Non-contact large transformer aging detection method, system, device and medium

Abstract

The invention provides a non-contact large transformer aging detection method, system and device and a medium, and relates to the technical field of transformer detection, and the method comprises the steps: collecting a three-dimensional image, surface temperature distribution and environment data of a transformer, and employing a GNSS synchronization timestamp to guarantee the data time sequence consistency. An accurate three-dimensional temperature field grid model is constructed based on the data, grid encryption processing is carried out on a key area, and meanwhile, an environmental compensation coefficient is calculated to correct the influence of environmental factors. A dynamic threshold is learned through a prediction model, temperature anomaly features are analyzed and extracted from a three-dimensional temperature field grid model according to the dynamic threshold, and then an aging risk level is evaluated by using a machine learning model, so that lossless and accurate monitoring and evaluation of the aging state of the large transformer are realized.

Description

Technical Field

[0001] The present application relates to the technical field of transformer detection, and in particular to a non-contact large-scale transformer aging detection method, system, device and medium. Background Art

[0002] Large outdoor transformers are core hub equipment in the power system, and their operating status is directly related to the stability of the regional power grid and the security of power supply. Large outdoor transformers have complex structures and severe load fluctuations, which are easily aggravated by the influence of the external environment, which may cause breakdown or even explosion accidents. These environmental factors include but are not limited to changes in temperature, humidity and wind speed. For example, in high temperature and high humidity environments, the insulation material inside the transformer may age faster. Existing transformer aging detection technologies mainly include contact detection and manual detection, but contact detection equipment is complex to install and has high maintenance costs, and manual detection is time-consuming and labor-intensive. These problems limit the effectiveness and reliability of traditional methods in large transformer aging detection. Summary of the Invention

[0003] The present application provides a non-contact large transformer aging detection method, system, device and medium to solve one or more technical problems existing in the prior art and at least provide a beneficial option or create conditions.

[0004] In one aspect, the present application provides a non-contact large transformer aging detection method, comprising the following steps: Collect 3D images, surface temperature distribution data, and environmental data of large transformers in a non-contact manner, and use GNSS synchronized timestamps for time alignment. Constructing a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and performing grid encryption processing on the winding area, the core area, and the heat dissipation area therein; Calculating an environmental compensation coefficient based on the environmental data and historical environmental data; Calculating a dynamic threshold value through a prediction model according to the environmental compensation coefficient, the surface temperature distribution data and a preset time sliding window; extracting temperature anomaly features from the three-dimensional temperature field grid model according to the dynamic threshold; The temperature anomaly characteristics are input into a machine learning model to output the aging risk level of the large transformer.

[0005] Furthermore, the three-dimensional temperature field grid model of the large transformer is constructed based on the three-dimensional image and the surface temperature distribution data, and the winding area, the core area and the heat dissipation area thereof are subjected to grid encryption processing, including the following steps: Generating an initial grid structure model of the large transformer using a three-dimensional reconstruction algorithm according to the three-dimensional image; Using an image recognition algorithm, identifying and marking the winding area, the core area, and the heat dissipation area in the initial grid structure model; Each data point in the surface temperature distribution data is matched and mapped to the initial grid structure model to construct a three-dimensional temperature field grid model of the large transformer, and local grid encryption processing is performed on the winding area, core area and heat dissipation area therein.

[0006] Furthermore, the environmental data includes real-time temperature, real-time humidity and real-time wind speed, and the historical environmental data includes historical average temperature, historical average humidity and historical average wind speed; The environmental compensation coefficient satisfies the following formula: ; in, represents the environmental compensation coefficient; represents the real-time temperature, represents the real-time humidity, Indicates the real-time wind speed; represents the historical average temperature, represents the historical average humidity, represents the historical average wind speed; represents the temperature weight coefficient, represents the humidity weight coefficient, represents the wind speed weight coefficient; .

[0007] Furthermore, the dynamic threshold includes a temperature gradient amplitude threshold and a temperature standard deviation threshold; The step of extracting temperature anomaly features from the three-dimensional temperature field grid model according to the dynamic threshold comprises the following steps: Calculating in real time the temperature gradient amplitude of each grid cell and the temperature standard deviation of adjacent grid cells in the three-dimensional temperature field grid model; Marking grid cells whose temperature gradient amplitude exceeds the temperature gradient amplitude threshold as first-level abnormal areas; Marking grid cells whose temperature standard deviation exceeds the temperature standard deviation threshold as secondary abnormal areas; The temperature gradient change rate of the primary abnormal area and the temperature fluctuation range of the secondary abnormal area are extracted as the temperature abnormality features.

