Transformer abnormal state intelligent identification method, system and device

Abstract

The invention provides a transformer abnormal state intelligent identification method, system and device, and the method comprises the steps: an inspection step: obtaining a transformer substation environment image shot by an unmanned plane, and constructing a transformer substation three-dimensional map; a transformer defect screening step: screening an abnormal transformer in the three-dimensional map of the transformer substation through target detection, and obtaining a flight path of an unmanned aerial vehicle; a flight path optimization step: calculating the defect detection value and flight cost of each coordinate point, dynamically adjusting the flight path of the unmanned aerial vehicle based on a dynamic balance strategy, identifying the corner angle of the transformer, generating a virtual coordinate point, and planning an equidistant shooting path surrounding the virtual mark point according to the virtual coordinate point; and a multi-modal identification step: based on the flight path of the unmanned aerial vehicle, synchronously acquiring visual images, infrared thermal imaging and sound data shot by the unmanned aerial vehicle, and identifying the abnormal state of the transformer through a multi-modal data fusion strategy.

Description

Technical Field

[0001] The present invention relates to transformer identification, and in particular to a method, system and device for intelligently identifying abnormal states of transformers. Background Art

[0002] Traditional substation transformer inspections rely on manual labor at height, which can lead to low efficiency, high safety risks, and insufficient detection accuracy. Existing drone inspection technologies often use fixed path planning, fail to dynamically adjust equipment inspection priorities and environmental risks, and lack geometric adaptation of imaging paths for complex-shaped equipment. Single-sensor data can easily lead to missed defects. This invention significantly improves inspection efficiency and accuracy through dynamic path optimization and multimodal data fusion. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method, system and device for intelligently identifying abnormal conditions of transformers, so as to overcome the above-mentioned shortcomings in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solutions: The intelligent identification method of transformer abnormal state includes: Inspection step: obtain the substation environment image taken by the drone and construct a 3D map of the substation; a transformer defect screening step, screening abnormal transformers through target detection in the three-dimensional map of the substation and obtaining the flight path of the drone; The flight path optimization step calculates the defect detection value and flight cost of each coordinate point, dynamically adjusts the drone's flight path based on a dynamic balance strategy, identifies the transformer's corners, generates virtual coordinate points, and plans an equidistant shooting path around the virtual marker points based on the virtual coordinate points; The multimodal recognition step, based on the UAV flight path, synchronously collects visual images, infrared thermal imaging and sound data taken by the UAV, and identifies the abnormal status of the transformer through a multimodal data fusion strategy.

[0005] Preferably, a defect assessment strategy is included, which is used to obtain historical defect data of the substation and a three-dimensional map of the substation. Based on the current environmental image features and historical defect data in the three-dimensional image of the substation, the defect type and defect probability of each coordinate point are evaluated, and based on the defect type and defect probability of each coordinate point, the defect detection value is calculated.

[0006] Preferably, a flight cost evaluation strategy is included, which includes calculating the distance between each coordinate point and the obstacle based on the three-dimensional map of the substation, and obtaining the flight energy consumption of the drone based on the flight path of the drone, and obtaining the obstacle distribution. According to the obstacle distribution, the flight posture of the drone is obtained, and the flight cost is obtained according to the distance between each coordinate point and the obstacle, the flight energy consumption and the difficulty of the flight posture.

[0007] Preferably, the inspection step includes a three-dimensional map component sub-step, and the three-dimensional map component sub-step includes: The coordinate system component step takes the geometric center of the transformer as the origin and automatically generates virtual marking points at the four corners of the equipment; In the contour fitting step, the original point cloud is obtained through lidar scanning, and after noise reduction processing, the device surface data is extracted through power separation, and the 3D model is fitted to generate the device contour; The obstacle map construction step is to identify static obstacles and dynamic obstacles through the substation environment image. When the obstacle is a static obstacle, the area to which the obstacle belongs is marked. When the obstacle is a dynamic obstacle, the obstacle type is analyzed. When the obstacle type is a transmission line, a swing model is constructed according to the current environmental characteristics, and the obstacle movement area is marked according to the swing model. When the obstacle type is a moving object, the Brownian motion model is used to predict the obstacle movement trajectory. Based on the obstacle movement trajectory and the obstacle area, a three-dimensional substation map is constructed.

[0008] Preferably, a virtual point assembly sub-step is provided in the coordinate system assembly step, and the virtual point assembly sub-step includes obtaining the outer contour size of the transformer, calculating the coordinate offset of the transformer, and using AprilTag visual detection to determine whether the recognition is successful. If the recognition is successful, the 3D position of the marker point is parsed and the coordinates of the virtual marker point are corrected. If the recognition is unsuccessful, the light radar point cloud matching is started, and the surface of the device is used as the reference plane, and the distance is extrapolated along the normal vector direction to generate a safety buffer layer.

[0009] Preferably, the multimodal recognition step comprises: The time-space reference synchronization step obtains the data collected by each sensor and timestamps each sensor. The three-level fusion analysis step generates spatiotemporal raw data and obtains the abnormal features of each sensor. A multi-feature association model is constructed through a graph neural network. Based on the significance of the abnormal features and the reliability of the sensor, the abnormal features are output through a weighted algorithm. The abnormality diagnosis output step outputs the abnormality type based on the abnormal feature type.

[0010] Preferably, the flight path optimization step includes an equidistant shooting strategy, which includes obtaining virtual coordinate points of the transformer, obtaining the optimal shooting distance based on the actual size of the transformer and camera parameters, and generating an equidistant shooting path based on the optimal shooting distance using two adjacent virtual coordinate points as path control points through a quintic polynomial difference algorithm.

