Partial discharge location method, device, storage medium and system for GIS equipment

Abstract

This invention discloses a method, apparatus, storage medium, and system for locating partial discharge in GIS equipment, relating to the field of power equipment technology. The method includes: acquiring partial discharge data from a GIS device, wherein the GIS device is a device associated with a gas-insulated switchgear in a power system; processing the partial discharge data to construct a target dataset; training an initial location model using the target dataset to obtain a target location model, wherein the initial location model and the target location model employ the same neural network architecture, constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and determining the partial discharge location result of the GIS device using real-time operational data of the GIS device and the target location model. This invention solves the technical problem of low accuracy in sensing and locating partial discharge faults in GIS equipment in related technologies.

Description

Technical Field

[0001] This invention relates to the field of power equipment technology, and more particularly to the field of power equipment operation status assessment technology. Specifically, it relates to a method, device, storage medium, and system for locating partial discharge in GIS equipment. Background Technology

[0002] Gas-Insulated Switchgear (GIS) refers to equipment associated with GIS in power systems. GIS equipment typically uses sulfur hexafluoride (SF6) gas as the insulating medium to encapsulate electrical components such as circuit breakers, disconnectors, grounding switches, voltage / current transformers, surge arresters, busbars, bushings, and cable terminals within a grounded metal casing. GIS equipment offers advantages such as small footprint, long maintenance cycles, and low susceptibility to external environmental interference, making it a critical control device widely used in modern power systems. However, defects inevitably exist in GIS equipment during manufacturing, transportation, installation, and use, potentially leading to varying degrees of partial discharge (PD) faults.

[0003] Partial discharge refers to the discharge phenomenon occurring between electrodes but not penetrating them, and it is the most common and typical operational fault in GIS equipment. The cumulative effect of partial discharge leads to the continuous deterioration of the dielectric properties of the insulating material, which may eventually lead to insulation breakdown accidents. Long-term partial discharge can cause local expansion of insulation deterioration in the equipment, and even cause insulation breakdown or surface flashover of combined electrical appliances, seriously affecting the safety and stability of power grid operation. After partial discharge occurs, high-frequency electromagnetic waves will repeatedly propagate within the pipes of GIS equipment, undergoing multiple refractions and reflections, making the propagation process extremely complex. At the same time, GIS equipment itself operates in a complex electromagnetic environment, and partial discharge signals couple and propagate with environmental interference signals, making it very difficult to locate partial discharge signals in GIS equipment. Due to the complexity of the partial discharge mechanism and the numerous causes of faults, current technologies for detecting and locating partial discharge faults in GIS equipment have low accuracy, far from meeting the current safe operation requirements of power systems.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method, apparatus, storage medium, and system for locating partial discharge faults in GIS equipment, thereby at least solving the technical problem of low accuracy in sensing and locating partial discharge faults in GIS in related technologies.

[0006] According to one aspect of the present invention, a method for locating partial discharge in a GIS device is provided, comprising: acquiring partial discharge data of the GIS device, wherein the GIS device is a device associated with a gas-insulated switch in a power system; processing the partial discharge data to construct a target dataset; training an initial location model using the target dataset to obtain a target location model, wherein the initial location model and the target location model adopt the same neural network architecture, the neural network architecture being constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and determining the partial discharge location result of the GIS device using real-time operating data of the GIS device and the target location model.

[0007] Optionally, in the GIS equipment, multiple first positions are determined on the pipe wall according to a first number corresponding to the pipe axis and a second number corresponding to the pipe circumference, and the multiple first positions are used to arrange multiple sensors.

[0008] Optionally, the pipe wall is also provided with multiple second positions corresponding to multiple preset fault categories, and the multiple second positions represent the location where the partial discharge fault occurs.

[0009] Optionally, acquiring partial discharge data from GIS equipment includes: using multiple sensors to acquire partial discharge data, wherein the partial discharge data includes field strength information corresponding to multiple second locations.

[0010] Optionally, data processing of the partial discharge data to construct the target dataset includes: preprocessing the field strength information in the partial discharge data based on the three-dimensional coordinate information of multiple second locations to obtain processed result data, wherein the preprocessing operations include: missing value completion, outlier correction, and data denoising; normalizing the processed result data to establish feature data tables corresponding to multiple second locations; and constructing the target dataset based on the feature data tables, wherein the target dataset includes multiple sets of sample label pairs, the samples in each set of sample label pairs are determined by the field strength information, and the labels in each set of sample label pairs are determined by the three-dimensional coordinate information.

[0011] Optionally, the target dataset includes multiple sets of sample label pairs; training the initial localization model using the target dataset to obtain the target localization model includes: inputting samples from the multiple sets of sample label pairs in the target dataset into the initial localization model to obtain prediction results; calculating the model training loss based on the prediction results and the labels in the multiple sets of sample label pairs; and adjusting the model hyperparameters of the initial localization model according to the model training loss to obtain the target localization model.

[0012] Optionally, the neural network architecture includes a convolutional subnetwork and an attention subnetwork; the samples from multiple sample label pairs in the target dataset are input into the initial localization model to obtain the prediction results, including: extracting features from the samples through the convolutional subnetwork to obtain the extraction results; selecting features from the extraction results through the attention subnetwork to obtain the selection results; and using the target activation function and the selection results to predict the location to obtain the prediction results.

[0013] Optionally, the neural network architecture also includes a genetic sub-network; adjusting the model hyperparameters of the initial localization model according to the model training loss to obtain the target localization model includes: performing hyperparameter optimization calculation on the initial localization model through the genetic sub-network and the model training loss to obtain the target hyperparameters; adjusting the model hyperparameters of the initial localization model according to the target hyperparameters to obtain the target localization model.

[0014] Optionally, the GIS equipment partial discharge location method further includes: using a regularization strategy to correct overfitting of the target location model and updating the target location model, wherein the regularization strategy is determined by preset regularization parameter terms and sparse penalty parameter terms, and the regularization strategy is used to adjust the target activation function in the target location model.

[0015] Optionally, determining the partial discharge location result of the GIS equipment by utilizing the real-time operating data of the GIS equipment and the target positioning model includes: collecting the real-time operating data of the GIS equipment, wherein the real-time operating data includes at least the real-time pipeline field strength data; inputting the real-time operating data into the target positioning model for positioning prediction to obtain the partial discharge location result.

[0016] According to another aspect of the present invention, a partial discharge location device for a GIS device is also provided, comprising: an acquisition module for acquiring partial discharge data of a GIS device, wherein the GIS device is a device associated with a gas-insulated switch in a power system; a processing module for processing the partial discharge data to construct a target dataset; a training module for training an initial location model using the target dataset to obtain a target location model, wherein the initial location model and the target location model adopt the same neural network architecture, and the neural network architecture is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and a location module for determining the partial discharge location result of the GIS device using real-time operating data of the GIS device and the target location model.

[0017] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored program, wherein, when the program is running, it controls the device where the storage medium is located to execute the GIS device partial discharge positioning method described above.

