Balance compensation method and related devices for partial discharge detection system

Abstract

This application relates to a balance compensation method and related apparatus for a partial discharge detection system. The method includes the following steps: inputting the image to be detected into a target detection model, determining the target bounding box based on target box detection; expanding the target bounding box by a preset pixel length, and setting the area within the target bounding box as the target region; obtaining the target object based on semantic segmentation, and setting the area within the target region that does not belong to the target object as the analysis region; inputting the portion of the image to be detected corresponding to the analysis region into a first pre-trained model, classifying the sensors into working sensors and abnormal sensors, and obtaining the abnormal classification corresponding to each abnormal sensor; obtaining the input signal based on the working sensor, preprocessing the input signal, inputting the preprocessed input signal and the abnormal type corresponding to the abnormal sensor into the pre-trained model, and obtaining the detection result. This application has the advantage of improving the accuracy of the detection result.

Description

Technical Field

[0001] This application relates to the field of electrical engineering, and in particular to a method and apparatus for balancing compensation of a partial discharge detection system. Background Technology

[0002] In the field of electrical engineering, partial discharge is a common phenomenon caused by excessively high local electric field strength in high-voltage electrical equipment, leading to electrical breakdown. Partial discharge can be classified into surface discharge, air gap discharge, and discharge in solid dielectrics. These discharge phenomena can cause equipment damage, insulation aging, or even equipment failure, thereby affecting the reliability and safety of power systems.

[0003] Detecting partial discharge is of great significance because it can promptly identify potential problems in high-voltage electrical equipment, improve equipment reliability and safety, prevent equipment failures, extend equipment lifespan, and reduce equipment maintenance costs and downtime. To achieve this goal, researchers and engineers have developed a variety of partial discharge detection technologies.

[0004] Current partial discharge detection technologies mainly include multi-sensor technology, which utilizes electromagnetic sensors (such as current transformers, voltage transformers, and electromagnetic field sensors), acoustic sensors (such as air-coupled acoustic sensors and contact acoustic sensors), and optical sensors (such as infrared thermal imagers and photomultiplier tubes) to capture and analyze partial discharge signals. By fusing data from different types of sensors, the accuracy and reliability of detection can be improved.

[0005] However, partial discharge detection technology still faces some challenges, such as poor contact and abnormal equipment posture, which can generate signal interference and lead to abnormal input signal acquisition. When these abnormal signals are treated as normal signals, they may affect the accuracy of the detection results. Summary of the Invention

[0006] To improve the accuracy of detection results, this application provides a balance compensation method and related apparatus for a partial discharge detection system.

[0007] Firstly, this application provides a balance compensation method for a partial discharge detection system, which adopts the following technical solution:

[0008] A method for balancing and compensating a partial discharge detection system includes the following steps:

[0009] S1. Input the image to be detected into the target detection model, and determine the target bounding box based on the target bounding box detection;

[0010] S2. Expand the target bounding box by a preset pixel length, and set the area within the target bounding box as the target area;

[0011] The target object is obtained based on semantic segmentation, and the region within the target area that does not belong to the target object is set as the analysis region.

[0012] S3. Input the portion of the image to be detected corresponding to the analysis region into the first pre-trained model, classify the sensor regions into working sensors and abnormal sensors, and obtain the abnormal classification corresponding to each abnormal sensor;

[0013] S4. Based on the input signal acquired by the working sensor, the input signal is preprocessed, wherein the input signal includes partial discharge signal and noise signal;

[0014] S5. Input the preprocessed input signal and the anomaly type corresponding to the anomaly sensor into the pre-trained model and obtain the detection result.

[0015] By employing the above technical solution, sensor regions can be effectively identified and located through the setting of the target detection model and the target bounding box. Expanding the target bounding box and setting the target region helps to more accurately capture the edge regions of the sensor. Obtaining the target object based on semantic segmentation and setting the region within the target region that does not belong to the target object as the analysis region helps to distinguish the sensor from its surrounding environment, thereby making subsequent analysis more accurate. By inputting the portion of the image to be detected corresponding to the analysis region into the first pre-trained model, working sensors and abnormal sensors can be identified, and the anomaly classification corresponding to each abnormal sensor can be obtained. This can reduce the influence of abnormal sensors on the detection results to a certain extent. Based on the input signal acquired by the working sensor, and by preprocessing the input signal, it helps to eliminate the interference of noise signals on the detection of partial discharge signals. Inputting the preprocessed input signal and the anomaly type corresponding to the abnormal sensor into the pre-trained model can yield the detection results. This method can further improve the accuracy and reliability of partial discharge signal detection.

[0016] Optionally, S1 includes the following steps:

[0017] S11. Input the image to be detected into the target detection model;

[0018] S12. Output the target bounding box;

[0019] S13. Determine the target bounding box based on the target bounding box detection.

[0020] By employing the above technical solution, the target object (e.g., an insulator) in the image can be automatically identified by inputting the image to be detected into the target detection model. This helps improve detection efficiency, reduce manual intervention, lower the possibility of misjudgment, and provide accurate target information for subsequent steps. The output target bounding box can accurately locate the position and range of the target object in the image, facilitating the extraction of the region of interest in subsequent steps. In addition, the target bounding box also helps to remove irrelevant information in the image, thereby improving the accuracy of subsequent analysis and recognition. Determining the target bounding box based on target bounding box detection can accurately identify and locate target objects (such as insulators of power facilities) in the image, providing reliable input data for subsequent steps. This helps improve the accuracy and robustness of the entire detection process, thereby improving the performance of power facility anomaly detection.

