Grounding grid corrosion fault detection method and device, storage medium and computer equipment

Abstract

The invention provides a grounding grid corrosion fault detection method and device, a storage medium and computer equipment. The method comprises the steps that the zero-sequence voltage and zero-sequence current of a grounding grid are monitored; when it is determined that the grounding grid has a corrosion fault according to the zero-sequence voltage and the zero-sequence current, synchronously acquiring a target zero-sequence voltage of a bus in the grounding grid and a target zero-sequence current of each feeder line, and generating a target zero-sequence voltage-zero-sequence current image of each feeder line; and each target zero-sequence voltage-zero-sequence current image is inputted to a pre-constructed grounding grid corrosion fault detection model, a fault detection result of each feeder line is obtained, and the grounding grid corrosion fault detection model comprises a space attention residual learning block. According to the method, complex equipment is not needed, only a data processing link needs to be added in an original monitoring system, the system is simplified, the cost and the maintenance difficulty are reduced, when large-scale data and multi-area real-time monitoring are carried out, image data can be efficiently processed, the computing power and the response speed can be improved, and the requirements of a modern electric power system are met.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a grounding grid corrosion fault detection method, device, storage medium, and computer equipment. Background Art

[0002] Current methods for detecting ground grid corrosion in power systems have numerous shortcomings. Traditional detection methods typically rely on direct measurements of current and voltage to assess the health of the ground grid. However, since changes in ground resistance caused by ground grid corrosion are typically minimal, these methods struggle to accurately and promptly identify corrosion faults. Traditional digital signal processing (DSP) techniques require fault signature extraction, but this approach is susceptible to noise interference and other environmental factors, resulting in unstable diagnostic results. This approach, particularly in complex electromagnetic environments, carries a high risk of misdiagnosis and missed diagnosis. Furthermore, existing ground grid corrosion detection techniques typically require complex equipment and high maintenance costs. Because current changes caused by corrosion are often subtle, existing technologies place high demands on equipment installation and signal processing, increasing system complexity, costs, and maintenance difficulties. Therefore, simply, quickly, and accurately detecting ground grid corrosion faults without adding additional equipment has become a pressing technical challenge in power systems.

[0003] As power systems expand and data volumes increase, traditional detection methods are becoming increasingly inefficient when processing large amounts of data. This is especially true when real-time monitoring of grounding grid corrosion across multiple regions is required. The computing power and response speed of a single processing method are no longer sufficient to meet the demands of modern power systems. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above technical deficiencies, especially the technical deficiencies of low efficiency in the prior art.

[0005] In a first aspect, the present application provides a grounding grid corrosion fault detection method, the method comprising:

[0006] Monitor the zero-sequence voltage and zero-sequence current of the grounding grid;

[0007] When a corrosion fault is determined in the grounding grid based on the zero-sequence voltage and zero-sequence current, the target zero-sequence voltage of the bus in the grounding grid and the target zero-sequence current of each feeder are synchronously collected, and a target zero-sequence voltage-zero-sequence current image of each feeder is generated;

[0008] Each target zero-sequence voltage-zero-sequence current image is input into a pre-built grounding grid corrosion fault detection model to obtain the fault detection result of each feeder. The grounding grid corrosion fault detection model includes a spatial attention residual learning block.

[0009] In one embodiment, the step of determining whether a grounding grid corrosion fault has occurred based on the zero-sequence voltage and the zero-sequence current includes:

[0010] When the zero-sequence voltage meets the preset zero-sequence voltage differential condition, the fault time of the grounding grid corrosion fault is determined;

[0011] Within the preset delay from the start of the fault time, if the zero-sequence voltage meets the preset zero-sequence voltage effective value condition, and the zero-sequence current meets the preset zero-sequence current differential condition and the preset zero-sequence current effective value condition, it is determined that a grounding grid fault has occurred.

[0012] In one embodiment, the formula corresponding to the preset zero-sequence voltage differential condition is:

[0013]

[0014] The formula corresponding to the preset zero-sequence voltage effective value condition is:

[0015]

[0016] The formula corresponding to the preset zero-sequence current differential condition is:

[0017]

[0018] The formula corresponding to the preset zero-sequence current effective value condition is:

[0019]

[0020] in, is the sampling zero sequence voltage, is the effective value of the sampled zero-sequence voltage, To calculate the differential threshold of the differential voltage, is the effective value threshold of the sampling zero-sequence voltage, is the sampled zero-sequence current of the nth feeder, To calculate the differential threshold of the differential current, is the effective value of the sampled zero-sequence current of the nth feeder, is the effective value threshold of the sampling zero-sequence current, is the failure time of grounding grid corrosion fault, The start time of fault feeder detection.

[0021] In one embodiment, the step of generating a target zero-sequence voltage-zero-sequence current image for each feeder includes:

[0022] The target zero-sequence voltage and each target zero-sequence current are normalized, and the normalized target zero-sequence voltage and each normalized target zero-sequence current are superimposed on the same image. After binarization of each image, the target zero-sequence voltage-zero-sequence current image of each feeder is obtained.

[0023] In one embodiment, the formula corresponding to the spatial attention residual learning block is:

[0024]

[0025] Among them, x and y represent the input and output of the residual block respectively, represents the residual mapping to be learned by the stacked layer, represents the weight of the convolution kernel, represents the ReLu function, represents the hyperparameter, Represents a feature map with attention weights.

[0026] In one embodiment, the grounding grid corrosion fault detection model consists of a convolutional layer, a maximum pooling layer, two spatial attention residual learning blocks, a global average pooling layer, and a bidirectional fully connected classification layer.

[0027] In one embodiment, the method further comprises:

[0028] Parallel computing technology is used to accelerate the generation of target zero-sequence voltage-zero-sequence current images for each feeder and to accelerate the training of grounding grid corrosion fault detection models.

[0029] In a second aspect, the present application provides a grounding grid corrosion fault detection device, the device comprising:

[0030] Voltage and current monitoring module, used to monitor the zero-sequence voltage and zero-sequence current of the grounding grid;

[0031] An image generation module is used to synchronously collect the target zero-sequence voltage of the bus in the grounding grid and the target zero-sequence current of each feeder when it is determined that a corrosion fault has occurred in the grounding grid based on the zero-sequence voltage and zero-sequence current, and to generate a target zero-sequence voltage-zero-sequence current image of each feeder;

[0032] The fault detection result determination module is used to input each target zero-sequence voltage-zero-sequence current image into a pre-built grounding grid corrosion fault detection model to obtain the fault detection result of each feeder. The grounding grid corrosion fault detection model includes a spatial attention residual learning block.

[0033] In a third aspect, the present application provides a storage medium: the storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the grounding grid corrosion fault detection method as described in any of the above embodiments.

[0034] In a fourth aspect, the present application provides a computer device, comprising: one or more processors, and a memory;

[0035] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the steps of the grounding grid corrosion fault detection method in any one of the above embodiments are performed.

