Partial discharge detection method and device, computer equipment and readable storage medium

Abstract

The invention relates to a partial discharge detection method and device, computer equipment and a readable storage medium. The method comprises the steps of collecting an image of a to-be-detected article, a partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field intensity, converting the partial discharge ultrasonic signal into a Mel spectrogram and an inverse Mel spectrogram so as to obtain a fusion spectrogram, and performing weighted calculation on feature vectors of the fusion spectrogram through a pre-trained deep learning network so as to obtain the maximum electric field intensity of the to-be-detected article. Obtaining a category result of the partial discharge ultrasonic signal, positioning the partial discharge ultrasonic signal to obtain a positioning result of the partial discharge position of the to-be-detected article, and carrying out image fusion on the maximum electric field intensity, the category result, the positioning result and the to-be-detected article image to obtain the partial discharge position of the to-be-detected article. And obtaining a target image containing the partial discharge type, the partial discharge position and the partial discharge size. By adopting the method, the speed and accuracy of partial discharge detection can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of partial discharge detection, in particular to a partial discharge detection method and device, computer equipment and a readable storage medium. BACKGROUND

[0002] In the operation of the power system, partial discharge is a common precursor phenomenon of faults when the internal insulation of the power equipment is deteriorated or defects exist. Effective detection of the partial discharge is a key technical means for evaluating the insulation state of the power equipment and preventing major faults, and has become an important monitoring method.

[0003] However, the occurrence and evolution of the partial discharge phenomenon are influenced by multiple factors such as electric field distribution, dielectric material characteristics, and equipment structure; meanwhile, the physical form of the partial discharge is complex and diverse, including air gap discharge, corona discharge, surface discharge, and floating potential discharge, each type having unique generation mechanism and discharge characteristics. In addition, the signal form excited by the partial discharge also has significant diversity, covering electromagnetic waves, sound waves, light waves, and other physical quantities. More importantly, the actual power equipment operating environment is usually extremely complex, and factors such as strong electromagnetic interference, temperature fluctuation, and humidity change seriously interfere with the detection of the partial discharge signal. The complexity of the above discharge influencing factors, the diversity of the discharge types, the diversity of the signal forms, and the high interference of the field environment together constitute the complex background faced by the partial discharge detection, making it extremely difficult to accurately identify the partial discharge category and precisely locate the partial discharge position.

[0004] The traditional partial discharge detection method is often difficult to balance efficiency, flexibility, and accuracy when dealing with such highly complex and strongly interfered actual scenarios, and is difficult to meet the needs of modern smart grids for lean control of the state of the power equipment. Therefore, how to use advanced technologies such as deep learning to effectively overcome the complex background interference, accurately identify the partial discharge category, and precisely locate the partial discharge position, so as to significantly reduce the operation and maintenance cost and risk of the power equipment, has become a key technical problem to be solved by the technical personnel in the field. SUMMARY

[0005] Therefore, it is necessary to provide a partial discharge detection method, device, computer equipment, computer readable storage medium, and computer program product capable of accurately identifying the partial discharge category and position of the power equipment in a complex scenario.

[0006] In a first aspect, the present application provides a partial discharge detection method, comprising:

[0007] acquiring an image of an object to be detected, a partial discharge ultrasonic signal of the object to be detected, and a maximum electric field strength;

[0008] convert the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram;

[0009] fuse the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram;

[0010] perform weighted calculation on a feature vector of the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal;

[0011] perform positioning on the partial discharge ultrasonic signal to obtain a positioning result of a partial discharge position of the to-be-detected article;

[0012] perform image fusion on the maximum electric field intensity, the category result, the positioning result, and the to-be-detected article image to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0013] In one of the embodiments, the fusing the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram comprises:

[0014] converting the mel spectrogram into a first feature matrix of a single channel and converting the inverse mel spectrogram into a second feature matrix of a double channel;

[0015] obtaining first channel features and second channel features according to the first feature matrix and the second feature matrix;

[0016] obtaining third channel features according to a specified proportion of first features in the first feature matrix and a specified proportion of second features in the second feature matrix;

[0017] fusing the first channel features, the second channel features, and the third channel features to obtain the fused spectrogram.

[0018] In one of the embodiments, the training method of the pre-trained deep learning network comprises:

[0019] obtaining an initial deep learning network;

[0020] performing average pooling on time series of each feature dimension of a feature map of the fused spectrogram through an average pooling layer of the initial deep learning network to obtain an average value feature matrix;

[0021] performing median pooling on time series of each feature dimension of the feature map of the fused spectrogram through a median pooling layer of the initial deep learning network to obtain a median value feature matrix;

[0022] maximizing the time sequence of each feature dimension of the feature map of the fusion spectrogram through a max-pooling layer of the initial deep learning network to obtain a maximum feature matrix;

[0023] randomly weighting the average feature matrix, the median feature matrix and the maximum feature matrix to obtain trainable parameters of a target pooling layer;

[0024] fusing the target pooling layer into the initial deep learning network for training to obtain the pre-trained deep learning network.

[0025] In one of the embodiments, the collecting of the image of the to-be-detected article, the partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field strength comprises:

[0026] collecting the image of the to-be-detected article through a camera with a fixed position;

[0027] collecting the partial discharge ultrasonic signal of the to-be-detected article through at least two ultrasonic sensors;

[0028] collecting the maximum electric field strength of the to-be-detected article through an electric field meter.

[0029] In one of the embodiments, the positioning of the partial discharge ultrasonic signal to obtain the positioning result of the partial discharge position of the to-be-detected article comprises:

[0030] determining the spatial coordinates of the partial discharge position based on the spatial coordinates of the at least two ultrasonic sensors, the propagation speed of the partial discharge ultrasonic signal and the propagation time of the partial discharge ultrasonic signal;

[0031] converting the spatial coordinates of the partial discharge position into spatial coordinates in a camera coordinate system of the camera;

[0032] projecting the spatial coordinates in the camera coordinate system to the image of the to-be-detected article to obtain the positioning result of the partial discharge position in the target image; the positioning result comprises the coordinates of the partial discharge position in the target image.