[0021] Optionally, S13 includes the following steps:

[0022] S131. Input the preprocessed image to be detected into the selected target detection model to perform target detection;

[0023] S132. From the output of the object detection model, select high-confidence target boxes based on the confidence threshold;

[0024] S133. Use the selected target boxes as the target bounding boxes.

[0025] By employing the above technical solutions, inputting preprocessed images into the target detection model helps improve the model's detection accuracy for target objects (such as specific components of power facilities). Preprocessing can include image enhancement, denoising, and other operations, which help reduce the impact of noise and environmental factors on the detection results, improving detection accuracy and robustness. Filtering high-confidence target boxes based on confidence thresholds helps reduce false positives and false negatives. This will improve the accuracy of target detection, providing more reliable input data for subsequent steps, thereby improving the performance of the entire anomaly detection process. Using the filtered target boxes as target bounding boxes can accurately pinpoint the location and extent of the target to be detected (such as specific components of power facilities) in the image. This helps subsequent steps to more accurately analyze and identify target regions, improving the accuracy and efficiency of the entire anomaly detection process.

[0026] Optionally, S2 includes the following steps:

[0027] S21. Calculate the coordinates of the expanded bounding box based on the coordinate information of the target bounding box and the preset pixel length;

[0028] S22. Based on the expanded bounding box coordinates, the corresponding regions in the image to be detected are divided into target regions and analysis regions;

[0029] S23. Store the information of the target area and the analysis area in a data structure.

[0030] By employing the above technical solution, determining the preset pixel length helps in appropriately expanding the target bounding box. A suitable preset pixel length ensures analysis accuracy while avoiding excessively large areas that increase computational complexity and analysis time. Expanding the target bounding box allows for the inclusion of environmental information surrounding the target object to a certain extent, aiding in further analysis and identification. This expansion improves the accuracy of target recognition and analysis, and also facilitates a more comprehensive evaluation of the target object's state in subsequent steps. Setting a target region helps separate areas related to the target object from other irrelevant areas in the image, allowing subsequent analysis and identification processes to focus more on the target object and its surrounding environment. This contributes to improving the accuracy and efficiency of the entire anomaly detection process, reducing the risk of false positives and false negatives.

[0031] Optionally, S3 includes the following steps:

[0032] S31. Select a semantic segmentation model;

[0033] S32. Input the target region into the semantic segmentation model;

[0034] S33. Obtain the segmentation result of the target object;

[0035] S34. Set the analysis area.

[0036] By employing the above technical solutions and selecting a suitable semantic segmentation model, the accuracy and efficiency of target object segmentation can be improved. A high-performance semantic segmentation model can more accurately identify and segment target objects, thus providing more reliable data for subsequent analysis and recognition processes. Inputting the target region into the semantic segmentation model allows the model to focus on processing image regions related to the target object, thereby improving segmentation accuracy and efficiency. Obtaining the segmentation results of the target object clearly identifies the target object and its boundaries, facilitating further analysis and recognition of the target object's state. The segmentation results provide accurate basic data for subsequent analysis and recognition processes, thereby improving the accuracy and efficiency of the entire anomaly detection process. Setting the analysis region allows for further filtering of image regions closely related to the target object based on the target region. Setting the analysis region helps subsequent steps to focus more on processing data related to the target object's state, thereby improving the accuracy and efficiency of the anomaly detection process.

[0037] Optionally, S4 includes the following steps:

[0038] S41. Select the first pre-trained model;

[0039] S42. Input the image of the analysis region into the first pre-trained model;

[0040] S43. Classification sensor area;

[0041] S44. Obtain the anomaly classification of the abnormal sensor.

[0042] By adopting the above technical solutions, selecting an appropriate first pre-trained model helps improve the accuracy and efficiency of sensor area classification. A high-performance pre-trained model can more accurately distinguish between working and abnormal sensors, thus providing more reliable data for subsequent analysis and diagnosis. Inputting the analysis area image into the first pre-trained model allows the model to focus on processing image regions related to sensor status, thereby improving classification accuracy and efficiency. This will help accurately identify abnormal sensors and reduce the risk of false positives and false negatives. Classifying sensor areas helps identify working and abnormal sensors, facilitating further analysis and diagnosis of anomalies. Sensor area classification results provide accurate basic data for subsequent anomaly detection and diagnosis, thereby improving the accuracy and efficiency of the entire anomaly identification process. Obtaining the anomaly classification of abnormal sensors can clarify the anomaly type, providing targeted information for subsequent anomaly handling and maintenance. Accurately identifying the anomaly type helps to take appropriate measures to solve the anomaly problem, thereby improving the operational stability and safety of power facilities.

[0043] Optionally, step S5 includes:

[0044] S51. Establish communication with the working sensor, wherein the working sensor is a UHF sensor located in a partial discharge risk area, and obstacles in the propagation path of the working sensor have been cleared to reduce signal attenuation and reflection;

[0045] S52. Improve the sampling rate and resolution of the working sensor, and increase signal synchronization;

[0046] S53. Perform real-time or offline noise reduction on the acquired signals and enhance the signals.