[0008] Furthermore, the machine learning model includes a support vector machine model and a random forest model; the machine learning model adapts to different transformers through transfer learning.

[0009] Furthermore, the length of the time sliding window is 6 hours, and the sliding step is 30 minutes.

[0010] Furthermore, the prediction model includes a temporal convolutional network model; the temporal convolutional network model includes 3 layers of dilated convolutional layers; wherein the dilation coefficient of the first dilated convolutional layer is 2, the dilation coefficient of the second dilated convolutional layer is 4, and the dilation coefficient of the third dilated convolutional layer is 8.

[0011] On the other hand, the present application provides a non-contact large-scale transformer aging detection system, including a data acquisition module, a three-dimensional temperature field construction module, a temperature anomaly feature extraction module, and an aging risk prediction module; The data acquisition module is used to collect three-dimensional images, surface temperature distribution data and environmental data of large transformers in a non-contact manner, and perform time alignment using GNSS synchronized timestamps; The three-dimensional temperature field construction module is used to construct a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and perform grid encryption processing on the winding area, the core area and the heat dissipation area therein; The temperature anomaly feature extraction module is used to calculate an environmental compensation coefficient based on the environmental data and historical environmental data; calculate a dynamic threshold value through a prediction model based on the environmental compensation coefficient, the surface temperature distribution data and a preset time sliding window; and extract temperature anomaly features from the three-dimensional temperature field grid model based on the dynamic threshold value; The aging risk prediction module is used to input the temperature anomaly characteristics into a machine learning model and output the aging risk level of the large transformer.

[0012] On the other hand, the present application provides a non-contact large-scale transformer aging detection device, which is movable and deployed on unmanned inspection equipment, including a GNSS unit, a thermal imager, a 3D depth camera, an environmental sensor group, an edge computing unit, and a communication unit; The thermal imager is used to collect surface temperature distribution data of large transformers; The 3D depth camera is used to collect a three-dimensional image of the large transformer; The environmental sensor group is used to collect environmental data of the large transformer; The GNSS unit is used to provide a synchronization time stamp to perform time sequence alignment on the surface temperature distribution data, the three-dimensional image, and the environmental data; The edge computing unit includes a memory and a processor, the memory stores a computer program, and the processor implements the aforementioned non-contact large transformer aging detection method when executing the computer program; The communication unit is used to communicate with the remote monitoring platform.

[0013] On the other hand, the present application provides a computer-readable storage medium storing a program executable by a processor. When the program is executed by the processor, it is used to implement the aforementioned non-contact large transformer aging detection method.

[0014] The beneficial effects of the present application are as follows: The present application proposes a non-contact large-scale transformer aging detection method, which collects three-dimensional images, surface temperature distribution and environmental data of the transformer, and uses GNSS synchronized timestamps to ensure data timing consistency. Based on these data, an accurate three-dimensional temperature field grid model is constructed, and grid encryption processing is performed on key areas. At the same time, the environmental compensation coefficient is calculated to correct the influence of environmental factors. The dynamic threshold is learned through the prediction model, and the temperature anomaly characteristics are analyzed and extracted from the three-dimensional temperature field grid model according to the dynamic threshold. Then, the aging risk level is evaluated using a machine learning model, thereby achieving non-destructive and accurate monitoring and evaluation of the aging status of large transformers. The present application also provides corresponding systems, devices and media. The beneficial effects of the systems, devices and media are similar to those of the method and will not be repeated here.

[0015] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be achieved and obtained through the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation to the technical solution of the present invention.