[0018] According to another aspect of the present invention, a partial discharge location system for GIS equipment is also provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the partial discharge location method for GIS equipment described above.

[0019] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the GIS device partial discharge positioning method described above.

[0020] In this embodiment of the invention, partial discharge data of a GIS device is acquired, wherein the GIS device is a device associated with a gas-insulated switch in a power system. The partial discharge data is processed to construct a target dataset. Using the target dataset, an initial positioning model is trained to obtain a target positioning model. The initial positioning model and the target positioning model employ the same neural network architecture, which is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism. Using the real-time operating data of the GIS device and the target positioning model, the partial discharge positioning result of the GIS device is determined. Thus, this invention achieves the goal of training a target positioning model to locate partial discharge in operating GIS devices, thereby improving the accuracy of partial discharge perception and positioning of GIS devices, and solving the problem of low accuracy in perceiving and locating partial discharge faults in GIS devices in related technologies. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0022] Figure 1 This is a hardware structure block diagram of an optional terminal device for a partial discharge location method for GIS equipment according to an embodiment of the present invention;

[0023] Figure 2 This is a flowchart of a partial discharge location method for GIS equipment according to an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram showing the locations of sensors distributed on the pipe wall in an optional GIS device according to an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of the longitudinal section of a pipe cavity in an optional GIS device according to an embodiment of the present invention;

[0026] Figure 5This is a schematic diagram of the cross-section of a pipe section cavity in an optional GIS device according to an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the cross-section of a pipe section cavity in another optional GIS device according to an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of an optional target dataset partitioning method according to an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the training process of an optional target localization model according to an embodiment of the present invention;

[0030] Figure 9 This is a schematic diagram of an optional densely connected convolutional neural network structure according to an embodiment of the present invention;

[0031] Figure 10 This is a schematic diagram of an optional attention subnetwork according to an embodiment of the present invention;

[0032] Figure 11 This is a schematic diagram of the calculation process corresponding to an optional genetic algorithm according to an embodiment of the present invention;

[0033] Figure 12 This is a schematic diagram of the model training process of an optional target localization model according to an embodiment of the present invention;

[0034] Figure 13 This is a schematic diagram of the model iteration process of an optional target localization model according to an embodiment of the present invention;

[0035] Figure 14 This is a structural block diagram of a partial discharge locating device for GIS equipment according to an embodiment of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] According to an embodiment of the present invention, an embodiment of a partial discharge location method for GIS equipment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] Figure 1 This is a hardware structure block diagram of an optional terminal device for a partial discharge location method for GIS equipment according to an embodiment of the present invention, such as... Figure 1 As shown, the terminal device may include one or more processors 102 (processor 102 may include, but is not limited to, a microprocessor (MCU) or a field-programmable gate array (FPGA) processing device), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display device 110, an input / output device 108, a Universal Serial Bus (USB) port (which may be included as one of the ports of a computer bus, not shown in the figure), a network interface (not shown in the figure), a power supply (not shown in the figure), and / or a camera (not shown in the figure). Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal device described above. For example, the terminal device may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0040] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits may be embodied, in whole or in part, as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the terminal device (or mobile device).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the partial discharge location method for GIS equipment in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned partial discharge location method for GIS equipment. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0042] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the terminal device. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0043] Under the above operating environment, the embodiments of the present invention provide as follows: Figure 2 The partial discharge location method for GIS equipment shown is as follows: Figure 2 This is a flowchart of a partial discharge location method for GIS equipment according to an embodiment of the present invention, as follows: Figure 2 As shown, the method includes the following implementation steps:

[0044] Step S201: Obtain partial discharge data of GIS equipment, wherein the GIS equipment is a device associated with gas-insulated combination switch in the power system;

[0045] Step S202: Process the partial discharge data to construct the target dataset;

[0046] Step S203: Using the target dataset, train the initial localization model to obtain the target localization model. The initial localization model and the target localization model use the same neural network architecture, which is constructed based on genetic algorithm, densely connected convolutional neural network and attention mechanism.

[0047] Step S204: Using the real-time operating data of the GIS equipment and the target positioning model, determine the partial discharge positioning result of the GIS equipment.

[0048] In this embodiment of the invention, partial discharge data of a GIS device is acquired, wherein the GIS device is a device associated with a gas-insulated switch in a power system. The partial discharge data is processed to construct a target dataset. Using the target dataset, an initial positioning model is trained to obtain a target positioning model. The initial positioning model and the target positioning model employ the same neural network architecture, which is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism. Using the real-time operating data of the GIS device and the target positioning model, the partial discharge positioning result of the GIS device is determined. Thus, this invention achieves the goal of training a target positioning model to locate partial discharge in operating GIS devices, thereby improving the accuracy of partial discharge perception and positioning of GIS devices, and solving the problem of low accuracy in perceiving and locating partial discharge faults in GIS devices in related technologies.

[0049] The methods described in the embodiments of the present invention will be further described below.

[0050] Optionally, in the above-mentioned partial discharge positioning method for GIS equipment, multiple first positions are determined on the pipe wall according to a first number corresponding to the pipe axial direction and a second number corresponding to the pipe circumference, and the multiple first positions are used to arrange multiple sensors.

[0051] In one application scenario, the aforementioned first quantity can be 8, and the aforementioned second quantity can be 3. For example, on the pipeline of a GIS device, 8 positions are determined along the axial direction, and 3 positions are determined along the circumferential direction (corresponding to 0°, 90°, and 180° of the cross-section, respectively). That is, a total of 24 first positions are set on the pipeline of the GIS device, and a sensor is set at each first position to receive the field strength amplitude information at the fault point.

[0052] In the above application scenarios, such as Figure 3 As shown, eight locations (the first number) are uniformly selected along the pipe axis (set as the Z-axis) on the inner wall of the GIS equipment's pipe. The cross-section corresponding to the pipe axis mentioned above can be called the XY section. In... Figure 3From the perspective shown, the top position is set as the 0° position corresponding to the pipe cross-section, and the bottom position is set as the 180° position corresponding to the pipe cross-section. Based on this, three positions (i.e., the second quantity) can be taken at 0°, 90°, and 180° on each pipe cross-section. That is to say, 8×3=24 first positions are taken inside the GIS equipment pipe, and 24 virtual sensors are set at the 24 first positions.

[0053] Furthermore, the three-dimensional perspective view of the pipeline is as follows: Figure 4 As shown, multiple cross-sectional positions can be taken in the axial direction of the pipe, or at a certain position (e.g., Figure 3 A basin-type insulator is installed at position 5 in the Z-axis direction shown. The basin-type insulator includes a convex surface and a concave surface.

[0054] Optionally, in the above-mentioned GIS equipment partial discharge location method, multiple second locations corresponding to multiple preset fault categories are also set on the pipe wall, and the multiple second locations represent the location where the partial discharge fault occurs.