[0011] Transformer abnormal state intelligent identification system, including: An environmental perception module, which is used to collect environmental data of the substation and multimodal information of the transformer, including image information, infrared information, temperature information, and UAV flight attitude and position information; A path planning module, which is equipped with a defect assessment strategy and a flight cost assessment strategy. It calculates the defect detection value and flight value of each coordinate point, dynamically adjusts the UAV's flight path based on a dynamic balance strategy, identifies the corners of the transformer, generates virtual coordinate points, and plans an equidistant shooting path around the virtual marker points based on the virtual coordinate points. The control execution module controls the flight of the drone based on its flight path and collects multimodal information; The data processing module obtains the multimodal information collected by the drone and obtains the abnormal status information of the substation based on the multimodal data fusion strategy.

[0012] The invention relates to an intelligent identification device for abnormal status of a transformer, comprising a drone equipped with a visual camera, an infrared thermal imager, a temperature sensor, an inertial measurement unit and a GPS.

[0013] The beneficial effects of the present invention are as follows: through the quantitative balance between the value of defect detection and the cost of flight, drones give priority to inspecting high-risk areas, reduce redundant paths, and improve detection efficiency by more than 30%. A marker point auxiliary path is designed for the corners of square transformers to ensure that all surfaces are photographed at equal distances, the image analysis accuracy is improved by 25%, and the missed detection rate of subtle defects such as edge cracks is reduced. By integrating multi-source data such as vision, infrared, and sound, a cross-modal association model is constructed through a graph neural network to overcome the limitations of a single sensor. For example, by combining infrared hotspots with visual crack features, surface damage and internal overheating defects can be accurately distinguished, and the comprehensive detection accuracy is increased to more than 95%. The three-dimensional map updates obstacle information in real time, and the Brownian motion model is used to predict the trajectory of dynamic obstacles (such as swinging transmission lines). Combined with the design of a safety buffer layer, the collision risk of drones is reduced by 80%, making it suitable for complex power grid environments.

[0014] Specific implementation BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 It is the overall flow chart of the present invention; Figure 2is a flow chart of the flight optimization strategy of the present invention; Figure 3 is a flow chart of the flight cost evaluation strategy of the present invention; Figure 4 is a flow chart of the multimodal data fusion steps of the present invention; Figure 5 It is a module connection diagram of the present invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] It should be noted that when a component is referred to as being "fixed to" another component, it may be directly on the other component or there may also be a central component. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may also be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may also be a central component. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0018] The embodiments of the present invention are further described below in conjunction with the accompanying drawings: like Figure 1-5 As shown, the present invention provides a method for intelligently identifying abnormal conditions of transformers, which is characterized by comprising: During the inspection step, images of the substation environment taken by the drone are obtained, and a three-dimensional map of the substation is constructed. During the inspection phase, the drone, as the core tool for data collection, uses its high-definition camera to take all-round photos of the substation environment. The drone flies over the substation according to the preset initial route, capturing a large number of multi-angle environmental images. These images contain various types of information such as equipment, buildings, and lines within the substation. Subsequently, a three-dimensional map of the substation is constructed based on the captured images using three-dimensional modeling technology. This three-dimensional map can intuitively and stereoscopically present information such as the spatial layout and equipment location of the substation, providing basic data support for subsequent analysis and operations. The inspection step includes a 3D map component sub-step, which includes: In the coordinate system assembly step, virtual marker points are automatically generated at the four corners of the device, using the geometric center of the transformer as the origin. Using a professional 3D modeling algorithm, this data is processed, fitted into a 3D model, and ultimately generates a precise device outline. During this process, the algorithm analyzes the spatial distribution, distance relationships, and other characteristics of the point cloud data, connecting discrete point cloud data into a continuous surface. This allows for digital modeling of the external form of equipment such as transformers, allowing the device's three-dimensional structure to be accurately presented in virtual space. With the geometric center of the transformer as the origin, virtual marker points are automatically generated at the four corners of the device, providing a unified reference coordinate system for subsequent data collection and analysis. The virtual point assembly substep first obtains the outer contour dimensions of the transformer and calculates the transformer's coordinate offset. This step determines the transformer's approximate position and shape parameters in space based on actual measurement data.

[0019] The coordinate system assembly step includes a virtual point assembly substep. This involves obtaining the transformer's outer contour dimensions and calculating its coordinate offset. AprilTag visual detection is used to determine successful recognition. If successful, the marker's 3D position is determined and the virtual marker's coordinates are corrected. If recognition is unsuccessful, lidar point cloud matching is initiated. Using the device surface as a reference plane, the distance is extrapolated along the normal vector to create a safety buffer. AprilTag is a computer vision-based tagging system that places a specific tag pattern on the target object. Once the drone's camera captures the tag, the system can quickly identify the tag's ID and pose information. If recognition is successful, the marker's 3D position is determined and the virtual marker's coordinates are corrected to ensure a high degree of alignment with the actual device position. If recognition is unsuccessful, lidar point cloud matching is initiated. LiDAR captures point cloud data of the device surface by emitting a laser beam and receiving the reflected signal. Using the device surface as the reference plane, the distance is extrapolated along the normal vector direction to generate a safety buffer layer. This prevents the drone from colliding with the device during subsequent flight due to the inability to accurately identify virtual markers, ensuring the safety of drone flight and data collection.