[0017] Figure 1 This is a flow chart of the non-contact large transformer aging detection method provided by the present application; Figure 2 This is a schematic diagram of the implementation process of determining a three-dimensional temperature field grid model provided by this application; Figure 3 This is a schematic diagram of the implementation process of the detection provided by this application to obtain the aging risk level; Figure 4 This is a structural diagram of the non-contact large-scale transformer aging detection system provided by this application; Figure 5 This is a structural diagram of the non-contact large transformer aging detection device provided by this application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0019] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be considered as limiting the present application. All other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0022] Large transformers are crucial components in power systems, and their operating status directly affects the safety and stability of the power grid. Therefore, it is of great significance to effectively monitor and evaluate the aging status of large transformers. In the existing technology, aging detection methods mainly include two categories: contact and non-contact. Contact detection usually involves physically connecting sensors to key parts of the transformer to obtain data. This method often requires downtime, which increases power outage time and costs, and direct contact may cause potential damage to the equipment. In addition, due to the limitations of installing sensors, contact detection is difficult to achieve comprehensive coverage monitoring of the entire equipment, and cannot provide real-time, continuous data flow.

[0023] Furthermore, large outdoor transformers, due to the unique nature of their operating environment, are significantly affected by numerous environmental factors. Specifically, outdoor transformers are directly exposed to the elements and experience dramatic temperature swings, ranging from freezing to scorching. Extreme temperatures not only affect the efficiency of the transformer's internal components but also accelerate the aging of insulation materials, increasing the risk of localized overheating, potentially leading to equipment failure or shortening service life. High humidity can cause condensation inside the transformer, potentially causing electrical shorts or corrosion, especially in inadequately sealed components. Prolonged exposure to humidity can also weaken insulation performance, increasing the likelihood of a voltage drop and threatening the safe and stable operation of the power grid.

[0024] Furthermore, while adequate wind speeds can help dissipate heat, strong winds or unstable airflow patterns can alter the cooling system's effectiveness, resulting in ineffective cooling in certain areas. Furthermore, inclement weather conditions such as sandstorms can cause mechanical wear or other damage to the transformer after foreign particles enter the transformer.

[0025] On the other hand, while existing non-contact detection technologies can be performed without interrupting service, most of these methods lack comprehensive analytical tools. For example, some rely solely on a single type of sensor (such as an infrared thermal imager) to collect surface temperature information, failing to fully integrate three-dimensional structural images and other environmental parameters for comprehensive analysis. This one-sided data collection approach can lead to inaccurate diagnostic results and the inability to accurately identify specific aging areas or types within the transformer. Furthermore, traditional methods perform poorly when dealing with complex environmental factors. For example, they fail to account for the impact of changes in ambient temperature and humidity on measurement results, further impacting the accuracy of aging status assessments.

[0026] Secondly, in the assessment method based on fixed thresholds, an alarm or shutdown operation is triggered when the temperature of a certain part of the transformer exceeds the preset safety limit. However, this method lacks the ability to predict the quantitative classification of the degree of aging. It relies only on a single indicator (such as temperature exceeding the limit) as the shutdown standard, which cannot effectively reflect the overall health status of the transformer, limiting its scope of application and effectiveness.

[0027] Based on the above-mentioned situation, the embodiments of the present application propose a non-contact large-scale transformer aging detection method, system, device and medium. The embodiments of the present application integrate multiple non-contact sensing technologies, including 3D depth cameras, thermal imagers and environmental sensor groups, and can simultaneously obtain the three-dimensional structure, surface temperature distribution and working environment conditions of the transformer. By constructing an accurate three-dimensional temperature field grid model and using machine learning algorithms, an accurate assessment of the degree of transformer aging is achieved, thereby overcoming the shortcomings of the existing technology. This method improves the accuracy and efficiency of aging detection, reduces interference with equipment, and supports the stable operation of the power system.

[0028] First, the non-contact large transformer aging detection method provided by the embodiment of the present application will be described in detail with reference to the accompanying drawings.

[0029] The non-contact large-scale transformer aging detection method proposed in the embodiments of the present application can be applied to a terminal or a server, or can be software running on the terminal or server. The terminal can be, but is not limited to, a tablet computer, a laptop computer, or a desktop computer. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks, and big data and artificial intelligence platforms.

[0030] Reference Figure 1 The implementation process of the non-contact large transformer aging detection method provided in the embodiment of the present application includes but is not limited to the following steps.

[0031] In step S110 , a three-dimensional image, surface temperature distribution data, and environmental data of a large transformer are collected in a non-contact manner, and time alignment is performed using a GNSS synchronized timestamp.