[0055] In one application scenario, the aforementioned preset fault categories can be Class A, Class B, and Class C faults. Class A faults represent faults occurring within the pipe space of the GIS equipment (not in contact with the pipe wall), Class B faults represent faults occurring on the surface of the central conductor of the GIS equipment, and Class C faults represent faults occurring on the concave and convex surfaces of the basin insulator in the GIS equipment. Further, multiple second locations are determined within the areas corresponding to the aforementioned preset fault categories, and these multiple second locations are designated as the fault point locations.

[0056] Optionally, in step S201 above, obtaining the partial discharge data of the GIS equipment may further include the following steps:

[0057] Step S211: Using multiple sensors, partial discharge data is acquired, wherein the partial discharge data includes field strength information corresponding to multiple second locations.

[0058] In the above application scenarios, the location of the fault point corresponding to a type A fault is called the type A location, the location of the fault point corresponding to a type B fault is called the type B location, and the location of the fault point corresponding to a type C fault is called the type C location. Based on this, such as Figure 5As shown, Class A positions are evenly distributed within the air chamber space, neither contacting the conductor surface nor the basin insulator surface. Along the pipe axis (i.e., the Z-axis direction), an XY plane (i.e., the cross-section of the GIS equipment pipe cavity) is cut at 100mm intervals. A polar coordinate system (r, θ) is constructed within the XY plane, where r represents the distance to the central conductor and θ represents the rotation angle. Four circles are defined within this plane: r = 230, r = 330, r = 430, and r = 530. Excitation sources are placed at eight positions on each circle: θ = 0°, θ = 45°, θ = 90°, θ = 135°, θ = 180°, θ = 225°, θ = 270°, and θ = 315°. In other words, a total of 32 Class A partial discharge signal sources (i.e., Class A positions) are set on each XY plane.

[0059] Still as Figure 5 As shown, Class B locations are set on the surface of the central conductor. Similar to Class A locations, on the intercepted XY plane, a circle with a polar coordinate of θ = 130° is taken. On this circle, eight locations are set at θ = 0°, θ = 45°, θ = 90°, θ = 135°, θ = 180°, θ = 225°, θ = 270°, and θ = 315°, for a total of eight partial discharge signal sources. In other words, a total of eight Class B partial discharge signal sources (i.e., Class B locations) are set on each XY plane.

[0060] Figure 6 This includes the projection of the basin-type insulator along the Z-axis (left image, i.e., the transverse projection of the basin-type insulator) and the projection of the basin-type insulator along the XY plane (right image, i.e., the longitudinal projection of the basin-type insulator), such as... Figure 6 As shown, Class C locations are distributed on the convex and concave surfaces of the basin insulator. A polar coordinate system similar to that described for Class A locations is established on the projection of the basin insulator in the Z-axis direction. Following the same partial discharge signal source setting method as for Class A locations, 32 Class C partial discharge signal sources (i.e., Class C locations) are determined on the concave and convex surfaces respectively. In other words, a total of 64 Class C partial discharge signal sources (i.e., Class C locations) are set on the surface of each basin insulator (including the concave and convex surfaces).

[0061] Based on the above methods and steps, the number of fault points corresponding to multiple areas and various fault locations in the pipeline of the GIS equipment is set as shown in Table 1 below. In Table 1, the XY plane inside the pipeline represents the set of multiple XY planes taken inside the pipeline.

[0062] Table 1

[0063]

[0064] By using the 24 virtual sensors positioned at the aforementioned 24 first locations, a total of 1144 fault location locations can be collected for field strength information. Table 2 below shows the field strength data collected by some sensors at some fault location locations. For example, in this case, sensor (0°, 1) represents... Figure 3 The sensor is positioned at the first location, determined by the 0° position corresponding to the mid-section and the Z-axis direction, with the number 1. Location 1 corresponding to the fault point represents the first location randomly selected from the aforementioned 1144 fault point locations; similarly, location 2 corresponding to the fault point represents the second location randomly selected from the aforementioned 1144 fault point locations.

[0065] Correspondingly, in application scenarios, the coordinate information of the above-mentioned multiple fault point locations can also be obtained. For example, the XYZ coordinates corresponding to positions 1 to 7 randomly selected from multiple fault point locations in Table 2 are shown in Table 3 below.

[0066] Table 2

[0067]

[0068]

[0069] Table 3

[0070]

[0071] It should be noted that in Table 2 above, the unit of the field strength data can be determined according to the application scenario requirements and the specific physical quantity corresponding to the field strength data. In Table 3 above, the length unit corresponding to the coordinates of the fault point location can be determined according to the application scenario requirements and the specific size of the GIS equipment. As shown in Table 3 above, in the application scenario, multiple fault point locations can also be recorded for corresponding fault types. For example, the fixed point location corresponding to location 1 above is a Class A location, that is, the fault point location corresponding to Class A faults.

[0072] Optionally, step S202 above, which involves processing the partial discharge data to construct the target dataset, may further include the following steps:

[0073] Step S221: Based on the three-dimensional coordinate information of multiple second positions, the field strength information in the partial discharge data is preprocessed to obtain the processed result data. The preprocessing operations include: missing value completion, outlier correction and data denoising.

[0074] Step S222: Normalize the processed data and establish multiple feature data tables corresponding to the second position;

[0075] Step S223: Based on the feature data table, construct the target dataset, wherein the target dataset includes multiple sets of sample label pairs, the samples in each set of sample label pairs are determined by field strength information, and the labels in each set of sample label pairs are determined by three-dimensional coordinate information.

[0076] Based on the above method steps, in the application scenario, data preprocessing is performed on the field strength data collected by the sensor and the three-dimensional coordinate information of the fault point. Data preprocessing operations may include handling missing values, correcting outliers, and removing noise. After data preprocessing, the processed data is normalized to establish a fault point feature data table. Further, based on the fault point feature data table, the processed data is divided into a training set, a validation set, and a test set. Each set of data in the training set, validation set, and test set includes the field strength information of the fault point collected by the sensor, and the data label for each set is the three-dimensional coordinate information of the fault point corresponding to the data sample. Each set of data can be referred to as a sample label pair.

[0077] Specifically, outlier handling in the aforementioned data preprocessing can include an identification process and an outlier correction process. During sensor operation, interference or noise from the environment may affect the sensor's measurement results, and signal distortion may also occur during transmission. For outliers generated under these two conditions, the box plot method can be used to identify partial discharge anomaly signals, and the median of the corresponding data can be used to replace the outlier based on the historical operating data of the GIS equipment.

[0078] Specifically, the normalization process in the above data preprocessing can scale the feature values ​​to a relatively small range, which helps to accelerate the convergence process of optimization algorithms such as gradient descent. It can also reduce the correlation between features, which helps the model to better capture the true distribution of the data. Therefore, the field strength data collected by the sensor and the three-dimensional coordinate information of the fault point are normalized respectively.

[0079] For example, the field strength data can be normalized using the calculation method shown in the following formula (1).

[0080]

[0081] In the above formula (1), E′ represents the normalized value of the field strength amplitude at the fault point, and E represents the field strength amplitude at the fault point. min E represents the minimum electric field amplitude. max This represents the maximum field strength amplitude.