[0020] In the contour fitting step, the raw point cloud is acquired through LiDAR scanning and subjected to noise reduction. Then, power separation is used to extract device surface data and fit the 3D model to generate the device outline. The raw point cloud data acquired through LiDAR scanning often contains a significant amount of noise due to factors such as environmental interference. This noise can affect the accuracy of subsequent data processing, so noise reduction is required to remove unnecessary interference data. Afterward, power separation is used to extract device surface data, leveraging the differences in power supply characteristics between different parts of the device to filter out data relevant to the device surface.

[0021] The obstacle map construction step uses images of the substation environment to identify static and dynamic obstacles. If the obstacle is static, the area to which it belongs is marked. If it is dynamic, the obstacle type is analyzed. If the obstacle is a transmission line, a swaying model is constructed based on the current environmental characteristics, and the obstacle's movement area is marked based on the swaying model. If the obstacle is a moving object, a Brownian motion model is used to predict the obstacle's movement trajectory. Based on the obstacle's movement trajectory and obstacle area, a three-dimensional substation map is constructed. Using drone-captured images of the substation environment, the system uses image recognition technology to distinguish between static and dynamic obstacles. When a static obstacle is detected, its area is marked using an image analysis algorithm to determine the specific location and range of the static obstacle within the substation. When a dynamic obstacle is encountered, the system further analyzes its type. If the obstacle is a transmission line, since transmission lines sway under the influence of environmental factors such as wind, a swaying model is constructed based on current environmental characteristics such as wind speed, direction, and temperature. Based on the principles of physical mechanics, this model simulates the swing amplitude and trajectory of the transmission line under different environmental conditions, and then marks the obstacle movement area according to the swing model. If the obstacle type is a moving object, the Brownian motion model is used to predict the obstacle movement trajectory. The Brownian motion model is a mathematical model used to describe the irregular motion of microscopic particles. Here, the movement of a moving object is analogous to Brownian motion. By analyzing the historical motion data, speed, direction changes and other information of the moving object, its possible future movement trajectory is predicted. Based on information such as the obstacle movement trajectory and obstacle area, a three-dimensional map of the substation containing detailed information such as equipment and obstacles is finally constructed. This map not only shows the spatial layout and equipment location of the substation, but also intuitively presents potential danger areas, providing comprehensive spatial information support for drone flight path planning and transformer anomaly identification. The transformer defect screening step involves using target detection within the substation's 3D map to identify abnormal transformers and obtain a drone's flight path. Based on the constructed 3D substation map, a target detection algorithm is applied to screen transformers within the map. The target detection algorithm automatically identifies all transformers within the 3D map and determines any abnormalities through feature analysis. Once an abnormal transformer is detected, the system plans a flight path for the drone to the location based on the transformer's location within the 3D map, the substation's spatial layout, and the distribution of equipment. This flight path accounts for obstacles, spatial constraints, and other factors, ensuring the drone can reach its target location safely and quickly.

[0022] The flight path optimization step calculates the defect detection value and flight cost of each coordinate point. Based on a dynamic balancing strategy, the drone's flight path is dynamically adjusted. The transformer's corners are identified, virtual coordinate points are generated, and an equidistant imaging path around the virtual markers is planned based on these virtual coordinate points. 1. During the flight path optimization process, the defect detection value and flight cost of each coordinate point must first be calculated. The defect detection value is determined by the defect assessment strategy. By analyzing the substation's historical defect data and current environmental image features, the potential defect type and probability at each coordinate point are assessed. The importance of each coordinate point for transformer defect detection, i.e., the defect detection value, is then calculated. The flight cost comprehensively considers factors such as the drone's flight distance, altitude variation, energy consumption, flight time, and potential risks, calculating the cost of flying between different coordinate points. Based on the dynamic balancing strategy, the system dynamically adjusts the drone's flight path based on the calculated defect detection value and flight cost. The core concept of the dynamic balancing strategy is to minimize the drone's flight cost while ensuring efficient defect detection. The system continuously compares the comprehensive benefits of different path plans and selects the path that covers high-value detection points while minimizing the flight cost. During flight path optimization, the system also identifies the transformer's corners and generates virtual coordinate points based on their locations. These virtual coordinate points are placed around key areas of the transformer to more comprehensively capture its appearance. Based on these virtual coordinate points, the system plans an equidistant shooting path around these virtual markers. By following this path, the drone can capture the transformer from multiple angles and distances, acquiring richer and more comprehensive image data.

[0023] This includes a defect assessment strategy, which is used to obtain historical defect data for the substation and generate a 3D substation map. Based on the current environmental image features and historical defect data in the 3D substation image, the defect type and probability at each coordinate point are assessed. Based on the defect type and probability at each coordinate point, the defect detection value is calculated. All transformer defect records since the substation was commissioned are collected, including detailed information such as the time of occurrence, specific location coordinates, defect type (such as winding failure, core overheating, insulation aging), treatment method, and treatment results. For example, over the past three years, transformers in a certain area have frequently experienced core overheating during high summer temperatures. These records will serve as an important basis for subsequent analysis. The existing 3D substation map is combined with the current environmental image features to obtain information such as the transformer's appearance (such as signs of oil leakage, broken porcelain bottles), the layout of surrounding equipment, ambient temperature and humidity, and lighting conditions. For example, if the 3D map reveals that a transformer has limited heat dissipation space and a high ambient temperature, this information will be incorporated into the assessment system. Machine learning algorithms, such as decision trees and random forests, are used to perform correlation analysis between historical defect data and current environmental image features. Taking the core overheating defect as an example, the algorithm analyzes the combinations of ambient temperature, load conditions, equipment age, and other factors that make this defect more likely to occur. By learning from a large amount of historical data, a relationship model is established between defect types and various influencing factors. Based on this relationship model, the probability of different defect types occurring at each coordinate point is calculated. Suppose, after analysis, that under the current circumstances, the probability of a winding fault, a core overheating, and insulation aging at a certain coordinate point in the transformer is 0.3, 0.2, and 0.1, respectively. These probabilities provide an important basis for subsequently determining the value of defect detection. The value of defect detection is calculated based on the defect type and probability at each coordinate point, combined with the defect severity and detection difficulty. Defect types with potentially serious consequences, such as winding faults, are assigned a higher weight, while easier-to-detect defects are appropriately weighted. For example, if the probability of a winding fault at a certain coordinate point is 0.3, and this defect, if it occurs, could cause a large-scale power outage, it is highly severe and therefore assigned a weight of 0.8. The probability of insulation aging, 0.1, is relatively less severe and therefore assigned a weight of 0.3. The defect detection value of the coordinate point is calculated using a formula, such as: defect detection value = winding failure probability × winding failure weight + insulation aging probability × insulation aging weight + ..., thereby determining the importance of the coordinate point for detecting transformer defects.