[0047] By employing the above technical solutions, establishing communication with the working sensor, and ensuring that obstacles in the propagation path are cleared, signal attenuation and reflection can be effectively reduced. This helps to acquire high-quality partial discharge signals, thereby improving the accuracy of detection and analysis. Increasing the sampling rate and resolution of the working sensor can capture more detailed signal characteristics, contributing to improved detection accuracy of partial discharge signals. Simultaneously, increasing signal synchronization ensures data consistency among multiple sensors, thereby improving the reliability and stability of the entire detection system. Real-time or offline denoising can eliminate noise components in the signal, reducing the risk of false positives and false negatives. Signal enhancement helps improve signal quality, making partial discharge signal characteristics more apparent, thus improving the accuracy of detection and analysis. This is crucial for ensuring the normal operation and safety of power facilities.

[0048] Secondly, the balance compensation device for a partial discharge detection system provided in this application adopts the following technical solution:

[0049] A balance compensation device for a partial discharge detection system, comprising:

[0050] The bounding box initialization module is used to input the image to be detected into the target detection model and determine the target bounding box based on the target bounding box detection.

[0051] The outer frame expansion module is used to expand the target outer frame by a preset pixel length and set the area within the target outer frame as the target area;

[0052] The semantic segmentation module is used to obtain the target object based on semantic segmentation and to set the regions within the target region that do not belong to the target object as the analysis region.

[0053] The classification module is used to input the part of the image to be detected corresponding to the analysis region into the first pre-trained model, classify the sensor region into working sensors and abnormal sensors, and obtain the abnormal classification corresponding to each abnormal sensor.

[0054] The preprocessing module is used to preprocess the input signal obtained from the working sensor, whereby the input signal includes partial discharge signal and noise signal.

[0055] The detection module is used to input the pre-processed input signal and the anomaly type corresponding to the anomaly sensor into the pre-trained model and obtain the detection result.

[0056] Thirdly, the computer device provided in this application adopts the following technical solution:

[0057] A computer device comprising:

[0058] One or more processors;

[0059] Memory;

[0060] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the balance compensation method of the partial discharge detection system described above.

[0061] Fourthly, the computer-readable storage medium provided in this application adopts the following technical solution:

[0062] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above.

[0063] The storage medium stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement: the balance compensation method of the partial discharge detection system described above. Attached Figure Description

[0064] Figure 1 A schematic diagram illustrating the application environment of the balance compensation method of the partial discharge detection system in one embodiment of the present invention is shown.

[0065] Figure 2 A flowchart illustrating the balance compensation method of a partial discharge detection system in one embodiment of the present invention is shown.

[0066] Figure 3 A schematic diagram of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0067] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application.

[0068] In the following description, numerous specific details are set forth for purposes of explanation in order to provide a thorough understanding of the inventive concept. As part of this specification, some of the accompanying drawings of this disclosure are block diagrams illustrating structures and devices to avoid complicating the disclosed principles. For clarity, not all features of the actual embodiment need to be described. Furthermore, the language used in this disclosure has been primarily chosen for readability and instructional purposes and may not have been chosen to define or limit the subject matter of the invention, thus requiring the necessary claims to determine such inventive subject matter. References to “an embodiment” or “an embodiment” in this disclosure mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment, and multiple references to “an embodiment” or “an embodiment” should not be construed as necessarily referring to the same embodiment.

[0069] Unless explicitly defined, the terms “a,” “an,” and “the” are not intended to refer to a singular entity, but rather to include a general category whose specific examples can be used for illustration. Therefore, the use of the terms “a” or “an” can mean any number of at least one, including “a,” “one or more,” “at least one,” and “one or more.” The term “or” means any of the options and any combination of the options, including all options unless explicitly indicated that the options are mutually exclusive. The phrase “at least one of” when combined with a list of items refers to a single item in the list or any combination of items in the list. The phrase does not require all items listed unless explicitly defined as such.

[0070] This application discloses a balance compensation method for a partial discharge detection system. The balance compensation method for a partial discharge detection system provided by this invention can be applied to, for example... Figure 1 In this application environment, the balance compensation method of the partial discharge detection system is applied to the balance compensation system of the partial discharge detection system. This system includes a client and a server, where the client communicates with the server via a network. The client, also known as the user terminal, refers to the program that provides local services to the client, corresponding to the server. Furthermore, the client can be a computer program, a smart device APP program, or a third-party app embedded in other APPs. The client can be installed on various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0071] In the power industry, partial discharge (PD) refers to the discharge phenomenon of local dielectrics in power equipment. Partial discharge detection is an important method for assessing the insulation condition of power equipment, which can effectively prevent faults and extend equipment life. In related technologies, partial discharge signals in the input signal that match the phase of the reference signal are typically detected by multiplying the input signal and the reference signal. This process is sometimes referred to as coherent signal detection.