[0033] In one of the embodiments, the image fusion of the maximum electric field strength, the category result, the positioning result and the image of the to-be-detected article to obtain a target image containing the partial discharge category, the partial discharge position and the partial discharge size comprises:

[0034] determining the actual partial discharge radius based on the maximum electric field strength and a preset empirical coefficient;

[0035] determine a partial discharge radius in the target image based on the actual partial discharge radius, a camera focal length, and a distance between the camera and the object to be detected;

[0036] The partial discharge position is the center of the circle, the partial discharge radius is the range of the simulated partial discharge, the ultrasonic wave of the partial discharge is presented in a visual form on the target image, and the partial discharge category and the partial discharge size are displayed on the target image.

[0037] In a second aspect, the present application also provides a partial discharge detection device, comprising:

[0038] The acquisition module is configured to acquire an object image, a partial discharge ultrasonic signal of the object to be detected, and a maximum electric field intensity.

[0039] The conversion module is configured to convert the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram.

[0040] The fusion module is configured to fuse the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram.

[0041] The recognition module is configured to perform weighted calculation on a feature vector of the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0042] The positioning module is configured to position the partial discharge ultrasonic signal to obtain a positioning result of a partial discharge position of the object to be detected.

[0043] The fusion module is further configured to perform image fusion on the maximum electric field intensity, the category result, the positioning result, and the object image to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0044] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0045] acquire an object image, a partial discharge ultrasonic signal of the object to be detected, and a maximum electric field intensity.

[0046] convert the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram.

[0047] fuse the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram.

[0048] The feature vector of the fusion spectrogram is calculated by a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0049] The partial discharge ultrasonic signal is positioned to obtain a positioning result of the partial discharge position of the to-be-detected article.

[0050] The maximum electric field intensity, the category result, the positioning result, and the to-be-detected article image are image fused to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0051] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:

[0052] The to-be-detected article image, the partial discharge ultrasonic signal of the to-be-detected article, and the maximum electric field intensity are collected.

[0053] The partial discharge ultrasonic signal is converted into a mel spectrogram and an inverse mel spectrogram.

[0054] The mel spectrogram and the inverse mel spectrogram are fused to obtain a fusion spectrogram.

[0055] The feature vector of the fusion spectrogram is calculated by a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0056] The partial discharge ultrasonic signal is positioned to obtain a positioning result of the partial discharge position of the to-be-detected article.

[0057] The maximum electric field intensity, the category result, the positioning result, and the to-be-detected article image are image fused to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0058] In a fifth aspect, the present application further provides a computer program product comprising a computer program, the computer program being executed by a processor to implement the following steps:

[0059] The to-be-detected article image, the partial discharge ultrasonic signal of the to-be-detected article, and the maximum electric field intensity are collected.

[0060] The partial discharge ultrasonic signal is converted into a mel spectrogram and an inverse mel spectrogram.

[0061] The mel spectrogram and the inverse mel spectrogram are fused to obtain a fusion spectrogram.

[0062] The feature vector of the fusion spectrogram is calculated by a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0063] The partial discharge ultrasonic signal is positioned to obtain a positioning result of the partial discharge position of the to-be-detected article.

[0064] The maximum electric field intensity, the category result, the positioning result, and the to-be-detected article image are image fused to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0065] The partial discharge detection method, device, computer equipment, computer readable storage medium, and computer program product described above, by collecting a to-be-detected article image, a partial discharge ultrasonic signal of the to-be-detected article, and a maximum electric field intensity, converting the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram, then fusing the mel spectrogram and the inverse mel spectrogram to obtain a fusion spectrogram, identifying the fusion spectrogram by a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal, positioning the partial discharge ultrasonic signal to obtain a positioning result of the partial discharge position, and fusing the partial discharge category, the partial discharge position, and the partial discharge size into the to-be-detected article image, realize partial discharge detection of power equipment in a complex background, reduce the cost of partial discharge detection, and use a mel spectrogram with strong feature representation and high resolution to significantly improve the speed and accuracy of partial discharge detection. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0067] Figure 1 An application environment diagram of the partial discharge detection method in an embodiment;

[0068] Figure 2 A flowchart of the partial discharge detection method in an embodiment;

[0069] Figure 3 A flowchart of the spectrogram fusion method in an embodiment;

[0070] Figure 4 A flowchart of the training method of the deep learning network in an embodiment;

[0071] Figure 5A flowchart of a method for determining a target image in an embodiment;

[0072] Figure 6 A block diagram of a partial discharge detection device in an embodiment;

[0073] Figure 7 An internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0074] For the purpose, technical solutions and advantages of the present application to be clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0075] It should be noted that the terms "first", "second" and the like used in the present application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "include" and "have" used in the present application and any variations thereof are intended to cover non-exclusive inclusion. The term "a plurality of" used in the present application means two or more. The term "and / or" used in the present application means one of the options or any combination of multiple options.

[0076] The partial discharge detection method provided by the embodiments of the present application can be applied to the application environment as shown in Figure 1 , and Figure 1 An application environment diagram of a partial discharge detection method in an embodiment. The terminal 102 communicates with the ultrasonic sensor, the electric field meter and the camera through the network in a wired or wireless manner. The terminal 102 stores the data to be processed through the data storage system. The terminal 102 can be a personal computer, a notebook computer, a smart phone, a tablet computer, a drone and an Internet of Things device, etc. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. Specifically, the camera is used to collect the image of the to-be-detected object, the ultrasonic sensor is used to collect the partial discharge ultrasonic signal of the to-be-detected object, and the electric field meter is used to collect the maximum electric field intensity of the to-be-detected object. Then the image of the to-be-detected object, the partial discharge ultrasonic signal and the maximum electric field intensity data are sent to the terminal 102, and the terminal 102 processes these data. The to-be-detected object can be, but is not limited to, a power device such as a transformer, a power cable, an insulator, a switch cabinet, etc.

[0077] In an exemplary embodiment, as shown in Figure 2 , a partial discharge detection method is provided, which is applied to a terminal as shown in Figure 1 for example, including the following steps 202 to 212. Wherein:

[0078] Step 202, collect the image of the to-be-detected article, the partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field intensity.

