Method, system and equipment for diagnosing overheating fault of power equipment and medium

Abstract

The invention discloses a power equipment overheating fault diagnosis method, system, equipment and medium, and relates to the field of fault diagnosis. According to the method, a visible light image and an infrared image are enhanced through a preset double-flow generative adversarial network to obtain a target visible light image and a target infrared image, and a device semantic mask graph is generated according to the target visible light image; projecting temperature data in the target infrared image to each segmented power device according to the device semantic mask map to form a device temperature distribution map; determining static temperature characteristics and dynamic temperature characteristics of each power device according to the device temperature distribution diagram, and generating a space-time temperature evolution diagram according to the static temperature characteristics, the dynamic temperature characteristics and the physical connection relationship among the power devices; and predicting an overheating fault probability, a fault type and a potential fault source of each power device according to the space-time temperature evolution diagram and the electrical parameters. By implementing the technical scheme provided by the invention, the overheating fault prediction accuracy of the power equipment is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a diagnosis method, system, device and medium for overheat fault of power equipment. BACKGROUND

[0002] With the continuous development of the power system, the scale and complexity of power equipment are increasing, and ensuring the safe and stable operation of power equipment is crucial to the reliability of the entire power system. Power equipment generates heat during operation. Once an overheat fault occurs, it may cause equipment damage, power outages and other serious consequences, not only causing huge economic losses, but also affecting the normal production and life order of society. Therefore, accurately and timely diagnosing the overheat fault of power equipment has high practical significance, which can effectively reduce the operation risk of the power system and improve the stability and safety of power supply.

[0003] In traditional diagnosis of overheat fault of power equipment, a single detection method is usually used. One common way is to regularly arrange manual inspection, and the staff observes the appearance of the power equipment by eye to see if there are any abnormal signs of heat, such as discoloration, smoke, etc. Another way is to use a simple temperature measuring tool, such as a contact thermometer, to measure the surface temperature of the power equipment at a single point to determine whether the equipment has an overheat problem.

[0004] However, the traditional diagnosis method has obvious defects. The manual inspection method is not only inefficient, but also greatly influenced by the experience and subjective factors of the staff, making it difficult to achieve real-time and comprehensive monitoring, and it is even more difficult to find some overheat faults hidden inside the equipment. The simple temperature measuring tool can only obtain single-point temperature information and cannot reflect the overall temperature distribution of the power equipment, which may miss potential overheat fault points. SUMMARY

[0005] The present application provides a diagnosis method, system, device and medium for overheat fault of power equipment, which accurately predicts the overheat fault of power equipment, gives early warning and disposal suggestions, and improves the operation efficiency and safety of power equipment.

[0006] In a first aspect of the present application, a diagnosis method for overheat fault of power equipment is provided, applied to a power equipment diagnosis platform, the method comprising: Collecting electrical parameters, visible light images and infrared images of power equipment, enhancing the visible light images and the infrared images through a preset dual-flow generative adversarial network to obtain target visible light images and target infrared images, segmenting various types of power equipment on the target visible light images to generate device semantic mask images; According to the device semantic mask map, temperature data in the target infrared image is projected onto each segmented power device to form a device temperature distribution map; According to the device temperature distribution map, static temperature features and dynamic temperature features of each power device are determined, and a space-time temperature evolution map is generated according to the static temperature features, the dynamic temperature features and a physical connection relationship between power devices, the static temperature features including temperature mean value, maximum value and variance, and the dynamic temperature features including temperature rising rate, fluctuation frequency and evolution mode of temperature distribution; According to the space-time temperature evolution map and the electrical parameters, overheat failure probability, failure type, potential failure source, multi-level early warning information and disposal suggestions of each power device are predicted.

[0007] Optionally, the enhancing the visible light image and the infrared image by the preset dual-stream generative adversarial network to obtain a target visible light image and a target infrared image comprises: A dual-stream generative adversarial network is constructed, the dual-stream generative adversarial network comprising a first generator, a second generator and a shared discriminator; A visible light training image is input into the first generator to obtain a first visible light image, an infrared training image is input into the second generator to obtain a first infrared image, a joint score is output by the discriminator according to the first visible light image and the first infrared image, parameters of the first generator and the second generator are adjusted according to the joint score for next iteration training until a loss function converges or a maximum number of iterations is reached to obtain a trained preset dual-stream generative adversarial network; The visible light image and the infrared image are input into the preset dual-stream generative adversarial network, a target visible light image is obtained by the first generator, and a target infrared image is obtained by the second generator.

[0008] Optionally, the segmenting each type of power device on the target visible light image to generate a device semantic mask map comprises: The target visible light image is input into a preset semantic segmentation model to obtain a class label of each pixel point, and a preliminary device semantic mask map is generated according to the class label; According to the shape and size of the power device, regions in the preliminary device semantic mask map are merged or segmented, and the outline of the power device is determined to obtain an optimized device semantic mask map; The optimized device semantic mask map is superimposed with the target visible light image to verify the optimized device semantic mask map to generate a device semantic mask map.

[0009] Optionally, the projecting the temperature data in the target infrared image onto each segmented power equipment according to the equipment semantic mask map to form an equipment temperature distribution map comprises: determining a target region participating in matching according to the equipment semantic mask map, obtaining a first feature point of the target region from the target infrared image, and obtaining a second feature point of the target region from the visible light image; matching the first feature point and the second feature point to obtain a plurality of matching point pairs, screening the plurality of matching point pairs using a first preset algorithm, and obtaining an affine transformation matrix according to the screened target matching point pairs; pixel-by-pixel mapping the temperature data in the target infrared image to the corresponding power equipment in the target visible light image according to the affine transformation matrix to form an equipment temperature distribution map.

[0010] Optionally, the determining the static temperature feature and the dynamic temperature feature of each power equipment according to the equipment temperature distribution map comprises: calculating the mean, maximum value and variance of the temperature in the target region corresponding to each power equipment in the equipment temperature distribution map, and constructing a static temperature feature according to the mean, maximum value and variance; obtaining the equipment temperature distribution map under a continuous time sequence, determining the temperature rising rate, fluctuation frequency and evolution mode of temperature distribution in the time sequence, and constructing a dynamic temperature feature according to the temperature rising rate, fluctuation frequency and evolution mode; and combining the static temperature feature and the dynamic temperature feature according to the power equipment to form a temperature feature vector of the target power equipment.

[0011] Optionally, the generating a space-time temperature evolution map according to the static temperature feature, the dynamic temperature feature and the physical connection relationship between the power equipments comprises: regarding each independent power equipment in the power system as a node, associating a corresponding temperature feature vector with each node to form a node set; establishing an edge between the nodes based on the actual physical connection relationship of the power system, and assigning a weight to the edge according to the closeness, heat conduction possibility or influence strength of the connection to form an edge set; combining the node set and the edge set to form a basic space map, and adding a time sequence feature sequence to each node in the basic space map to form a space-time temperature evolution map.

[0012] Optionally, the predicting the overheating fault probability, fault type, potential fault source and multi-level warning information and disposal suggestion of each power equipment according to the space-time temperature evolution map and the electrical parameters comprises: extracting key features from the spatio-temporal temperature evolution map, using a second preset algorithm to perform pattern recognition on the key features to obtain a feature pattern related to the overheating fault, the key features including an occurrence frequency of temperature abnormal points, a rate change of temperature rise, and an abnormal aggregation pattern of temperature distribution; inputting the feature pattern into a preset prediction model to obtain an overheating fault probability and a predicted fault type of each power equipment; combining physical connection relationships between the power equipments and equipment operation state information to analyze potential sources of the overheating fault, generating multi-level early warning information according to the overheating fault probability and the predicted fault type, and generating corresponding disposal suggestions according to each level of early warning information.