[0072] For a specific example:

[0073] Suppose we have an input signal that contains a partial discharge signal and a noise signal:

[0074] Input signal = Partial discharge signal + Noise signal

[0075] There is also a reference signal, which is a known partial discharge signal:

[0076] Reference signal = known partial discharge signal

[0077] Phase-sensitive multiplication refers to multiplying the input signal with a reference signal:

[0078] Multiplication result = Input signal × Reference signal

[0079] When the partial discharge signal in the input signal matches the phase of the reference signal, the multiplication result will have a large output value. Conversely, when the noise signal in the input signal does not match the reference signal, the multiplication result will tend towards zero. In other words, phase-sensitive multiplication can highlight the partial discharge signal in the input signal that matches the reference signal, thereby improving the signal-to-noise ratio.

[0080] Next, the multiplication result is low-pass filtered to remove high-frequency noise components and retain the output signal that is phase-correlated with the reference signal. Based on the filtered output signal, it can be determined whether there is a partial discharge signal in the input signal.

[0081] This process helps suppress random noise interference in the input signal, improving the sensitivity and accuracy of partial discharge detection. However, it requires a reference signal beforehand, and in practical applications, obtaining a reference signal that matches the partial discharge signal in the input signal can be very difficult. Reference signals are usually obtained through experimental or theoretical analysis, but because partial discharge phenomena can vary depending on factors such as equipment type, operating conditions, discharge type, and discharge location, it is difficult to obtain an accurate and universal reference signal. Specifically, this can manifest in the following ways:

[0082] Partial discharge signal characteristics change: The frequency, amplitude, and phase of the partial discharge signal may change with the operating status of the equipment and environmental conditions. When the reference signal cannot adapt to these changes in real time, the detection performance of the phase-sensitive detection method may be affected.

[0083] Coexistence of multiple discharge types: In practical applications, multiple types of partial discharge phenomena may exist inside the device, and their signal characteristics may differ. In such cases, using a single reference signal may not meet the detection requirements.

[0084] Reference signal error: The acquisition of the reference signal may be affected by factors such as experimental conditions, equipment errors, and environmental interference, which can limit the accuracy and reliability of the reference signal. This will directly affect the detection performance of the phase-sensitive detection method.

[0085] Reference Figure 2 The balance compensation method of the partial discharge detection system includes the following steps S1-S6.

[0086] S1. Input the image to be detected into the target detection model, and determine the target bounding box based on the target bounding box detection.

[0087] Optionally, in one embodiment, S1 includes the following steps S11-S13.

[0088] S11. Input the image to be detected into the target detection model.

[0089] By setting the target detection model and the target bounding box, the sensor area can be effectively identified and located.

[0090] Optionally, in one embodiment, S11 includes the following steps S111-S112.

[0091] S111. Adjust the resolution and size of the image to be detected to suit the input requirements of the target detection model.

[0092] This example uses YOLOv5 for object detection, which requires an input size of 640x640 pixels. In this case, the resolution of the image to be detected needs to be adjusted to 640x640 pixels before being input into the YOLOv5 model. This can be achieved using an image processing library, such as Python's OpenCV library, by calling the `cv2.resize()` function.

[0093] For example, consider the scenario of partial discharge detection in a power transformer. The image to be detected might be a high-resolution image containing multiple sensors. To accommodate the input requirements of the selected target detection model, the original image needs to be resized to the size required by the model for accurate sensor identification and positioning.

[0094] S112. Normalize the image to be detected so that its pixel values ​​are distributed within a suitable range to improve the training and detection performance of the model.

[0095] Normalization typically maps the pixel values ​​of an image from their original range (e.g., 0-255) to a smaller range, such as 0-1 or -1 to 1. This can be achieved by dividing each pixel value of the image by 255 (for the 0-1 range) or subtracting 127.5 and then dividing by 127.5 (for the -1 to 1 range). This can be done using image processing libraries, such as Python's OpenCV library, via the `cv2.normalize()` function.

[0096] For example, in the same scenario of partial discharge detection of power transformers, normalization can distribute the pixel values ​​of the image to be detected within a suitable range. This can improve the training and detection performance of the model because normalization can reduce computation and training time, while also reducing the risk of optimization problems such as vanishing and exploding gradients.

[0097] S12. Output target bounding box.

[0098] Outputting the target bounding box can accurately locate the position and extent of the target object in the image, facilitating subsequent steps to extract the region of interest. Furthermore, the target bounding box helps to remove irrelevant information from the image, thereby improving the accuracy of subsequent analysis and recognition.

[0099] Optionally, in one embodiment, S12 includes S121-S122.

[0100] S121. Based on the actual application scenario and performance requirements, select a target detection model suitable for sensor detection, such as YOLO, Faster R-CNN, SSD, etc.

[0101] When applications require rapid sensor detection with high accuracy, the YOLO (You Only Look Once) model is a good choice due to its high accuracy and speed in real-time object detection tasks. Alternatively, other object detection models can be considered, such as Faster R-CNN (for high accuracy requirements) or SSD (for higher speed requirements). In the scenario of partial discharge detection in power transformers, it may be necessary to detect multiple sensors in a short period. In this case, a suitable object detection model can be selected based on the needs, as different models offer different trade-offs between speed and accuracy.

[0102] S122. Load the pre-trained object detection model weights to facilitate object detection on the image to be detected.

[0103] After selecting YOLO as the object detection model, pre-trained weights are loaded for detection. For example, pre-trained weight files suitable for YOLO can be downloaded from the internet. Then, deep learning frameworks (such as PyTorch or TensorFlow) can be used to load the weight files and apply them to the YOLO model.