[0032] It should be noted that GNSS refers to the Global Navigation Satellite System (GNSS). GNSS synchronized timestamps refer to obtaining high-precision time signals through global navigation satellite systems (such as GPS, BeiDou, and Galileo), generating unified time tags for devices or systems, and achieving cross-regional and cross-device time synchronization. GNSS satellites are equipped with highly stable atomic clocks and continuously broadcast navigation messages containing precise time information. Receiving terminals calculate satellite signals to obtain the current Coordinated Universal Time (UTC) or system reference time, and then generate a timestamp. This timestamp has global consistency and nanosecond-level accuracy, and is widely used in scenarios requiring high-precision time synchronization, such as communication network synchronization, precision measurement, IoT device collaboration, and power system phase alignment. It ensures the consistency of the time base of each node and improves the accuracy of data recording and event sorting.

[0033] In step S110, a 3D image, surface temperature distribution data, and environmental data of the large transformer are collected non-contactly, and time alignment is performed using GNSS synchronized timestamps. This step provides basic data support for subsequent analysis, ensuring temporal consistency and avoiding data analysis errors caused by time asynchrony. This non-contact collection method not only reduces operational interference with equipment but also enables comprehensive data acquisition without affecting normal transformer operation.

[0034] Step S120 , constructing a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and performing grid encryption processing on the winding area, the core area, and the heat dissipation area.

[0035] It's important to note that the winding area is a core component of large transformers. Composed primarily of coils made of conductive materials such as copper or aluminum, it transfers electrical energy between the primary and secondary sides. The operating condition of this area directly impacts the transformer's electrical performance. Issues such as aging, overheating, or partial discharge can lead to reduced efficiency or even failure. Therefore, monitoring its temperature changes and insulation condition is crucial.

[0036] The iron core, typically constructed from stacked silicon steel laminations, provides a low-reluctance path to enhance the magnetic field and minimize energy loss. As the energy conversion center of the transformer, the efficient operation of the iron core is crucial to maintaining the stability and efficiency of the entire system. Any anomalies due to overheating, magnetic saturation, or manufacturing defects can result in additional energy loss and potentially damage the windings and other components, making this area a key monitoring target.

[0037] The heat dissipation area, including coolers, fans, and oil pumps, dissipates heat generated by the transformer during operation, ensuring that all components remain within a safe operating temperature range. Good heat dissipation design and functional effectiveness are critical to extending transformer life and improving operational efficiency. If the heat dissipation system fails to function effectively, internal equipment temperatures will rise, accelerating the aging of insulation materials and increasing the risk of failure. Therefore, monitoring the status of the heat dissipation area helps promptly identify and resolve potential temperature control issues.

[0038] In step S120, a 3D temperature field mesh model of the large transformer is constructed based on the acquired 3D image and surface temperature distribution data. Mesh refinement is performed on the winding, core, and heat dissipation areas. This step helps to more accurately simulate and analyze the temperature distribution within the transformer. Meshing improves the model's resolution, particularly in critical areas, allowing for more accurate capture of temperature changes in these areas and providing a detailed thermodynamic basis for aging assessment.

[0039] Step S130: Calculate the environmental compensation coefficient based on the environmental data and historical environmental data.

[0040] In step S130, an environmental compensation coefficient is calculated based on current and historical environmental data. This step accounts for the impact of environmental factors such as temperature, humidity, and wind speed on the transformer surface temperature. This coefficient is used to correct for these factors, ensuring the authenticity and accuracy of the temperature measurement results. This is done to eliminate bias in transformer temperature monitoring caused by environmental changes, making aging detection more reliable.

[0041] Step S140 , calculating a dynamic threshold value through a prediction model according to the environmental compensation coefficient, the surface temperature distribution data and a preset time sliding window.

[0042] In step S140, a dynamic threshold is calculated using a predictive model using the environmental compensation coefficient, surface temperature distribution data, and a preset time sliding window. This process aims to adjust the anomaly detection criteria based on real-time environmental conditions, making the detection criteria more flexible and adaptable. Setting a dynamic threshold can better identify abnormal temperature changes that may occur in specific environments, improving the sensitivity and accuracy of early fault warnings.

[0043] Optionally, set the length of the time sliding window to 6 hours and the sliding step to 30 minutes.

[0044] Specifically, a 6-hour sliding window allows the system to capture trends in transformer surface temperature over time and across the environment, while a 30-minute sliding step ensures highly sensitive monitoring of these trends. This setup not only identifies short-term temperature fluctuations but also effectively monitors long-term temperature trends, enabling a more comprehensive assessment of transformer health.

