of 6, 64 output channels, a module repetition count of 4, and a step size of 2.

[0142] The seventh layer is the inverted residual layer, with dimensions of 14×14×64, an upsizing factor of 6, 96 output channels, a module repetition count of 3, and a step size of 1.

[0143] Eighth layer: Inverted residual layer, with dimensions of 14×14×96, a dimensionality increase of 6, 160 output channels, module repetition times of 3, and a step size of 2.

[0144] Layer 9: Inverted residual layer, size 7×7×160, with a dimensionality increase of 6, 320 output channels, module repetition once, and stride of 1. Layer 10: Convolutional layer, size 7×7×320, 1280 output channels, module repetition once, and stride of 1.

[0145] Eleventh layer: Average pooling layer, size 7×7×1280, module repetition 1 time.

[0146] The twelfth layer is a convolutional layer with a size of 1×1×1280. Since there are 10 different fault types in this example, the final number of output channels is 10.

[0147] The thirteenth layer: the output layer, with 10 output channels.

[0148] Table 1. MobileNetV2 Intelligent Diagnostic Network Architecture Configuration Table

[0149] It should be noted that the first convolutional layer is the convolutional layer of the second layer, the first inverse residual layer is the inverse residual layer of the third layer, the second inverse residual layer is the inverse residual layer of the fourth to ninth layers, the second convolutional layer is the convolutional layer of the tenth layer, and the third convolutional layer is the convolutional layer of the twelfth layer.

[0150] The first feature map is generated by convolution in the second convolutional layer. The second feature map is generated by dimensionality increase in the third inverse residual layer. The third feature map is generated by convolution in the fourth to ninth inverse residual layers. The fourth feature map is generated by convolution in the tenth convolutional layer. The fifth feature map is generated by dimensionality reduction in the eleventh average pooling layer. The sixth convolutional layer is generated by convolution in the twelfth convolutional layer. The target feature map is generated by classification in the thirteenth output layer.

[0151] Each inverse residual layer is equivalent to a module, which contains the three parts shown in Figure 4. The dimensionality increase ratio described in the steps only refers to the parameter settings for dimensionality increase in the corresponding layer. Each inverse residual layer goes through dimensionality increase, feature extraction, and dimensionality reduction steps. Therefore, each inverse residual layer does not only perform dimensionality increase, but also includes three steps.

[0152] In practice, the target sample Bowditch curves from the training set are input into the initial MobileNetV2 intelligent diagnostic network model to train the network model, and the target feature map of the target sample Bowditch curves is extracted through the initial MobileNetV2 intelligent diagnostic network model.

[0153] Optionally, this method further includes the following steps S41-S44:

[0154] S41. Determine whether the running state category corresponding to each target feature map is consistent with the preset running state category;

[0155] S42. If the running status categories are consistent, the target sample Bowditch curve of the test set is input into the embedded initial diagnostic network model for testing.

[0156] S43. When the test results are consistent with the training results, the embedded initial diagnostic network model is determined as the embedded target diagnostic network model.

[0157] S44. If the running state categories are inconsistent, optimize the model parameters of the embedded initial diagnostic network model, and jump to the step of inputting the target sample Bowditch curve of the training sample into the input layer until the running state category corresponding to the target feature map is consistent with the preset running state category, and then execute the next step.

[0158] It should be noted that the preset operating status categories are operating status categories set according to actual conditions or manually.

[0159] In practice, the running state category corresponding to each target feature map is compared with the preset running state category. For example, if one target feature map corresponds to the running state category "Outer Encapsulation Fault - Minor Fault," which matches the preset fault state "Outer Encapsulation Fault - Minor Fault," and another target feature map corresponds to the running state category "Outer Encapsulation Fault - Moderate Fault," which also matches the preset fault state "Outer Encapsulation Fault - Moderate Fault," training ends if the running states of most or all target feature maps match the preset running state categories. The Bowditch curves of the target samples from the test set are then input into the current diagnostic network model.

[0160] The model's classification accuracy is calculated using the formula "number of correctly classified samples / total number of test samples." To evaluate a model's performance, the classification accuracy is measured. If the accuracy exceeds a pre-defined threshold (e.g., above 85%), the model is considered to have good performance; if it falls below 85%, the model is considered to have poor performance. Therefore, when the model's classification accuracy reaches 85%, the current MobileNetV2 intelligent diagnostic network model is selected as the target MobileNetV2 intelligent diagnostic network model. If the model's classification accuracy falls below 85%, the activation function and other parameters of the MobileNetV2 intelligent diagnostic network model are optimized, and then step S32 is executed.