[0104] S13. Determine the target bounding box based on the target bounding box detection.

[0105] Determining the bounding box of a target based on bounding box detection can accurately identify and locate target objects (such as insulators of power facilities) in an image, providing reliable input data for subsequent steps. This helps improve the accuracy and robustness of the entire detection process, thereby enhancing the performance of power facility anomaly detection.

[0106] Optionally, in one embodiment, S13 includes the following steps:

[0107] S131. Input the preprocessed image to be detected into the selected target detection model to perform target detection.

[0108] After selecting the YOLO model and loading pre-trained weights, the pre-processed image to be detected is input into the model for object detection. This can be achieved by calling the model's prediction or inference functions using a deep learning framework such as PyTorch or TensorFlow.

[0109] For example, in the scenario of partial discharge detection of power transformers, the target detection model needs to identify and locate sensors in the image. The pre-processed image to be detected is input into the model, and the model will output the detection results, including the bounding box, category, and confidence score.

[0110] S132. From the output of the target detection model, select target boxes with high confidence based on the confidence threshold.

[0111] After obtaining the object detection results output by the model, it is necessary to filter out the high-confidence bounding boxes based on a confidence threshold. For example, a confidence threshold of 0.5 can be set, meaning that only bounding boxes with a confidence level greater than 0.5 will be retained. This can be done using library functions in programming languages, such as Python's NumPy library, to filter out high-confidence bounding boxes.

[0112] For example, in the scenario of partial discharge detection of power transformers, screening high-confidence target boxes helps to eliminate false detections and improve the accuracy of detection results.

[0113] S133. Use the selected target boxes as the target bounding boxes.

[0114] After selecting high-confidence target boxes, these target boxes are used as target bounding boxes for region setting and analysis in subsequent steps.

[0115] S2. Expand the target bounding box by a preset pixel length, and set the area within the target bounding box as the target area.

[0116] Expanding the target bounding box and setting the target area helps to capture the edge areas of the sensor more accurately.

[0117] Optionally, in one embodiment, S2 includes S21-S23.

[0118] S21. Calculate the coordinates of the expanded bounding box based on the coordinate information of the target bounding box and the preset pixel length.

[0119] Calculate the coordinates of the expanded bounding box based on the coordinates of the target bounding box and the preset pixel length.

[0120] Example: Given the coordinates of the target bounding box (x1, y1, x2, y2) and a preset pixel length d, the coordinates of the expanded bounding box can be calculated using the following formula: (x1-d, y1-d, x2+d, y2+d). This calculation can be performed using library functions of a programming language, such as Python's NumPy library.

[0121] For example, in the scenario of partial discharge detection of power transformers, expanding the outer frame helps to capture partial discharge signals around the sensor more accurately, while avoiding the loss of partial discharge signals due to the target outer frame being too narrow.

[0122] S22. Based on the expanded bounding box coordinates, the corresponding regions in the image to be detected are divided into target regions and analysis regions.

[0123] After obtaining the coordinates of the expanded bounding box, the corresponding region in the image to be detected can be divided into the target region and the analysis region. The target region refers to the area containing the sensor, while the analysis region refers to the area within the expanded bounding box that does not belong to the target object. This can be achieved using image processing libraries, such as Python's OpenCV library, by using the `cv2.rectangle()` function to divide the region.

[0124] In the scenario of partial discharge detection in power transformers, by dividing the image to be detected into a target area and an analysis area, we can better focus on the sensor part of interest, while eliminating irrelevant background noise and improving the accuracy of the detection results.

[0125] S23. Store the information of the target area and the analysis area in a data structure for further processing and analysis in subsequent steps.

[0126] After defining the target and analysis regions, the information of these regions needs to be stored in a data structure for further processing and analysis in subsequent steps. Specifically, the coordinate information of the regions can be stored in a data structure within a programming language, such as a list or dictionary in Python. For example, in the scenario of partial discharge detection of power transformers, storing information about the target and analysis regions helps to quickly locate and process these regions in subsequent steps, thereby improving the efficiency of the detection process.

[0127] S3. Obtain the target object based on semantic segmentation, and set the region within the target area that does not belong to the target object as the analysis region.

[0128] Obtaining the target object based on semantic segmentation and setting the area within the target area that does not belong to the target object as the analysis area helps to distinguish the sensor from its surrounding environment, thereby making subsequent analysis more accurate.

[0129] Optionally, in one embodiment, S3 includes the following steps S31-S34.

[0130] S31. Select a semantic segmentation model.

[0131] Choose an appropriate semantic segmentation model based on the specific application scenario and object detection requirements. Common semantic segmentation models include U-Net and DeepLab. These models may perform differently in different tasks, so the selection should be based on the actual needs.

[0132] S32. Input the target region into the semantic segmentation model.

[0133] The target region obtained from step S2 is input into the selected semantic segmentation model. The input target region can be cropped and resized to fit the model input using an image processing library in a programming language (such as Python's OpenCV library).

[0134] S33. Obtain the segmentation result of the target object.

[0135] After inputting the target region into the semantic segmentation model, the segmentation result of the target object is obtained. The segmentation result is usually represented as a mask image of the same size as the input image, where the pixel value represents the probability that each pixel belongs to the target object. By applying a thresholding operation to the mask image, pixels can be classified as belonging to the target object or not.