[0013] In a second aspect of the present application, a diagnosis system for overheating fault of power equipment is provided, comprising a collection module, a projection module, a temperature module, and a prediction module, wherein: The collection module is configured to collect electrical parameters, visible light images, and infrared images of the power equipment, enhance the visible light images and the infrared images by using a preset dual-stream generative adversarial network to obtain target visible light images and target infrared images, segment various types of power equipment on the target visible light images to generate a device semantic mask map; The projection module is configured to project temperature data in the target infrared images onto each segmented power equipment according to the device semantic mask map to form a device temperature distribution map; The temperature module is configured to determine static temperature features and dynamic temperature features of each power equipment according to the device temperature distribution map, generate a spatio-temporal temperature evolution map according to the static temperature features, the dynamic temperature features, and physical connection relationships between the power equipments, the static temperature features including a temperature mean value, a maximum value, and a variance, and the dynamic temperature features including a temperature rise rate, a fluctuation frequency, and an evolution pattern of temperature distribution; The prediction module is configured to predict an overheating fault probability, a fault type, a potential fault source, multi-level early warning information, and disposal suggestions of each power equipment according to the spatio-temporal temperature evolution map and the electrical parameters.

[0014] In a third aspect of the present application, an electronic device is provided, comprising a processor, a memory, a user interface, and a network interface, the memory being configured to store instructions, the user interface and the network interface both being configured to communicate with other devices, and the processor being configured to execute the instructions stored in the memory to enable the electronic device to perform the method according to any one of the above.

[0015] In a fourth aspect of the present application, a computer-readable storage medium is provided, the computer-readable storage medium storing instructions, when the instructions are executed, performing the method according to any one of the above.

[0016] To sum up, the one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. By presetting a dual-flow generative adversarial network to enhance visible light images and infrared images, the clarity and contrast of the images can be significantly improved, noise and blur can be reduced, and power equipment in the images can be more clearly distinguished. By segmenting various power equipment on the target visible light image and generating equipment semantic mask images, accurate positioning and recognition of power equipment are achieved, providing an accurate basis for subsequent temperature data projection and fault diagnosis; 2. According to the equipment semantic mask image, the temperature data in the target infrared image is projected onto each segmented power equipment, ensuring accurate correspondence between the temperature data and the specific equipment and avoiding confusion and misjudgment of the temperature data. The formed equipment temperature distribution map can intuitively show the temperature distribution of each power equipment, facilitating the rapid identification of temperature abnormal areas by operation and maintenance personnel and providing an intuitive basis for fault diagnosis; 3. Static temperature features (such as temperature mean, maximum value, and variance) and dynamic temperature features (such as temperature rise rate, fluctuation frequency, and temperature distribution evolution mode) are extracted from the equipment temperature distribution map, comprehensively reflecting the temperature state and change trend of the power equipment. Combined with the physical connection relationship between power equipment, a spatiotemporal temperature evolution map is generated, which can capture the spatiotemporal dynamic relationship of temperature features between equipment, reveal potential temperature abnormal propagation paths and fault development rules; 4. Based on the spatiotemporal temperature evolution map, the overheat fault probability and fault type of each power equipment can be accurately predicted, providing timely fault warning for operation and maintenance personnel and helping to take measures in advance to prevent fault expansion. By analyzing the temperature abnormal patterns in the spatiotemporal temperature evolution map and the physical connection relationship between equipment, the potential fault source can be located, providing targeted guidance for fault troubleshooting and repair. The generated multi-level warning information and disposal suggestions can be processed according to the severity and urgency of the fault, ensuring that operation and maintenance personnel can take reasonable disposal measures according to the actual situation, improving the efficiency and effectiveness of fault handling. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow diagram of a power equipment overheat fault diagnosis method disclosed in an embodiment of the present application; Figure 2 is a module diagram of a power equipment overheat fault diagnosis system disclosed in an embodiment of the present application; Figure 3 is a structural diagram of an electronic device disclosed in an embodiment of the present application.

[0018] Explanation of reference signs: 201, acquisition module; 202, projection module; 203, temperature module; 204, prediction module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. DETAILED DESCRIPTION

[0019] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in combination with the drawings in the specification. Obviously, the described embodiments are only some of the embodiments of the present application, not all.

[0020] In the description of the embodiments of the present application, the words such as "for example" or "for instance" are used to represent an example, illustration or description. Any embodiment or design scheme described as "for example" or "for instance" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as "for example" or "for instance" are intended to present the relevant concept in a specific manner.

[0021] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are used for description purposes only and should not be interpreted as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. The terms "include", "contain", "have" and their variants mean "include but are not limited to", unless otherwise specifically emphasized.

[0022] The present embodiment discloses a diagnosis method for overheat fault of power equipment, Figure 1 is a flowchart of a diagnosis method for overheat fault of power equipment disclosed by the present embodiment, which is applied to a power equipment diagnosis platform, such as Figure 1 As shown in the figure, the method comprises the following steps: S101, acquiring electrical parameters, visible light images and infrared images of power equipment, enhancing the visible light images and the infrared images through a preset dual-flow generative adversarial network to obtain target visible light images and target infrared images, segmenting various types of power equipment on the target visible light images, and generating a device semantic mask map; S102, projecting temperature data in the target infrared images onto each segmented power equipment according to the device semantic mask map to form a device temperature distribution map; S103, determining static temperature characteristics and dynamic temperature characteristics of each power equipment according to the equipment temperature distribution map, generating a space-time temperature evolution map according to the static temperature characteristics, the dynamic temperature characteristics and the physical connection relationship between the power equipments, wherein the static temperature characteristics include temperature mean value, maximum value and variance, and the dynamic temperature characteristics include temperature rising rate, fluctuation frequency and evolution mode of temperature distribution; S104, predicting the overheat failure probability, failure type, potential failure source, multi-level warning information and disposal suggestion of each power equipment according to the space-time temperature evolution map and the electrical parameters.

[0023] Key electrical parameters in the operation process of power equipment, such as current size, voltage value, power factor, etc., can be obtained in real time by various sensors (such as current sensors, voltage sensors, etc.). These parameters can reflect the running state and load condition of the equipment, providing important electrical characteristic information for subsequent fault diagnosis. Visible light images can be obtained by taking appearance images of power equipment using a visible light camera. The visible light images can present information such as the physical form, surface condition, and surrounding environment of the equipment, for example, whether there is obvious damage, foreign matter attachment, etc. Infrared images can be collected by an infrared thermal imager. The infrared images can directly reflect the temperature distribution of the equipment surface, because different temperatures of objects will present different colors or brightness in the infrared images. The visible light images and infrared images are enhanced by a pre-set dual-flow generative adversarial network. For visible light images, the enhancement can improve the clarity and contrast of the images, reduce noise interference, and make the details of the equipment more clearly visible. For infrared images, the enhancement can highlight temperature differences and make the temperature distribution more obvious, which is convenient for subsequent temperature analysis and fault diagnosis. Image segmentation algorithms (such as semantic segmentation algorithms based on deep learning) are used to segment various power equipment in the target visible light image. The algorithm can identify the boundaries of different equipment in the image and divide the image into multiple regions, each region corresponding to a power equipment. A corresponding semantic mask is generated for each segmented equipment region. The semantic mask is a binary image, where the value of the equipment region is 1 and the value of the non-equipment region is 0. Through the semantic mask, the position and range of each equipment in the image can be clearly determined, providing accurate positioning information for projecting temperature data onto the equipment in the subsequent step. According to the equipment semantic mask, the temperature data in the target infrared image is accurately projected onto each segmented power equipment. Specifically, for each pixel point in the infrared image, its position information is used to determine whether it belongs to the target equipment region (determined by the semantic mask), and if so, the temperature value of the pixel point is assigned to the corresponding equipment. This projection method ensures the accurate correspondence between temperature data and specific equipment, avoiding the confusion of temperature data between different equipment, and provides a reliable data basis for subsequent analysis of the temperature condition of each equipment. The temperature data projected onto each equipment is presented in a visualized manner to form an equipment temperature distribution map. In the temperature distribution map, different colors or brightness represent different temperature values, and the temperature distribution of the equipment surface can be directly observed through the change of color or brightness. The operation and maintenance personnel can quickly identify the temperature abnormal area, such as local temperature being too high or temperature distribution being uneven, by observing the equipment temperature distribution map, which provides clues for further fault diagnosis.