[0082] Furthermore, the three-dimensional coordinate information of the fault point is normalized according to the calculation method of the following formula (2).

[0083]

[0084] In formula (2), x(y / z) represents the three-dimensional coordinate value of the fault point; x(y / z)′ represents the normalized three-dimensional coordinate value of the fault point; x(y / z)min represents the minimum value corresponding to each coordinate component in the three-dimensional coordinates; and x(y / z)max represents the maximum value corresponding to each coordinate component in the three-dimensional coordinates.

[0085] Furthermore, based on the normalized data, data samples are obtained by organizing the data according to the following formula (3), and data labels corresponding to the data samples are obtained by organizing the data according to the following formula (4).

[0086]

[0087] label i =[x i ,y i ,z i ] Formula (4)

[0088] In the above formula (3), data i This represents the i-th data sample. This represents the field strength data collected by the k-th sensor in the i-th data sample. In the above formula (4), label i Indicates data i The corresponding data label, [x i ,y i ,z i ] represents label i The three-dimensional coordinates of the corresponding fault location.

[0089] After obtaining the processed target dataset (including multiple sets of data in the training, validation, and test sets, each set including data samples and data labels) through the above data preprocessing, the multiple sets of data in the target dataset are randomly shuffled, and 70% of the target dataset is used as the training set, 15% as the validation set, and 15% as the test set. In the above application scenario, the method for partitioning the full sample set (unordered) corresponding to the target dataset can be as follows: Figure 7 As shown, the full sample set is divided into three parts: training set, validation set, and test set.

[0090] The training set described above is the dataset used to actually train the neural network model (i.e., the object localization model). The neural network model learns the features and patterns of the data on the training set, adjusting its internal parameters (such as weights and biases). The training goal of the neural network model is to achieve good performance on the training data. The validation set is used for hyperparameter tuning of the neural network model. During training, the model learns on the training set and is evaluated on the validation set to observe its performance on unseen data. The test set further evaluates the neural network model. After training and validation, the test set is used to evaluate the generalization ability of the neural network model, that is, to evaluate its performance on entirely new data.

[0091] Optionally, the target dataset includes multiple sets of sample label pairs; in step S203 above, training the initial localization model using the target dataset to obtain the target localization model may further include the following execution steps:

[0092] Step S231: Input the samples from multiple sample label pairs in the target dataset into the initial localization model to obtain the prediction results;

[0093] Step S232: Calculate the model training loss based on the prediction results and the labels in multiple sample label pairs;

[0094] Step S233: Adjust the hyperparameters of the initial localization model according to the model training loss to obtain the target localization model.

[0095] Based on the above method steps, in application scenarios, it can be based on, for example... Figure 8 The model training process shown yields a target localization model. For example... Figure 8 As shown, in the application scenario, the partial discharge localization requirements of GIS equipment are clearly defined, and a localization plan is formulated based on these requirements. Sensor installation points and fault points are set in the GIS equipment, and partial discharge data corresponding to the fault points inside the GIS equipment are collected using the sensors. The collected partial discharge data is preprocessed to obtain the target dataset. Then, a densely connected attention neural network model based on a genetic algorithm (which can also be regarded as the initial model) is built. The target dataset is used to train the neural network model to obtain the training result (which can be regarded as the first intermediate model). Regularization technology is used to optimize the training result to obtain the optimized result (which can be regarded as the second intermediate model). The optimized result is further trained and tested, and the results are analyzed and verified to determine the verification result (which can be regarded as the final target localization model).

[0096] It should be noted that the above-mentioned densely connected attention neural network model based on genetic algorithm can be a neural network model that uses genetic algorithm, densely connected convolutional neural network structure (also known as 1DenseNet), and attention mechanism.

[0097] Optionally, the neural network architecture includes a convolutional sub-network and an attention sub-network; in step S231 above, inputting samples from multiple sets of sample label pairs in the target dataset into the initial localization model to obtain prediction results may further include the following execution steps:

[0098] Step S2311: Extract features from the samples using a convolutional sub-network to obtain the extraction results;

[0099] Step S2312: Feature selection is performed on the extracted results through an attention subnetwork to obtain the selection result;

[0100] Step S2313: Use the target activation function and the selection result to predict the location and obtain the prediction result.

[0101] The aforementioned convolutional subnetwork can be a partial model corresponding to the aforementioned densely connected convolutional neural network structure (i.e., DenseNet). The aforementioned attention subnetwork can be a partial model related to the aforementioned attention mechanism.

[0102] In deep learning, Convolutional Neural Networks (CNNs) have a wide range of applications. The DenseNet network within CNNs establishes dense connections between all preceding layers and subsequent layers, achieving feature reuse through connections along feature channels. Therefore, the DenseNet network can be trained to achieve greater convolutional depth with fewer parameters and lower computational costs, resulting in higher accuracy and making it suitable for processing one-dimensional data such as fault point magnitudes.

[0103] Specifically, the convolutional sub-network in this embodiment of the invention adopts the DenseNet network architecture, such as... Figure 9 As shown, the convolutional subnetwork includes dense blocks and transition layers, and the transition layers may include convolutional layers. Figure 9 The diagram shows two one-dimensional convolutional layers (1×3), a pooling layer, and a linear layer.

[0104] like Figure 9As shown, in DenseBlock, the feature dimensions of each layer are consistent. The nonlinear combination function H(·) in DenseBlock adopts a BN-ReLu-Conv structure. The BN-ReLu-Conv structure represents a structure that combines Batch Normalization (BN), Rectified Linear Unit (ReLu), and Convolution (Conv).

[0105] In DenseNet, all preceding layers are connected as input, as shown in the following formula (5).

[0106] out l =H l ([out1,out2,…,out l-1 ]) Formula (5)

[0107] In the above formula (5), H l (·) represents the nonlinear transformation function of the l-th layer network, out l This represents the output of the l-th layer network. The operations corresponding to the nonlinear combination function can include a series of BN operations, ReLU operations, and Conv operations.

[0108] For example, the specific implementation of the BN operation can be shown in the following formula (6).

[0109]

[0110] In the above formula (6), N l () represents the result of batch normalization of the data at level l. The table shows the output sum of the first l-layer network, where γ represents the scaling factor, β represents the offset parameter, μ represents the mean of the input data, and σ... 2 ζ represents the variance of the input data, and ζ represents a preset constant (used to avoid cases where the denominator is 0).

[0111] For example, the specific implementation of the ReLU operation can be shown in Equations (7) and (8) below.

[0112] ReLU(x weight_sum ) = max(0, x weight_sum ) Formula (7)

[0113] Pooling(X)i,j=max m,n (X i×S+m,j×S+n ) Formula (8)

[0114] In the above formula (7), x weight_sumThe result is obtained by weighted summing the outputs of neurons in the previous layer with their weights, and then adding a bias term. ReLU(x) weight_sum ) represents the result of linear correction. Pooling(x)i,j represents the pooling result of the i-th row and j-th column of the current feature matrix X. The pooling window size in the pooling operation is represented as F×F, and the operation step size of the pooling operation is represented as S. m,n are the indices in the pooling window, used to traverse the elements in the pooling window.