[0036] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0037] The grounding grid corrosion fault detection method provided in this application determines corrosion faults by monitoring zero-sequence voltage and zero-sequence current. Compared with traditional methods, it can more sensitively capture the tiny electrical signal changes in the early stage of corrosion and achieve early detection. When determining the corrosion fault, the target zero-sequence voltage-zero-sequence current image is synchronously collected and generated to provide a basis for subsequent accurate analysis. The grounding grid corrosion fault detection model containing a spatial attention residual learning block is used to process the image. The spatial attention mechanism focuses on key areas, highlights the characteristics of corrosion faults, and suppresses noise interference. The residual learning block enhances the model learning ability, improves detection accuracy and stability, and reduces the risk of misdiagnosis and missed diagnosis. This method does not require complex equipment. It only needs to add a data processing link to the original monitoring system, simplify the system, reduce costs and maintenance difficulties, and can efficiently process image data when dealing with large-scale data and multi-region real-time monitoring, improve computing power and response speed, and meet the needs of modern power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0039] Figure 1 An example diagram of zero-sequence component fault analysis provided in an embodiment of the present application;

[0040] Figure 2 A schematic diagram of a flow chart of a grounding grid corrosion fault detection method provided in an embodiment of the present application;

[0041] Figure 3 An example diagram of creating a ZSV-ZSC image provided in an embodiment of the present application;

[0042] Figure 4 An example diagram of image recognition of the generated ZSV-ZSC waveform provided in an embodiment of the present application;

[0043] Figure 5 An example diagram of two learning blocks provided in an embodiment of the present application;

[0044] Figure 6 An example diagram of the model structure of the SARL-CNN-based detection method provided in an embodiment of the present application;

[0045] Figure 7 A schematic diagram of the structure of a grounding grid corrosion fault detection device provided in an embodiment of the present application;

[0046] Figure 8 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0048] like Figure 1 As shown in the figure, (a) is the compensation system diagram, (b) is the zero-sequence equivalent network with a ground grid corrosion fault in the compensation system, and (c) is the simplified zero-sequence equivalent network ignoring the feeder impedance. In addition, it is assumed that the ZSV (Zero Sequence Voltage) on the bus is , then the ZSV and ZSC (Zero Sequence Current) equations can be expressed as:

[0049]

[0050] by As a benchmark, Laplace transform is used to calculate, and As shown below. Considering that the reference direction of ZSC is from bus to feeder, The sign of should be negative.

[0051]

[0052] For a healthy feeder k, The phase of For the fault feeder 1, since the compensation of the transient component by the Peterson coil can be ignored in the fault transient, in the initial stage of the grounding grid corrosion fault, The phase lags behind Therefore, the phase difference between the ZSV and the ZSC can be used to detect the faulty feeder. However, the phase difference is affected by parameters, topology, and fault conditions, such as large transition resistance, fault time variation, and noise interference. This poses a significant challenge to faulty feeder detection based on ZSVs and local ZSCs.

[0053] In fact, due to the Fault feeder ZSC The resulting superimposed waveform is ZSC with healthy feeders The generated superimposed waveforms are quite different, so image recognition based on ZSV-ZSC waveforms is a feasible solution to improve the accuracy of fault feeder detection in distribution networks.

[0054] Based on this, the present application provides a grounding grid corrosion fault detection method. The following embodiments are described using the method applied to computer equipment as an example. It can be understood that the computer equipment can be any device with data processing functions, including but not limited to a single server, a server cluster, a personal laptop computer, a desktop computer, etc. Figure 2 As shown, the present application provides a chat corpus annotation method, the method comprising:

[0055] S101: Monitor the zero-sequence voltage and zero-sequence current of the grounding grid.

[0056] The grounding grid is a crucial component of the grounding system for power equipment. It consists of interconnected, buried conductors (such as metal wires and flat steel) forming a network. Its primary function is to conduct abnormal currents, voltages, and other abnormalities in various equipment in the power system to the ground, ensuring the safety of equipment and personnel. Corrosion in the grounding grid can seriously affect its grounding performance, threatening the stable operation of the power system. Zero-sequence voltage is a symmetrical component in the power system that describes the symmetry of the voltages within a three-phase system. Under normal operation, the three-phase voltages are symmetrical, and the zero-sequence voltage is typically close to zero. However, when a ground fault or other asymmetrical fault occurs in the system, the zero-sequence voltage exhibits abnormal variations, with its amplitude and phase deviating from the normal range. Monitoring zero-sequence voltage is crucial for detecting corrosion in the grounding grid, as corrosion can degrade the grounding performance of the grounding grid, leading to variations in the zero-sequence voltage. Zero-sequence current is also a symmetrical component in the power system, corresponding to the zero-sequence voltage. Under normal operation, the three-phase currents are balanced, and the zero-sequence current is nearly zero. However, corrosion in the grounding grid alters the impedance characteristics of the grounding grid, causing abnormal zero-sequence current. By monitoring the changes in zero-sequence current, key clues can be provided for detecting grounding grid corrosion faults.

[0057] In this step, in actual power system applications, the computer connects to the power system's monitoring devices to acquire real-time zero-sequence voltage data from the grounding grid. These monitoring devices are typically installed at key nodes on distribution lines and can accurately measure the amplitude and phase of the zero-sequence voltage. After receiving the zero-sequence voltage data, the computer uses data processing algorithms to analyze and process it. For example, filtering techniques are used to remove noise interference and ensure data accuracy.

[0058] Similar to zero-sequence voltage monitoring, computer equipment collects real-time zero-sequence current data from the grounding grid by connecting to current transformers or other current monitoring devices. Current transformers convert high currents into low currents at a certain ratio, enabling the computer equipment to accurately measure them. After acquiring the zero-sequence current data, the computer equipment also performs data preprocessing, such as filtering and normalization. It then determines whether any zero-sequence current anomalies exist by comparing the data with the zero-sequence current characteristics during normal operation or by utilizing advanced data analysis methods such as machine learning models. If an anomaly is detected, the computer equipment will take appropriate measures, such as issuing an alarm and recording fault information, to ensure the safe operation of the power system.

[0059] It is understandable that changes in zero-sequence voltage and zero-sequence current, as important symmetrical components of the power system, can keenly reflect corrosion faults in the grounding grid. By monitoring these two parameters in real time, computer equipment can capture abnormal signals in the early stages of a fault, promptly identifying corrosion problems in the grounding grid, thereby achieving early warning and rapid response. Compared with traditional detection methods, this method does not require complex equipment and high maintenance costs. It only requires the addition of corresponding data processing modules to the existing power monitoring system to effectively monitor grounding grid corrosion faults. At the same time, because the acquisition of zero-sequence voltage and zero-sequence current data is relatively simple and stable and is not significantly affected by environmental factors, it can improve the accuracy and reliability of detection results, reduce the risk of misdiagnosis and missed diagnosis, and better meet the high efficiency and high precision requirements of modern power systems for grounding grid corrosion fault detection.

[0060] S102: When it is determined based on the zero-sequence voltage and the zero-sequence current that a corrosion fault occurs in the grounding grid, a target zero-sequence voltage of a bus in the grounding grid and a target zero-sequence current of each feeder are synchronously collected, and a target zero-sequence voltage-zero-sequence current image of each feeder is generated.

[0061] Corrosion failures occur when the metal materials of the grounding grid undergo electrochemical reactions under the long-term influence of environmental factors such as soil chemicals, moisture, and oxygen. This causes the metal materials to gradually degrade in structure and performance, thereby compromising the grid's proper grounding function. This failure can alter the electrical properties of the grounding grid, such as increasing grounding resistance and decreasing conductivity, thereby threatening the safe operation of the power system. A bus is a main line connecting multiple power devices or components, collecting and distributing electrical energy. In a grounding grid, a bus typically refers to the main line that connects individual feeders and ultimately connects to the grounding device. It is responsible for directing current from the entire grounding grid to the ground, ensuring the safe operation of the power system. Feeders are branch lines extending from the bus and are responsible for transmitting electrical energy to specific power devices or users. In a grounding grid, each feeder is connected to the bus and to the ground through the grounding device. The grounding condition of a feeder is directly related to the safety of equipment and personnel within its power supply area.