[0045] Step S150 : extracting temperature anomaly features from the three-dimensional temperature field grid model according to the dynamic threshold.

[0046] In step S150, temperature anomaly features are extracted from the three-dimensional temperature field grid model based on a dynamic threshold. This step is key to achieving a detailed analysis of the transformer's aging status. By calculating the temperature gradient amplitude and temperature standard deviation, potential aging hotspots or abnormal areas can be effectively identified. This multi-dimensional feature-based anomaly detection method helps accurately locate potential problems within the transformer, laying the foundation for subsequent risk assessment.

[0047] Step S160: Input the temperature anomaly characteristics into the machine learning model and output the aging risk level of the large transformer.

[0048] In step S160, the extracted temperature anomaly signatures are fed into a machine learning model to output the aging risk level of the large transformer. This step applies the machine learning model, combined with a large amount of existing training data, to automatically assess the transformer's current aging condition. This approach not only improves diagnostic efficiency but also enables adaptive adjustments based on different transformer characteristics, providing more personalized aging risk assessment results.

[0049] In some embodiments of the present application, reference is made to Figure 2In step S120, a three-dimensional temperature field grid model of a large transformer is constructed based on the three-dimensional image and surface temperature distribution data, and the winding area, core area and heat dissipation area are meshed. The implementation process includes but is not limited to the following steps.

[0050] Step S210 : generating an initial grid structure model of the large transformer based on the three-dimensional image using a three-dimensional reconstruction algorithm.

[0051] In step S210, a preliminary geometric mesh structure model is constructed using the non-contact collected 3D image data of the large transformer and a 3D reconstruction algorithm. This model provides the spatial framework for subsequent temperature field mapping and analysis, ensuring a one-to-one correspondence between temperature distribution information and the actual physical structure. Establishing this initial mesh structure model is the foundation for constructing the 3D temperature field, helping to transform abstract data into a visual physical model and supporting detailed analysis.

[0052] Step S220 , using an image recognition algorithm, identifying and marking the winding area, the core area, and the heat dissipation area in the initial grid structure model.

[0053] In step S220, based on the existing initial mesh model, image recognition techniques, such as convolutional neural networks (CNNs), edge detection algorithms, or feature matching algorithms, are combined to automatically identify and segment key functional areas within the model. These areas include the transformer's core components, namely the winding area, the core area, and the heat sink area. Accurately identifying and marking these areas provides a basis for subsequent local encryption and targeted analysis, improving the model's engineering practicality and diagnostic accuracy.

[0054] Step S230 , mapping each data point in the surface temperature distribution data to the initial grid structure model, constructing a three-dimensional temperature field grid model of the large transformer, and performing local grid encryption processing on the winding area, core area and heat dissipation area.

[0055] In step S230, the collected infrared thermal imaging or other surface temperature distribution data is accurately superimposed on the three-dimensional grid structure model using coordinate mapping technology, thereby forming a three-dimensional temperature field grid model containing temperature information. Based on this, local mesh refinement is performed on identified key areas (winding area, core area, and heat dissipation area). This involves further refining the mesh cells on the basis of the original mesh to improve the spatial resolution of the model. This refinement strategy significantly improves the ability to capture temperature changes in key areas, helping to detect even minor abnormal temperature rises, and thus more accurately reflecting the transformer's aging status and potential failure risks.

[0056] In some embodiments of the present application, the three-dimensional reconstruction algorithm adopts a depth map fusion algorithm, such as the KinectFusion algorithm. This algorithm constructs an accurate large-scale transformer geometric structure model by continuously capturing depth images of the scene and fusing these images into a gradually refined three-dimensional model in real time. Depth map fusion algorithms such as KinectFusion are particularly suitable for non-contact three-dimensional reconstruction tasks because they can not only process complex object surfaces, but also effectively reduce noise and voids, and improve the quality and resolution of the reconstructed model. Using such algorithms for three-dimensional modeling of transformers can provide a high-precision spatial framework for subsequent steps such as temperature field analysis and aging assessment, ensuring the accuracy and reliability of data mapping. In addition, due to its efficient data processing capabilities, good reconstruction effects can be guaranteed even in dynamic environments, increasing the flexibility and applicability of the detection process.