[0161] Step 205: Calculate the characteristic parameters of the target Bowditch curve.

[0162] Optionally, step 205 includes the following step S51:

[0163] S51. Extract the characteristic parameters of the target Bowditch curve of the reactor under test; among which, the characteristic parameters include the tilt angle, major axis and minor axis;

[0164] The formula for calculating the tilt angle is:

[0165] The formula for extracting the major axis is:

[0166] The formula for extracting the minor axis is:

[0167] In the formula, a is the major axis, b is the minor axis, θ is the inclination angle, and U c I represents the voltage amplitude. c φ is the current amplitude. u For voltage phase, φ i This represents the current phase.

[0168] It should be noted that the tilt angle, major axis, and minor axis of the target Bowditch curve are calculated using the following formula.

[0169] The formula for calculating the tilt angle is:

[0170] The formula for extracting the major axis is:

[0171] The formula for extracting the minor axis is:

[0172] In the formula, a is the major axis, b is the minor axis, θ is the inclination angle, and U c I represents the voltage amplitude. c φ is the current amplitude. uThe initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

[0173] Step 206: Input the target Bowditch curve into the preset embedded target diagnostic network model and extract the feature map of the target Bowditch curve.

[0174] It should be noted that the embedded target diagnostic network model can extract the feature map of the target Bowditch curve.

[0175] In practice, the target Bowditch curve is input into a trained and tested target diagnostic network model (i.e., the MobileNetV2 intelligent diagnostic network model) to extract the feature map of the target Bowditch curve.

[0176] Step 207: Use the feature parameters and feature map of the target Bowditch curve to generate feature information.

[0177] In this embodiment of the invention, the specific implementation of step 207 is similar to that of step 104, and will not be described again here.

[0178] Step 208: Compare whether the feature information is consistent with the preset feature information, and determine the fault state of the reactor under test based on the comparison result.

[0179] Optionally, step 208 includes the following steps S61-S64:

[0180] S61. Compare whether the characteristic parameters and characteristic map corresponding to the target Bowditch curve are consistent with the characteristic parameters and characteristic map of the Bowditch curve under the preset healthy state.

[0181] S62. If the characteristic parameters and characteristic diagrams are consistent, then it is determined that the reactor under test is not faulty.

[0182] S63. If the characteristic parameters and characteristic map are inconsistent, determine whether the characteristic parameters and characteristic map are consistent with the characteristic parameters and characteristic map of the Bowditch curve under the preset fault state.

[0183] S64. If the characteristic parameters and characteristic diagrams are consistent, the current fault state of the reactor under test shall be determined according to the fault state corresponding to the consistency of the characteristic parameters and characteristic diagrams.

[0184] It should be noted that an intelligent diagnostic system for winding faults in dry-type air-core reactors was built, and the MobileNetV2 intelligent diagnostic network model was embedded. The main functions of this system include: real-time display of voltage and current waveforms measured by the transformers; real-time display of Bowditch curve comparison charts and characteristic parameters plotted from data in the sample set and measurement data of the reactor under test; the intelligent diagnostic system for the winding status of dry-type air-core reactors also includes functions such as database import and data saving. Based on the comparison of system interface images and the inversion results of intelligent algorithms, the detection and diagnosis of inter-turn short-circuit faults in dry-type air-core reactors are achieved.

[0185] In practical implementation, as shown in Figure 5, the fault diagnosis system stores fingerprint data, which refers to the characteristic map and characteristic parameter data of the reactor under healthy conditions. After collecting the reactor voltage and current under operating conditions, drawing its characteristic map, and calculating its characteristic parameters, the data is compared and displayed on the waveform graph with the fingerprint data. For example, red represents the "fingerprint image," while orange represents the "curve image drawn after acquisition." If there is no fault, the images are almost identical, and their characteristic parameters are approximately equal. If there is a fault, the images and characteristic parameters change significantly. At the same time, the drawn images are fed into the trained MobileNetV2 intelligent diagnosis network model and compared with the characteristic maps under multiple fault conditions. If there is no fault, the fault diagnosis section displays "no fault." If a fault occurs, fault information such as "outer encapsulation fault_minor fault" is displayed.