[0062] The target zero-sequence voltage refers to the zero-sequence voltage signal with specific characteristics collected from the bus by computer equipment through specific monitoring and processing methods when a grounding grid corrosion fault occurs. Compared to the zero-sequence voltage during normal operation, the target zero-sequence voltage may vary in amplitude, phase, or waveform. These changes can reflect information such as the extent and location of the grounding grid corrosion fault. The target zero-sequence current refers to the zero-sequence current signal with specific characteristics collected from each feeder by computer equipment through monitoring when a grounding grid corrosion fault occurs. Changes in the target zero-sequence current can also provide key clues for detecting and locating grounding grid corrosion faults. The target zero-sequence voltage-zero-sequence current image is an image generated by using specific data processing and visualization techniques to collect the target zero-sequence voltage and target zero-sequence current data. This image can intuitively display the relationship between the two, such as phase difference, amplitude ratio, waveform characteristics, etc., which helps further analyze the nature and severity of the grounding grid corrosion fault.

[0063] In this step, while monitoring the grounding grid's zero-sequence voltage and current in real time, the computer can use pre-set algorithms and models to determine whether a corrosion fault has occurred. For example, the computer can use statistical methods to analyze historical zero-sequence voltage and current data and calculate statistical characteristics such as their mean and variance. If the currently collected zero-sequence voltage or current data exceeds the normal range by a certain multiple, such as the mean plus or minus three times the variance, a preliminary judgment is made that a corrosion fault may exist. Furthermore, the computer can use machine learning models, such as support vector machines and neural networks, to classify the zero-sequence voltage and current data. By pre-training the model with a large number of known normal and faulty samples, it can automatically identify the characteristic patterns of corrosion faults. In actual monitoring, the collected data is input into the model, which then outputs a judgment result. If the model determines a corrosion fault, the computer will execute subsequent actions.

[0064] After confirming a corrosion fault in the grounding grid, the computer equipment must synchronously collect the target zero-sequence voltage of the bus and the target zero-sequence current of each feeder. To achieve synchronous collection, the computer equipment can employ high-precision clock synchronization technology to ensure that all monitoring points collect data at the same time. For example, GPS-synchronized clocks or network time protocol servers can be used to synchronize the time of monitoring devices connected to the bus and each feeder. To collect the target zero-sequence voltage of the bus, the computer equipment connects to voltage transformers or voltage sensors installed on the bus to obtain high-precision zero-sequence voltage signals. These sensors convert high-voltage signals into low-voltage signals suitable for processing by the computer equipment. Hardware devices such as data acquisition cards convert the analog signals into digital signals and transmit them to the computer equipment. To collect the target zero-sequence current of each feeder, the computer equipment connects to current transformers or current sensors installed on each feeder. These sensors similarly convert high-current signals into low-current signals and transmit them to the computer equipment via the data acquisition equipment. During the collection process, the computer equipment labels and distinguishes the zero-sequence current data for each feeder to ensure accurate mapping to the specific feeder.

[0065] After collecting the target zero-sequence voltage of the bus and the target zero-sequence current data for each feeder, the computer needs to generate a target zero-sequence voltage-current image for each feeder. This process includes multiple steps, including data preprocessing, feature extraction, and image generation. First, the collected data undergoes preprocessing, such as filtering to remove noise and normalization. Filtering can employ digital filtering algorithms, such as Butterworth filters and Chebyshev filters, to remove high-frequency noise and interference from the signal. Normalization converts data of different dimensions and amplitudes to a common scale, facilitating subsequent processing and comparison. Next, feature extraction is performed. The computer can extract features such as amplitude, phase, and frequency of the zero-sequence voltage and current. For example, parameters such as the effective value, peak value, and phase angle of the zero-sequence voltage and current can be calculated and used as the basis for image generation. Finally, a specific image generation algorithm is used to convert the extracted feature data into a visual image. A scatter plot or waveform graph can be plotted using a two-dimensional coordinate system, with the zero-sequence voltage as the horizontal axis and the zero-sequence current as the vertical axis. The differences between different features can also be further highlighted through color, line thickness, etc., making the image more intuitive and informative.

[0066] It's understandable that when a corrosion fault occurs in a grounding grid, synchronously collecting data from the bus and individual feeders can fully capture the electrical state changes of the entire grounding grid at the moment of the fault. The bus's target zero-sequence voltage reflects the overall voltage anomaly of the grounding grid, while the target zero-sequence current of each feeder indicates the current variation characteristics of each branch line. By generating a target zero-sequence voltage-zero-sequence current image from this data, the relationship between the two, such as phase difference and amplitude ratio, can be intuitively displayed. These relationship characteristics are important for accurately determining the type, severity, and specific location of the corrosion fault. Compared with analyzing zero-sequence voltage or zero-sequence current separately, this comprehensive analysis method can provide more comprehensive and accurate information, helping to accurately locate and assess corrosion faults, thereby improving the efficiency and reliability of grounding grid corrosion fault detection and better ensuring the safe operation of the power system.

[0067] S103: Input each target zero-sequence voltage-zero-sequence current image into a pre-built grounding grid corrosion fault detection model to obtain a fault detection result for each feeder. The grounding grid corrosion fault detection model includes a spatial attention residual learning block.

[0068] The grounding grid corrosion fault detection model is a model built using artificial intelligence or machine learning techniques, specifically designed to analyze and identify characteristic patterns of grounding grid corrosion faults. The model is trained using a large amount of historical data and known fault samples to learn the differences in the electrical characteristics of the grounding grid under normal and corrosion fault conditions. In practical applications, the model automatically extracts and analyzes features from the target zero-sequence voltage-zero-sequence current image and outputs fault detection results for each feeder, including information such as whether a corrosion fault has occurred and the severity of the fault. The spatial attention residual learning block is a key component of the grounding grid corrosion fault detection model, consisting of a spatial attention mechanism and a residual learning architecture. The spatial attention mechanism automatically focuses on key regions and important features in the target zero-sequence voltage-zero-sequence current image, suppressing the influence of irrelevant or interfering information, thereby highlighting areas highly correlated with corrosion faults and improving the model's sensitivity to fault characteristics. The residual learning architecture, by introducing residual connections, addresses the vanishing gradient problem in deep neural networks, enabling the model to more effectively learn complex feature representations in the image, enhancing the model's learning and generalization capabilities, and further improving the accuracy and stability of fault detection.

[0069] In this step, in the practical application scenario of ground grid corrosion fault detection, after the computer device generates the target zero-sequence voltage-zero-sequence current image, it needs to input it into a pre-built ground grid corrosion fault detection model. This process first involves format conversion and preprocessing of the image data to ensure that the image can be correctly received and processed by the model. The computer device adjusts the generated image data to the input format required by the model, for example, standardizing image parameters such as resolution, pixel depth, and color channels, and performs normalization to map pixel values to the numerical range required by the model. Next, the computer device transmits the preprocessed image data to the computing module or server where the ground grid corrosion fault detection model resides via a data transmission interface. During the transmission process, to ensure data integrity and accuracy, a reliable data transmission protocol, such as TCP / IP, can be used to ensure that image data is not lost or corrupted during transmission. Furthermore, to improve data transmission efficiency, especially when processing large amounts of image data, data compression technology can be used to reduce data volume and speed up transmission without losing critical information.

[0070] After receiving the target zero-sequence voltage and current image, the grounding grid corrosion fault detection model uses its internal spatial attention residual learning block to conduct in-depth image analysis and processing. The spatial attention mechanism first performs a global scan of the image to identify areas highly correlated with corrosion fault characteristics, such as those with dramatic zero-sequence voltage and current variations and areas with significant waveform distortion. By adjusting attention weights, the model focuses more computing resources on these key areas, meticulously extracting and analyzing their features while relatively ignoring background information or noise that contributes less to fault detection.