[0024] The average temperature of each power equipment in a period of time or at a certain moment is calculated, and the temperature average can reflect the overall temperature level of the equipment, which is one of the important indicators for evaluating the running state of the equipment. The highest temperature value in the equipment area is found, and the temperature maximum can reflect the highest temperature point that may appear in the equipment, which is of great significance for judging whether the equipment exists overheating risk. The variance of temperature values in the equipment area is calculated, and the variance reflects the dispersion degree of temperature data. If the variance is large, it means that the surface temperature of the equipment is unevenly distributed, which may cause local overheating or poor heat dissipation. The trend of the temperature of the equipment over time is analyzed, and the temperature rise per unit time, i.e. the temperature rise rate, is calculated. If the temperature rise rate is too fast, it may indicate that the equipment has abnormal heating, such as short circuit, overload, etc. Through Fourier transform or other frequency spectrum analysis methods, the fluctuation frequency of the temperature data of the equipment is determined. Abnormal temperature fluctuation frequency may reflect the instability of the running state of the equipment, such as temperature fluctuation caused by load change, poor contact, etc. The temperature distribution map is observed over time, and the evolution mode of the temperature distribution is identified, such as heat spot diffusion, temperature gradient change, etc. These evolution modes can provide more in-depth information for fault diagnosis and help to judge the development trend of the fault. The physical connection relationship between the power equipment is considered, and the static temperature characteristics and dynamic temperature characteristics of each equipment are integrated in time and space dimensions. For example, if there is a circuit connection between two devices, their temperature changes may affect each other. The generated space-time temperature evolution map takes time as the horizontal axis and power equipment as the vertical axis, and represents the temperature characteristics of each equipment at different time points through visual elements such as color, size or shape. The graph can intuitively show the space-time dynamic change of the equipment temperature and reveal the mutual relationship of the temperature characteristics between the equipment and the fault propagation law.

[0025] The temperature feature information in the spatio-temporal temperature evolution map is fused and analyzed with the collected electrical parameters. For example, combined with the changes in electrical parameters such as current and voltage, and the temperature change trend of the equipment, a fault prediction model is established using machine learning or deep learning algorithms (such as neural networks, support vector machines, etc.). Through the fault prediction model, the probability of each power equipment overheating failure is calculated. The higher the fault probability, the greater the risk of equipment failure, which needs to be paid attention to. According to the features extracted from the spatio-temporal temperature evolution map and electrical parameters, match with known fault type features. Different fault types (such as short circuit, overload, poor contact, etc.) often have different temperature characteristics and electrical parameter change rules. Through feature matching and pattern recognition algorithms, the possible fault type of each power equipment is predicted, providing specific direction for subsequent fault handling. Combined with the physical connection relationship between power equipment and the temperature abnormal propagation path in the spatio-temporal temperature evolution map, the potential source of the overheating fault is analyzed. For example, if a device has a temperature anomaly and the devices connected to it also have a temperature rise, there may be a common fault source. Through correlation analysis and reasoning, the location of the potential fault source is determined, providing targeted guidance for fault troubleshooting and repair. According to the overheating fault probability and severity, the warning information is divided into multiple levels, such as low risk, medium risk, high risk, etc. Different levels of warning information correspond to different handling measures and response times. For each warning level and predicted fault type, generate corresponding disposal suggestions. Disposal suggestions may include immediate shutdown inspection, enhanced monitoring, adjustment of operating parameters, scheduling of maintenance plans, and other specific measures to help operations personnel handle faults in a timely and effective manner, ensuring the safe operation of power equipment.

[0026] Optionally, the enhancing the visible light image and the infrared image by the preset dual-stream generative adversarial network comprises: A dual-stream generative adversarial network is constructed, which includes a first generator, a second generator, and a shared discriminator; A visible light training image is input into the first generator to obtain a first visible light image, and an infrared training image is input into the second generator to obtain a first infrared image. The discriminator outputs a joint score according to the first visible light image and the first infrared image. The parameters of the first generator and the second generator are adjusted according to the joint score for next iteration training until the loss function converges or the maximum number of iterations is reached to obtain a trained preset dual-stream generative adversarial network; The visible light image and the infrared image are input into the preset dual-stream generative adversarial network, and the target visible light image is obtained by the first generator, and the target infrared image is obtained by the second generator.

[0027] The constructed dual-flow generative adversarial network includes a first generator, a second generator, and a shared discriminator. The design of this structure is to process visible light images and infrared images respectively, while realizing some correlation or contrast learning between the two images at the feature level through the shared discriminator. The first generator is specifically designed for generating visible light images, and its task is to convert the input visible light image (visible light training image in the training stage) into an enhanced visible light image (first visible light image in the training stage). The generator is usually composed of multiple layers of neural networks, such as convolutional layers, deconvolutional layers, batch normalization layers, etc. Through a series of transformations and feature extraction on the input image, an image with better quality is generated. The second generator corresponds to the processing of infrared images, and its task is to convert the input infrared training image into an enhanced first infrared image. Its network structure is similar to that of the first generator, but it is optimized for the characteristics of infrared images, such as focusing more on the extraction and processing of temperature distribution-related features. The shared discriminator is used to judge whether the input image is a real image (from the original data set) or a generated image by the generator. The shared discriminator can receive both the first visible light image and the first infrared image and output a joint score. This joint scoring mechanism enables the discriminator to consider the features of both images, prompting the generator to consider not only the quality of each image but also their correlation when generating images, such as in the power equipment scene, where both visible light images and infrared images should reflect the same state of the equipment. In the training stage, the visible light training image is input into the first generator, which processes it and outputs the first visible light image; at the same time, the infrared training image is input into the second generator, which outputs the first infrared image. The discriminator receives the first visible light image and the first infrared image and outputs a joint score based on their features. This score reflects the discriminator's judgment of the authenticity of the input image. If the score is high, it means that the discriminator considers the input image to be closer to the real image; if the score is low, it means that the image may be generated by the generator. Based on the joint score output by the discriminator, the parameters of the first generator and the second generator are adjusted. The purpose of adjusting the parameters is to make the generated images more realistic to deceive the discriminator. For example, if the discriminator's score for the first visible light image is low, it means that the quality of the visible light image generated by the first generator is not high, so the parameters of the first generator will be adjusted to generate better quality images in the next iteration. Through repeated input of images, generation of images, discriminator scoring, and parameter adjustment, the loss function converges (i.e., the performance of the generator and the discriminator reaches a balanced state, and the generated images can better deceive the discriminator, while the discriminator can also better distinguish between real images and generated images) or reaches the maximum number of iterations, thereby obtaining the trained preset dual-flow generative adversarial network. In practical applications, the collected visible light images and infrared images are input into the trained preset dual-flow generative adversarial network.At this time, the first generator processes the visible light image to generate a target visible light image, and the second generator processes the infrared image to generate a target infrared image. After the enhancement processing of the dual-flow generative adversarial network, the quality of the target visible light image and the target infrared image is improved. For the visible light image, the clarity and contrast may be improved, noise and blur may be reduced, and the details of the equipment may be more clearly visible; for the infrared image, the temperature difference may be highlighted, and the temperature distribution may be more obvious, facilitating subsequent temperature analysis and fault diagnosis.

[0028] In the process of image enhancement, the first generator tries to preserve the key features related to power equipment in the visible light image, such as the shape, color, and identification of the equipment. These features are crucial for accurately identifying the type and state of the equipment, and through the training of the generative adversarial network, it can be ensured that these important information is not lost while enhancing the image quality. The second generator preserves the key features reflecting the temperature distribution of the equipment in the infrared image, such as temperature hotspots and temperature gradients. These features are important evidence for determining whether the equipment has an overheating fault, and after enhancement processing, these features will be more prominent and accurate. The generative adversarial network can strengthen the feature expression of the equipment in the visible light image, making the features of the equipment in the image more prominent. For example, for some equipment with similar appearance but different models, the subtle differences between them can be better highlighted through enhancement processing, improving the accuracy of equipment identification. For infrared images, strengthening feature expression can make temperature abnormal areas more obvious, facilitating the timely detection of potential equipment faults. For example, in the target infrared image, a temperature rise area that is not very obvious may be strengthened, making it easier to detect.