[0079] Wherein, the to-be-detected article can be, but is not limited to, power equipment, which can include transformers, gas insulated switchgear (GIS) combined electrical appliances, power cables, insulators, switch cabinets, etc.; can also be wind turbine windings, photovoltaic inverter power modules and other new energy field equipment; can also be high-voltage motor stators, frequency converter insulation grids and other industrial equipment.

[0080] The image of the to-be-detected article is an image including the to-be-detected article, which can be, but is not limited to: a visible light image, such as using a camera equipped with a high-definition camera to take an image of the to-be-detected article; an infrared image, which captures the overheating area of the to-be-detected article by a thermal imager; an ultraviolet image, which is used to record the ultraviolet light spots generated by the surface corona discharge of the to-be-detected article.

[0081] The partial discharge ultrasonic signal is an acoustic emission signal generated by the partial discharge of the to-be-detected article, which is collected by an ultrasonic sensor (frequency band 40 kHz (kilohertz)-300 kHz), and the signal reflects the discharge intensity and type characteristics; for example, an ultrasonic sensor is used to receive 150 kHz ultrasonic signals of overhead line corona discharge.

[0082] The maximum electric field intensity is obtained by an electric field meter or an electric field probe to obtain the peak electric field distribution of the to-be-detected article; for example, a portable electric field meter (such as an accuracy of ±1 kV / m (kilovolt / meter)) is used to detect the surface electric field of the cable terminal, and the peak value of 80 kV / m is recorded.

[0083] In one embodiment, a high-voltage switch cabinet (to-be-detected article) can be taken by a camera equipped with a high-definition camera to obtain a high-voltage switch cabinet image; an ultrasonic sensor is attached to the side wall of the switch cabinet, and is aligned with the direction of the infrared abnormal point to detect the pulse partial discharge ultrasonic signal; and the electric field meter is inserted into the switch cabinet to scan the surface of the insulating partition in the cabinet, and the peak value of the electric field intensity is measured at the infrared abnormal point to detect the partial discharge fault of the insulating partition in the cabinet.

[0084] In the above embodiment, the to-be-detected article image is obtained by a high-definition camera, the time-frequency characteristics of the partial discharge ultrasonic signal are obtained by an ultrasonic sensor, and the peak value of the electric field intensity is obtained by an electric field meter. The three-dimensional data fusion of the to-be-detected article image, the partial discharge ultrasonic signal and the peak value of the electric field intensity can accurately control the positioning error of the partial discharge source, and provide a data basis for the subsequent steps, and solve the positioning difficulty problem in complex environment.

[0085] Step 204, converting the partial discharge ultrasonic signal into a Mel spectrogram and an inverse Mel spectrogram.

[0086] The Mel spectrogram is a time-frequency representation method based on the human auditory perception characteristics. The physical frequency (Hz) is converted into a Mel scale by simulating the nonlinear response of the human auditory system to frequency, and the energy distribution characteristics of the signal are extracted.

[0087] The inverse Mel spectrogram (IMel) is a time-frequency representation method for high-frequency physical details. By inversely reconstructing the high-frequency information compressed in the Mel scale, the problem of detail loss in the ultrahigh frequency band (> 150 kHz) of the traditional Mel spectrogram is solved.

[0088] In one embodiment, the partial discharge ultrasonic signal can be pre-processed, framed, windowed, fast Fourier transformed (FFT), and Mel / IMel filtered, etc. to obtain the Mel spectrogram and the IMel spectrogram. Exemplarily, in the partial discharge ultrasonic signal processing, the original ultrasonic signal is first pre-processed, for example, a first-order high-pass filter H(z) = 1-0.97z -1 is used to compensate for the attenuation of high-frequency components in propagation; then the partial discharge ultrasonic signal is framed, for example, the frame length is 1024 points, the frame shift is 256 points, covering a 40 ms (millisecond) pulse, and a Hanning window is applied to each frame to reduce spectral leakage; then the windowed partial discharge ultrasonic signal is subjected to fast Fourier transform (FFT, 4096-point zero padding) to obtain a high-resolution linear spectrum. For the Mel spectrogram, the spectrum is passed through a set of 128 triangular Mel filters, with a nonlinear distribution of 40-300 kHz. After calculating the energy of each filter and taking the logarithmic compression, a time-frequency graph highlighting the perceptual characteristics is formed. At the same time, for the inverse Mel spectrogram, the same spectrum is processed using a set of 128 linear octave filters (logarithmic equally spaced distribution of 150-300 kHz) to retain high-frequency physical details (such as a 284 kHz resonance peak). After logarithmic scaling, a complementary time-frequency representation with high time resolution (0.25 ms) is generated. For example, a certain transformer floating discharge signal is processed by this process, which presents a 152 kHz wide peak (2450 Mel) in the Mel spectrogram, and the IMel spectrogram accurately captures the 284 kHz sharp peak (the signal-to-noise ratio is improved compared to the original spectrum).

[0089] The nonlinear conversion formula of the Mel spectrogram is:

[0090]

[0091] The nonlinear conversion formula of the IMel spectrogram is:

[0092]

[0093] wherein M represents a Mel scale, IM represents an IMel scale, and f is a frequency of the partial discharge ultrasonic signal.

[0094] In the above embodiment, by converting the partial discharge ultrasonic signal into a Mel spectrogram and an IMel spectrogram, the cross-scale synergistic enhancement of perceptual features and physical high-frequency details is realized, and a data basis is provided for subsequent deep learning networks, thereby improving the accuracy of partial discharge detection.

[0095] Step 206, the Mel spectrogram and the IMel spectrogram are fused to obtain a fused spectrogram.

[0096] wherein the fused spectrogram is a three-dimensional tensor fused by channel stacking and attention weighting of the Mel spectrogram and the IMel spectrogram, including time, frequency and channel three dimensions.

[0097] Exemplarily, the Mel spectrogram is converted into a first feature matrix, the IMel spectrogram is converted into a second feature matrix, and a new n-dimensional matrix is obtained according to part of the features in the first feature matrix and part of the features in the second feature matrix; the first feature matrix, the second feature matrix and the new n-dimensional matrix are fused to obtain the fused spectrogram.