[0071] The residual learning block plays a crucial role in feature extraction. Through residual connections, it enables the model to more effectively utilize gradient information for backpropagation when learning complex nonlinear feature maps, avoiding the vanishing gradient problem that can hinder model training. Specifically, the residual learning block applies a series of convolution and activation operations to the input feature map, then adds the residuals of the resulting feature map to the original input feature map before passing it to the next layer. This structure enables the model to maintain robust learning capabilities even in deep networks, enabling it to more accurately capture subtle differences in the target zero-sequence voltage and zero-sequence current images, thereby improving the accuracy of corrosion fault detection.

[0072] After image analysis and processing, the grounding grid corrosion fault detection model uses the learned characteristic patterns and fault classification knowledge to perform fault detection on the target zero-sequence voltage-zero-sequence current image for each feeder. The model outputs a fault detection result for each feeder, typically in the form of a probability value or binary classification label indicating whether a corrosion fault has occurred on the feeder and the severity of the fault. These results are further processed and utilized by computer equipment, such as generating fault reports, triggering alarms, and guiding maintenance decisions. In one example, the grounding grid corrosion fault detection model inputs the ZSV-ZSC image of each feeder and outputs its identification results. Finally, the identification results for all feeders are obtained, and the faulty feeder is detected. A feeder with an identification result of "1" indicates a faulty feeder, indicating corrosion at that location. A feeder with an identification result of "0" indicates a healthy feeder, indicating no corrosion at that location.

[0073] It can be understood that by inputting the target zero-sequence voltage-zero-sequence current image into the grounding grid corrosion fault detection model, the rich electrical feature information in the image can be fully utilized to achieve a comprehensive analysis of the corrosion fault. The spatial attention mechanism enables the model to focus on key areas in the image, highlighting features closely related to the corrosion fault, effectively suppressing the influence of noise interference and other irrelevant factors, thereby improving the accuracy of fault detection. The residual learning block enhances the model's learning ability, enabling it to more effectively learn complex feature representations, further improving the reliability of detection results. Compared with traditional detection methods, this method does not require complex equipment and high maintenance costs. It only requires the deployment of the corresponding grounding grid corrosion fault detection model in computer equipment to achieve efficient detection of grounding grid corrosion faults. Furthermore, this method can quickly process large amounts of data, meeting the requirements of modern power systems for real-time monitoring and rapid response, and better ensuring the safe and stable operation of power systems.

[0074] In the above-described embodiment, corrosion faults are identified by monitoring zero-sequence voltage and zero-sequence current. Compared to traditional methods, this method can more sensitively capture subtle electrical signal changes in the early stages of corrosion, enabling early detection. When a corrosion fault is identified, a target zero-sequence voltage-zero-sequence current image is simultaneously acquired and generated, providing a foundation for subsequent precise analysis. The image is processed using a grounding grid corrosion fault detection model that includes a spatial attention residual learning block. The spatial attention mechanism focuses on key areas, highlighting corrosion fault characteristics and suppressing noise interference. The residual learning block enhances the model's learning capabilities, improving detection accuracy and stability, and reducing the risk of misdiagnosis or missed diagnosis. This method does not require complex equipment and only requires adding a data processing step to the existing monitoring system, simplifying the system and reducing costs and maintenance. When dealing with large-scale data and multi-region real-time monitoring, it can efficiently process image data, improve computing power and response speed, and meet the needs of modern power systems.

[0075] In one embodiment, the step of determining whether a grounding grid corrosion fault has occurred based on the zero-sequence voltage and the zero-sequence current includes:

[0076] When the zero-sequence voltage meets the preset zero-sequence voltage differential condition, the fault time of the grounding grid corrosion fault is determined;

[0077] Within the preset delay from the start of the fault time, if the zero-sequence voltage meets the preset zero-sequence voltage effective value condition, and the zero-sequence current meets the preset zero-sequence current differential condition and the preset zero-sequence current effective value condition, it is determined that a grounding grid fault has occurred.

[0078] The zero-sequence voltage differential condition refers to a pre-set threshold for zero-sequence voltage change used to determine whether a grounding grid corrosion fault has occurred. It defines the upper and lower limits of the zero-sequence voltage change within a certain time interval. When the actual monitored change exceeds this range, the computer device preliminarily determines that a corrosion fault may have occurred. This condition is based on the stability of the zero-sequence voltage during normal operation of the power system and the abnormal fluctuation patterns during corrosion faults. The fault time refers to the specific time at which the computer device determines that a grounding grid corrosion fault may have occurred, based on the preset zero-sequence voltage differential condition. This is determined through precise time synchronization and timestamping of monitoring data, providing an accurate time reference for subsequent fault analysis and location. The preset delay is a pre-set interval from the time of fault determination. During this time, the device continuously monitors the zero-sequence voltage and current to ensure they meet other preset fault judgment conditions. The preset delay takes into account the dynamic process of fault development in the power system and the stability and reliability of monitoring data. It provides an observation window for the device to collect more fault characteristic information and avoids misjudgments due to transient signal fluctuations or interference.

[0079] The zero-sequence voltage RMS condition is a pre-set threshold used to determine whether the zero-sequence voltage RMS value is abnormal. The zero-sequence voltage RMS value reflects the average energy level of the zero-sequence voltage over a period of time. When it exceeds a preset range, it may indicate that the grounding performance of the grounding grid is affected by corrosion. This condition is set based on the statistical characteristics of the zero-sequence voltage RMS value during normal power system operation and the variation pattern during corrosion faults. The zero-sequence current differential condition is a pre-set threshold used to determine whether the zero-sequence current variation is abnormal. It defines the upper and lower limits of the zero-sequence current variation within a certain time interval. When the actual monitored variation exceeds this range, the device deems the zero-sequence current of the grounding grid to have abnormal fluctuations. This condition is set based on the stability of the zero-sequence current during normal power system operation and the abnormal fluctuation pattern during corrosion faults. The zero-sequence current RMS condition is a pre-set threshold used to determine whether the zero-sequence current RMS value is abnormal. The zero-sequence current RMS value reflects the average energy level of the zero-sequence current over a period of time. When it exceeds a preset range, it may indicate that the grounding performance of the grounding grid has been affected by corrosion. This condition is set based on the statistical characteristics of the zero-sequence current RMS value during normal power system operation and the variation pattern during corrosion faults.

[0080] Specifically, the computer monitors the grounding grid's zero-sequence voltage in real time, continuously collecting data at preset intervals. For each new sampled value, the difference between it and the zero-sequence voltage value at the previous sampling moment is calculated using a simple subtraction operation. This difference is then compared with a preset threshold. If it exceeds an upper or lower limit, the preset zero-sequence voltage differential condition is considered met. To improve judgment accuracy, the device may pre-process the raw data using algorithms such as sliding average filtering or median filtering to remove noise.

[0081] When it is determined that the zero-sequence voltage meets the preset zero-sequence voltage differential condition, the computer device immediately records the current system time as the fault time. The device relies on an internal high-precision clock chip and is calibrated with the power system time synchronization device to ensure the accuracy of the fault time record. The determination of the fault time provides an accurate time reference for subsequent fault analysis and is used to generate fault reports and alarm information. After determining the fault time, the computer device starts the timer and enters the preset delay phase. During this period, the device continuously monitors the relevant parameters of the zero-sequence voltage and zero-sequence current, collects data at a preset sampling frequency, and processes and analyzes it in real time. To ensure data integrity and continuity, the device may use data structures such as buffers or queues to store monitoring data for subsequent comprehensive analysis.