[0029] Optionally, the segmenting each type of power equipment on the target visible light image to generate an equipment semantic mask image comprises: inputting the target visible light image into a preset semantic segmentation model to obtain a class label of each pixel point, and generating a preliminary equipment semantic mask image according to the class label; merging or segmenting regions in the preliminary equipment semantic mask image according to the shape and size of the power equipment, and determining the contour of the power equipment to obtain an optimized equipment semantic mask image; overlaying the optimized device semantic mask map with the target visible light image to validate the optimized device semantic mask map to generate a device semantic mask map.

[0030] The target visible light image processed by the dual-stream generative adversarial network is input into a preset semantic segmentation model. The target visible light image contains rich visual information such as the appearance, structure, and surrounding environment of the power equipment, which is an important basis for equipment segmentation. The preset semantic segmentation model is a deep learning model trained on a large number of labeled visible light image data. The model can classify each pixel in the input image and determine its class, such as whether it belongs to power equipment (such as transformers, switch cabinets, etc.) or background (such as sky, ground, wall, etc.). Through the processing of the model, a class label is assigned to each pixel in the image. According to the class label of each pixel, a preliminary equipment semantic mask image is generated. In the preliminary equipment semantic mask image, different classes of pixels are represented by different values or colors. For example, pixels belonging to power equipment may be labeled as 1, and pixels belonging to the background may be labeled as 0. In this way, a binary image with the same size as the target visible light image is formed, where the white area (or other specific color area) represents the power equipment, and the black area represents the background, preliminarily realizing the separation of power equipment and background. Due to factors such as image noise, lighting changes, and equipment occlusion, the preliminary equipment semantic mask image may not be accurate, for example, adjacent multiple devices may be segmented into one region, or one device may be segmented into multiple small regions. Therefore, the preliminary equipment semantic mask image needs to be optimized according to the actual shape and size characteristics of the power equipment. If there are multiple adjacent regions in the preliminary equipment semantic mask image that belong to the same device but are incorrectly segmented, these regions can be merged according to the shape and size information of the device. For example, by calculating the distance and shape similarity between adjacent regions, it can be determined whether they belong to the same device, and if so, they can be merged into one region. If the preliminary equipment semantic mask image incorrectly identifies a device as a large region, but the device may actually consist of multiple parts or overlap with other devices, the region can be segmented according to the structural characteristics and boundary information of the device. For example, edge detection algorithms can be used to determine the boundaries of the device, and the large region can be segmented into multiple small regions that conform to the actual shape of the device. After merging or segmenting the preliminary equipment semantic mask image, the contours of each power equipment are further determined. The precise contours of the equipment can be obtained through contour extraction algorithms such as the Canny edge detection algorithm combined with the contour tracking algorithm. The contour information can more accurately describe the shape and boundaries of the device, providing more detailed information for subsequent fault diagnosis and equipment management. The optimized equipment semantic mask image is superimposed on the target visible light image. The superimposition method is to overlay the semantic mask image on the target visible light image in a semi-transparent or other appropriate manner, so that the maintenance personnel can intuitively see the correspondence between the semantic mask image and the actual image.Through the superimposed display, the operation and maintenance personnel can intuitively check the accuracy of the optimized device semantic mask graph. For example, whether the contour of the device in the semantic mask graph is consistent with the actual contour of the device in the target visible light image, whether there is a segmentation error or a missing situation. If it is found that there is inaccuracy, the semantic segmentation model can be further adjusted or re-optimized. After verification and confirmation that the optimized device semantic mask graph is accurate, it is used as the final device semantic mask graph. The semantic mask graph can accurately identify the position and range of various types of power equipment in the target visible light image, and provide a reliable basis for subsequent temperature data projection, feature extraction, fault diagnosis and other steps.

[0031] The target visible light image is input into a pre-set semantic segmentation model. The model is trained by a large number of visible light images of power equipment and can learn the characteristics of different power equipment. By classifying each pixel point in the image, a class label is assigned to each pixel point, thereby preliminarily determining the device category to which each pixel point belongs, and generating a preliminary device semantic mask. This semantic segmentation method based on deep learning can accurately identify the approximate location and range of different devices in the image, providing a basis for subsequent optimization processing. According to the shape and size characteristics of the power equipment, the regions in the preliminary device semantic mask are merged or segmented. For example, for some devices with similar shapes but close distances, they may be merged according to the actual size and shape of the device; and for some multiple devices that are misjudged as one region, they may be segmented according to the boundary characteristics of the device. Through this optimization processing, the contour of the power equipment can be more accurately determined, eliminating possible errors and noise in the preliminary segmentation and further improving the accuracy of the segmentation. In actual power equipment scenarios, there may be problems such as device occlusion and uneven lighting, which can result in unsatisfactory preliminary segmentation results. Through the optimization processing based on shape and size, the segmentation robustness can be improved to better adapt to these complex scenarios. In the preliminary device semantic mask, each pixel point has a corresponding class label, which clearly identifies the power equipment category to which the pixel point belongs. This makes the generated device semantic mask not only contain spatial location information of the device, but also contain rich semantic information, which can clearly represent the distribution of different types of power equipment in the image. The optimized device semantic mask is superimposed on the target visible light image, and the accuracy of the device semantic mask can be intuitively verified through visual comparison. If the device contour in the mask is basically consistent with the actual device contour in the visible light image, it means that the segmentation result is accurate and reliable; if there is a significant mismatch, it can be found and further adjusted. This verification mechanism can timely detect errors in the segmentation process, such as device contour deviation, missegmentation or missed segmentation. Through timely feedback and adjustment, the quality of the device semantic mask can be further improved to ensure that it accurately reflects the actual situation of the power equipment in the image. The device semantic mask generated through the verification mechanism has higher credibility and can provide reliable data support for subsequent applications. In the operation and maintenance of power equipment, accurate and reliable device semantic masks can help operation and maintenance personnel more accurately understand the distribution and status of the equipment and make more reasonable maintenance plans and decisions.

[0032] Optionally, the projecting the temperature data in the target infrared image onto each segmented power equipment according to the device semantic mask to form a device temperature distribution map comprises: determining a target region involved in matching according to the device semantic mask map, acquiring first feature points of the target region from the target infrared image, and acquiring second feature points of the target region from the visible light image; matching the first feature points and the second feature points to obtain a plurality of matching point pairs, screening the plurality of matching point pairs using a first preset algorithm, and obtaining an affine transformation matrix according to the target matching point pairs after screening; mapping temperature data in the target infrared image to corresponding power equipment in the target visible light image pixel by pixel according to the affine transformation matrix to form a device temperature distribution map.

[0033] The device semantic mask map explicitly identifies the location and range of each power device in the target visible light image. By analyzing the semantic mask map, the target regions that need to be involved in matching can be determined, which correspond to each segmented power device. First feature points are extracted from the target region of the target infrared image. Feature points are pixel points in an image that have unique properties (such as corner points, edge points, etc.), which can represent the local features of the image. In the infrared image, the first feature points may be related to the temperature changes, edge profiles, etc. of the device. For example, the edge of the high-temperature area of the device, the boundary of different temperature areas, etc. can all become feature points. Second feature points are extracted from the same target region of the visible light image. The second feature points of the visible light image are usually related to the appearance shape, texture, edge, etc. of the device. For example, the contour edge of the device, the corner points of the identification pattern, etc. can all be feature points. A variety of feature point extraction algorithms can be used, such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), etc. These algorithms can stably extract feature points under different scales, rotations, and lighting conditions, providing a reliable basis for subsequent feature matching. The first feature points and the second feature points are matched to obtain a plurality of matching point pairs. The matching algorithm will judge whether two feature points match according to the similarity between their descriptors (feature vectors that describe the local region around the feature point). For example, the Euclidean distance between the descriptors of two feature points is calculated, the smaller the distance, the higher the similarity, and the more likely it is to become a matching point pair. Since there may be some mis-matched point pairs in the feature point matching process, a first preset algorithm is needed to screen the plurality of matching point pairs. Common screening algorithms include the RANSAC (Random Sample Consensus) algorithm. The RANSAC algorithm estimates a transformation model (here, an affine transformation model) by randomly selecting a set of matching point pairs, and then calculates whether other matching point pairs conform to the model, thereby screening out inliers (correct matching point pairs) that conform to the model and removing outliers (mis-matched point pairs). Affine transformation is a two-dimensional linear transformation that can perform translation, rotation, scaling, and shearing operations on an image. According to the screened target matching point pairs, an affine transformation matrix can be solved. The affine transformation matrix contains the transformation parameters from the target infrared image coordinate system to the target visible light image coordinate system. The affine transformation matrix can usually be solved by methods such as least squares. Assuming that the affine transformation matrix is A, and the matching point pair is (x1, y1) (a point in the infrared image) and (x2, y2) (a point in the visible light image), then Through multiple matching point pairs, a system of equations can be listed, and the matrix A is solved. According to the calculated affine transformation matrix, the temperature data in the target infrared image is mapped pixel by pixel to the corresponding power equipment in the target visible light image. For each pixel point (x1, y1) in the infrared image, the corresponding position (x2, y2) in the visible light image is calculated through the affine transformation matrix A, and then the temperature value of the pixel point is assigned to the device area at the corresponding position in the visible light image. In programming implementation, all pixel points of the infrared image can be traversed, affine transformation is performed on each pixel point, and the temperature value is filled into the corresponding position of the visible light image according to the transformed coordinates. After temperature data mapping, each power equipment region in the target visible light image corresponds to the corresponding temperature data, forming a device temperature distribution map. In the temperature distribution map, different colors or brightness can be used to represent different temperature values, for example, red represents high temperature area and blue represents low temperature area, and through the change of color or brightness, the temperature distribution of the device surface can be directly observed.