[0186] Example 3

[0187] Please refer to Figure 6, which is a structural block diagram of a reactor inter-turn short-circuit fault diagnosis system provided in Embodiment 3 of the present invention.

[0188] This invention provides a reactor inter-turn short-circuit fault diagnosis system, comprising:

[0189] The plotting module 601 is used to plot the target Bowditch curve of the reactor under test using the operating data of the reactor under test.

[0190] Calculation module 602 is used to calculate the characteristic parameters of the target Bowditch curve;

[0191] Feature map module 603 is used to input the target Bowditch curve into a preset target diagnostic network model and extract the feature map of the target Bowditch curve.

[0192] Feature information module 604 is used to generate feature information by using the feature parameters and feature map of the target Bowditch curve;

[0193] The comparison module 605 is used to compare whether the feature information is consistent with the preset feature information, and to determine the fault state of the reactor under test based on the comparison result.

[0194] Optionally, the drawing module 601 includes:

[0195] The data acquisition submodule is used to acquire the operating data of the reactor under test; the operating data includes current data and voltage data.

[0196] The reconstruction submodule is used to extract the power frequency components corresponding to the voltage and current data, and to reconstruct the voltage and current signals using the power frequency components.

[0197] The plotting submodule is used to plot the initial Bowditch curve of the reactor under test using the reconstructed voltage signal and the reconstructed current signal;

[0198] The processing submodule is used to preprocess the initial Bowditch curve to generate the target Bowditch curve.

[0199] Optionally, the processing submodule includes:

[0200] The noise reduction submodule is used to perform image denoising on the initial Bowditch curve based on the Gaussian noise reduction method.

[0201] The enhancement submodule is used to perform image enhancement processing on the denoised Bowditch curve using a local sharpening method;

[0202] The shrinking submodule is used to sequentially perform grayscale conversion, image rotation, image stretching, and image shrinking on the enhanced Bowditch curve to generate the target Bowditch curve.

[0203] Optionally, it also includes:

[0204] The acquisition submodule is used to acquire normal operating data of reactors in a healthy state and fault operating data of reactors in multiple different fault states;

[0205] A submodule is built to construct a sample set using normal operation data and fault operation data.

[0206] The Bowditch curve submodule is used to plot the initial sample Bowditch curve corresponding to each set of running data using the running data in the sample set.

[0207] The classification submodule is used to classify the initial sample Bowditch curves according to the running status of each running data in the sample set, and add status labels to the initial sample Bowditch curves of each running status category based on the classification results, and generate updated sample Bowditch curves.

[0208] The image preprocessing submodule is used to preprocess the updated sample Bowditch curves to generate the target sample Bowditch curves.

[0209] The partitioning submodule is used to divide the Bowditch curves of each target sample into training and testing sets.

[0210] Optionally, it also includes:

[0211] The model construction submodule is used to construct the embedded initial diagnostic network model; wherein, the embedded initial diagnostic network model includes an input layer, a first convolutional layer, a first inverse residual layer and multiple second inverse residual layers, a second convolutional layer, an average pooling layer, a third convolutional layer and an output layer;

[0212] The input layer submodule is used to input the target sample Bowditch curve from the training set into the input layer;

[0213] The first convolutional layer submodule is used to normalize the image size of the target sample Bowditch curve through the input layer and input it into the first convolutional layer;

[0214] The first inverse residual layer submodule is used to perform convolution processing on the size-normalized target sample Bowditch curve through the first convolution layer to generate the first feature map and input it into the first inverse residual layer in sequence.

[0215] The second inverse residual layer submodule is used to perform dimensionality upscaling on the first feature map through the first inverse residual layer to generate a second feature map and then input it into multiple second inverse residual layers in sequence.

[0216] The second convolutional layer submodule is used to perform convolution processing on the second feature map through each second inverted residual layer to generate the third feature map and input it into the second convolutional layer.

[0217] The average pooling layer submodule is used to perform convolution processing on the third feature map through the second convolutional layer to generate the fourth feature map and input it into the average pooling layer.

[0218] The third convolutional layer submodule is used to reduce the dimensionality of the fourth feature map through the average pooling layer, generate the fifth feature map, and input it into the third convolutional layer.