[0082] Within a preset delay, the computer calculates the effective value of the collected zero-sequence voltage data. This is achieved through mathematical operations such as squaring, integration, and square root extraction. Each sampled value is first squared, then integrated and summed within a preset time window, and finally the square root is taken to obtain the effective value. The effective value is compared with a preset threshold. If it exceeds the range, the preset zero-sequence voltage effective value condition is considered met. The computer analyzes the collected zero-sequence current data. First, the zero-sequence current differential is calculated to determine whether the preset zero-sequence current differential condition is met. This process is similar to determining the zero-sequence voltage differential condition. Simultaneously, the effective value of the zero-sequence current is calculated using a similar method as the zero-sequence voltage effective value calculation. This effective value is compared with the preset threshold to determine whether the condition is met. During this determination, the device also performs pre-processing operations such as filtering on the raw zero-sequence current data to improve data accuracy and reliability. Within the preset delay, if the computer determines that the zero-sequence voltage meets the preset zero-sequence voltage effective value condition and the zero-sequence current meets both the preset zero-sequence current differential condition and the preset zero-sequence current effective value condition, the device integrates these results and ultimately determines that a corrosion fault has occurred in the grounding grid. The comprehensive judgment process is implemented through logical AND operations. Only when all relevant conditions are met will the final fault determination signal be output for subsequent operations such as triggering alarms, recording fault information, and generating fault reports.

[0083] In this embodiment, the fault time is determined by preliminarily judging the zero-sequence voltage differential condition, and abnormal changes when the corrosion fault occurs are captured in a timely manner to provide a time reference for subsequent confirmation. The differential and effective value conditions of the zero-sequence voltage RMS and the zero-sequence current are comprehensively judged within the preset delay to avoid misjudgment due to single signal fluctuations or interference. This comprehensive judgment method takes into account the multiple characteristics of the zero-sequence voltage and zero-sequence current when the corrosion fault occurs, making the detection results more comprehensive and accurate. Compared with the traditional single-condition judgment method, it can better adapt to the complex power system environment, reduce the risk of false alarms and missed alarms, and improve the safety and stability of power system operation. At the same time, the preset delay setting provides a time window for the full development and stable collection of fault characteristics, further ensuring the reliability of fault judgment.

[0084] In one embodiment, the formula corresponding to the preset zero-sequence voltage differential condition is:

[0085]

[0086] The formula corresponding to the preset zero-sequence voltage effective value condition is:

[0087]

[0088] The formula corresponding to the preset zero-sequence current differential condition is:

[0089]

[0090] The formula corresponding to the preset zero-sequence current effective value condition is:

[0091]

[0092] in, is the sampling zero sequence voltage, is the effective value of the sampled zero-sequence voltage, To calculate the differential threshold of the differential voltage, is the effective value threshold of the sampling zero-sequence voltage, is the sampled zero-sequence current of the nth feeder, To calculate the differential threshold of the differential current, is the effective value of the sampled zero-sequence current of the nth feeder, is the effective value threshold of the sampling zero-sequence current, is the failure time of grounding grid corrosion fault, The start time of fault feeder detection.

[0093] Specifically, the preset zero-sequence voltage differential condition indicates that at time and Whether the change of zero sequence voltage exceeds the threshold If it exceeds, it means that the zero-sequence voltage has changed significantly and a corrosion fault may have occurred. The preset zero-sequence voltage effective value condition means: in the time interval [ , ], whether the effective value of the zero sequence voltage exceeds the threshold If it exceeds, it means that the effective value of zero-sequence voltage is abnormal, and further confirms the corrosion fault. The preset zero-sequence current differential condition means: in the time interval [ , ], the zero sequence current of the nth feeder is and Whether the change between If it exceeds, it means that the zero-sequence current has changed significantly and a corrosion fault may have occurred. The preset zero-sequence current effective value condition means: in the time interval [ , ], whether the zero sequence current effective value of the nth feeder exceeds the threshold If it exceeds, it means that the effective value of the zero-sequence current is abnormal, and corrosion fault is further confirmed.

[0094] In this embodiment, by comprehensively determining the differential and effective value conditions of zero-sequence voltage and current, this method significantly improves the accuracy and reliability of grounding grid corrosion fault detection. After the fault time is determined, continuous monitoring within a preset delay can promptly capture fault characteristics and avoid misjudgments caused by transient interference. Furthermore, by combining the multi-dimensional characteristics of voltage and current, it comprehensively reflects the grounding grid status, effectively reducing the risk of false alarms and missed alarms, ensuring the safe and stable operation of the power system, and providing maintenance personnel with accurate and timely fault information, facilitating rapid response and resolution.

[0095] In one example, to detect feeder faults under high-resistance ground faults and avoid false operations caused by interference, a startup module is constructed based on the coordination of the ZSV and ZSC standards:

[0096]

[0097] The ZSV's differential criterion records the onset of a ground grid corrosion fault, while other criteria are used to confirm the ground grid corrosion fault. If any ZSC connected to the same busbar meets both the differential and effective value criteria, the ZSC-based criterion is considered satisfied. The ZSV's differential threshold can be set to 2-5V, and the ZSV's effective value threshold can be set to 15% of the rated voltage. The ZSC's differential threshold can be set to 1% of the rated current, and the ZSC's effective value threshold can be set to 4% of the rated current, and can be further reduced to identify extreme high-impedance faults (HIFs). Only when all criteria meet their thresholds within a set delay will the protection device initiate further faulted feeder detection, thereby distinguishing ground grid corrosion faults from other disturbances.

[0098] In one embodiment, the step of generating a target zero-sequence voltage-zero-sequence current image for each feeder includes:

[0099] The target zero-sequence voltage and each target zero-sequence current are normalized, and the normalized target zero-sequence voltage and each normalized target zero-sequence current are superimposed on the same image. After binarization of each image, the target zero-sequence voltage-zero-sequence current image of each feeder is obtained.

[0100] Among them, normalization is a data processing method used to convert data with different dimensions and different numerical ranges into the same scale range. In this application, the target zero-sequence voltage and target zero-sequence current are normalized to map their values into a preset range. This can eliminate the influence of dimensions and dimensions, make the data comparable, and facilitate subsequent image superposition and analysis. Superposition refers to the merging of multiple different data sets or images into the same coordinate system for comprehensive analysis and comparison. In this application, the normalized target zero-sequence voltage and each normalized target zero-sequence current image are superimposed on the same image, which can intuitively display the relationship between the two, such as phase difference, amplitude ratio, etc., which helps to further analyze the characteristics of grounding grid corrosion faults. Binarization is an image processing technology that converts a grayscale image into an image with only black and white pixel values. In this application, the superimposed image is binarized to divide the pixels in the image into foreground and background according to a preset threshold. Through binarization processing, the key features in the image can be highlighted and the image content can be simplified, making it easier for subsequent image recognition and analysis algorithms to extract fault feature information more efficiently.

[0101] Specifically, in the practical application of ground grid corrosion fault detection, after acquiring the target zero-sequence voltage and current data, the computer must first normalize this data. Normalization typically involves performing a linear transformation on each data point, mapping it to a preset interval, such as [0, 1]. This normalization process can convert target zero-sequence voltage and current data of different dimensions and numerical ranges to the same scale, providing a foundation for subsequent image overlay and analysis.