[0098] In the above embodiment, by adaptively weighting the Mel spectrogram and the IMel spectrogram, the inherent defects of the contradiction between time-frequency resolution in traditional single spectrogram representation are solved, on the one hand, the perceptual enhancement characteristics of the Mel scale to the discharge fundamental harmonic are inherited, and on the other hand, the high-frequency physical details captured by the IMel scale are retained, solving the technical problem of insufficient feature extraction caused by the diversity of partial discharge ultrasonic signals.

[0099] Step 208, the feature vector of the fused spectrogram is calculated by a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0100] wherein the pre-trained deep learning network is a SEResNet (Squeeze-and-Excitation Residual Network) model architecture pre-trained based on a large number of partial discharge ultrasonic signal samples, used for calculating the feature vector of the fused spectrogram and outputting the category result of the partial discharge ultrasonic signal. The feature vector refers to a fixed-length high-dimensional numerical array extracted from the pre-trained deep learning network after the fused spectrogram is input into the network. The category result can be the type of partial discharge and the probability corresponding to each type; for example, the type of partial discharge can be air gap discharge, corona discharge, surface discharge, suspended potential discharge and the like.

[0101] Exemplarily, the ultrasonic sensor is attached to the side wall of the switch cabinet, a pulse partial discharge ultrasonic signal is detected, a Mel spectrogram and an IMel spectrogram are obtained based on the partial discharge ultrasonic signal, and then a fusion spectrogram is obtained, the fusion spectrogram is identified through a pre-trained SEResNet model, and the probability that the partial discharge type of the switch cabinet is surface discharge is 95%, and the probability that the partial discharge type of the switch cabinet is corona discharge is 5%.

[0102] In the above embodiment, the feature vector of the fusion spectrogram is weighted and calculated through the pre-trained deep learning network, the accurate and robust classification of the partial discharge type is realized, and the technical problem that the complex partial discharge type is difficult to accurately distinguish is solved.

[0103] In step 210, the partial discharge ultrasonic signal is located to obtain a positioning result of the partial discharge position of the to-be-detected object.

[0104] The positioning result represents the partial discharge position of the to-be-detected object, and can be the spatial coordinates of the partial discharge position of the to-be-detected object. The partial discharge position is the region of the partial discharge of the to-be-detected object.

[0105] Exemplarily, the partial discharge ultrasonic signal emitted by the to-be-detected object can be collected by at least two ultrasonic sensors, the spatial coordinates and mutual distance of the two ultrasonic sensors are known, and the spatial coordinates of the partial discharge position can be determined according to the spatial coordinates of the two ultrasonic sensors, the propagation speed and propagation time of the partial discharge ultrasonic signal.

[0106] In the above embodiment, the spatial coordinates of the discharge position are accurately calculated, the accurate positioning of the partial discharge position of the hidden danger power equipment is realized, the breakdown accident caused by the continuous development of the partial discharge is effectively prevented, the reliability of the power equipment is improved, and the operation and maintenance cost is reduced.

[0107] In step 212, the maximum electric field intensity, the category result, the positioning result, and the to-be-detected object image are image fused to obtain a target image containing the partial discharge category, the partial discharge position, and the partial discharge size.

[0108] The partial discharge category is the type of the partial discharge, for example, the type of the partial discharge can be air gap discharge, corona discharge, surface discharge, suspended potential discharge, and the like. The partial discharge size is represented by the maximum electric field intensity, and the physical meaning is that it directly reflects the severity of the insulation deterioration of the partial discharge region. For example, when the air gap discharge inside the transformer is detected, if the target image after fusion shows that the maximum electric field intensity at a certain position reaches 15 kV / mm (kilovolt / millimeter), which is much higher than the withstand limit of insulating oil 8 kV / mm, it can be determined that there is a high-risk partial discharge at the position; if the electric field intensity of the GIS cavity suspended potential discharge region is 6 kV / mm, it indicates that the discharge degree is relatively light.

[0109] Exemplarily, taking high-voltage cable partial discharge detection as an example, the terminal takes the cable infrared image as a substrate, accurately locates the cable partial discharge position in the image, superimposes the discharge type identifier as “creepage discharge”, and visualizes the maximum electric field strength 12 kV / mm of the discharge position with the color spot size and color depth, such as a red large color spot representing a high-risk discharge. For example, the finally generated target image intuitively presents “there is a 12 kV / mm high-risk creepage discharge at the cable terminal flange”.

[0110] In the above embodiment, by fusing the multi-modal data of electric field strength, discharge category, spatial coordinates and device body image, the discharge parameters are displayed on the device real scene image, so that the operation and maintenance personnel can intuitively identify the partial discharge parameters without cross comparison of multiple reports, and the maintenance level of the power equipment is significantly improved.

[0111] The above partial discharge detection method, by collecting the image of the to-be-detected article, the partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field strength, converting the partial discharge ultrasonic signal into a mel spectrum graph and an inverse mel spectrum graph, and then fusing the mel spectrum graph and the inverse mel spectrum graph to obtain a fused spectrum graph, identifying the category result of the partial discharge ultrasonic signal through the pre-trained deep learning network, positioning the partial discharge ultrasonic signal to obtain the positioning result of the partial discharge position, and fusing the partial discharge category, the partial discharge position and the partial discharge size into the image of the to-be-detected article, realizes the partial discharge detection of the power equipment under complex background, reduces the cost of partial discharge detection, and significantly improves the speed and accuracy of partial discharge detection by using the mel spectrum graph with strong feature representation ability and high resolution.

[0112] In an exemplary embodiment, the collection of the image of the to-be-detected article, the partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field strength in step 202 can be: collecting the image of the to-be-detected article by a fixed camera, collecting the partial discharge ultrasonic signal of the to-be-detected article by at least two ultrasonic sensors, and collecting the maximum electric field strength of the to-be-detected article by an electric field meter.

[0113] Exemplarily, a camera equipped with a high-definition camera can be placed around the to-be-detected article to ensure that the to-be-detected article is photographed, and the image of the to-be-detected article is collected through the camera; the ultrasonic sensor can be a high-sensitivity sensor with known spatial coordinates and mutual distance, for example, four ultrasonic sensors are placed around the to-be-detected article, respectively receiving the partial discharge ultrasonic signal emitted by the to-be-detected article, and respectively recording the emission time and reception time of the partial discharge ultrasonic signal; the peak electric field distribution of the to-be-detected article can also be obtained by an electric field meter or an electric field probe.