[0034] According to the device semantic mask, the target region participating in matching is determined, and the region of the corresponding power device in the target infrared image and the visible light image can be accurately located. The problem of inaccurate local device region matching caused by overall image matching is avoided, and the pertinence and accuracy of subsequent feature point matching are improved. Only the target region is used for feature point extraction and matching, and the interference of other irrelevant regions in the image is excluded, so that the matching process is more efficient and accurate, which helps to improve the accuracy of temperature data projection. After obtaining a plurality of matching point pairs by matching the first feature points obtained from the target infrared image and the second feature points obtained from the visible light image, the matching point pairs are screened using a first preset algorithm. The algorithm can remove incorrect matching point pairs and retain reliable matching point pairs, thereby improving the quality of the target matching point pairs. The target matching point pairs after screening are used to calculate an affine transformation matrix, and the high-quality matching point pairs can ensure that the calculated affine transformation matrix is more accurate, thereby ensuring the accuracy of temperature data mapping. According to the affine transformation matrix obtained from the target matching point pairs after screening, the geometric transformation relationship between the corresponding power device regions in the target infrared image and the target visible light image can be accurately described. Through the affine transformation matrix, the temperature data in the target infrared image can be pixel by pixel mapped to the corresponding power device in the target visible light image, and the accurate correspondence between the temperature data and the device position is realized. In actual application, the target infrared image and the visible light image may have geometric differences due to factors such as shooting angle, distance, device posture, etc. The affine transformation matrix can effectively compensate for these differences, ensuring that the temperature data can be accurately projected onto the device and improving the reliability of temperature data mapping. Pixel-by-pixel mapping of temperature data can retain the detailed information of temperature distribution in the infrared image, such as temperature gradient, local hot spots, etc. This enables the generated device temperature distribution map to accurately reflect the temperature changes on the surface of the device, providing rich information for subsequent fault diagnosis and analysis. Accurate temperature data mapping helps to improve the accuracy of overheating fault diagnosis and can more accurately identify the temperature abnormal region of the device and determine the location and severity of the fault.

[0035] Optionally, the determining, according to the device temperature distribution map, of the static temperature feature and the dynamic temperature feature of each power device comprises: calculating the mean, maximum value and variance of the temperature in the target region corresponding to each power device in the device temperature distribution map, and constructing a static temperature feature according to the mean, the maximum value and the variance; obtaining device temperature distribution maps in a continuous time sequence, determining the temperature rise rate, fluctuation frequency and evolution mode of temperature distribution in the time sequence, and constructing a dynamic temperature feature according to the temperature rise rate, the fluctuation frequency and the evolution mode; and combining the static temperature feature and the dynamic temperature feature according to the power device to form a temperature feature vector of the target power device.

[0036] In the device temperature distribution map, for each power device corresponding to the target area, the average value of all pixel temperature values in the region is calculated. The average value can reflect the overall temperature level of the device at the current time or in a certain static scene. For example, in a substation, by calculating the temperature average of a transformer, the average temperature condition of the transformer under normal operation or specific working conditions can be quickly understood, providing basic data for judging whether it is within the normal temperature range. Find the temperature value corresponding to the pixel point with the highest temperature in the target area. The maximum value usually represents the local overheating situation that may occur in the device, and is one of the important indicators for device fault warning. If the maximum value exceeds the safety temperature threshold of the device, it may mean that the device has local faults or abnormal operating conditions, which needs to be checked and handled in time. Variance is a measure of the dispersion of temperature values in the target area. The larger the variance, the more uneven the temperature distribution on the device surface, and there may be a large local temperature difference. This may be due to internal device faults, uneven heat dissipation, or other factors. By analyzing the variance, the stability of the device temperature distribution can be understood, which helps to judge whether the device has potential problems. Combine the calculated average, maximum and variance to form a static temperature feature vector for each power device. This vector can comprehensively describe the temperature characteristics of the device at the current time or in a static scene, providing important static information for subsequent fault diagnosis, state evaluation, etc. To obtain dynamic temperature features, the device temperature distribution map under continuous time series needs to be obtained. This can be achieved by regularly collecting infrared images of the device over a period of time and generating corresponding temperature distribution maps. Continuous time series data can reflect the change of device temperature over time, providing a basis for analyzing the dynamic characteristics of the device. Calculate the rate of change of device temperature over time, i.e. temperature rise rate. The temperature rise rate can reflect whether the heating of the device is normal. If the temperature rise rate is too fast, it may mean that the device has faults such as overload or short circuit, which needs to be handled in time. Analyze the fluctuation of device temperature in the time series and calculate its fluctuation frequency. Fluctuation frequency can reflect the stability of the device operating state. If the fluctuation frequency is too high, it may indicate that the device is disturbed by external interference or has internal instability factors, which needs to be further checked and analyzed. Observe the change trend of the device temperature distribution over time and summarize its evolution pattern. For example, the temperature distribution of some devices may show certain regular changes during normal operation, while the evolution pattern of the temperature distribution may change when a fault occurs. By analyzing the evolution pattern of the temperature distribution, potential faults of the device can be detected in advance. Combine dynamic indicators such as temperature rise rate, fluctuation frequency and evolution pattern of temperature distribution to form a dynamic temperature feature vector for each power device. This vector can describe the temperature change characteristics of the device over time, providing dynamic information for judging the operating state and predicting faults of the device.The static temperature feature vector and the dynamic temperature feature vector of each power equipment are combined to form a complete temperature feature vector of the target power equipment. This combined feature vector contains the static and dynamic temperature information of the equipment, which can more comprehensively reflect the temperature characteristics and operating state of the equipment.

[0037] The static temperature feature provides an important basis for preliminary judgment of the fault of the power equipment. The operating and maintenance personnel can monitor and analyze the static temperature feature of the equipment in real time according to historical data or preset thresholds. Once it is found that the mean, maximum value or variance of the temperature of a certain equipment exceeds the normal range, a warning can be issued in time to arrange further inspection and maintenance, and to avoid further expansion of the fault. Compared with the static temperature feature, the dynamic temperature feature can more timely reflect the abnormal change of the equipment. Through real-time monitoring and analysis of the temperature rise rate, fluctuation frequency and evolution mode, a warning signal can be issued before the fault is obviously manifested, so as to gain more processing time for the operating and maintenance personnel and reduce the impact of the fault on the power system. The static temperature feature and the dynamic temperature feature are combined according to the power equipment to form a temperature feature vector of the target power equipment, which can completely describe the temperature characteristics of the equipment. The temperature feature vector contains the temperature information of the equipment at different time points and in different states, which reflects not only the static temperature level of the equipment, but also the dynamic temperature change trend of the equipment. For example, for a large generator set, its temperature feature vector can include the static temperature mean, maximum value and variance of each key part under different working conditions, as well as the temperature rise rate, fluctuation frequency and evolution mode, etc., which provides rich data support for a comprehensive understanding of the temperature state of the generator set. The temperature feature vector provides strong support for accurate fault diagnosis and state assessment of power equipment. Based on the temperature feature vector, machine learning, deep learning and other algorithms can be used to establish fault diagnosis models and state assessment models. These models can automatically identify the normal state and fault state of the equipment through learning a large amount of historical data, and accurately judge the type, location and severity of the fault. At the same time, the temperature feature vector can also be used for health state assessment of the equipment to provide a scientific basis for maintenance and replacement of the equipment.