[0136] S34. Set the analysis area.

[0137] Based on the target object segmentation result obtained in step S33, regions within the target area that do not belong to the target object are defined as analysis regions. Analysis regions can be obtained by applying a mask image of the target object to the target region image. In this way, the analysis region will include background information within the target area, but not the target object itself.

[0138] S4. Input the portion of the image to be detected corresponding to the analysis region into the first pre-trained model, classify the sensor regions into working sensors and abnormal sensors, and obtain the abnormal classification corresponding to each abnormal sensor.

[0139] By inputting the portion of the image to be detected corresponding to the analysis region into the first pre-trained model, the working and anomalous sensors can be identified, and the anomaly classification corresponding to each anomalous sensor can be obtained. This can reduce the impact of anomalous sensors on the detection results to some extent.

[0140] Optionally, in one embodiment, S4 includes S41-S44.

[0141] S41. Select the first pre-trained model.

[0142] Choose an appropriate first pre-trained model based on the actual application scenario and the requirements of sensor anomaly detection. Common deep learning models include Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). These models may perform differently in different tasks, so the selection needs to be based on the specific requirements.

[0143] S42. Input the image of the analysis region into the first pre-trained model.

[0144] The analysis region image obtained from step S3 is input into the selected first pre-trained model. The input analysis region image can be cropped and resized to fit the model input using an image processing library in a programming language (such as Python's OpenCV library).

[0145] S43. Classification sensor area.

[0146] After inputting the image of the analysis area into the first pre-trained model, the sensor regions are classified into working sensors and abnormal sensors based on the model's output. The output typically contains the probability that each sensor region belongs to an abnormal or working state. By applying a threshold operation to the probability results, the sensors can be distinguished as working sensors or abnormal sensors.

[0147] S44. Obtain the anomaly classification of the abnormal sensor.

[0148] Based on the output of the first pre-trained model, an anomaly classification is assigned to each anomalous sensor. The anomaly classification can be various possible fault types, such as poor contact, abnormal device posture, etc. Anomaly classification helps to further understand and resolve sensor anomaly problems.

[0149] S5. Based on the input signal acquired by the working sensor, the input signal is preprocessed, wherein the input signal includes partial discharge signal and noise signal.

[0150] The input signal is acquired by the working sensor and preprocessed, which helps to eliminate the interference of noise signal on the detection of partial discharge signal.

[0151] Optionally, in one embodiment, S5 includes the following steps S51-S53.

[0152] S51. Establish communication with the working sensor, wherein the working sensor is a UHF sensor located in a partial discharge risk area, and obstacles in the propagation path of the working sensor have been cleared to reduce signal attenuation and reflection.

[0153] In GIS equipment, partial discharge mainly occurs in critical components such as switches, busbars, and connectors. UHF sensors can be installed on the housing of these components or placed inside the equipment to be closer to the source of the partial discharge. When deploying sensors, obstacles and reflective surfaces in the signal propagation path should be avoided as much as possible. Properly configuring the location and number of sensors helps improve signal reception quality. At the same time, avoid placing sensors near obstacles in the signal propagation path to reduce the adverse effects of signal attenuation and reflection on signal quality.

[0154] S52. Improve the sampling rate and resolution of the working sensor and increase signal synchronization.

[0155] Using a higher sampling rate and resolution allows for more accurate capture of signal details.

[0156] S53. Perform real-time or offline noise reduction on the acquired signals and enhance the signals.

[0157] This can improve the signal-to-noise ratio of the signal, making feature extraction more accurate.

[0158] S6. Input the preprocessed input signal and the anomaly type corresponding to the anomaly sensor into the pre-trained model and obtain the detection result.

[0159] In practical applications, predetermined partial discharge signals can be collected in advance as preprocessing input signals to extract features for training machine learning models. Feature extraction is the process of converting the original signal into representative feature values. These features may include time-domain features (such as peak value, mean, standard deviation, etc.), frequency-domain features (such as spectral features, energy distribution, etc.), and time-frequency features (such as wavelet analysis, short-time Fourier transform, etc.).

[0160] Next, a suitable machine learning algorithm needs to be selected to train the model. In different implementations, various classifiers can be tried, such as Support Vector Machines (SVM), neural networks, random forests, k-nearest neighbors, etc. Through methods such as cross-validation, the performance of different models can be evaluated and compared to select the best model.

[0161] Once the model is trained, it can be applied to actual partial discharge signal detection. When a new input signal enters the system, the model can automatically identify and classify the partial discharge signal without relying on a pre-acquired reference signal. This overcomes the difficulty of obtaining reference signals to some extent, thereby improving detection accuracy and reliability.

[0162] It is worth noting that machine learning and artificial intelligence methods typically require large amounts of training data to achieve good performance. Therefore, in practical applications, it may be necessary to continuously update and expand the dataset to adapt to the needs of different devices and scenarios.

[0163] Specifically, in one embodiment, S6 may include the following steps S61-S68.

[0164] S61. Collect partial discharge signals and related information and generate a dataset, wherein the related information includes temperature detection signals, humidity detection signals, GIS equipment operating load level information, GIS equipment model information, sensor signal information and equipment working duration information corresponding to each partial discharge signal, as well as abnormal discharge signals and abnormal types corresponding to abnormal sensors.