[0114] In an exemplary embodiment, Figure 3This is a flowchart illustrating a spectrogram fusion method in one embodiment, such as... Figure 3 The spectrogram fusion method shown involves a specific implementation of fusing a Mel spectrogram and an inverse Mel spectrogram to obtain a fused spectrogram. Figure 2 Based on the illustrated embodiment, for any details not described in this embodiment, please refer to [the relevant documentation / reference needed]. Figure 2 The description in the document, applied to this method Figure 1 The following steps, 2061 to 2064, are used as an example to illustrate the application environment.

[0115] in:

[0116] Step 2061: Convert the Mel spectrogram into a single-channel first feature matrix and the inverse Mel spectrogram into a dual-channel second feature matrix.

[0117] For example, the Mel spectrogram is transformed into a first feature matrix A∈R1 for a single channel. n×n×1 The inverse Mel spectrogram is transformed into a two-channel second characteristic matrix B∈R2. n×n×1 Where n is the dimension of the first feature matrix or the dimension of the second feature matrix; R1 is the data structure of the Mel spectrogram, specifically including time, frequency band and number of channels; R2 is the data structure of the IMel spectrogram, specifically including time, frequency band and number of channels.

[0118] In one embodiment, the energy pixel value representing each specific moment and Mel frequency point in the Mel spectrogram can be converted into a 128×128 single-channel matrix as the first feature matrix; the inverse Mel spectrogram can be split into a real and imaginary dual-channel matrix as the second feature matrix.

[0119] Step 2062: Obtain the first channel features and the second channel features based on the first feature matrix and the second feature matrix.

[0120] For example, the first feature matrix and the second feature matrix are used as inputs. A convolution operation is performed on the first feature matrix of a single channel to extract Mel frequency domain features as the first channel features; a complex convolution is performed on the second feature matrix of a dual channel to extract the original frequency domain phase features as the second channel features.

[0121] Step 2063: Obtain the third channel feature based on the first feature in the first feature matrix at a specified proportion and the second feature in the second feature matrix at a specified proportion.

[0122] For example, the specified ratio can be, but is not limited to, 1 / 2. For instance, features from the first feature matrix A and the second feature matrix B are each randomly selected. As the first feature and the second feature respectively, a new n-dimensional matrix is formed according to the first feature and the second feature as a third channel feature.

[0123] In step 2064, the first channel feature, the second channel feature and the third channel feature are fused to obtain a fused spectrogram.

[0124] Exemplarily, the first channel feature, the second channel feature and the third channel feature can be superimposed in weights to obtain the fused spectrogram, for example, the weight ratio can be but is not limited to 0.5:0.3:0.2.

[0125] In the above embodiment, by adaptively weighting and fusing the mel spectrogram and the inverse mel spectrogram, more features in the spectrogram can be retained, and the technical problem of insufficient feature extraction caused by the diversity of partial discharge ultrasonic signals is solved.

[0126] In an exemplary embodiment, Figure 4 The flowchart of the training method of the deep learning network in an embodiment is shown in FIG. 8. Figure 4 The training method of the deep learning network shown in FIG. 8 involves the specific acquisition method of the deep learning network. For the parts not described in detail in the above embodiment, please refer to the description in Figure 2 The training method of the deep learning network shown in FIG. 8 involves the specific acquisition method of the deep learning network. For the parts not described in detail in the above embodiment, please refer to the description in Figure 2 Figure 1 The training method of the deep learning network shown in FIG. 8 involves the specific acquisition method of the deep learning network. For the parts not described in detail in the above embodiment, please refer to the description in

[0127] In step 2081, an initial deep learning network is obtained.

[0128] The initial deep learning network refers to a basic model that has a complete network architecture before training starts, but has not been optimized in parameters by partial discharge ultrasonic signal sample data.

[0129] In an exemplary embodiment, the initial deep learning network can be but is not limited to SEResNet (Squeeze-and-Excitation Residual Network). SEResNet is an improved network structure based on ResNet (Deep Residual Network) by introducing an SE (Squeeze-and-Excitation) module. The core idea is to explicitly model the interdependence between feature channels, adaptively recalibrate the importance of channel features, and significantly improve the representation ability of the network while only increasing a small amount of computational overhead.

[0130] In step 2082, the average value feature matrix is obtained by taking the average value of the time sequence in each feature dimension of the feature map of the fused spectrogram through the average pooling layer of the initial deep learning network. ​

[0131] Exemplarily, the key features in the fusion spectrogram are extracted through the convolutional layer of the initial deep learning network first to obtain a plurality of feature maps, and the time sequence of each feature dimension of each feature map is averaged through the average pooling layer to combine the average values to obtain an average value feature matrix.

[0132] In step 2083, the time sequence of each feature dimension of the feature map of the fusion spectrogram is taken as a median value through the median pooling layer of the initial deep learning network to obtain a median value feature matrix.

[0133] Exemplarily, the key features in the fusion spectrogram are extracted through the convolutional layer of the initial deep learning network first to obtain a plurality of feature maps, and the time sequence of each feature dimension of each feature map is taken as a median value through the median pooling layer to combine the median values to obtain a median value feature matrix.

[0134] In step 2084, the time sequence of each feature dimension of the feature map of the fusion spectrogram is taken as a maximum value through the maximum pooling layer of the initial deep learning network to obtain a maximum value feature matrix.

[0135] Exemplarily, the key features in the fusion spectrogram are extracted through the convolutional layer of the initial deep learning network first to obtain a plurality of feature maps, and the time sequence of each feature dimension of each feature map is taken as a maximum value through the maximum pooling layer to combine the maximum values to obtain a maximum value feature matrix.

[0136] In step 2085, the trainable parameters of the target pooling layer are obtained according to the average value feature matrix, the median value feature matrix and the maximum value feature matrix.

[0137] The target pooling layer is a self-defined deep learning pooling layer, which can generate more robust feature representation by dynamically weighting and fusing multi-time sequence statistical features (average value feature matrix / median value feature matrix / maximum value feature matrix), and the weight parameters of the target pooling layer can be optimized by training. The trainable parameters are parameters to be optimized by parameter optimization of partial discharge ultrasonic signal sample data, i.e., parameters to be optimized.