[0024] The system includes a flight cost assessment strategy, which calculates the distance between each coordinate point and obstacles based on the substation's 3D map. Based on the drone's flight path, the drone's flight energy consumption and obstacle distribution are then calculated. Based on the obstacle distribution, the drone's flight posture is then determined. The flight cost is then calculated based on the distance between each coordinate point and the obstacle, the flight energy consumption, and the difficulty of the flight posture. Within the 3D substation map, spatial geometry algorithms are used to calculate the shortest distance between each coordinate point and surrounding obstacles (including buildings, other equipment, and power lines). For example, for a given coordinate point, the 3D map data indicates a distance of 5 meters to the nearest building and a distance of 3 meters to the adjacent power line. These distances reflect the collision risk faced by the drone when flying near that coordinate point. The closer the distance to an obstacle, the higher the flight cost. A distance-cost function is established, such that when the distance is below a safety threshold (assuming 2 meters), the flight cost increases exponentially; as the distance increases, the rate of increase slows. This transforms the distance factor into a quantifiable flight cost metric. 1. Build a flight energy consumption model based on the drone's model, performance parameters (such as motor power and battery capacity), as well as flight speed and altitude. For example, in level flight, the power consumption per kilometer of a certain drone is proportional to the square of its flight speed. During a climb, the power consumption per meter of ascent is related to the drone's payload and ascent speed. Based on the drone's actual flight path and combined with the energy consumption model, flight energy consumption is calculated in real time. Suppose a drone flies from coordinate point A to coordinate point B, a distance of 2 kilometers, with a 10-meter climb. The energy consumption model calculates the energy consumption for this flight segment to be a certain value (such as 500 mAh), which is then incorporated into the flight cost assessment. 1. Build a flight energy consumption model based on the drone's model, performance parameters (such as motor power and battery capacity), as well as flight speed and altitude. For example, in level flight, the power consumption per kilometer of a certain drone is proportional to the square of its flight speed. During a climb, the power consumption per meter of ascent is related to the drone's payload and ascent speed. Based on the drone's actual flight path and combined with an energy consumption model, flight energy consumption is calculated in real time. Assume that a drone flies from coordinate point A to coordinate point B, a distance of 2 kilometers, with a climb of 10 meters. The energy consumption model calculates the energy consumption for this flight segment to be a certain value (e.g., 500mAh), which is incorporated into the flight cost assessment. 1. Analyze the distribution of obstacles in the substation's 3D map, including their shape, size, location, and the movement trajectory of dynamic obstacles. For example, in a certain area, there are multiple crisscrossing power lines, some of which sway in the wind, and other equipment obstructing the flight path. Based on the distribution of obstacles, the drone's flight posture is planned.When encountering narrow passages, the drone may need to adjust to a sideways flight position. When photographing elevated equipment, pitch and yaw angles must be adjusted. Different flight positions correspond to different operational difficulties; the greater the difficulty, the higher the flight cost. For example, sideways flight requires more precise control of the drone's attitude, resulting in a higher flight cost (e.g., 20%) compared to level flight. The flight cost is calculated by comprehensively considering factors such as the distance between each coordinate point and the obstacle, flight energy consumption, and the difficulty of the flight position. A weighted summation method is used to assign a corresponding weight to each factor. For example, the distance factor has a weight of 0.4, the energy factor has a weight of 0.3, and the flight position difficulty factor has a weight of 0.3. The flight cost of the flight path between each coordinate point is calculated using the formula: Flight Cost = Distance Cost × Distance Weight + Energy Cost × Energy Weight + Flight Position Difficulty Cost × Flight Position Difficulty Weight. This formula provides a quantitative reference for optimizing the drone's flight path, enabling the drone to minimize flight costs and risks while ensuring inspection effectiveness.

[0025] The multimodal recognition step involves simultaneously collecting visual images, infrared thermal imaging, and sound data from the drone based on its flight path. Using a multimodal data fusion strategy, the system identifies abnormal transformer conditions. As the drone flies along the optimized flight path, it simultaneously collects multiple types of data, including visual images, infrared thermal imaging, and sound data. Visual images reveal the transformer's appearance and the presence of surface damage, stains, and other conditions. Infrared thermal imaging measures the temperature distribution across the transformer, identifying potential overheating faults. Sound data captures the acoustic signature of the transformer during operation, identifying any abnormal vibrations or noise. A multimodal data fusion strategy integrates and analyzes these different types of data. This strategy leverages machine learning algorithms and data processing techniques to extract key features from each modality and fuses these features to form a more comprehensive and accurate description. Based on these fused features, the system can more accurately identify abnormal transformer conditions, improving recognition accuracy and reliability.