[0057] In some embodiments of the present application, the environmental data includes real-time temperature, real-time humidity, and real-time wind speed, and the historical environmental data includes historical average temperature, historical average humidity, and historical average wind speed. In step S130, the environmental compensation coefficient satisfies the following formula (1): (1); In formula (1), represents the environmental compensation coefficient; Indicates the real-time temperature. Indicates real-time humidity. Indicates real-time wind speed; represents the historical average temperature, represents the historical average humidity, represents the historical average wind speed; represents the temperature weight coefficient, represents the humidity weight coefficient, represents the wind speed weight coefficient; .

[0058] Formula (1) dynamically compares real-time monitoring environmental data (temperature, humidity, wind speed) with historical averages and calculates the environmental compensation coefficient based on preset weight coefficients to accurately quantify the impact of environmental factors on equipment operation. Specifically, first, dynamic calibration eliminates measurement deviations caused by environmental fluctuations, such as temperature distortion in high-temperature and high-humidity environments. Second, multi-factor synergistic compensation is performed, including temperature weighting to dominate the heat transfer effect, humidity weighting to correct insulation material performance degradation, and wind speed weighting to quantify changes in heat dissipation efficiency. In addition, normalization ensures that the sum of the weight coefficients is 1, maintaining the consistency of the physical meaning of the compensation coefficients and allowing the weight combination to be adjusted according to different regional climates. Furthermore, benchmark adaptation uses the historical environmental average as a reference system to identify the offset of the current environment relative to the normal state. This compensation mechanism enables the aging detection system to eliminate environmental interference and focus on changes in the equipment's state. For example, in hot weather, the sensitivity of temperature-related parameters is reduced to avoid misjudgment, and in strong wind conditions, the wind speed weighting is used to quickly respond to changes in heat dissipation conditions and improve the reliability of temperature anomaly identification. Therefore, this dynamic compensation strategy is one of the core technical supports for achieving accurate aging detection.

[0059] In some embodiments of the present application, a dynamic threshold is calculated through a prediction model based on an environmental compensation coefficient, surface temperature distribution data, and a preset time sliding window; wherein the prediction model includes a lightweight timing model, such as a temporal convolutional network (TCN).

[0060] TCN plays a key role in calculating dynamic thresholds based on environmental compensation coefficients, surface temperature distribution data, and a preset time sliding window. As a lightweight time series model, TCN can effectively capture long-term dependencies in time series data. Through its unique dilated convolutional layer, it can significantly reduce the number of model parameters and computational complexity without sacrificing performance. This enables TCN to accurately identify trends and abnormal fluctuations in temperature changes when processing tasks such as large-scale transformer aging detection that require analyzing long-span data, thereby providing a reliable basis for calculating dynamic thresholds and enhancing the accuracy and timeliness of transformer status assessments. Therefore, TCN not only improves the efficiency of the prediction model, but also enhances the system's adaptability to environmental changes, making it an important tool for achieving accurate aging detection.

[0061] In some embodiments of the present application, the TCN includes three dilated convolutional layers; wherein the dilation coefficient of the first dilated convolutional layer is 2, the dilation coefficient of the second dilated convolutional layer is 4, and the dilation coefficient of the third dilated convolutional layer is 8.

[0062] The first dilated convolutional layer allows the network to view the time series of the input data at regular intervals (i.e., with a dilation factor of 2), meaning it can cover a relatively long timeframe without increasing the filter size. This design helps capture recent but not immediate temperature trends, making it particularly useful for identifying short-term fluctuations that could affect transformer health.

[0063] As the number of layers increases, the expansion factor of the second dilated convolutional layer increases to 4, further expanding the network's receptive field, enabling it to consider data changes over longer time periods. This layer can identify trends or patterns that span hours or even longer, which is crucial for understanding how environmental factors affect the operating status of transformers over the long term.

[0064] The third dilated convolutional layer has an expansion factor of 8, significantly expanding the network's receptive field and enabling it to capture information on a more macroscopic timescale. This is particularly important for analyzing transformer aging, as aging is a gradual process involving the cumulative effects of multiple factors over a long period of time. In this way, this layer can help identify potential risk points that may lead to equipment aging and provide a basis for subsequent risk assessment.