[0138] Exemplarily, the calculation formula of the target pooling layer includes:

[0139]

[0140] Wherein, TPooling(X) represents the target pooling layer, X represents the time sequence data of the input fusion spectrogram, wherein the dimension of X is TxD, wherein T represents the number of time steps, D represents the feature dimension of each time step, θ represents the random parameter of the target pooling layer between 0 and 1, X t represents the value of X at the tth time step, represents the average value feature matrix, a median value feature matrix, a maximum value feature matrix.

[0141] In step 2086, the target pooling layer is fused into the initial deep learning network for training to obtain a pre-trained deep learning network.

[0142] Exemplarily, the target pooling layer is inserted before the global pooling layer of SEResNet, and the trainable parameters of the target pooling layer are set as trainable. After 10000 groups of GIS partial discharge ultrasonic signal samples are input and 500 rounds of training, the weight optimization is as follows: the weight of the average value feature matrix is 0.15, the weight of the median value feature matrix is 0.25, and the weight of the maximum value feature matrix is 0.60, indicating that the network pays more attention to the maximum value feature, and the discharge pulse peak is sensitive to GIS fault. Then, the pre-trained deep learning network is output.

[0143] In the above embodiment, the fusion of SEResNet and TPooling(X) forms a new model, which can enhance the feature extraction capability of the fusion sonogram formed by the ultrasonic signal in the deep learning process. The new model transmits the extracted fusion sonogram features to the fully connected layer (FC), and finally outputs the partial discharge recognition result of the detected object by the classifier Softmax.

[0144] In an exemplary embodiment, the positioning of the partial discharge ultrasonic signal in step 210 can obtain the positioning result of the partial discharge position of the detected object, which can be: determining the spatial coordinates of the partial discharge position based on the spatial coordinates of at least two ultrasonic sensors, the propagation speed of the partial discharge ultrasonic signal, and the propagation time of the partial discharge ultrasonic signal, then converting the spatial coordinates of the partial discharge position into spatial coordinates in the camera coordinate system of the camera, and finally projecting the spatial coordinates in the camera coordinate system to the detected object image to obtain the positioning result of the partial discharge position in the target image, which includes the coordinates of the partial discharge position in the target image. Exemplarily, the partial discharge ultrasonic signal emitted by the detected object can be collected by four ultrasonic sensors. Assuming that the spatial coordinates of the partial discharge position are O(x, y, z), the spatial coordinates of the four ultrasonic sensors are O1(x1, y1, z1), O2(x2, y2, z2), O3(x3, y3, z3), and O4(x4, y4, z4), the distances from the four ultrasonic sensors to the partial discharge position O of the detected object are L1, L2, L3, and L4, respectively; the time difference of the partial discharge ultrasonic signal emitted by the partial discharge position to the four ultrasonic sensors is T1, T2, T3, and the propagation speed v of the partial discharge ultrasonic signal is constant. The calculation formula of the partial discharge position includes:

[0145]

[0146] By the above formula, the spatial coordinates of the partial discharge position can be obtained as O(x, y, z). Then, the spatial coordinates of the partial discharge signal are converted into spatial coordinates O'(x', y', z') in the camera coordinate system of the fixed point position, and the spatial coordinates O'(x', y', z') in the camera coordinate system are projected onto the image plane P(v, w) of the to-be-detected object, and the projection of the spatial coordinates in the camera coordinate system onto the image of the to-be-detected object is as follows:

[0147] [x', y', z'] T = R[x, y, z] T + U, [v, w, u] T = K[x', y', z'] T

[0148] wherein R is a rotation matrix, U is a translation vector, v and w are used to represent the image plane of the to-be-detected object, u is a scaling factor of the homogeneous coordinates, K is a camera intrinsic parameter matrix, and T is a transpose matrix.

[0149] In an exemplary embodiment, Figure 5 a flowchart of a determination method of a target image in an embodiment is shown in FIG. 12, and a determination method of a target image is shown in FIG. 13, which involves image fusion of the maximum electric field intensity, the category result, the positioning result, and the image of the to-be-detected object to obtain a target image containing the partial discharge category, the partial discharge position, and the partial discharge size. The specific implementation of the determination method of the target image is based on the embodiment shown in FIG. 12, and the description of the embodiment not described in detail can be referred to the description in FIG. 12. The application environment of the method applied in FIG. 12 is taken as an example for description, which includes the following steps 2121 to 2123. Figure 5 Figure 2 Figure 2 Figure 1

[0150] wherein:

[0151] Step 2121, determining the actual partial discharge radius based on the maximum electric field intensity and a preset empirical coefficient.

[0152] wherein the actual partial discharge radius is the radius of the actual discharge area of the partial discharge of the to-be-detected object.

[0153] Exemplarily, the calculation formula of the actual partial discharge radius can include:

[0154]

[0155] wherein R is the actual partial discharge radius, E max ​​​​is the maximum electric field intensity; n and H are preset empirical coefficients, which are usually derived based on prior experience and are commonly used in the industry; for example, for corona discharge: H≈0.6-1.5, n≈1.0-1.6; for surface discharge: H≈0.3-0.7, n≈1.6-2.2; for air gap discharge: H≈0.25-0.6, n≈1.7-2.3; for floating discharge: H≈0.5-1.1, n≈1.4-1.9, and the specific values of the preset empirical coefficients can be determined according to the electric field intensity (5-40 kV / mm).

[0156] In step 2122, the partial discharge radius in the target image is determined based on the actual partial discharge radius, the camera focal length, and the distance between the camera and the to-be-detected object.

[0157] In step 2122, the partial discharge radius in the target image is determined based on the actual partial discharge radius, the camera focal length, and the distance between the camera and the to-be-detected object.

[0158] Exemplarily, the calculation formula of the partial discharge radius in the target image can include:

[0159]

[0160] wherein f is the camera focal length in millimeters; R1 is the partial discharge radius in the target image in pixels, representing the radius of the partial discharge region; and D is the distance between the camera and the to-be-detected object in millimeters.