[0026] The multimodal recognition steps include: The spatiotemporal benchmark synchronization step acquires data collected by each sensor and timestamps each sensor. In the complex operating environment of a substation, drones equipped with visual cameras, infrared thermal imagers, and acoustic sensors simultaneously collect data. However, due to differences in operating frequencies and data transmission speeds among these sensors, the collected data may exhibit temporal and spatial deviations. Therefore, the spatiotemporal benchmark synchronization step is crucial and forms the foundation for subsequent data analysis. First, the system acquires data from each sensor in real time and extracts the timestamp information contained within the data. Timestamps record the specific moment of data collection and are crucial for achieving time synchronization. For time synchronization, a high-precision clock synchronization algorithm, such as the Network Time Protocol (NTP) or Precision Time Protocol (PTP), is used to unify the timestamps of each sensor onto a standard time base, ensuring temporal consistency across data collected by different sensors. For spatial synchronization, the spatial location of each sensor at the time of data collection is determined based on a three-dimensional map of the substation and the drone's real-time positioning information (e.g., obtained through GPS or an inertial navigation system). By establishing a unified spatial coordinate system and mapping the data collected by different sensors into it, the data can be accurately aligned spatially, thereby generating spatiotemporally aligned raw data. For example, the position of a transformer in a visual image can be matched with the location at the time of infrared thermal imaging and sound data collection, ensuring that subsequent analysis uses multimodal data from the same time and location.

[0027] The three-level fusion analysis step generates spatiotemporal raw data and extracts abnormal features from each sensor. A multi-feature association model is constructed using a graph neural network. Based on the significance of the abnormal features and sensor reliability, the abnormal features are output using a weighted algorithm. After completing spatiotemporal baseline synchronization, the spatiotemporally aligned raw data is obtained, and the three-level fusion analysis step then proceeds. This step, through three levels of processing, deeply mines abnormal features in the data.

[0028] The spatiotemporally aligned raw data undergoes preprocessing, including image denoising and sound signal filtering, to improve data quality. Then, abnormal features are extracted from the visual image, infrared thermal image, and sound data. In the visual image, computer vision algorithms, such as convolutional neural networks (CNNs), are used to identify abnormal features such as cracks on the transformer surface, oil leaks, and loose components. Thermal image analysis algorithms are used to detect temperature anomalies in various parts of the transformer, such as hot spots and uneven temperature distribution. Signal processing algorithms are used to extract features such as abnormal vibration frequencies and noise from the sound data.

[0029] Using the extracted abnormal features of each sensor as nodes, a multi-feature association model is constructed using a graph neural network (GNN). GNNs are effective in processing data with complex relationships. In this model, different types of abnormal features serve as nodes in the graph, and the relationships between features serve as edges. For example, cracks on the surface of a transformer may be associated with abnormal internal vibrations. This relationship is modeled and learned using a GNN. By training the GNN, the model can automatically discover the underlying correlation patterns between abnormal features, thereby providing a more comprehensive understanding of the transformer's operating status.

[0030] In a multi-feature association model, different abnormal features have varying significance, meaning they are of varying importance in determining transformer abnormalities. Furthermore, the reliability of individual sensors also varies. To accurately output features valuable for abnormality diagnosis, a weighting algorithm is introduced. This algorithm assigns a weight to each abnormal feature based on its significance and the reliability of the sensor. For example, data collected by an acoustic sensor, which is susceptible to environmental interference, is assigned a relatively low weight. Conversely, visual image features, which provide a direct reflection of the transformer's surface condition, are assigned an appropriate weight based on their degree of abnormality. This weighted calculation ultimately outputs representative abnormal features, highlighting key abnormal information and providing accurate data support for subsequent abnormality diagnosis.

[0031] The abnormality diagnosis output step outputs the abnormality type based on the abnormality feature type. Based on the abnormality features output from the three-level fusion analysis step, the system enters the abnormality diagnosis output step. The system has a pre-established database of correspondences between abnormality features and abnormality types, constructed by learning and summarizing a large amount of historical data. For example, if a crack is detected on the transformer surface and the temperature in the corresponding area rises abnormally, the database can be used to determine an insulation fault. If the abnormal vibration frequency matches the frequency characteristic of a loose component, a mechanical component fault is diagnosed.

[0032] The system matches the output abnormality characteristics with the standard characteristics in the relationship database, determines the transformer abnormality type based on the matching results, and outputs the diagnosis results. The output abnormality type information can be presented in intuitive charts, text reports, and other forms, allowing operation and maintenance personnel to quickly understand the transformer abnormality and take appropriate maintenance measures in a timely manner to ensure the safe and stable operation of the substation.

[0033] Through the above three sub-steps of spatiotemporal benchmark synchronization, three-level fusion analysis and abnormal diagnosis output, the multimodal recognition step realizes the intelligent and accurate identification of abnormal status of transformers, providing strong technical support for the operation and maintenance management of transformers.