[0165] Each dataset of partial discharge signals includes different types of partial discharges (e.g., surface discharge, surface discharge, and internal discharge) as well as noise interference. Features need to be extracted from these signals to train a machine learning model.

[0166] It is worth noting that in this embodiment, the dataset specifically includes information related to partial discharge signals and abnormal discharge signals to facilitate accurate classification of these signals. This is because different environmental factors can have varying effects on partial discharge signals. For example, temperature has a significant impact on the propagation and capture of discharge signals, primarily in the following aspects:

[0167] Impact on partial discharge: High temperatures may degrade the performance of insulating materials, increasing the probability of partial discharge. Simultaneously, temperature changes may cause thermal expansion or contraction of the internal structure of the equipment, thus affecting the intensity and distribution of partial discharge.

[0168] Impact on signal propagation: Temperature affects the propagation speed and attenuation characteristics of electromagnetic waves. In partial discharge detection, this means that temperature changes can lead to variations in signal propagation time and attenuation, thus affecting the signal strength received by the sensor.

[0169] Impact on sensor performance: Temperature changes can affect sensor performance. For example, some sensors may experience drift at high temperatures, leading to changes in signal amplitude and phase. Furthermore, the noise characteristics of the sensor may also be affected by temperature.

[0170] Impact on Feature Vectors: Temperature affects the generation, propagation, and sensor performance of partial discharge signals, potentially causing changes in the extracted feature values. For example, the peak value, spectral characteristics, and statistical features of partial discharge signals may differ at different temperatures. This can negatively impact the performance of machine learning models and reduce detection accuracy.

[0171] To address this issue, temperature can be incorporated as an additional feature into the feature vector. This allows machine learning models to more accurately classify signals at different temperatures based on temperature characteristics. Simultaneously, the training dataset should include samples at different temperatures to help the model learn the impact of temperature on signal features. Furthermore, during the model evaluation phase, data at different temperatures can be analyzed to ensure the model has good generalization ability under various temperature conditions.

[0172] In addition, abnormal discharge signals can also be processed based on the known abnormality type to obtain effective internal information.

[0173] S62. Perform baseline drift elimination, noise filtering, and normalization on the data within the dataset.

[0174] To achieve higher recognition accuracy, the dataset needs to be preprocessed, such as by removing baseline drift, filtering, and normalizing. Preprocessing can reduce data redundancy and noise interference, improving the effectiveness of subsequent feature extraction and model training.

[0175] S63. Extract features from the data within the dataset.

[0176] Multiple features can be extracted from partial discharge signals, including time-domain, frequency-domain, and time-frequency-domain features.

[0177] S64. Select a subset of features with high discriminative power from the extracted features. This can be done using methods such as correlation coefficient, mutual information, principal component analysis (PCA), and / or linear discriminant analysis (LDA).

[0178] From the features extracted above, a subset of features with high discriminative power for classification is selected. This can be achieved through different feature selection methods, such as correlation coefficient, mutual information, principal component analysis (PCA), and linear discriminant analysis (LDA). The purpose of feature selection is to reduce feature dimensionality, decrease redundant information, and retain information useful for classification.

[0179] S65. Construct a feature matrix based on the selected feature combination, where each row represents a discharge signal sample and each column corresponds to a feature value.

[0180] The selected features are combined into a feature matrix, where each row represents a partial discharge signal sample and each column corresponds to a feature value.

[0181] S66. Assign corresponding category labels to the feature matrix to indicate the type of partial discharge or abnormal discharge.

[0182] Each sample is assigned a class label representing its type of partial discharge. The feature matrix and class labels will serve as input data for the machine learning model. It's important to note that the class labels here include both normal and abnormal categories.

[0183] S67. Train the machine learning model to use as input class labels and feature matrices.

[0184] Using the constructed feature matrix and class labels, select an appropriate machine learning algorithm (such as support vector machine, neural network, random forest, etc.) to train the classification model. The model's performance can be evaluated using methods such as cross-validation to select the optimal model.

[0185] S68. Input the actual partial discharge signal based on the trained model and output the detection result.

[0186] The trained model is then applied to actual partial discharge signal detection. For new input signals, the model can automatically identify and classify partial discharge signals or abnormal discharge types without relying on pre-acquired reference signals.

[0187] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0188] In one embodiment, a balance compensation device for a partial discharge detection system is provided, which corresponds one-to-one with the balance compensation method for the partial discharge detection system described in the above embodiments. The balance compensation device for the partial discharge detection system includes a bounding box initialization module, a bounding box expansion module, a semantic segmentation module, a classification module, a preprocessing module, and a detection module. Detailed descriptions of each functional module are as follows:

[0189] The bounding box initialization module is used to input the image to be detected into the target detection model and determine the target bounding box based on the target bounding box detection.

[0190] The outer frame expansion module is used to expand the target outer frame by a preset pixel length and set the area within the target outer frame as the target area;

[0191] The semantic segmentation module is used to obtain the target object based on semantic segmentation and to set the regions within the target region that do not belong to the target object as the analysis region.

[0192] The classification module is used to input the part of the image to be detected corresponding to the analysis region into the first pre-trained model, classify the sensor region into working sensors and abnormal sensors, and obtain the abnormal classification corresponding to each abnormal sensor.