[0115] For example, the specific implementation of the Conv operation can be shown in the following formula (9).

[0116]

[0117] In the above formula (9), X represents the input feature matrix of the current convolutional layer, p represents the current position in the one-dimensional feature matrix corresponding to the convolution operation, and x p+r W represents the value at position p+r in the input sequence during one-dimensional convolution. r represents the weight value at position r in the convolution kernel, and b represents the bias term of the convolutional layer.

[0118] As shown above, DenseBlock can yield the results of a series of BN, ReLU, and Conv operations. DenseBlock can effectively extract and combine input features, enhance the network's expressive power, and help accelerate training and improve model performance.

[0119] like Figure 9 As shown, a transition can be used to connect two adjacent DenseBlocks, reducing the size of the feature dimension. Each transition can include a 1×3 convolutional layer and a pooling layer (e.g., an average pooling layer). In other words, the structure of the DenseNet network architecture can include: a BN operation part, a ReLU operation part, a 1×3 convolutional part, and an average pooling part.

[0120] Furthermore, the attention subnetwork in this embodiment can be used for feature selection. In practical applications, different inputs may have different importance. Therefore, this embodiment utilizes the attention mechanism in the attention subnetwork to enable the model to dynamically allocate weights based on the importance of the inputs, thereby improving the model's performance and generalization ability. The contribution of the features extracted by the aforementioned DenseNet to the localization task is unknown. The attention mechanism can provide weights for different features, achieving feature selection and thus improving the accuracy of fault localization tasks.

[0121] Specifically, the calculation method of the corresponding attention mechanism in the attention sub-network can be expressed as the following formulas (10) and (11).

[0122]

[0123] In the above formula (10), Attention(·) represents the calculation result of the attention mechanism, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and D represents the value matrix. k represents the scaling factor, whose size is equal to the dimension of K, and Softmax(·) represents an activation function used for multi-class classification problems.

[0124] In the above formula (11), X represents the input matrix, W Q W represents the weight value corresponding to the query matrix. K W represents the weight values ​​corresponding to the key matrix. V This represents the weight values ​​corresponding to the value matrix.

[0125] Therefore, in application scenarios, it can be done according to the following... Figure 10 The computation process corresponding to the attention subnetwork shown implements the attention mechanism computation corresponding to the above formulas (10) and (11). For example... Figure 10 As shown, N represents the dimension of the neural network input features, and D... i (where i = x, q, k, v) represents the data dimensions of the current input feature matrices X, K, Q, and V, and H represents the final output matrix. The similarity is calculated using the Q and K matrices, and the weight value is obtained after normalization of the similarity. The weight value is multiplied by V to obtain the final output feature H.

[0126] Based on the above method steps, the neural network architecture of the target localization model in this embodiment of the invention employs an attention subnetwork to implement a multi-head attention mechanism. Compared to the traditional self-attention mechanism, the multi-head attention mechanism in this embodiment enables the model to focus on multiple aspects simultaneously, thereby improving its ability to understand complex information. By decomposing the attention mechanism into multiple heads, the model can learn richer and more representative feature representations in different subspaces, and then fuse the features from different subspaces to improve the overall model's generalization ability and robustness.

[0127] Optionally, the neural network architecture also includes a genetic sub-network; in step S233 above, adjusting the model hyperparameters of the initial localization model based on the model training loss to obtain the target localization model may further include the following execution steps:

[0128] Step S233 1: Calculate the hyperparameters of the initial localization model by using the genetic sub-network and model training loss to obtain the target hyperparameters;

[0129] Step S2332: Adjust the model hyperparameters of the initial positioning model according to the target hyperparameters to obtain the target positioning model.

[0130] The aforementioned genetic subnetwork can be a partial model corresponding to a pre-defined genetic algorithm, and this genetic subnetwork is used to further optimize the model's hyperparameters.

[0131] The selection of weights in a neural network has a significant impact on its training performance. This embodiment of the invention employs a genetic algorithm for global weight search. A genetic algorithm is an optimization search algorithm that simulates natural selection and genetic mechanisms, mimicking the process of biological evolution. By performing genetic operations on a population of candidate solutions, solutions better suited to the target are evolved generation by generation. Its basic idea is to simulate the mechanisms of heredity, mutation, and selection in biological evolution. The fitness function is the reciprocal of the mean square error between the network output value and the actual coordinates of the fault point. Traditional genetic algorithms suffer from slow convergence speed and a tendency to get trapped in local optima. Therefore, after the elite retention step, a local search operator is applied to iteratively search the optimal chromosome generated in each iteration.

[0132] Specifically, the calculation process corresponding to the genetic algorithm used in this embodiment of the invention is as follows: Figure 11 As shown. Figure 11 As shown, after assessing the fitness of individuals in the population, the model enters the parameter optimization process. The parameter optimization process includes three stages: encoding, population initialization, population evaluation and selection, crossover, and mutation.

[0133] In the encoding stage, the weight parameters of the neural network are encoded into the form of chromosomes or genomes using binary encoding.

[0134] Initializing the population refers to randomly generating a set of initial individuals (solutions) at the start of the genetic algorithm. This set of individuals constitutes the initial population. The quality of the initial population directly affects the convergence and search efficiency of the subsequent genetic algorithm. This invention chooses to randomly generate individuals based on the solution space of the problem, typically by randomly selecting initial values ​​within the valid range of the parameters.

[0135] Population evaluation refers to the process of assessing the fitness of each individual in the population. Fitness assessment is usually accomplished through a defined fitness function, which measures the quality of each individual in solving problems. The fitness function can be expressed as the following formula (12).

[0136]

[0137] In the above formula (12), N represents the number of fault points in the GIS equipment, (x i y i , z i(x') represents the true coordinates of the i-th fault point. i y' i , z' i ) represents the model predicted coordinate value of the i-th fault point.

[0138] In the selection phase, the fitness function of each individual in the population is fitted. i Perform mutual comparisons. Evaluate the fitness function `fit` of individual `i`. i This is used to determine the individual's superiority or inferiority within the population. The fitness function is generally considered to be... i The higher the value, the better the trait of individual i, and therefore the greater the probability that the genes of individual i will be passed on to the next generation.

[0139] The crossover process facilitates the exchange of genetic information between individuals of the previous generation. A portion of the chromosomes from the father and mother are exchanged, resulting in a new offspring carrying entirely new genes. This process preserves the superior genes of both parents, allowing the overall traits of the population to continuously evolve in a direction that adapts to the environment. The calculation process corresponding to the crossover process can be expressed as the following formula (13).

[0140] Crossover(m, n) = k Formula (13)

[0141] In the above formula (13), Crossover(·) represents the crossover algorithm, m and n represent the parent individuals, and k represents the new offspring individuals after crossover.