[0038] Optionally, the generating a space-time temperature evolution graph according to the static temperature feature, the dynamic temperature feature and the physical connection relationship between the power equipments comprises: Each independent power equipment in the power system is taken as a node, and a corresponding temperature feature vector is associated with each node to form a node set; Based on the actual physical connection relationship of the power system, edges are established between the nodes, and weights are assigned to the edges according to the closeness, heat conduction possibility or influence strength of the connection to form an edge set; combining the set of nodes and the set of edges to form a base space graph, and adding a time series of features in each node in the base space graph to form a spatiotemporal temperature evolution graph.

[0039] Each individual power equipment in the power system is considered as a node. For example, in a substation, transformers, circuit breakers, disconnectors, etc. can be considered as nodes respectively. This abstraction allows the description of the components of the power system in discrete units. A temperature feature vector is associated with each node, which is composed of static temperature features (such as mean, maximum, variance) and dynamic temperature features (such as temperature rise rate, fluctuation frequency, evolution pattern) extracted previously. In this way, each node carries rich temperature information, reflecting the temperature characteristics of the equipment at different times and states. All nodes with temperature feature vectors are combined to form a node set. This set contains all power equipment in the power system that needs to be monitored and analyzed, laying the foundation for subsequent construction of the spatio-temporal temperature evolution graph. Based on the actual physical connection relationship of the power system, edges are established between nodes. For example, in power lines, there is a physical connection between adjacent towers and conductors, so edges need to be established between the corresponding nodes. The establishment of such edges reflects the actual topology between devices, allowing the description of the spatial relationship between devices. The edges are assigned weights according to the degree of connection, the possibility of heat conduction, or the strength of influence. The degree of connection can be measured by the physical distance, contact area, etc. between devices; the possibility of heat conduction or the strength of influence is related to the material, thermal resistance, etc. of the device. For example, the edge weight between two directly connected devices with a large contact area may be larger, as the heat conduction between them is more direct and stronger. By assigning edge weights, the degree of temperature influence between devices can be more accurately described. All edges with weights are combined to form an edge set. This set describes the connection relationship and mutual influence degree between devices in the power system, providing information for the spatial structure of the spatio-temporal temperature evolution graph. The node set and edge set are combined to form a basic spatial graph. This graph is a static topology that reflects the spatial distribution and connection relationship of devices in the power system. In the basic spatial graph, each node represents a power device, and each edge represents the physical connection and heat conduction relationship between devices. Time series features are added to each node in the basic spatial graph. Time series features are composed of temperature feature vectors at different time points, which record the changes in device temperature over time. By adding time series features, the basic spatial graph becomes a spatio-temporal temperature evolution graph, which can reflect the spatial distribution and temporal changes of device temperature. The spatio-temporal temperature evolution graph provides a comprehensive perspective to observe and analyze the temperature state of the power system. By analyzing the temperature features of nodes and the weights of edges, the temperature mutual influence relationship between devices can be understood; by observing the time series features, the temperature change trend and abnormal conditions of devices can be found. This is of great significance for fault diagnosis, state assessment, and operation optimization of the power system.For example, when the temperature characteristics of a certain node are abnormal, the temperature changes of the surrounding nodes can be analyzed through the space-time temperature evolution graph to determine the propagation path and impact range of the fault, so that appropriate measures can be taken for processing.

[0040] Each independent power device in the power system is regarded as a node, and a corresponding temperature characteristic vector is associated with each node to form a node set. This way accurately constructs the network topology of the power device, clearly showing the location and relationship of each device in the power system. The temperature characteristic vector of each node contains static and dynamic temperature information of the device, which can accurately reflect the temperature characteristics of each device. This makes it possible to not only understand the spatial relationship between devices when analyzing the temperature state of the power system, but also to deeply understand the temperature changes of each device. Based on the actual physical connection relationship of the power system, edges are established between nodes, and weights are assigned to edges according to the closeness of the connection, the possibility of heat conduction or the strength of the influence to form an edge set. This quantitative method reasonably describes the connection relationship between devices, and the weight size reflects the degree of heat conduction or mutual influence between devices. The construction of the edge set makes it possible to accurately analyze the heat interaction between devices in a complex power system, providing strong support for in-depth study of the temperature distribution and variation law of the power system. In each node of the basic space graph, a time sequence feature sequence is added to integrate the time dimension information into the space structure to form a space-time temperature evolution graph. The time sequence feature sequence records the temperature characteristics of each device at different time points, and completely records the temperature change process of the device. Through the space-time temperature evolution graph, historical temperature data can be easily analyzed and reviewed to understand the temperature performance of the device in different time periods, providing important historical basis for fault diagnosis, device maintenance and performance evaluation. The space-time temperature evolution graph realizes the organic combination of space structure and time sequence information, making it possible to jointly analyze the temperature of the power device from the time and space dimensions. By observing the temperature changes of the nodes on the time axis and the mutual influence of the nodes due to heat conduction, the dynamic evolution law of the device temperature can be captured. The space-time temperature evolution graph contains the spatial position, temperature characteristics and time sequence change information of the power device, providing comprehensive data support for fault diagnosis.

[0041] Optionally, the method further comprises: extracting key features from the space-time temperature evolution graph, and using a second preset algorithm to perform pattern recognition on the key features to obtain a feature pattern related to the overheat fault, the key features including the frequency of temperature abnormal points, the rate change of temperature rise, and the abnormal aggregation pattern of temperature distribution; inputting the feature mode into a preset prediction model to obtain an overheat failure probability and a predicted failure type of each power equipment; combining a physical connection relationship between the power equipments and equipment operation state information to analyze potential sources causing the overheat failure; generating multi-level early warning information according to the overheat failure probability and the predicted failure type, and generating a corresponding disposal suggestion according to each level of early warning information.

[0042] In the spatiotemporal temperature evolution graph, temperature anomaly points refer to areas or nodes where the temperature exceeds the normal range. By counting the frequency of these anomaly points, we can reflect the frequency of abnormal temperature conditions in a certain period of time. For example, if a device has multiple temperature anomaly points in a short period of time, it may indicate that the device has potential problems and needs to be closely monitored. Temperature rise rate refers to the speed at which the device temperature rises over time. By analyzing the changes in temperature rise rate, we can understand whether the device temperature rise trend is abnormal. For example, under normal circumstances, the device temperature rise rate is relatively stable, but if it suddenly rises sharply or fluctuates greatly, it may indicate that the device is about to overheat. Observe the abnormal aggregation of temperature distribution in the spatiotemporal temperature evolution graph, such as some areas with temperatures consistently higher than surrounding areas or irregularly aggregated temperature distribution patterns. This abnormal aggregation pattern may indicate local overheating or poor heat dissipation within the device. Use the second preset algorithm to perform pattern recognition on the extracted key features. The second preset algorithm may be a machine learning or deep learning-based method that can identify feature patterns related to overheating failure by learning from a large amount of historical data. For example, the algorithm can learn that when a device is about to overheat, specific combinations of temperature anomaly point frequency, temperature rise rate change, and temperature distribution abnormal aggregation pattern occur, thereby identifying these patterns in new data and providing a basis for subsequent fault prediction. Input the feature patterns related to overheating failure obtained through pattern recognition into the preset prediction model. This prediction model is trained on a large amount of historical fault data and can accurately estimate the overheating failure probability and predicted fault type of each power device based on the input feature patterns. The prediction model outputs a probability value representing the likelihood of each power device overheating. For example, the probability value is between 0 and 1, and the larger the value, the higher the risk of overheating failure. At the same time, the model also predicts the type of overheating failure that may occur, such as winding overheating or local overheating caused by poor contact. This helps maintenance personnel understand the nature of the fault in advance and provides guidance for subsequent fault handling. Analyze the potential sources of overheating failure by combining the physical connection relationship between power devices and device operating state information. Power devices are usually physically connected, and a device failure may affect other devices connected to it. For example, in a power line, if a connection point has poor contact, it may cause the temperature at that point to rise, affecting the temperature distribution of adjacent devices. At the same time, device operating state information such as load size and operating time also affects the occurrence of overheating failure. By considering these factors comprehensively, we can more accurately determine the potential source of failure. Based on the overheating failure probability and predicted fault type, generate multi-level warning information. For example, the warning information can be divided into levels (emergency), level 2 (serious), and level 3 (general).When the overheat fault probability is high and the predicted fault type is serious, a first-level warning information is generated; when the fault probability and type are at a medium level, a second-level warning information is generated; and when the fault risk is relatively low, a third-level warning information is generated. The multi-level warning information can enable the operation and maintenance personnel to take corresponding measures according to the severity of the fault. According to each level of warning information, a corresponding treatment suggestion is generated. For the first-level warning information, the treatment suggestion can include immediate shutdown inspection, organization of professional personnel for emergency repair, etc.; for the second-level warning information, the suggestion can be to arrange maintenance at a suitable time, strengthen monitoring of the equipment, etc.; and for the third-level warning information, the suggestion can be to record the equipment state, regularly perform inspection, etc. These treatment suggestions aim to provide specific operation guidance for the operation and maintenance personnel, helping them to timely and effectively handle potential overheat faults and ensure the safe and stable operation of the power equipment.