[0161] In step 2123, the ultrasonic waves of the partial discharge are presented in a visual form on the target image, with the highest value of the frequency of the partial discharge ultrasonic signal as the reference, the partial discharge position as the center, and the partial discharge radius as the range of the simulated partial discharge, and the partial discharge category and the partial discharge size are displayed on the target image.

[0162] Exemplarily, the ultrasonic waves of the partial discharge are presented in a visual form on the target image, with the highest value of the frequency of the partial discharge ultrasonic signal as the reference, the partial discharge position as the center, and the partial discharge radius R1 as the range of the simulated partial discharge, and the partial discharge effect is obtained by simulation, the partial discharge effect, the partial discharge category, the partial discharge position, and the partial discharge size obtained in the foregoing are fused on the image, and the final target image is output.

[0163] In the above embodiment, by taking the highest frequency of the partial discharge ultrasonic signal as the reference, taking the detected discharge position as the center, and accurately defining the simulation discharge influence range with the calculated discharge radius, the ultrasonic signal generated by the partial discharge is intuitively and dynamically superimposed and mapped onto the target image. This visual presentation greatly improves the intuitiveness and efficiency of condition monitoring, enabling maintenance personnel to instantly and accurately locate potential fault points, intuitively assess the severity and spatial distribution range of the discharge, and quickly identify the type of discharge, thereby significantly shortening the fault diagnosis time and significantly improving the early warning capability of equipment hazards. It provides strong technical support for formulating accurate maintenance strategies, preventing catastrophic failures, optimizing equipment operation reliability, and ultimately ensuring the safe and stable operation of power equipment.

[0164] The above partial discharge detection method, by collecting the image of the to-be-detected article, the partial discharge ultrasonic signal of the to-be-detected article and the maximum electric field strength, converting the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram, and then fusing the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram, identifying the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal, positioning the partial discharge ultrasonic signal to obtain a positioning result of the partial discharge position, and fusing the partial discharge category, the partial discharge position and the partial discharge size into the image of the to-be-detected article, realizes the partial discharge detection of the power equipment in a complex background. While reducing the cost of partial discharge detection, the mel spectrogram with strong feature representation ability and high resolution is used to significantly improve the speed and accuracy of partial discharge detection.

[0165] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by combination are within the scope of protection of the present application.

[0166] Based on the same inventive concept, the application further provides a partial discharge detection device for implementing the above-mentioned partial discharge detection method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more partial discharge detection device embodiments provided below can refer to the limitations of the partial discharge detection method described above, which will not be described here again.

[0167] In one exemplary embodiment, as shown in Figure 6 A partial discharge detection device 600 is provided, comprising: an acquisition module 610, a conversion module 620, a fusion module 630, an identification module 640, and a positioning module 650, wherein:

[0168] The acquisition module 610 is configured to acquire an image of an object to be detected, a partial discharge ultrasonic signal of the object to be detected, and a maximum electric field intensity.

[0169] The conversion module 620 is configured to convert the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram.

[0170] The fusion module 630 is configured to fuse the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram.

[0171] The identification module 640 is configured to perform weighted calculation on a feature vector of the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal.

[0172] The positioning module 650 is configured to position the partial discharge ultrasonic signal to obtain a positioning result of a partial discharge position of the object to be detected.

[0173] The fusion module 630 is further configured to perform image fusion on the maximum electric field intensity, the category result, the positioning result, and the image of the object to be detected to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

[0174] In one exemplary embodiment, the fusion module 630 is further configured to convert the mel spectrogram into a first feature matrix of a single channel and convert the inverse mel spectrogram into a second feature matrix of a double channel.

[0175] The first feature matrix and the second feature matrix are used to obtain first channel features and second channel features.

[0176] The first channel features and the second channel features are used to obtain third channel features according to a specified proportion of first features in the first feature matrix and a specified proportion of second features in the second feature matrix.

[0177] The first channel feature, the second channel feature and the third channel feature are fused to obtain the fused spectrogram.

[0178] In an exemplary embodiment, the identification module 640 is further configured to obtain an initial deep learning network;

[0179] The time sequence of each feature dimension of the feature map of the fused spectrogram is averaged by an average pooling layer of the initial deep learning network to obtain an average value feature matrix;

[0180] The time sequence of each feature dimension of the feature map of the fused spectrogram is taken as a median value by a median pooling layer of the initial deep learning network to obtain a median value feature matrix;

[0181] The time sequence of each feature dimension of the feature map of the fused spectrogram is taken as a maximum value by a maximum pooling layer of the initial deep learning network to obtain a maximum value feature matrix;

[0182] Random weighting is performed according to the average value feature matrix, the median value feature matrix and the maximum value feature matrix to obtain trainable parameters of a target pooling layer;

[0183] The target pooling layer is fused into the initial deep learning network for training to obtain the pre-trained deep learning network.

[0184] In an exemplary embodiment, the acquisition module 610 is further configured to acquire the image of the to-be-detected object by a camera with a fixed position;

[0185] The partial discharge ultrasonic signal of the to-be-detected object is acquired by at least two ultrasonic sensors;

[0186] The maximum electric field intensity of the to-be-detected object is acquired by an electric field meter.

[0187] In an exemplary embodiment, the positioning module 650 is further configured to determine the spatial coordinates of the partial discharge position based on the spatial coordinates of the at least two ultrasonic sensors, the propagation speed of the partial discharge ultrasonic signal and the propagation time of the partial discharge ultrasonic signal;

[0188] The spatial coordinates of the partial discharge position are converted into spatial coordinates in a camera coordinate system of the camera;

[0189] The spatial coordinates in the camera coordinate system are projected onto the image of the to-be-detected object to obtain a positioning result of the partial discharge position in the target image; the positioning result includes the coordinates of the partial discharge position in the target image.

[0190] In an exemplary embodiment, the fusion module 630 is further configured to determine an actual partial discharge radius based on the maximum electric field intensity and a preset empirical coefficient;

[0191] determine a partial discharge radius in the target image based on the actual partial discharge radius, a camera focal length, and a distance between the camera and the to-be-detected article;

[0192] take the highest frequency value of the partial discharge ultrasonic signal as a reference, take the partial discharge position as a center, take the partial discharge radius as a range of simulated partial discharge, present the ultrasonic wave of the partial discharge in a visual form on the target image, and display the partial discharge category and the partial discharge size on the target image.