[0102] After normalization, the computer needs to overlay the normalized target zero-sequence voltage and each normalized target zero-sequence current image into a single image. Specifically, a two-dimensional coordinate system can be constructed with the target zero-sequence voltage as the horizontal axis and the target zero-sequence current as the vertical axis. Then, within this coordinate system, the normalized target zero-sequence voltage and the corresponding target zero-sequence current at each moment are plotted as a point, forming a scatter plot. To more clearly demonstrate the relationship between the two, these points can be connected to form a waveform graph. During the actual image overlay process, the computer can use image fusion technology to fuse the pixel values of the two datasets (normalized target zero-sequence voltage and target zero-sequence current). For example, the pixel values of the two datasets can be added together according to certain weights to generate a new fused image. The weight setting can be adjusted according to actual application requirements to highlight different features.

[0103] After obtaining the overlaid image, the computer needs to perform binarization processing to highlight features related to the ground grid corrosion fault. The key to binarization is determining an appropriate threshold. This threshold can be determined based on various methods, such as fixed threshold and adaptive threshold. In the fixed threshold method, a fixed threshold is set based on experience or prior knowledge; in the adaptive threshold method, the threshold is automatically adjusted based on the local characteristics of the image to adapt to different image conditions. During the binarization process, the computer will iterate over each pixel in the image and compare the pixel value with the threshold. If the pixel value is greater than the threshold, it is set to the foreground (white); otherwise, it is set to the background (black). This method can highlight key features in the overlaid image (such as abnormal phase relationships and amplitude ratios), simplify the image content, and facilitate more efficient extraction of fault feature information by subsequent image recognition and analysis algorithms.

[0104] After the normalization, overlay, and binarization processes described above, the computer generates target zero-sequence voltage-zero-sequence current images for each feeder. These images visually demonstrate the relationship between the target zero-sequence voltage and target zero-sequence current for each feeder when a ground grid corrosion fault occurs, highlighting the fault characteristics. The resulting images can be used for subsequent fault detection and location analysis, providing maintenance personnel with clear and concise fault information.

[0105] In this embodiment, for ground grid corrosion fault detection, the target zero-sequence voltage and each target zero-sequence current are normalized, and the normalized target zero-sequence voltage and each normalized target zero-sequence current are superimposed on the same image. Each image is then binarized to produce a target zero-sequence voltage-zero-sequence current image. This significantly improves the efficiency and accuracy of fault feature extraction. Normalization eliminates differences in dimensions and numerical ranges between different data, making the data comparable and providing a foundation for subsequent image superposition. Image superposition intuitively displays the relationship between the target zero-sequence voltage and target zero-sequence current, such as phase difference and amplitude ratio, facilitating the detection of abnormal features. Binarization further highlights key features and simplifies image content, facilitating more efficient fault information extraction by subsequent image recognition and analysis algorithms. This method more clearly presents the characteristics of ground grid corrosion faults, improves the accuracy and reliability of fault detection, reduces the risk of false positives and missed positives, and better ensures the safe and stable operation of the power system.

[0106] In one example, when a ground grid corrosion fault occurs, the first half-cycle ZSV of the busbar and local ZSC on each feeder can be sampled and collected. To implement the detection solution, the ZSV and local ZSC need to be processed and superimposed on the same graph to generate a ZSV-ZSC image for each feeder. The entire process of creating a ZSV-ZSC image is as follows: Figure 3 shown.

[0107] For ZSV on the bus, its amplitude is directly normalized to [-1, 1]:

[0108]

[0109] Where T is the period of ZSV, is the normalized ZSV.

[0110] Similarly, the ZSC of the nth feeder is also handled by the following formula.

[0111]

[0112] Where N is the number of feeders, Set to 0.7.

[0113] After the above ZSV and ZSC processing, the ZSV-ZSC image of each feeder can be drawn. At the same time, the 0th axis, 1th axis, and -1th axis are added to each image. In addition, each image is binarized and cropped to a size of 128×128.

[0114] In one embodiment, the formula corresponding to the spatial attention residual learning block is:

[0115]

[0116] Among them, x and y represent the input and output of the residual block respectively, represents the residual mapping to be learned by the stacked layer, represents the weight of the convolution kernel, represents the ReLu function, represents the hyperparameter, Represents a feature map with attention weights.

[0117] Specifically, first calculate , that is, the input x is extracted and transformed through operations such as convolutional layers to obtain a residual map. Add to the input x, that is , realizing residual connection, helps the gradient flow directly to the previous layer during the back propagation process, alleviating the gradient disappearance problem. and hyperparameters Multiplying together, we get , and then add it to the result of the residual connection, that is , so that the output feature map pays more attention to the important areas. Finally, the above results are activated by the ReLU function , and the final output feature map y is obtained. The activation function introduces nonlinearity, enabling the network to learn complex feature representations.

[0118] In this embodiment, the spatial attention residual learning block formula effectively improves the learning ability and feature expression ability of the convolutional neural network by combining residual connections and attention mechanisms. Residual connections help alleviate the gradient vanishing problem, allowing the network to be trained deeper and enhancing the model's expressive power. The attention mechanism enables the network to automatically focus on the task-critical areas in the input feature map, improving the ability to capture key information. At the same time, the introduction of hyperparameters provides flexible control over the degree of influence of the attention feature map, further optimizing the feature fusion effect. This structure not only improves the adaptability and stability of the model to complex scenes, but also enhances its robustness to noise and interference, thereby significantly improving the performance of the model in various visual tasks.

[0119] As an example, in image recognition, different regions of an image should have different weights, with more attention paid to the meaningful parts. Since ZSC decays over time, it is necessary to focus on the initial failure time region of the generated ZSV-ZSC image. Figure 4 A ZSV-ZSC image is shown, which is different from the image Compared with the region, However, due to the variety of fault conditions, it is impossible to determine the best Value, if CNN only focuses on The features of the region may hinder image recognition. Therefore, the recognition of ZSV-ZSC images requires a CNN model with strong attention ability.

[0120] Deep convolutional neural networks (DCNNs) have been widely used for image recognition. As network depth increases, DCNNs can achieve state-of-the-art performance in early visual recognition tasks. However, deep networks are more difficult to train, and detection accuracy can saturate or even rapidly degrade. This phenomenon is known as the degradation problem.

[0121] To avoid the degradation problem that occurs as the network depth increases, residual learning is widely used in DCNN. A common DCNN block consists of several stacked convolutional layers, which fit the desired underlying mapping denoted as H(x). In residual learning, instead of directly approximating H(x), the stacked layers learn a residual mapping The residual block is as follows Figure 5(a), which is defined as:

[0122]

[0123] In order to improve the representation ability of DCNN, attention mechanisms are usually embedded in convolutional layers. For image recognition, especially ZSV-ZSC image recognition, it is very important to tell DCNNs where to focus. Therefore, a spatial attention module (SAM) is proposed to emphasize the meaningful features of the image. Given a feature map As input, the spatial information of the feature map is aggregated through two pooling operations: maximum pooling and average pooling. They represent maximum pooling, average pooling, and average pooling, respectively. and average pooling features . Then, the spatial attention formula as the output of SAM is:

[0124]

[0125] in, represents the sigmoid function, Represents the learned weights of the feature map M.

[0126] The feature map with attention weights can be obtained as follows:

[0127]

[0128] in Represents element-wise multiplication.

[0129] In order to embed residual learning and spatial attention learning simultaneously, this paper proposes a spatial attention residual learning (SARL) module, such as Figure 5 (b) As shown in Figure 2, it can be seen that identity mapping and SAM are used simultaneously in the proposed SARL module. Finally, the output of the SARL module can be calculated as:

[0130]

[0131] Among them, α is a hyperparameter and is set to 0.5.