[0142] Based on the crossover process, the chromosomes exchanged by the parents undergo gene mutations for some reason, resulting in a completely new gene slice that does not exist on the chromosomes of either parent. The probability of mutation in the genetic algorithm (GA) is relatively small. Therefore, a small mutation probability is set in the GA. In the embodiments of this application, the mutation probability can be set to 1%.

[0143] After the mutation stage, a genetic computation is completed through elite preservation and local search. If the preset maximum number of iterations is not reached, the fitness of individuals in the population is evaluated again and the parameter optimization process is carried out again. If the preset maximum number of iterations is reached, the computation process corresponding to the genetic algorithm ends, the model hyperparameters determined at this time are determined as target hyperparameters, and the target localization model is obtained based on the target hyperparameters.

[0144] Through the above methods and steps, this embodiment of the application combines genetic algorithms, densely connected convolutional neural network structures, and attention mechanisms to construct a neural network architecture, and uses this neural network architecture to train and obtain a target localization model. The overall network architecture of the above target localization model is as follows: Figure 12As shown, the network architecture in this example may include (in order of data flow): DensetNet-1, Attention Mechanism-1, DensetNet-2, Attention Mechanism-2, pooling layers, and linear layers. The genetic algorithm optimization process can optimize the hyperparameters of the above DensetNet-1, Attention Mechanism-1, DensetNet-2, Attention Mechanism-2, pooling layers, and linear layers.

[0145] Optionally, the above-mentioned partial discharge location method for GIS equipment may further include the following steps:

[0146] Step S205: Apply a regularization strategy to correct overfitting of the target localization model and update the target localization model. The regularization strategy is determined by preset regularization parameters and sparse penalty parameters. The regularization strategy is used to adjust the target activation function in the target localization model.

[0147] Based on the above method steps, in application scenarios, regularization techniques are used to address the potential overfitting problem in target localization models. Overfitting refers to the phenomenon where a model performs well on training data but poorly on new data. This is usually caused by the model overlearning noise or specific samples in the training data. Excessive model parameters can also lead to overfitting when training deep neural networks. To address this, this embodiment of the invention employs Dropout, L2 regularization, and sparse penalty parameters to solve the overfitting problem.

[0148] Specifically, Dropout is a technique that randomly discards some neurons during the training of a neural network (by setting the weights of those neurons to 0) to prevent overfitting. By randomly discarding some neurons during training, the neural network can learn without relying on any single neuron, reducing collaboration between neurons, increasing the network's robustness and generalization ability, and lowering the risk of overfitting.

[0149] Specifically, L2 regularization is a technique that penalizes weights during neural network training to prevent overfitting. In this embodiment, L2 regularization is mainly used for weight decay, which penalizes the model's complexity by adding the square of the L2 norm (i.e., Euclidean norm) of the parameters to the loss function. The specific implementation of the above weight decay can be expressed as the following formula (14).

[0150]

[0151] In the above formula (14), Ω weights Indicates the weight after decay. This represents the square of the L2 norm of the weight parameters in a neural network.

[0152] By adding an L2 regularization term to the loss function, the weights can be made smoother, reducing their volatility and thus lowering the risk of overfitting. L2 regularization can be seen as a constraint on the weights, making the model more stable and having stronger generalization ability.

[0153] Specifically, the introduction of a sparse penalty parameter for neuronal activity is usually intended to encourage certain neurons in the neural network to remain in a low-activation state, thereby improving the sparsity of the model and reducing the risk of overfitting. The neuronal activity level corresponding to the aforementioned sparse penalty parameter can be calculated using the following formula (15).

[0154]

[0155] In the above formula (15), ρ i ^ represents the output of the i-th neuron, n represents the number of data samples in the training set, l represents the l-th hidden layer, and x j H represents the output of the j-th hidden layer. (l) i Let represent the activation function of the i-th hidden node in the l-th layer.

[0156] Furthermore, in this embodiment of the invention, the sparse penalty parameter term of the KL divergence function (i.e., the KULLBACK-Leibler divergence function) is used, and the specific calculation method is shown in the following formula (16).

[0157]

[0158] In the above formula (16), Ω sparsity Let KL(·) denote the sparsity penalty parameter term, KL(·) denote the KL divergence function, and ρ denote the sparsity penalty factor (usually approaching 0). If ρ and ρ i If ^ are equal, then Ω sparsity The value is 0 if ρ and ρ i If they are not equal, then Ω sparsity The value will increase.

[0159] In this embodiment of the application, in combination with the above regularization technique, the loss function corresponding to the neural network architecture of the above target localization model is modified, and the modification method can be expressed as the following formula (17).

[0160] J reg (x)=J(x)+λ×Ω weights +β×Ω sparsity Formula (17)

[0161] In the above formula (17), J(x) represents the original loss function, J reg(x) represents the corrected loss function. The original loss function is determined by the difference between the predicted coordinates output by the neural network and the actual coordinates of the fault point. Ω weights Let λ represent the weight norm of the neural network, and let λ represent Ω. weights The corresponding regularization parameter, Ω sparsity β represents the activity level of neurons in a neural network, and β represents Ω. sparsity Corresponding regularization parameters.

[0162] Furthermore, based on the aforementioned model training scheme, the initial localization model is trained using training, validation, and test sets divided from the target dataset to obtain the target localization model. In one application scenario, the parameter values ​​involved in training the target localization model are shown in Table 4 below.

[0163] Table 4

[0164] Parameter name Value Number of iterations 200 Data batch size 16 Learning rate 0.001 Loss function Activation function ReLU Regularization parameters λ = 0.0001, β = 0.001 DenseNet layers 3 Number of DenseNet convolutional kernels 32 Attention head count 8

[0165] In one application scenario, according to... Figure 13 The model training process shown involves iterative training of the target localization model. For example... Figure 13 As shown, the model is trained in multiple rounds based on the training set data. After each round of model training, the training results are evaluated using validation data. If the evaluation results show that the current model is ineffective (i.e., the metric does not reach the preset threshold), the next round of model training is performed based on the training set data. If the evaluation results show that the current model is ineffective (i.e., the metric reaches the preset threshold), the model training is considered complete. At this point, the trained model is tested using the test set data, and the final effect of the model is determined based on the test results, that is, the target localization model is obtained.

[0166] This invention provides a method for locating partial discharge faults in GIS equipment based on a genetic algorithm, combining a DenseNet network and an Attention mechanism. Compared with traditional neural network models, the method provided in this invention has higher accuracy. By combining the powerful feature extraction capability of the DenseNet network and the focused attention capability of the Attention mechanism, the fault location method of this invention can more accurately capture the features of partial discharge signals, thereby significantly improving the accuracy of fault location. The introduction of a genetic algorithm gives the model a strong global search capability, and the optimization strategy and network structure design adopted can converge to the optimal solution faster, reducing training time and improving training efficiency. In addition, the method of this invention exhibits high adaptability in handling different types of partial discharge faults and can effectively handle fault location problems under various complex operating conditions.