[0193] The partial discharge detection device described above collects the article image, the partial discharge ultrasonic signal of the to-be-detected article, and the maximum electric field intensity through the collection module 610, converts the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram through the conversion module 620, and then fuses the mel spectrogram and the inverse mel spectrogram to obtain a fusion spectrogram through the fusion module 630. The category result of the partial discharge ultrasonic signal is obtained through the recognition module 640 pre-trained deep learning network, the positioning module 650 is used to position the partial discharge ultrasonic signal to obtain the positioning result of the partial discharge position, and the fusion module 630 is used to fuse the partial discharge category, the partial discharge position, and the partial discharge size into the to-be-detected article image. The partial discharge detection of the power equipment in a complex background is realized, the cost of the partial discharge detection is reduced, the mel spectrogram with strong feature representation ability and high resolution is used, and the speed and accuracy of the partial discharge detection are significantly improved.

[0194] Each module in the partial discharge detection device described above can be realized by software, hardware, and a combination thereof in whole or in part. Each module described above can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0195] In an exemplary embodiment, a computer device is provided, which can be a terminal, and an internal structure diagram of the computer device can be as shown in Figure 7The computer device shown in the figure includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, mobile cellular network, near field communication (Near Field Communication, NFC) or other technologies. The computer program is executed by the processor to realize a partial discharge detection method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0196] Those skilled in the art can understand that, Figure 7 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0197] In an exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0198] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0199] In an exemplary embodiment, a computer program product is provided, including a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0200] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0201] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of each method. In the embodiments provided in the present application, any reference to memory, database or other medium can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (Resistive Random Access Memory, ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (Artificial Intelligence, AI) processor, etc., without being limited thereto.

[0202] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, any combination of these technical features is deemed to be within the scope of the present application.

[0203] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A partial discharge detection method characterized by, The method comprises: Collecting an image of an object to be detected, partial discharge ultrasonic signals of the object to be detected, and maximum electric field intensity; Converting the partial discharge ultrasonic signals into a mel spectrogram and an inverse mel spectrogram; Fusing the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram; Performing weighted calculation on a feature vector of the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signals; Positioning the partial discharge ultrasonic signals to obtain a positioning result of a partial discharge position of the object to be detected; Performing image fusion on the maximum electric field intensity, the category result, the positioning result, and the image of the object to be detected to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

2. The method of claim 1, wherein, The method further comprises: Converting the mel spectrogram into a first feature matrix of a single channel and converting the inverse mel spectrogram into a second feature matrix of double channels; Obtaining first channel features and second channel features according to the first feature matrix and the second feature matrix; Obtaining third channel features according to a specified proportion of first features in the first feature matrix and a specified proportion of second features in the second feature matrix; Fusing the first channel features, the second channel features, and the third channel features to obtain the fused spectrogram.

3. The method of claim 1, wherein, The training method of the pre-trained deep learning network comprises: Obtaining an initial deep learning network; Taking average values of time sequences in each feature dimension of a feature map of the fused spectrogram through an average pooling layer of the initial deep learning network to obtain an average value feature matrix; Taking median values of time sequences in each feature dimension of the feature map of the fused spectrogram through a median pooling layer of the initial deep learning network to obtain a median value feature matrix; Taking maximum values of time sequences in each feature dimension of the feature map of the fused spectrogram through a maximum pooling layer of the initial deep learning network to obtain a maximum value feature matrix; Randomly weighting the average value feature matrix, the median value feature matrix, and the maximum value feature matrix to obtain trainable parameters of a target pooling layer; Fusing the target pooling layer into the initial deep learning network for training to obtain the pre-trained deep learning network.

4. The method of claim 1, wherein, The method further comprises: Collecting the image of the object to be detected through a camera with a fixed position; Collecting the partial discharge ultrasonic signals of the object to be detected through at least two ultrasonic sensors; Collecting the maximum electric field intensity of the object to be detected through an electric field meter.

5. The method of claim 4, wherein, The method further comprises: Determining spatial coordinates of the partial discharge position based on spatial coordinates of the at least two ultrasonic sensors, a propagation speed of the partial discharge ultrasonic signals, and a propagation time of the partial discharge ultrasonic signals. convert the spatial coordinates of the partial discharge position into spatial coordinates in a camera coordinate system of the camera; project the spatial coordinates in the camera coordinate system to the image of the to-be-detected object to obtain a positioning result of the partial discharge position in the target image; the positioning result includes coordinates of the partial discharge position in the target image.

6. The method of claim 4, wherein, The image fusion of the maximum electric field intensity, the category result, the positioning result, and the image of the to-be-detected object obtains a target image containing a partial discharge category, a partial discharge position, and a partial discharge size, including: determining an actual partial discharge radius based on the maximum electric field intensity and a preset empirical coefficient; determining a partial discharge radius in the target image based on the actual partial discharge radius, a camera focal length, and a distance between the camera and the to-be-detected object; taking the highest frequency value of the partial discharge ultrasonic signal as a reference, taking the partial discharge position as a circle center, taking the partial discharge radius as a range of simulated partial discharge, presenting the ultrasonic wave of the partial discharge in a visual form on the target image, and displaying the partial discharge category and the partial discharge size on the target image.

7. A partial discharge detection device, characterized by, The device includes: a collection module configured to collect an image of a to-be-detected object, a partial discharge ultrasonic signal of the to-be-detected object, and a maximum electric field intensity; a conversion module configured to convert the partial discharge ultrasonic signal into a mel spectrogram and an inverse mel spectrogram; a fusion module configured to fuse the mel spectrogram and the inverse mel spectrogram to obtain a fused spectrogram; an identification module configured to perform weighted calculation on a feature vector of the fused spectrogram through a pre-trained deep learning network to obtain a category result of the partial discharge ultrasonic signal; a positioning module configured to position the partial discharge ultrasonic signal to obtain a positioning result of a partial discharge position of the to-be-detected object; The fusion module is further configured to perform image fusion on the maximum electric field intensity, the category result, the positioning result, and the image of the to-be-detected object to obtain a target image containing a partial discharge category, a partial discharge position, and a partial discharge size.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 6.

Images