[0132] In one embodiment, the grounding grid corrosion fault detection model consists of a convolutional layer, a maximum pooling layer, two spatial attention residual learning blocks, a global average pooling layer, and a bidirectional fully connected classification layer.

[0133] like Figure 6As shown in the figure, the convolutional layer, as the initial layer of the model, extracts local features from the input zero-sequence voltage and current images through convolution operations. The convolutional layer automatically learns important features from the data, providing a foundation for subsequent classification and recognition tasks. Following the convolutional layer, the max pooling layer reduces the data dimensionality through downsampling, lowering computational complexity while retaining the most important feature information. This max pooling layer helps improve the model's robustness and adaptability to data variations. The spatial attention residual learning block combines residual connections with the attention mechanism to enhance the model's focus on and learning of key feature regions. The first spatial attention residual learning block further extracts and enhances features, while the second spatial attention residual learning block performs further feature learning and optimization based on this. The global average pooling layer performs a global average operation on the feature map, taking the global average of each feature channel as the final feature representation. This helps further reduce data dimensionality while retaining important global feature information. As the final layer of the model, the bidirectional fully connected classification layer performs classification based on the previously extracted and learned features. This bidirectional fully connected classification layer comprehensively considers both forward and backward feature information, improving classification accuracy and reliability.

[0134] In this embodiment, the grounding grid corrosion fault detection model automatically extracts and retains important features from the input data through convolutional and maximum pooling layers, reducing data dimensionality and improving computational efficiency. The spatial attention residual learning block, combined with residual connections and an attention mechanism, further enhances the model's focus on and learning capabilities for key feature regions, resulting in richer and more accurate feature representation. Maximum and global average pooling layers improve the model's adaptability and robustness to data changes, reducing the risk of overfitting. The bidirectional fully connected classification layer integrates forward and reverse feature information to enhance classification accuracy and reliability. The entire model structure works synergistically, achieving efficient computation and rational resource utilization, making it suitable for practical power system applications. It can significantly improve the performance of grounding grid corrosion fault detection and ensure the safe and stable operation of power systems.

[0135] In one embodiment, the method further comprises:

[0136] Parallel computing technology is used to accelerate the generation of target zero-sequence voltage-zero-sequence current images for each feeder and to accelerate the training of grounding grid corrosion fault detection models.

[0137] Parallel computing is a method of executing computing tasks by simultaneously using multiple computing resources (such as multi-core CPUs and GPUs). It breaks down complex computing problems into multiple subtasks that can be processed simultaneously on different computing units, greatly speeding up the calculation.

[0138] Specifically, in ground grid corrosion fault detection, computer equipment needs to process zero-sequence voltage and current data from multiple feeders. To accelerate the generation of the target zero-sequence voltage-zero-sequence current image, parallel computing technology is used. Specifically, the data from different feeders is distributed to multiple computing units for parallel processing. First, the zero-sequence voltage and current data of each feeder are normalized. Then, the normalized data are superimposed on the same image, and finally, the superimposed image is binarized. Through parallel computing technology, the image generation tasks for multiple feeders can be carried out simultaneously, significantly reducing the total image generation time.

[0139] Parallel computing technology also plays a crucial role in training grounding grid corrosion fault detection models. The training process involves numerous computational tasks, which can be broken down and assigned to multiple computing units for parallel processing. For example, in a convolutional layer, different convolution kernels can be assigned to different computing units. In a fully connected layer, computations for different neurons can also be performed in parallel. Furthermore, the training dataset can be divided into multiple subsets and assigned to different computing units for parallel training. During the actual parallel training process, the computer equipment needs to synchronize and update the model parameters. At the end of each training iteration, each computing unit aggregates the calculated gradients to the master computing unit, which then updates the model parameters based on the aggregated gradients. The master then distributes the updated parameters to each computing unit, beginning the next training iteration.

[0140] In this embodiment, parallel computing technology is used to accelerate the generation of target zero-sequence voltage-zero-sequence current images and the training of ground grid corrosion fault detection models during ground grid corrosion fault detection, significantly improving the efficiency and performance of the entire detection process. By processing data from multiple feeders in parallel, image generation time can be significantly shortened, making fault detection more timely and rapid. Furthermore, during the model training phase, parallel computing can fully utilize computing resources, accelerate model convergence, and improve model training efficiency. Furthermore, the application of parallel computing technology can reduce the overall operating cost of the system and improve resource utilization.

[0141] The proposed grounding grid corrosion fault detection method, based on voltage and current waveform image recognition, has been widely used in multiple power distribution networks and achieved remarkable results. By applying this method to actual power systems, the authors comprehensively evaluated its accuracy, real-time performance, and adaptability, and compared it with traditional fault detection methods.

[0142] The proposed grounding grid corrosion fault detection method, based on voltage and current waveform image recognition, has been tested in real-world scenarios. By collecting zero-sequence voltage (ZSV) and zero-sequence current (ZSC) waveform data from the grounding grid and analyzing it using image recognition, it can effectively monitor faults caused by grounding grid corrosion.

[0143] In real-world experiments, the proposed method surpassed traditional detection methods in both fault detection accuracy and real-time performance. Multiple tests have verified that the image recognition-based grounding grid corrosion fault detection method has an accuracy rate of 98.6%, an improvement of approximately 18% over traditional digital signal processing (DSP) methods. Traditional methods have a high misdiagnosis rate, particularly in high-resistance grounding and complex electromagnetic environments. However, the proposed method can effectively identify subtle current changes and accurately determine corrosion faults.

[0144] By accelerating image generation and model training through parallel computing technology, this method can complete fault diagnosis for each feeder within 0.5 seconds, ensuring that a response can be made in the real-time monitoring system in the shortest possible time, thereby ensuring the stable operation of the power system.

[0145] This method is highly adaptable to different types of ground grid corrosion and can handle a wide range of current waveform variations. In practical applications, it can effectively identify and accurately judge both minor changes in the early stages of corrosion and waveform variations caused by extreme high-impedance faults (HIFs).

[0146] Compared to traditional fault detection methods based on direct current and voltage measurements, the proposed method demonstrates higher accuracy and faster diagnostic speed in detecting ground grid corrosion. Traditional methods rely on static current and voltage measurements, making it difficult to detect subtle current changes caused by ground grid corrosion. However, the proposed method, through image recognition technology, extracts more effective information from ZSV-ZSC waveform images, achieving high-precision and efficient fault detection.

[0147] The following describes the grounding grid corrosion fault detection device provided by the embodiment of the present application. The grounding grid corrosion fault detection device described below and the grounding grid corrosion fault detection method described above can be used for reference. Figure 7 As shown, the present application provides a grounding grid corrosion fault detection device, the device comprising:

[0148] The voltage and current monitoring module 201 is used to monitor the zero-sequence voltage and zero-sequence current of the grounding grid;

[0149] An image generation module 202 is configured to synchronously collect a target zero-sequence voltage of a bus in the grounding grid and a target zero-sequence current of each feeder when a corrosion fault is determined to have occurred in the grounding grid based on the zero-sequence voltage and the zero-sequence current, and generate a target zero-sequence voltage-zero-sequence current image of each feeder;

[0150] The fault detection result determination module 203 is used to input each target zero-sequence voltage-zero-sequence current image into a pre-built grounding grid corrosion fault detection model to obtain a fault detection result for each feeder. The grounding grid corrosion fault detection model includes a spatial attention residual learning block.