[0167] Optionally, in step S204 above, determining the partial discharge location result of the GIS equipment using the real-time operating data and target positioning model may further include the following execution steps:

[0168] Step S241: Collect real-time operating data of GIS equipment, wherein the real-time operating data includes at least real-time pipeline field strength data;

[0169] Step S242: Input the real-time running data into the target positioning model for positioning prediction to obtain the partial discharge positioning result.

[0170] According to the above execution steps, in the application scenario, multiple sensors installed in the GIS equipment are used to collect real-time operating data of the GIS equipment. This real-time operating data is organized into input data, and the data format of the output data can adopt the data sample format shown in formula (3) above. Further, the above input data is input into the target positioning model for positioning prediction. The target positioning model will output the predicted coordinate values ​​of the partial discharge fault point (i.e., the above partial discharge positioning result). The data format of the above predicted coordinate values ​​can adopt the data label format shown in formula (4) above. Thus, the task of locating the partial discharge fault point of the GIS equipment in the power scenario is completed through the target positioning model, ensuring the operational safety of the GIS equipment and the safety, stability and reliability of the entire power system.

[0171] In this embodiment, a partial discharge locating device for GIS equipment is also provided. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, a "module" is a combination of software and / or hardware that can perform a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0172] Figure 14 This is a structural block diagram of a partial discharge locating device for GIS equipment according to an embodiment of the present invention, as shown below. Figure 14 As shown, the device includes:

[0173] The acquisition module 1401 is used to acquire partial discharge data of GIS equipment, wherein the GIS equipment is a device associated with gas-insulated combination switch in the power system;

[0174] Processing module 1402 is used to process partial discharge data and construct a target dataset;

[0175] Training module 1403 is used to train the initial localization model to obtain the target localization model using the target dataset. The initial localization model and the target localization model adopt the same neural network architecture, which is constructed based on genetic algorithm, densely connected convolutional neural network and attention mechanism.

[0176] The positioning module 1404 is used to determine the partial discharge positioning result of the GIS equipment by utilizing the real-time operating data of the GIS equipment and the target positioning model.

[0177] Optionally, in the above-mentioned GIS equipment partial discharge positioning device, the GIS equipment determines multiple first positions on the pipe wall according to a first number corresponding to the pipe axial direction and a second number corresponding to the pipe circumference, and the multiple first positions are used to arrange multiple sensors.

[0178] Optionally, in the above-mentioned GIS equipment partial discharge positioning device, multiple second positions corresponding to multiple preset fault categories are also provided on the pipe wall, and the multiple second positions represent the location where the partial discharge fault occurs.

[0179] Optionally, the acquisition module 1401 is further configured to: acquire partial discharge data using multiple sensors, wherein the partial discharge data includes field strength information corresponding to multiple second locations.

[0180] Optionally, the processing module 1402 is further configured to: preprocess the field strength information in the partial discharge data based on the three-dimensional coordinate information of multiple second positions to obtain the processed result data, wherein the processing operations corresponding to the preprocessing include: missing value completion operation, outlier correction operation and data denoising operation; normalize the processed result data to establish a feature data table corresponding to multiple second positions; and construct a target dataset based on the feature data table, wherein the target dataset includes multiple sets of sample label pairs, the samples in each set of sample label pairs are determined by the field strength information, and the labels in each set of sample label pairs are determined by the three-dimensional coordinate information.

[0181] Optionally, the target dataset includes multiple sets of sample label pairs; the training module 1403 is further configured to: input samples from the multiple sets of sample label pairs in the target dataset into the initial localization model to obtain prediction results; calculate the model training loss based on the prediction results and the labels in the multiple sets of sample label pairs; and adjust the model hyperparameters of the initial localization model according to the model training loss to obtain the target localization model.

[0182] Optionally, the neural network architecture includes a convolutional subnetwork and an attention subnetwork; the training module 1403 is further used to: extract features from samples through the convolutional subnetwork to obtain extraction results; select features from the extraction results through the attention subnetwork to obtain selection results; and predict the location using the target activation function and the selection results to obtain prediction results.

[0183] Optionally, the neural network architecture also includes a genetic sub-network; the training module 1403 is further used to: perform hyperparameter optimization calculation on the initial localization model through the genetic sub-network and model training loss to obtain target hyperparameters; and adjust the model hyperparameters of the initial localization model according to the target hyperparameters to obtain the target localization model.

[0184] Optionally, the above-mentioned GIS equipment partial discharge positioning device includes, in addition to all the above modules, a correction module 1405 (not shown in the figure), which is used to: correct overfitting of the target positioning model by adopting a regularization strategy and update the target positioning model, wherein the regularization strategy is determined by a preset regularization parameter term and a sparse penalty parameter term, and the regularization strategy is used to adjust the target activation function in the target positioning model.

[0185] Optionally, the positioning module 1404 is further configured to: collect real-time operating data of the GIS equipment, wherein the real-time operating data includes at least real-time pipeline field strength data; input the real-time operating data into the target positioning model for positioning prediction, and obtain the partial discharge positioning result.

[0186] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0187] According to another aspect of the present invention, a computer-readable storage medium is also provided, the storage medium including a stored program, wherein, when the program is running, it controls the device where the storage medium is located to execute any of the aforementioned GIS device partial discharge positioning methods.

[0188] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps: acquiring partial discharge data of a GIS device, wherein the GIS device is a device associated with a gas-insulated combination switch in a power system; processing the partial discharge data to construct a target dataset; using the target dataset to train an initial positioning model to obtain a target positioning model, wherein the initial positioning model and the target positioning model adopt the same neural network architecture, which is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and using the real-time operating data of the GIS device and the target positioning model to determine the partial discharge positioning result of the GIS device.

[0189] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0190] According to another aspect of the present invention, a partial discharge location system for GIS equipment is also provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the partial discharge location method for GIS equipment as described above.

[0191] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program: acquiring partial discharge data of a GIS device, wherein the GIS device is a device associated with a gas-insulated combination switch in a power system; processing the partial discharge data to construct a target dataset; training an initial positioning model using the target dataset to obtain a target positioning model, wherein the initial positioning model and the target positioning model adopt the same neural network architecture, which is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and determining the partial discharge positioning result of the GIS device using the real-time operating data of the GIS device and the target positioning model.

[0192] According to embodiments of the present invention, a computer program product is also provided. Optionally, in this embodiment, the computer program product can provide GIS device partial discharge location services based on the above-described GIS device partial discharge location method.

[0193] Optionally, in this embodiment, the computer program product described above may be a set of instructions and codes pre-written according to the partial discharge location method for GIS equipment. This computer program product can run on various computer platforms, including personal computers, servers, mobile devices, etc.

[0194] Optionally, in this embodiment, the instructions and code corresponding to the computer program product are used to implement the following method steps: acquiring partial discharge data of the GIS device, wherein the GIS device is a device associated with a gas-insulated combination switch in the power system; processing the partial discharge data to construct a target dataset; using the target dataset to train the initial positioning model to obtain a target positioning model, wherein the initial positioning model and the target positioning model adopt the same neural network architecture, and the neural network architecture is constructed based on a genetic algorithm, a densely connected convolutional neural network, and an attention mechanism; and using the real-time operating data of the GIS device and the target positioning model to determine the partial discharge positioning result of the GIS device.