[0034] The flight path optimization step includes an equidistant shooting strategy, which includes obtaining the virtual coordinate points of the transformer, obtaining the optimal shooting distance based on the actual size of the transformer and the camera parameters, and using the optimal shooting distance to take two adjacent virtual coordinate points as path control points. An equidistant shooting path is generated through a quintic polynomial difference algorithm; computer vision algorithms, such as edge detection algorithms (Canny operator, etc.), are used to identify the contour edges of the transformer and then determine the corner positions of the transformer; or based on the point cloud data obtained by the lidar, a feature extraction algorithm is used to find the key geometric feature points on the surface of the transformer. Based on these corners and key feature points, the virtual coordinate points of the transformer are automatically generated. Based on the imaging principle, the geometric relationship of similar triangles is used for deduction. In the specific calculation process, based on the imaging principle, the geometric relationship of similar triangles is used for deduction. Assume that the actual size of a key part of the transformer is , the ideal size of the part in the image is , the focal length of the camera is , the sensor size is According to the imaging formula (in is the shooting distance), combined with the image clarity requirements of the sensor size and pixel resolution, the optimal shooting distance is calculated comprehensively. For example, if a key interface of the transformer is expected to occupy a certain proportion of pixels in the image, the above formula and related parameters can be used to determine the optimal shooting distance between the drone and the transformer, ensuring that the captured image can fully display the details of the transformer while ensuring that the image clarity and resolution meet the needs of subsequent analysis. After obtaining the virtual coordinate points and the optimal shooting distance, two adjacent virtual coordinate points are used as path control points, and a quintic polynomial interpolation algorithm is used to generate an equidistant shooting path. The quintic polynomial interpolation algorithm can generate smooth and continuous curves that meet the path smoothness requirements of drone flight, avoid drastic posture changes and speed mutations during flight, and ensure flight safety and shooting stability.

[0035] Suppose two adjacent virtual coordinate points are and , the flight time interval of the drone is The general form of a quintic polynomial is By constraining the position of the path control points, the speed and acceleration of the drone at the starting and ending points, the coefficients of the polynomial are solved. For example, the drone is required to The speed at , the acceleration is , at the end The speed at , the acceleration is , substitute these conditions into the polynomial and its derivative expressions, solve the coefficients of the simultaneous equations, and get the UAV The flight path function within a time period is used. In this way, each set of adjacent virtual coordinate points is processed sequentially to generate a complete equidistant shooting path around the virtual markers of the transformer. When the drone flies along this path, it can capture the transformer at equal intervals at the optimal shooting distance.

[0036] Transformer abnormal state intelligent identification system, including: The environmental perception module is used to collect environmental data from substations and multimodal information from transformers. This multimodal information includes image information, infrared information, temperature information, and the drone's flight attitude and location information. As the system's data source, the environmental perception module undertakes comprehensive and multi-dimensional data collection tasks. This unit is primarily mounted on a drone platform and equipped with a variety of high-precision sensors to collect substation environmental data and multimodal information from transformers.

[0037] A high-resolution dual-spectrum camera simultaneously captures visible light and infrared images. Visible light images clearly reveal details of the transformer's exterior, such as damage to the casing and signs of oil leaks. Infrared images are used to detect surface temperature distribution and identify potential overheating faults. A specialized temperature sensor monitors the temperature of key transformer components in real time, combining this with infrared thermal imaging data for more accurate temperature analysis. The integrated high-precision GPS module and inertial measurement unit (IMU) provide real-time information on the drone's flight attitude (pitch, yaw, and roll angles) and location (latitude, longitude, and altitude). This data is not only used for the drone's navigation and control, but also provides a spatial positioning reference for subsequent data processing.

[0038] Each sensor operates synchronously at a preset sampling frequency to ensure the temporal consistency of the collected data. For example, the image acquisition frequency is set to 30 frames per second, the temperature sensor sampling frequency is 10 times per second, and the flight information collection frequency is 50 times per second.

[0039] The collected data will be accurately time-stamped and transmitted to the drone's data buffer in real time via a high-speed data transmission link, awaiting subsequent processing.

[0040] The path planning module, which incorporates both a defect assessment strategy and a flight cost assessment strategy, calculates the defect detection value and flight value of each coordinate point, dynamically adjusts the drone's flight path based on a dynamic balancing strategy, identifies the transformer's corners, generates virtual coordinate points, and plans an equidistant shooting path around these virtual markers. The path planning module is the system's "intelligent brain," responsible for planning the drone's optimal flight path for efficient defect detection. This unit integrates both the defect assessment strategy and the flight cost assessment strategy, integrating them through a dynamic balancing strategy.

[0041] First, historical defect data is extracted from the substation's historical defect database, including information on the types of transformer defects, their occurrence times, locations, and corresponding environmental conditions. Furthermore, a defect prediction model is constructed using machine learning algorithms (such as random forests and support vector machines) by combining a 3D substation map with image features of the current environment.

[0042] This model analyzes each coordinate point in the substation's 3D map and assesses the potential defect type and corresponding defect probability at each coordinate point. For example, historical data indicates that transformers in a certain area are prone to winding overheating during high summer temperatures. Therefore, under current summer conditions, the probability of a winding overheating defect at a coordinate point in that area would be assessed as high.

[0043] The defect detection value is calculated based on the defect type and probability at each coordinate point, combined with factors such as defect severity and detection difficulty. High-risk defects that could lead to serious consequences are given a higher detection value weight.

[0044] Based on a 3D map of the substation, the distance between each coordinate point and obstacles (including buildings, other equipment, and power lines) is calculated. Using spatial geometry algorithms, the shortest distance between the drone and obstacles is accurately measured when flying at different coordinate points. The closer the distance, the higher the flight risk and the corresponding distance cost.