[0151] In one embodiment, the image generation module 202 includes:

[0152] a fault time determination unit, for determining a fault time of a corrosion fault occurring in the grounding grid when the zero-sequence voltage satisfies a preset zero-sequence voltage differential condition;

[0153] The corrosion fault determination unit is used to determine that a grounding grid fault occurs if the zero-sequence voltage meets a preset zero-sequence voltage effective value condition and the zero-sequence current meets a preset zero-sequence current differential condition and a preset zero-sequence current effective value condition within a preset delay from the start of the fault time.

[0154] In one embodiment, the formula corresponding to the preset zero-sequence voltage differential condition is:

[0155]

[0156] The formula corresponding to the preset zero-sequence voltage effective value condition is:

[0157]

[0158] The formula corresponding to the preset zero-sequence current differential condition is:

[0159]

[0160] The formula corresponding to the preset zero-sequence current effective value condition is:

[0161]

[0162] in, is the sampling zero sequence voltage, is the effective value of the sampled zero-sequence voltage, To calculate the differential threshold of the differential voltage, is the effective value threshold of the sampling zero-sequence voltage, is the sampled zero-sequence current of the nth feeder, To calculate the differential threshold of the differential current, is the effective value of the sampled zero-sequence current of the nth feeder, is the effective value threshold of the sampling zero-sequence current, is the failure time of grounding grid corrosion fault, The start time of fault feeder detection.

[0163] In one embodiment, the image generation module 202 includes:

[0164] The image generation unit is used to normalize the target zero-sequence voltage and each target zero-sequence current, and superimpose the normalized target zero-sequence voltage and each normalized target zero-sequence current in the same image. After binarization processing is performed on each image, the target zero-sequence voltage-zero-sequence current image of each feeder is obtained.

[0165] In one embodiment, the formula corresponding to the spatial attention residual learning block is:

[0166]

[0167] Among them, x and y represent the input and output of the residual block respectively, represents the residual mapping to be learned by the stacked layer, represents the weight of the convolution kernel, represents the ReLu function, represents the hyperparameter, Represents a feature map with attention weights.

[0168] In one embodiment, the grounding grid corrosion fault detection model consists of a convolutional layer, a maximum pooling layer, two spatial attention residual learning blocks, a global average pooling layer, and a bidirectional fully connected classification layer.

[0169] In one embodiment, the apparatus further comprises:

[0170] The parallel computing module is used to accelerate the generation of the target zero-sequence voltage-zero-sequence current image of each feeder and accelerate the training of the grounding grid corrosion fault detection model using parallel computing technology.

[0171] In one embodiment, the present application also provides a storage medium, which stores computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the grounding grid corrosion fault detection method as described in any of the above embodiments.

[0172] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the grounding grid corrosion fault detection method as described in any one of the above embodiments.

[0173] Schematically, as Figure 8 As shown, Figure 8 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 can be provided as a server. Figure 8 Computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions executable by processing component 302, such as application programs. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 302 is configured to execute the instructions to perform the grounding grid corrosion fault detection method according to any of the above-described embodiments.

[0174] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0175] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0176] Finally, it should be noted that, in this article, relational terms such as first and second are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. Without further restriction, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element. Herein, "one," "said," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. A plurality refers to at least two, such as 2, 3, 5, or 8. "And / or" includes any and all combinations of the relevant listed items.

[0177] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0178] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A grounding grid corrosion fault detection method, characterized in that: The method comprises: Monitor the zero-sequence voltage and zero-sequence current of the grounding grid; When it is determined that a corrosion fault occurs in the grounding grid according to the zero-sequence voltage and the zero-sequence current, a target zero-sequence voltage of a bus in the grounding grid and a target zero-sequence current of each feeder are synchronously collected, and a target zero-sequence voltage-zero-sequence current image of each feeder is generated; Each of the target zero-sequence voltage-zero-sequence current images is input into a pre-built grounding grid corrosion fault detection model to obtain a fault detection result for each feeder, wherein the grounding grid corrosion fault detection model includes a spatial attention residual learning block.

2. The grounding grid corrosion fault detection method according to claim 1, characterized in that: The step of determining that a fault occurs in the grounding grid according to the zero-sequence voltage and the zero-sequence current comprises: When the zero-sequence voltage satisfies a preset zero-sequence voltage differential condition, determining a fault time when a corrosion fault occurs in the grounding grid; Within a preset delay from the start of the fault time, if the zero-sequence voltage satisfies a preset zero-sequence voltage effective value condition, and the zero-sequence current satisfies a preset zero-sequence current differential condition and a preset zero-sequence current effective value condition, it is determined that a fault has occurred in the grounding grid.

3. The grounding grid corrosion fault detection method according to claim 2, characterized in that: The formula corresponding to the preset zero-sequence voltage differential condition is: The formula corresponding to the preset zero-sequence voltage effective value condition is: The formula corresponding to the preset zero-sequence current differential condition is: The formula corresponding to the preset zero-sequence current effective value condition is: in, is the sampling zero sequence voltage, is the effective value of the sampled zero-sequence voltage, To calculate the differential threshold of the differential voltage, is the effective value threshold of the sampling zero-sequence voltage, is the sampled zero-sequence current of the nth feeder, To calculate the differential threshold of the differential current, is the effective value of the sampled zero-sequence current of the nth feeder, is the effective value threshold of the sampling zero-sequence current, is the failure time of grounding grid corrosion fault, The start time of fault feeder detection.

4. The grounding grid corrosion fault detection method according to claim 1, characterized in that: The step of generating a target zero-sequence voltage-zero-sequence current image for each feeder includes: The target zero-sequence voltage and each target zero-sequence current are normalized, and the normalized target zero-sequence voltage and each normalized target zero-sequence current are superimposed on the same image. After binarization processing is performed on each of the images, the target zero-sequence voltage-zero-sequence current image of each feeder is obtained.

5. The grounding grid corrosion fault detection method according to claim 1, characterized in that: The formula corresponding to the spatial attention residual learning block is: Among them, x and y represent the input and output of the residual block respectively, represents the residual mapping to be learned by the stacked layer, represents the weight of the convolution kernel, represents the ReLu function, represents the hyperparameter, Represents a feature map with attention weights.

6. The grounding grid corrosion fault detection method according to claim 1, characterized in that: The grounding grid corrosion fault detection model consists of a convolutional layer, a maximum pooling layer, two spatial attention residual learning blocks, a global average pooling layer and a bidirectional fully connected classification layer in sequence.

7. The grounding grid corrosion fault detection method according to any one of claims 1 to 6, characterized in that: The method further comprises: Parallel computing technology is used to accelerate the generation of the target zero-sequence voltage-zero-sequence current image of each feeder and to accelerate the training of the grounding grid corrosion fault detection model.

8. A grounding grid corrosion fault detection device, characterized in that: The device comprises: Voltage and current monitoring module, used to monitor the zero-sequence voltage and zero-sequence current of the grounding grid; an image generation module, configured to, when it is determined based on the zero-sequence voltage and the zero-sequence current that a corrosion fault has occurred in the grounding grid, synchronously collect a target zero-sequence voltage of a bus in the grounding grid and a target zero-sequence current of each feeder, and generate a target zero-sequence voltage-zero-sequence current image of each feeder; A fault detection result determination module is used to input each of the target zero-sequence voltage-zero-sequence current images into a pre-built grounding grid corrosion fault detection model to obtain a fault detection result for each feeder, wherein the grounding grid corrosion fault detection model includes a spatial attention residual learning block.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the grounding grid corrosion fault detection method according to any one of claims 1 to 7.

10. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the grounding grid corrosion fault detection method according to any one of claims 1 to 7 are performed.

Images