[0065] This multi-layered design with a gradually increasing expansion coefficient enables the TCN to effectively extract features at different time scales, thereby improving its ability to perceive transformer state changes. Combined with a 6-hour sliding window and a 30-minute sliding step, the TCN can not only accurately identify abnormal temperature fluctuations in the short term, but also capture aging trends over longer timeframes. This provides strong support for the accurate calculation of dynamic thresholds, thereby improving the sensitivity and reliability of the entire aging detection system. Furthermore, due to the TCN's high efficiency and lightweight nature, real-time monitoring and early warning can be achieved even in resource-constrained edge computing environments, greatly enhancing the system's practical value.

[0066] In some embodiments of the present application, the dynamic threshold value includes a temperature gradient amplitude threshold value and a temperature standard deviation threshold value. Figure 3 ,According to the dynamic threshold, the implementation process of extracting temperature anomaly ,features from the three-dimensional temperature field grid model includes but is not limited to the ,following steps.

[0067] Step S310 , calculating in real time the temperature gradient amplitude of each grid cell and the temperature standard deviation of adjacent grid cells in the three-dimensional temperature field grid model.

[0068] In step S310, based on the constructed three-dimensional temperature field grid model, the temperature gradient amplitude (i.e., the rate of temperature change in space) and the temperature standard deviation between that grid cell and its neighbors are calculated for each grid cell. This process quantifies the spatial inhomogeneity and local fluctuations in temperature distribution, providing critical data for subsequent identification of abnormal areas. Real-time calculation ensures the dynamic and responsive nature of aging status assessment, helping to promptly identify potential risk points.

[0069] Step S320 : Marking the grid cells whose temperature gradient amplitude exceeds the temperature gradient amplitude threshold as first-level abnormal areas.

[0070] In step S320, each grid cell is evaluated using a pre-set temperature gradient amplitude threshold. If the temperature gradient of a grid cell is significantly higher than normal, it indicates a sudden temperature change in that area, possibly due to a local hotspot or material degradation. Marking these grid cells as Level 1 abnormal areas helps quickly locate key areas that may have structural defects or early failures, providing a basis for subsequent risk level assessment.

[0071] Step S330 : Marking the grid cells whose temperature standard deviation exceeds the temperature standard deviation threshold as secondary abnormal areas. In step S330, the temperature standard deviation of each grid cell is compared with the surrounding area to see if it exceeds the set dynamic standard deviation threshold, thereby identifying areas with large temperature fluctuations. Large temperature standard deviations generally indicate unstable local thermal distribution, which may be affected by aging, poor contact, or other non-uniform factors. Marking such areas as secondary abnormal areas can further refine the abnormality type, help distinguish between the stability and randomness of local temperature rise, and improve the comprehensiveness and accuracy of diagnosis.

[0072] Step S340 , extracting the temperature gradient change rate of the first-level abnormal area and the temperature fluctuation range of the second-level abnormal area as temperature anomaly features.

[0073] In step S340, after completing the delineation of primary and secondary anomaly regions, representative temperature anomaly features are further extracted, including the temperature gradient change rate (reflecting the severity of the temperature change in the primary anomaly region) and the temperature fluctuation range (reflecting the magnitude of temperature changes over time and space) in the secondary anomaly region. These features not only enrich the data dimensions of aging analysis but also provide quantitative input parameters for subsequent machine learning models for aging risk assessment, enhancing the scientific nature and robustness of the model's judgments.

[0074] In some embodiments of the present application, machine learning models include support vector machine models and random forest models. These models are used to process temperature gradient characteristics and insulation oil degradation characteristics data collected from transformers to assess the aging degree of the transformers.

[0075] Support vector machines classify data by finding the optimal splitting hyperplane, while random forests use multiple decision trees to make voting decisions. Combining the two can improve the accuracy and robustness of predictions. In addition, transfer learning technology is used to make the model adaptable to different types of transformers.

[0076] Machine learning models are adapted to different transformers through transfer learning. Transfer learning allows a model to apply knowledge learned from one or more source tasks to a target task, even if the target task's data distribution differs from the source tasks. This approach is particularly suitable for large equipment such as transformers, as different transformer models may have different structures and operating parameters, but their aging mechanisms are similar. Through transfer learning, a trained model can be effectively adapted to new transformer models, reducing the amount of data and time required for retraining while maintaining high prediction accuracy.