[0193] The preprocessing module is used to preprocess the input signal obtained from the working sensor, whereby the input signal includes partial discharge signal and noise signal.

[0194] The detection module is used to input the pre-processed input signal and the anomaly type corresponding to the anomaly sensor into the pre-trained model and obtain the detection result.

[0195] Specific limitations regarding the balancing compensation device of the partial discharge detection system can be found in the limitations of the balancing compensation method for the partial discharge detection system described above, and will not be repeated here. Each module in the balancing compensation device of the aforementioned partial discharge detection system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the corresponding operations of each module.

[0196] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database contains data related to the balancing compensation method for the partial discharge detection system. The network interface is used for communication with external terminals via a network connection. When the processor executes the computer program, it implements a balancing compensation method for the partial discharge detection system.

[0197] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the balance compensation method of the partial discharge detection system described in the above embodiment, for example... Figure 2 S1 to S6 are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit of the balancing compensation device of the partial discharge detection system in the above embodiments. To avoid repetition, these will not be described again here.

[0198] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the balance compensation method of the partial discharge detection system described in the above embodiment, for example... Figure 2 S1 to S6 are shown. Alternatively, when the computer program is executed by the processor, it implements the functions of each module / unit in the balance compensation device of the partial discharge detection system in the above-described device embodiments. To avoid repetition, further details are omitted here.

[0199] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments of this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0200] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0201] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method of balance compensation for partial discharge detection systems, characterized in that, The method comprises the following steps: S1. inputting an image to be detected into a target detection model, and determining a target bounding box based on target box detection; S2. expanding the target bounding box by a preset pixel length, and setting a region within the target bounding box as a target region; S3. obtaining a target object based on semantic segmentation, and setting a region within the target region that does not belong to the target object as an analysis region; S4. inputting a part of the image to be detected corresponding to the analysis region into a first pre-trained model, classifying sensor regions into working sensors and abnormal sensors, and obtaining an abnormal classification corresponding to each abnormal sensor; S5. collecting input signals based on the working sensors, and pre-processing the input signals, wherein the input signals include partial discharge signals and noise signals; S6. inputting the pre-processed input signals and the abnormal types corresponding to the abnormal sensors into a pre-trained model, and obtaining a detection result; The S6 comprises the following sub-steps: S61. collecting partial discharge signals and related information and generating a data set, wherein the related information includes temperature detection signals, humidity detection signals, GIS device operating load level information, GIS device model information, sensor signal information and device working time information corresponding to each partial discharge signal, and abnormal discharge signals and abnormal types corresponding to the abnormal sensors; S62. performing baseline drift elimination, noise filtering and normalization on the data in the data set; S63. performing feature extraction on the data in the data set; S64. selecting a feature subset with high discriminability from the extracted features; S65. constructing a feature matrix based on the selected feature combination, wherein each row represents a discharge signal sample and each column corresponds to a feature value; S66. assigning corresponding class labels to the feature matrix to represent the partial discharge type or abnormal discharge type to which it belongs; S67. training a machine learning model for inputting class labels and feature matrices; S68. inputting actual partial discharge signals based on the trained model and outputting detection results.

2. The balance compensation method of partial discharge detection system according to claim 1, characterized in that, The S1 comprises the following steps: S11. inputting an image to be detected into a target detection model; S12. outputting a target bounding box; S13. determining a target bounding box based on target box detection.

3. The balance compensation method of partial discharge detection system according to claim 2, characterized in that, The S13 comprises the following steps: S131. inputting a pre-processed image to be detected into a selected target detection model for target detection; S132. selecting high-confidence target boxes from the output results of the target detection model based on a confidence threshold; S133. selecting the selected target boxes as target bounding boxes.

4. The balance compensation method of partial discharge detection system according to claim 3, characterized in that, The S2 comprises the following steps: S21. determining a preset pixel length; S22. performing an expansion operation on the target bounding box; S23. setting a target region.

5. The balance compensation method of partial discharge detection system according to claim 4, wherein, The S3 comprises the following steps: S31. selecting a semantic segmentation model; S32. inputting the target region into the semantic segmentation model; S33. obtaining a segmentation result of the target object; S34. setting an analysis region.

6. The balance compensation method of partial discharge detection system according to claim 5, wherein, The S4 comprises the following steps: S41. selecting a first pre-trained model; S42. inputting an analysis region image into the first pre-trained model; S43. classifying sensor regions; S44. Obtain the abnormal classification of the abnormal sensor.

7. The balance compensation method of partial discharge detection system according to claim 6, wherein, The step of S5 comprises: S51. Establish communication with the working sensor, wherein the working sensor is a UHF sensor arranged in the partial discharge risk area, and the obstacles on the propagation path of the working sensor have been cleared to reduce signal attenuation and reflection; S52. Increase the sampling rate and resolution of the working sensor, and increase signal synchronization; S53. Real-time or offline denoising and signal enhancement are performed on the collected signals.

8. A computer device, comprising: It comprises: one or more processors; a memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more application programs are configured to execute the balance compensation method of the partial discharge detection system according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, a code set or an instruction set, which is loaded and executed by the processor to realize the balance compensation method of the partial discharge detection system according to any one of claims 1 to 7.

Images