[0043] By extracting key features and performing pattern recognition, combined with a preset prediction model, the overheat fault probability and type of the power equipment can be more accurately predicted, and potential fault hidden dangers can be discovered in advance. Comprehensive consideration of the physical connection relationship and operation state information between the equipment helps to accurately analyze the potential source of the overheat fault and provide a clear direction for fault handling. The multi-level warning information and the corresponding treatment suggestions can enable the operation and maintenance personnel to take different measures according to the severity of the fault, improve the efficiency and pertinence of fault handling, and reduce the impact of the fault on the power system.

[0044] The embodiment also discloses a diagnosis system for overheat faults of power equipment, Figure 2 is a module schematic diagram of the diagnosis system for overheat faults of power equipment disclosed by the embodiment of the application, as Figure 2 shown, the system comprises an acquisition module 201, a projection module 202, a temperature module 203 and a prediction module 204, wherein: The acquisition module 201 is configured to acquire electrical parameters, a visible light image and an infrared image of the power equipment, enhance the visible light image and the infrared image by a preset dual-stream generative adversarial network to obtain a target visible light image and a target infrared image, segment various types of power equipment on the target visible light image, and generate a device semantic mask image; The projection module 202 is configured to project temperature data in the target infrared image onto each segmented power equipment according to the device semantic mask image to form a device temperature distribution map; The temperature module 203 is configured to determine static temperature features and dynamic temperature features of each power equipment according to the device temperature distribution map, generate a space-time temperature evolution map according to the static temperature features, the dynamic temperature features and a physical connection relationship between the power equipment, the static temperature features comprising a temperature mean value, a maximum value and a variance, and the dynamic temperature features comprising a temperature rising rate, a fluctuation frequency and an evolution mode of temperature distribution. The prediction module 204 is configured to predict, according to the spatio-temporal temperature evolution map and the electrical parameter, a probability of overheating failure, a failure type, a potential failure source, multi-level early warning information and disposal suggestions of each power equipment.

[0045] Optionally, the acquisition module 201 is configured to: construct a dual-flow generative adversarial network, the dual-flow generative adversarial network comprising a first generator, a second generator and a shared discriminator; input a visible light training image into the first generator to obtain a first visible light image, input an infrared training image into the second generator to obtain a first infrared image, output a joint score according to the first visible light image and the first infrared image through the discriminator, and adjust parameters of the first generator and the second generator according to the joint score for next iteration training until a loss function converges or a maximum number of iterations is reached to obtain a trained preset dual-flow generative adversarial network; input the visible light image and the infrared image into the preset dual-flow generative adversarial network, obtain a target visible light image through the first generator, and obtain a target infrared image through the second generator.

[0046] Optionally, the acquisition module 201 is configured to: input the target visible light image into a preset semantic segmentation model to obtain a class label of each pixel point, and generate a preliminary equipment semantic mask map according to the class label; merge or segment regions in the preliminary equipment semantic mask map according to shapes and sizes of the power equipment, and determine an outline of the power equipment to obtain an optimized equipment semantic mask map; superimpose the optimized equipment semantic mask map and the target visible light image to verify the optimized equipment semantic mask map to generate an equipment semantic mask map.

[0047] Optionally, the projection module 202 is configured to: determine a target region participating in matching according to the equipment semantic mask map, obtain first feature points of the target region from the target infrared image, and obtain second feature points of the target region from the visible light image; match the first feature points and the second feature points to obtain a plurality of matching point pairs, screen the plurality of matching point pairs using a first preset algorithm, and obtain an affine transformation matrix according to the screened target matching point pairs; map temperature data in the target infrared image to corresponding power equipment in the target visible light image pixel by pixel according to the affine transformation matrix to form an equipment temperature distribution map.

[0048] Optionally, the temperature module 203 is configured to: calculate the mean, maximum and variance of the temperature in the target area corresponding to each power device in the device temperature distribution map, and construct static temperature features according to the mean, the maximum and the variance; obtain the device temperature distribution map in a continuous time sequence, determine the temperature rising rate, fluctuation frequency and evolution mode of temperature distribution in the time sequence, and construct dynamic temperature features according to the temperature rising rate, the fluctuation frequency and the evolution mode; combine the static temperature features and the dynamic temperature features according to the power devices to form a temperature feature vector of the target power device.

[0049] Optionally, the temperature module 203 is configured to: regard each independent power device in the power system as a node, associate a corresponding temperature feature vector with each node to form a node set; establish edges between the nodes based on the actual physical connection relationship of the power system, and assign weights to the edges according to the closeness of the connection, the possibility or influence strength of heat conduction to form an edge set; combine the node set and the edge set to form a basic space graph, and add a time sequence feature sequence to each node in the basic space graph to form a space-time temperature evolution graph.

[0050] Optionally, the prediction module 204 is configured to: extract key features from the space-time temperature evolution graph, use a second preset algorithm to perform pattern recognition on the key features to obtain a feature mode related to the overheating fault, and the key features include the appearance frequency of temperature abnormal points, the rate change of temperature rise and the abnormal aggregation mode of temperature distribution; input the feature mode into a preset prediction model to obtain the overheating fault probability and the predicted fault type of each power device, analyze the potential source of the overheating fault in combination with the physical connection relationship between the power devices and the device operation state information, generate multi-level warning information according to the overheating fault probability and the predicted fault type, and generate corresponding disposal suggestions according to each level of warning information.

[0051] It should be noted that the apparatus provided in the above embodiments is only used as an example to illustrate the division of the above functional modules in realizing its functions, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above described functions. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is described in detail in the method embodiments, which will not be repeated here.

[0052] The embodiment also discloses an electronic device, referring to Figure 3 The electronic device can include at least one processor 301, at least one communication bus 302, a user interface 303, a network interface 304, and at least one memory 305.

[0053] The communication bus 302 is used to realize the connection communication between the components.

[0054] The user interface 303 can include a display screen (Display) and a camera (Camera), and the optional user interface 303 can further include a standard wired interface and a wireless interface.

[0055] The network interface 304 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0056] The processor 301 can include one or more processing cores. The processor 301 connects various parts in the server through various interfaces and lines, executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be realized in at least one of the hardware forms of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 301 can be integrated with a combination of one or several of a central processing unit (CPU), a graphics processor (GPU), and a modem. The CPU is mainly used to process an operating system, a user interface, and an application program; the GPU is used to render and draw the content to be displayed on the display screen; and the modem is used to process wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 301, but can be realized by a separate chip.