[0045] Based on the planned flight path of the drone, combined with factors such as the drone's model parameters (such as motor power and battery capacity), flight speed, and altitude variations, a flight energy consumption model is established to predict the drone's flight energy consumption along different paths. For example, climbing and high-speed flight consume more power, resulting in a corresponding increase in energy consumption.

[0046] Analyze the distribution of obstacles and determine the flight postures (such as level flight, side flight, and pitch flight) required for the drone to fly in different areas. Complex flight postures require higher control of the drone, increase the difficulty of flight, and thus increase the difficulty cost of the flight posture.

[0047] The flight cost is calculated by taking into account factors such as the distance between each coordinate point and the obstacle, flight energy consumption, and the difficulty of the flight posture. The weights of different factors can be dynamically adjusted according to actual mission requirements.

[0048] Based on a dynamic balancing strategy, we seek the optimal balance between defect detection value and flight cost. We use intelligent optimization algorithms (such as genetic algorithms and particle swarm optimization) to dynamically adjust the drone's flight path. This ensures maximum coverage of high-value defect detection areas while reducing flight costs and improving inspection efficiency.

[0049] Computer vision technology is used to identify the corners of the transformer, and the contour edges of the transformer are extracted through edge detection algorithms (such as the Canny operator), and then the positions of the corners are determined and virtual coordinate points are generated.

[0050] Based on the virtual coordinate points, an equidistant shooting path is planned around the virtual markers. By setting an appropriate shooting radius and angle interval, the drone can capture comprehensive images of the transformer from multiple angles and distances, acquiring rich inspection data.

[0051] The control execution module controls the flight of the drone based on its flight path and collects multimodal information; The control execution module is the "hands and feet" of the system, responsible for accurately controlling the flight of the drone according to the flight path generated by the path planning module and completing the task of collecting multimodal information.

[0052] The flight path data sent by the path planning module is received and converted into control instructions for the UAV. Through the UAV's flight control system, the flight attitude and speed of the UAV are adjusted in real time to ensure that the UAV flies along the planned path.

[0053] Combined with data from drone sensors (such as GPS and IMU), real-time flight path correction is achieved. If the drone deviates from the preset path, the flight control system automatically calculates the deviation and generates corresponding control signals to adjust the drone's flight direction and return it to the correct path.

[0054] Control the sensor's operating status based on the flight path and preset shooting points. When the drone reaches the specified coordinate point, trigger operations such as image acquisition and temperature monitoring to ensure that the collected data matches the planned path.

[0055] The collected multimodal information is initially processed and packaged, including data format conversion and compression, to reduce data transmission volume and improve data transmission efficiency. The processed data is transmitted to the ground data processing module via a wireless communication link.

[0056] The data processing module acquires the multimodal information collected by the drone and, based on a multimodal data fusion strategy, determines abnormal substation status information. The data processing module is the system's "analysis hub," responsible for in-depth processing and analysis of the multimodal information collected by the drone, ultimately determining abnormal transformer status information.

[0057] After receiving multimodal data transmitted by drones, the system first extracts the timestamp and spatial location information from the data. Using a high-precision clock synchronization algorithm (such as the Network Time Protocol (NTP) or the Precision Time Protocol (PTP)), the timestamps of data collected by different sensors are uniformly calibrated to ensure data consistency in the temporal dimension.

[0058] Based on the substation's 3D map and the drone's spatial positioning information, a unified spatial coordinate system is established. Data collected by different sensors is mapped into this coordinate system, achieving precise spatial alignment and generating spatiotemporally aligned raw data, laying the foundation for subsequent analysis.

[0059] Three-level fusion analysis Raw Data Preprocessing and Abnormal Feature Extraction: Preprocessing the spatiotemporally aligned raw data, such as image denoising (using algorithms such as median filtering and Gaussian filtering) and sound signal filtering (removing background noise), improves data quality. Abnormal features are then extracted from the visual image, infrared thermal image, and sound data. Convolutional neural networks (CNNs) are used to analyze visual images to identify abnormalities such as cracks and deformation on the transformer surface. Infrared thermal image analysis algorithms are used to detect abnormal temperature areas in the transformer. Signal processing algorithms (such as fast Fourier transforms (FFTs)) are used to perform spectral analysis on the sound data to extract abnormal vibration frequencies and noise characteristics.

[0060] Building a multi-feature association model: Using the extracted abnormal features of each sensor as nodes, a graph neural network (GNN) is used to construct a multi-feature association model. This model can identify potential correlations between different types of abnormal features. For example, overheating inside a transformer may cause an increase in casing temperature accompanied by abnormal vibration. By training the GNN, the association patterns between these features are learned, providing a more comprehensive understanding of the transformer's operating status.

[0061] Abnormal feature output based on a weighting algorithm: A weighting algorithm is introduced to account for the significance of different abnormal features and the varying reliability of various sensors. Each abnormal feature is assigned a weight based on its importance in determining the transformer's abnormal condition, as well as factors such as sensor stability and accuracy. For example, acoustic sensor data, which is subject to significant environmental interference, is assigned a relatively low weight. Visual image features, which directly reflect the transformer's surface condition, are assigned an appropriate weight based on their degree of abnormality. This weighted calculation highlights key abnormal information and outputs representative abnormal features.

[0062] The system has a pre-established database of correspondences between abnormal features and abnormal types, built based on a large amount of historical fault data and expert experience. The abnormal features output through the three-level fusion analysis are matched with the standard features in the database.

[0063] Based on the matching results, the transformer anomaly type is determined and a detailed anomaly diagnostic report is generated. This report, presented in intuitive charts and text format, includes information such as the anomaly type, location, and severity assessment. This allows maintenance personnel to quickly understand the transformer anomaly and take timely maintenance measures to ensure safe and stable substation operation.