[0195] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and their optional implementations, and will not be repeated here.

[0196] The sequence numbers of the above embodiments of the present invention are for description only and do not represent the superiority or inferiority of the embodiments.

[0197] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0198] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be 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 displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces; the indirect coupling or communication connection of units or modules can be electrical or other forms.

[0199] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0200] 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.

[0201] 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, ROM, RAM, portable hard drives, magnetic disks, or optical disks.

[0202] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for locating partial discharge in GIS equipment, characterized in that, include: Acquire partial discharge data of a GIS device, wherein the GIS device is a device associated with a gas-insulated combination switch in a power system; The partial discharge data is processed to construct a target dataset, wherein the target dataset includes multiple sets of sample label pairs; Using the target dataset, the initial localization model is trained to obtain the target localization model. The initial localization model and the target localization model adopt the same neural network architecture. The neural network architecture is constructed based on genetic algorithm, densely connected convolutional neural network and attention mechanism. The neural network architecture includes convolutional sub-network, attention sub-network and genetic sub-network. Using the target dataset, an initial localization model is trained to obtain a target localization model, including: extracting features from samples in the multiple sets of sample label pairs using the convolutional sub-network to obtain extraction results, wherein the convolutional sub-network is a partial model corresponding to a densely connected convolutional neural network structure, the convolutional sub-network establishes dense connections between all preceding layers and subsequent layers, and achieves feature reuse through connections of data features on feature channels; performing feature selection on the extraction results using the attention sub-network to obtain selection results; performing location prediction using the target activation function and the selection results to obtain prediction results; calculating the model training loss based on the prediction results and the labels in the multiple sets of sample label pairs; and performing hyperparameter optimization calculation on the initial localization model using the genetic sub-network and the model training loss to obtain target hyperparameters, wherein the hyperparameter optimization process includes encoding, initializing the population, evaluating the population and selection, crossover and mutation, the evaluation of the population refers to the process of evaluating the fitness of each individual in the population, the fitness evaluation is performed through a fitness function. To complete this, in the fitness function This indicates the number of fault points in the GIS equipment. This represents the actual coordinates of the i-th fault point. The model predicted coordinates of the i-th fault point are represented; the model hyperparameters of the initial positioning model are adjusted according to the target hyperparameters to obtain the target positioning model; Using the real-time operating data of the GIS device and the target positioning model, the partial discharge positioning result of the GIS device is determined.

2. The partial discharge location method for GIS equipment according to claim 1, characterized in that, In the GIS equipment, according to a first number corresponding to the pipe axis and a second number corresponding to the pipe circumference, multiple first positions are determined on the pipe wall, and the multiple first positions are used to arrange multiple sensors.

3. The partial discharge location method for GIS equipment according to claim 2, characterized in that, The pipe wall is also provided with multiple second locations corresponding to multiple preset fault categories, and the multiple second locations represent the location where the partial discharge fault occurs.

4. The partial discharge location method for GIS equipment according to claim 3, characterized in that, Obtaining the partial discharge data from the GIS device includes: The partial discharge data is acquired using the plurality of sensors, wherein the partial discharge data includes field strength information corresponding to the plurality of second locations.

5. The partial discharge location method for GIS equipment according to claim 4, characterized in that, Data processing of the partial discharge data to construct the target dataset includes: Based on the three-dimensional coordinate information of the multiple second positions, the field strength information in the partial discharge data is preprocessed to obtain the processed result data. The processing operations corresponding to the preprocessing include: missing value completion operation, outlier correction operation, and data denoising operation. The processing result data is normalized to establish a feature data table corresponding to the multiple second positions; Based on the feature data table, the target dataset is constructed, wherein the target dataset includes multiple sets of sample label pairs, the samples in each set of sample label pairs are determined by the field strength information, and the labels in each set of sample label pairs are determined by the three-dimensional coordinate information.

6. The partial discharge location method for GIS equipment according to claim 1, characterized in that, The partial discharge location method for GIS equipment also includes: The target localization model is overfitted by employing a regularization strategy to update the target localization model. The regularization strategy is determined by preset regularization parameters and sparse penalty parameters, and is used to adjust the target activation function in the target localization model.

7. The partial discharge location method for GIS equipment according to claim 1, characterized in that, Using the real-time operating data of the GIS device and the target positioning model, the partial discharge positioning result of the GIS device is determined as follows: Collect the real-time operating data of the GIS equipment, wherein the real-time operating data includes at least real-time pipeline field strength data; The real-time operating data is input into the target positioning model for positioning prediction to obtain the partial discharge positioning result.

8. A partial discharge locating device for GIS equipment, characterized in that, include: An acquisition module is used to acquire partial discharge data of a GIS device, wherein the GIS device is a device associated with a gas-insulated combination switch in a power system; The processing module is used to process the partial discharge data and construct a target dataset, wherein the target dataset includes multiple sets of sample label pairs; The training module is used to train the initial localization model using the target dataset to obtain the target localization model. The initial localization model and the target localization model adopt the same neural network architecture. The neural network architecture is constructed based on genetic algorithm, densely connected convolutional neural network and attention mechanism. The neural network architecture includes convolutional sub-network, attention sub-network and genetic sub-network. The training module is further configured to extract features from samples in the multiple sets of sample label pairs using the convolutional sub-network to obtain extraction results. The convolutional sub-network is a partial model corresponding to a densely connected convolutional neural network structure. The convolutional sub-network establishes dense connections between all preceding layers and subsequent layers, achieving feature reuse through connections of data features in feature channels. The attention sub-network performs feature selection on the extraction results to obtain selection results. A target activation function and the selection results are used for position prediction to obtain prediction results. Based on the prediction results and the labels in the multiple sets of sample label pairs, the model training loss is calculated. The initial localization model is subjected to hyperparameter optimization calculation using the genetic subnetwork and the model training loss to obtain target hyperparameters. The hyperparameter optimization process includes encoding, population initialization, population evaluation and selection, crossover, and mutation. Population evaluation refers to the process of assessing the fitness of each individual in the population, and the fitness assessment is performed using a fitness function. To complete this, in the fitness function This indicates the number of fault points in the GIS equipment. This represents the actual coordinates of the i-th fault point. The model predicted coordinates of the i-th fault point are represented; the model hyperparameters of the initial positioning model are adjusted according to the target hyperparameters to obtain the target positioning model; The positioning module is used to determine the partial discharge positioning result of the GIS device by utilizing the real-time operating data of the GIS device and the target positioning model.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the storage medium is located to perform the partial discharge location method of any one of claims 1 to 7.

10. A partial discharge positioning system for GIS equipment, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the partial discharge location method for a GIS device according to any one of claims 1 to 7.

11. A computer program product, characterized in that, The method includes a computer program that, when executed by a processor, implements the partial discharge location method for GIS equipment according to any one of claims 1 to 7.

Images