[0219] The output layer submodule is used to perform convolution processing on the fifth feature map through the third convolutional layer to generate multiple sixth feature maps and input them into the output layer.

[0220] The classification submodule is used to classify each sixth feature map through the output layer and generate multiple target feature maps.

[0221] Optionally, it also includes:

[0222] The judgment submodule is used to determine whether the running state category corresponding to each target feature map is consistent with the preset running state category;

[0223] The testing submodule is used to input the target sample Bowditch curve of the test set into the embedded initial diagnostic network model for testing if the running state categories are consistent.

[0224] The target diagnostic network model submodule is used to determine the embedded initial diagnostic network model as the embedded target diagnostic network model when the test results are consistent with the training results.

[0225] The jump-rotor module is used to optimize the model parameters of the embedded initial diagnostic network model if the running state categories are inconsistent, and then jump to the step of inputting the target sample Bowditch curve of the training sample into the input layer until the running state category corresponding to the target feature map is consistent with the preset running state category, and then execute the next step.

[0226] Optionally, the computing module 602 includes:

[0227] The parameter extraction submodule is used to extract the characteristic parameters of the target Bowditch curve of the reactor under test; among which, the characteristic parameters include the tilt angle, major axis and minor axis;

[0228] The formula for calculating the tilt angle is:

[0229] The formula for extracting the major axis is:

[0230] The formula for extracting the minor axis is:

[0231] In the formula, a is the major axis, b is the minor axis, θ is the inclination angle, and U c I represents the voltage amplitude. c φ is the current amplitude. u The initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

[0232] Optionally, the comparison module 605 includes:

[0233] The first comparison submodule is used to compare whether the feature parameters and feature map corresponding to the target Bowditch curve are consistent with the feature parameters and feature map of the Bowditch curve in the preset healthy state.

[0234] The "No Fault" submodule is used to determine that the reactor under test is not faulty if the characteristic parameters and characteristic diagrams are consistent.

[0235] The parameter inconsistency submodule is used to determine whether the feature parameters and feature map are consistent with the feature parameters and feature map of the Bowditch curve under the preset fault state if the feature parameters and feature map are inconsistent.

[0236] The characteristic parameter consistency submodule is used to determine the current fault state of the reactor under test according to the fault state corresponding to the consistency of characteristic parameters and characteristic map if both characteristic parameters and characteristic map are consistent.

[0237] Example 4

[0238] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, it implements the reactor inter-turn short-circuit fault diagnosis method as described in any embodiment of the present invention.

[0239] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0240] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0241] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0242] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0243] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0244] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing inter-turn short-circuit faults in a reactor, characterized in that, include: Using the operating data of the reactor under test, the target Bowditch curve of the reactor under test is plotted. Calculate the characteristic parameters of the target Bowditch curve; The target Bowditch curve is input into a preset embedded target diagnostic network model to extract the feature map of the target Bowditch curve; Feature information is generated using the feature parameters and feature map of the target Bowditch curve; The fault state of the reactor under test is determined based on the comparison result by comparing whether the feature information is consistent with the preset feature information.

2. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 1, characterized in that, The step of plotting the Bowditch curve of the reactor under test using its operating data includes: Collect operating data of the reactor under test; wherein, the operating data includes current data and voltage data; Extract the power frequency components corresponding to the voltage data and the current data respectively, and reconstruct the voltage signal and the current signal using the power frequency components; The initial Bowditch curve of the reactor under test is plotted using the reconstructed voltage signal and the reconstructed current signal. The initial Bowditch curve is preprocessed to generate the target Bowditch curve.

3. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 2, characterized in that, The step of performing image preprocessing on the initial Bowditch curve to generate the target Bowditch curve includes: The initial Bowditch curve is denoised using Gaussian denoising method. A local sharpening method was used to enhance the image of the denoised Bowditch curve. The enhanced Bowditch curve is subjected to grayscale conversion, image rotation, image stretching, and image shrinking in sequence to generate the target Bowditch curve.

4. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 1, characterized in that, Also includes: Acquire normal operating data of reactors in a healthy state and fault operating data of reactors under multiple different fault states; A sample set is constructed using the normal operation data and the fault operation data. Using the running data from the sample set, plot the initial sample Bowditch curve corresponding to each of the running data sets; The initial sample Bowditch curves are classified according to the running status of each running data in the sample set, and status labels are added to the initial sample Bowditch curves of each running status category according to the classification results to generate updated sample Bowditch curves. Image preprocessing is performed on the updated sample Bowditch curves to generate the target sample Bowditch curves; The Bowditch curves of each target sample are divided into a training set and a test set.

5. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 4, characterized in that, Also includes: Construct an embedded initial diagnostic network model; wherein the embedded initial diagnostic network model includes an input layer, a first convolutional layer, a first inverse residual layer and multiple second inverse residual layers, second convolutional layers, average pooling layers, a third convolutional layer and an output layer; The target sample Bowditch curve of the training set is input into the input layer; The image size of the target sample Bowditch curve is normalized through the input layer and then input into the first convolutional layer; The first convolutional layer performs convolution processing on the size-normalized Bowditch curve of the target sample to generate a first feature map, which is then sequentially input into the first inverse residual layer. The first feature map is up-dimensionalized by the first inverse residual layer to generate a second feature map, which is then sequentially input into multiple second inverse residual layers. The second feature map is convolved by each of the second inverted residual layers to generate a third feature map, which is then input into the second convolutional layer. The third feature map is convolved by the second convolutional layer to generate a fourth feature map, which is then input into the average pooling layer. The fourth feature map is reduced in dimensionality by the average pooling layer to generate a fifth feature map, which is then input into the third convolutional layer. The fifth feature map is convolved by the third convolutional layer to generate multiple sixth feature maps, which are then input into the output layer. The output layer classifies each of the sixth feature maps to generate multiple target feature maps.

6. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 5, characterized in that, Also includes: Determine whether the running state category corresponding to each target feature map is consistent with the preset running state category; If the running status categories are consistent, the target sample Bowditch curve of the test set is input into the embedded initial diagnostic network model for testing; When the test results are consistent with the training results, the embedded initial diagnostic network model is determined as the embedded target diagnostic network model. If the running state categories are inconsistent, the model parameters of the embedded initial diagnostic network model are optimized, and the process jumps to the step of inputting the target sample Bowditch curve of the training sample into the input layer until the running state category corresponding to the target feature map is consistent with the preset running state category, and then the next step is executed.

7. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 1, characterized in that, The step of calculating the characteristic parameters of the target Bowditch curve includes: Extract the characteristic parameters of the target Bowditch curve of the reactor under test; wherein, the characteristic parameters include the tilt angle, major axis and minor axis; The formula for calculating the tilt angle is as follows: The formula for extracting the major axis is: The formula for extracting the minor axis is: In the formula, a is the major axis, b is the minor axis, θ is the inclination angle, and U c I represents the voltage amplitude. c φ is the current amplitude. u The initial phase of the voltage signal, φ i This represents the initial phase of the current signal.

8. The method for diagnosing inter-turn short-circuit faults in reactors according to claim 7, characterized in that, The step of comparing whether the feature information is consistent with preset feature information and determining the fault state of the reactor under test based on the comparison result includes: Compare whether the feature parameters and feature map corresponding to the target Bowditch curve are consistent with the feature parameters and feature map of the Bowditch curve under the preset healthy state; If the characteristic parameters and characteristic diagrams are consistent, then it is determined that the reactor under test is not faulty. If the feature parameters and feature map are inconsistent, then determine whether the feature parameters and feature map are consistent with the feature parameters and feature map of the Bowditch curve under the preset fault state, respectively. If the characteristic parameters and the characteristic map are consistent, then the current fault state of the reactor under test is determined according to the fault state corresponding to the consistency of the characteristic parameters and the characteristic map.

9. A reactor inter-turn short-circuit fault diagnosis system, characterized in that, include: The plotting module is used to plot the target Bowditch curve of the reactor under test using the operating data of the reactor under test. The calculation module is used to calculate the characteristic parameters of the target Bowditch curve; The feature map module is used to input the target Bowditch curve into a preset embedded target diagnostic network model and extract the feature map of the target Bowditch curve. The feature information module is used to generate feature information using the feature parameters and feature map of the target Bowditch curve; The comparison module is used to compare whether the feature information is consistent with the preset feature information, and to determine the fault state of the reactor under test based on the comparison result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the reactor inter-turn short-circuit fault diagnosis method as described in any one of claims 1-8.

Images