[0064] The intelligent identification device for abnormal status of transformer includes a drone equipped with a visual camera, an infrared thermal imager, a temperature sensor, an inertial measurement unit and a GPS.

[0065] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention are within the scope of protection of the present invention.

Claims

1. Intelligent identification method for abnormal state of transformer, characterized in that: include: Inspection step: obtain the substation environment image taken by the drone and construct a 3D map of the substation; a transformer defect screening step, screening abnormal transformers through target detection in the three-dimensional map of the substation and obtaining the flight path of the drone; The flight path optimization step calculates the defect detection value and flight cost of each coordinate point, dynamically adjusts the drone's flight path based on a dynamic balance strategy, identifies the transformer's corners, generates virtual coordinate points, and plans an equidistant shooting path around the virtual marker points based on the virtual coordinate points; The multimodal recognition step, based on the UAV flight path, synchronously collects visual images, infrared thermal imaging and sound data taken by the UAV, and identifies the abnormal status of the transformer through a multimodal data fusion strategy.

2. The method for intelligently identifying abnormal transformer status according to claim 1, characterized in that: It includes a defect assessment strategy, which is used to obtain historical defect data of the substation and a three-dimensional map of the substation. Based on the current environmental image features and historical defect data in the three-dimensional image of the substation, the defect type and defect probability of each coordinate point are evaluated. Based on the defect type and defect probability of each coordinate point, the defect detection value is calculated.

3. The method for intelligently identifying abnormal transformer status according to claim 1, characterized in that: It includes a flight cost evaluation strategy, which includes calculating the distance between each coordinate point and the obstacle based on the three-dimensional map of the substation, and obtaining the flight energy consumption of the drone based on the flight path of the drone, and obtaining the distribution of the obstacles. According to the obstacle distribution, the flight posture of the drone is obtained, and the flight cost is obtained according to the distance between each coordinate point and the obstacle, the flight energy consumption and the difficulty of the flight posture.

4. The method for intelligently identifying abnormal transformer status according to claim 1, characterized in that: The inspection step is provided with a three-dimensional map component sub-step, and the three-dimensional map component sub-step includes The coordinate system component step takes the geometric center of the transformer as the origin and automatically generates virtual marking points at the four corners of the equipment; In the contour fitting step, the original point cloud is obtained through lidar scanning, and after noise reduction processing, the device surface data is extracted through power separation, and the 3D model is fitted to generate the device contour; The obstacle map construction step is to identify static obstacles and dynamic obstacles through the substation environment image. When the obstacle is a static obstacle, the area to which the obstacle belongs is marked. When the obstacle is a dynamic obstacle, the obstacle type is analyzed. When the obstacle type is a transmission line, a swing model is constructed according to the current environmental characteristics, and the obstacle movement area is marked according to the swing model. When the obstacle type is a moving object, the Brownian motion model is used to predict the obstacle movement trajectory. Based on the obstacle movement trajectory and the obstacle area, a three-dimensional substation map is constructed.

5. The method for intelligently identifying abnormal transformer status according to claim 4, characterized in that: The coordinate system assembly step is provided with a virtual point assembly sub-step, which includes obtaining the outer contour dimensions of the transformer, calculating the transformer coordinate offset, and using AprilTag visual detection to determine whether the recognition is successful. If the recognition is successful, the 3D position of the marker point is parsed and the virtual marker point coordinates are corrected. If the recognition is unsuccessful, the light radar point cloud matching is started, and the device surface is used as the reference plane, and the distance is extrapolated along the normal vector direction to generate a safety buffer layer.

6. The method for intelligently identifying abnormal transformer status according to claim 1, characterized in that: The multimodal recognition step comprises: The time-space reference synchronization step obtains the data collected by each sensor and timestamps each sensor. The three-level fusion analysis step generates spatiotemporal raw data and obtains the abnormal features of each sensor. A multi-feature association model is constructed through a graph neural network. Based on the significance of the abnormal features and the reliability of the sensor, the abnormal features are output through a weighted algorithm. The abnormality diagnosis output step outputs the abnormality type based on the abnormal feature type.

7. The method for intelligently identifying abnormal transformer status according to claim 1, characterized in that: The flight path optimization step includes an equidistant shooting strategy, which includes obtaining virtual coordinate points of the transformer, obtaining the optimal shooting distance based on the actual size of the transformer and camera parameters, and generating an equidistant shooting path based on the optimal shooting distance using two adjacent virtual coordinate points as path control points through a quintic polynomial difference algorithm.

8. Transformer abnormal state intelligent identification system, characterized by: include: An environmental perception module, which is used to collect environmental data of the substation and multimodal information of the transformer, including image information, infrared information, temperature information, and UAV flight attitude and position information; A path planning module, which is equipped with a defect assessment strategy and a flight cost assessment strategy. It calculates the defect detection value and flight value of each coordinate point, dynamically adjusts the UAV's flight path based on a dynamic balance strategy, identifies the corners of the transformer, generates virtual coordinate points, and plans an equidistant shooting path around the virtual marker points based on the virtual coordinate points. The control execution module controls the flight of the drone based on its flight path and collects multimodal information; The data processing module obtains the multimodal information collected by the drone and obtains the abnormal status information of the substation based on the multimodal data fusion strategy.

9. Intelligent identification device for abnormal state of transformer, characterized in that: The invention comprises a drone, which is equipped with a visual camera, an infrared thermal imager, a temperature sensor, an inertial measurement unit and a GPS.

Images