[0057] The memory 305 can include a random access memory (RAM) and can also include a read-only memory (ROM). Optionally, the memory 305 includes a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, codes, code sets, or instruction sets. The memory 305 can include a program storage area and a data storage area, where the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area can store data involved in the various method embodiments described above, etc. The memory 305 can also optionally be at least one storage device located away from the aforementioned processor 301. As shown in Figure 3 The memory 305, as a computer storage medium, can include an operating system, a network communication module, a user interface module, and an application program of a diagnosis method of a power equipment overheating fault.

[0058] In the electronic device shown in Figure 3 In the electronic device shown in

[0059] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0060] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0061] In several embodiments provided in the present application, it should be understood that the disclosed apparatus can be implemented in other manners. For example, the division of the apparatus embodiments is merely illustrative, and for example, the division of units can be changed, and for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some intervening devices, and can be in electric, mechanical or other forms.

[0062] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0063] If the integrated units are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable memory. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory 305 and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned memory 305 includes: a U disk, a mobile hard disk, a magnetic or optical disk and various program code storage media.

[0064] The above is merely exemplary embodiments of the present disclosure, and cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon considering the disclosure herein. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure that follow the general principles of the present disclosure and include common knowledge or conventional techniques in the art not disclosed in the present disclosure.

Claims

1. A method for diagnosing overheating faults of power equipment, characterized in that: Applied to a power equipment diagnostic platform, the method includes: Collect electrical parameters, visible light images, and infrared images of power equipment, enhance the visible light images and infrared images using a preset two-stream generative adversarial network to obtain target visible light images and target infrared images, segment various types of power equipment on the target visible light image, and generate a device semantic mask map; Projecting the temperature data in the target infrared image onto each segmented power device according to the device semantic mask to form a device temperature distribution map; Determining static temperature characteristics and dynamic temperature characteristics of each electrical device based on the device temperature distribution diagram, and generating a spatiotemporal temperature evolution diagram based on the static temperature characteristics, the dynamic temperature characteristics, and the physical connection relationship between the electrical devices, wherein the static temperature characteristics include the temperature mean, maximum value, and variance, and the dynamic temperature characteristics include the temperature rise rate, fluctuation frequency, and evolution pattern of the temperature distribution; The overheating failure probability, failure type, potential failure source, multi-level warning information and disposal suggestions of each power equipment are predicted based on the spatiotemporal temperature evolution diagram and the electrical parameters.

2. The method for diagnosing an overheating fault of an electric power device according to claim 1, characterized in that: The enhancing the visible light image and the infrared image by using a preset dual-stream generative adversarial network to obtain a target visible light image and a target infrared image includes: Constructing a two-stream generative adversarial network, wherein the two-stream generative adversarial network includes a first generator, a second generator, and a shared discriminator; Inputting a visible light training image into a first generator to obtain a first visible light image, and inputting an infrared training image into a second generator to obtain a first infrared image, outputting a joint score based on the first visible light image and the first infrared image by the discriminator, and adjusting the parameters of the first generator and the second generator respectively according to the joint score to perform the next iterative training until the loss function converges or the maximum number of iterations is reached to obtain a trained preset two-stream generative adversarial network; The visible light image and the infrared image are input into the preset two-stream generative adversarial network, a target visible light image is obtained by the first generator, and a target infrared image is obtained by the second generator.

3. The method for diagnosing an overheating fault of an electric power device according to claim 2, characterized in that: The step of segmenting various types of power equipment on the target visible light image and generating a device semantic mask includes: Inputting the target visible light image into a preset semantic segmentation model to obtain a category label for each pixel, and generating a preliminary device semantic mask map based on the category label; Merging or segmenting regions in the preliminary device semantic mask according to the shape and size of the power equipment, and determining the outline of the power equipment to obtain an optimized device semantic mask; The optimized device semantic mask map is superimposed on the target visible light image to verify the optimized device semantic mask map, so as to generate a device semantic mask map.

4. The method for diagnosing an overheating fault of an electric power device according to claim 1, characterized in that: The step of projecting the temperature data in the target infrared image onto each segmented power device according to the device semantic mask to form a device temperature distribution map includes: Determining a target area to be matched according to the device semantic mask, obtaining a first feature point of the target area from the target infrared image, and obtaining a second feature point of the target area from the visible light image; Matching the first feature point and the second feature point to obtain a plurality of matching point pairs, screening the plurality of matching point pairs using a first preset algorithm, and obtaining an affine transformation matrix according to the screened target matching point pairs; The temperature data in the target infrared image is mapped pixel by pixel to the corresponding power equipment in the target visible light image according to the affine transformation matrix to form a device temperature distribution map.

5. The method for diagnosing an overheating fault of an electric power device according to claim 1, characterized in that: Determining the static temperature characteristics and dynamic temperature characteristics of each power device according to the device temperature distribution diagram includes: Calculating the mean, maximum, and variance of the temperature in the target area corresponding to each electrical device in the device temperature distribution map, and constructing a static temperature feature based on the mean, maximum, and variance; Obtaining a device temperature distribution diagram in a continuous time series, determining a temperature rise rate, a fluctuation frequency, and an evolution pattern of the temperature distribution in the time series, and constructing a dynamic temperature feature based on the temperature rise rate, the fluctuation frequency, and the evolution pattern; The static temperature characteristics and the dynamic temperature characteristics are combined according to the power equipment to form a temperature characteristic vector of the target power equipment.

6. The method for diagnosing an overheating fault of an electric power device according to claim 5, characterized in that: Generating a spatiotemporal temperature evolution diagram according to the static temperature characteristics, the dynamic temperature characteristics, and the physical connection relationship between the power equipment includes: Taking each independent power device in the power system as a node, and associating the corresponding temperature characteristic vector with each node to form a node set; Based on the actual physical connection relationship of the power system, edges are established between nodes, and weights are assigned to the edges according to the tightness of the connection, the possibility of heat conduction, or the intensity of the influence to form an edge set; The node set and the edge set are combined to form a basic spatial graph, and a time series feature sequence is added to each node in the basic spatial graph to form a spatiotemporal temperature evolution graph.

7. The method for diagnosing an overheating fault of an electric power device according to claim 1, characterized in that: The prediction of the overheating failure probability, failure type, potential failure source, multi-level warning information and handling suggestions of each power device based on the spatiotemporal temperature evolution diagram and the electrical parameters includes: Extracting key features from the spatiotemporal temperature evolution graph, and performing pattern recognition on the key features using a second preset algorithm to obtain characteristic patterns related to overheating faults, the key features including the frequency of occurrence of temperature anomalies, the rate of temperature rise, and the abnormal aggregation pattern of temperature distribution; Inputting the characteristic pattern and the electrical parameters into a preset prediction model to obtain the overheating failure probability and predicted failure type of each electrical device; Combined with the physical connection relationship between power equipment and the equipment operating status information, the potential sources of overheating failures are analyzed, and multi-level warning information is generated based on the overheating failure probability and the predicted failure type, and corresponding handling suggestions are generated based on each level of warning information.

8. A diagnostic system for overheating faults of power equipment, characterized in that: It includes acquisition module, projection module, temperature module and prediction module, among which: an acquisition module configured to collect electrical parameters, visible light images, and infrared images of power equipment, enhance the visible light images and infrared images using a preset two-stream generative adversarial network to obtain target visible light images and target infrared images, segment various types of power equipment on the target visible light image, and generate a device semantic mask; a projection module configured to project the temperature data in the target infrared image onto each segmented power device according to the device semantic mask image to form a device temperature distribution map; a temperature module configured to determine static temperature characteristics and dynamic temperature characteristics of each electrical device based on the device temperature distribution diagram, and generate a spatiotemporal temperature evolution diagram based on the static temperature characteristics, the dynamic temperature characteristics, and the physical connection relationship between the electrical devices, wherein the static temperature characteristics include the temperature mean, maximum value, and variance, and the dynamic temperature characteristics include the temperature rise rate, fluctuation frequency, and evolution pattern of the temperature distribution; The prediction module is configured to predict the overheating failure probability, failure type, potential failure source and multi-level warning information and disposal suggestions of each power device based on the spatiotemporal temperature evolution diagram and the electrical parameters.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images