[0077] In some embodiments of the present application, the aging risk level includes a low risk level, a medium risk level, a high risk level, and an emergency risk level.

[0078] A low aging risk level means the transformer's degree of polymerization is greater than 500, indicating the insulation paper is close to new, with less than 30% loss in mechanical strength and a slow aging rate. In this state, there are no significant temperature gradient anomalies, the furfural concentration in the oil is extremely low, the acidity of the insulating oil is normal, and chemical decay has not yet begun.

[0079] A medium aging risk level means the transformer's degree of polymerization has dropped below 500 but remains above 400. At this point, the insulation paper enters an accelerated aging phase, experiencing a loss of mechanical strength between 30% and 50%. Particular attention should be paid to the risk of localized overheating. Typical signs include elevated localized temperature gradients or excessive furfural concentrations. Additionally, the acid value may slightly increase, accelerating cellulose degradation.

[0080] A high aging risk level indicates that the transformer's degree of polymerization (DOP) has dropped below 400, but remains above 250. At this point, the insulation paper's mechanical strength has lost more than 50%, indicating it is nearing the end of its lifespan (a DOP of 250 is considered end-of-life), making it susceptible to breakdown under short-circuit conditions. Abnormalities in multiple indicators indicate an accelerated aging rate. It is recommended that a power outage and disassembly be scheduled within three months, with partial insulation replacement or oil filtration to mitigate further deterioration.

[0081] A high aging risk level means the transformer's degree of polymerization has dropped below 250. At this point, the insulation paper's mechanical strength has almost completely been lost, and the transformer could collapse at any time due to electromagnetic forces or thermal stress. In this situation, the transformer should be shut down immediately, the load should be transferred, the winding insulation should be replaced, and the emergency response plan should be activated to prevent a potential major accident.

[0082] In some embodiments of the present application, the process of calculating and outputting the remaining useful life of a large transformer based on the aging risk level, environmental compensation coefficient and temperature anomaly characteristics, combined with the Weibull distribution model and the expected value method, includes but is not limited to the following steps.

[0083] Step S410 : According to the aging risk level, a risk parameter mapping table is called to obtain shape parameters and scale parameters of the Weibull distribution model.

[0084] In step S410, the corresponding Weibull distribution model parameters—namely, shape and scale parameters—are automatically matched based on the aging risk level output by the machine learning model using a pre-established risk parameter mapping table. Different aging levels represent differences in the degree of aging of the transformer's internal insulation materials, conductive components, and other components. The parameters of the Weibull distribution are closely related to its failure mode. This step provides a statistical model foundation appropriate for the current aging state for subsequent lifespan predictions, thereby improving the accuracy and personalization of remaining useful life calculations.

[0085] Step S420 , calculating and outputting the remaining service life of the large transformer based on the environmental compensation coefficient and the temperature anomaly characteristics in combination with the Weibull distribution model and the expected value method.

[0086] In step S420, based on the obtained Weibull distribution parameters, this step further incorporates the impact of environmental compensation coefficients and temperature anomaly characteristics on the equipment's operating status. This comprehensively considers the effects of the actual operating environment (such as temperature, humidity, and wind speed) on the acceleration or inhibition of the transformer's aging rate, as well as the additional thermal stress caused by local hot spots. Using the expected value method to integrate the Weibull distribution, the expected time until the transformer fails or reaches the end of its life, i.e., the remaining useful life (RUL), is calculated. This method not only integrates statistical reliability theory but also incorporates real-time monitoring data, making the life prediction results more practical for engineering applications. This helps operations and maintenance personnel formulate scientific maintenance and replacement plans, thereby improving the safety and economic efficiency of power system operations.

[0087] It should be noted that the Weibull distribution model is a probabilistic distribution model used to describe the life characteristics of equipment or systems. It is particularly suitable for estimating the remaining life of electrical equipment with different aging patterns. It uses shape parameters and scale parameters to characterize the changing trend of the failure rate over time and can flexibly adapt to various types of failure modes, such as early failures, random failures, and wear and tear failures. Combined with the expected value method, that is, calculating the mean of the Weibull distribution to estimate the average remaining life of the equipment, it can provide scientific aging risk assessment and remaining life prediction for key equipment such as large transformers. This method not only takes into account the inherent aging characteristics of the equipment, but also combines the impact of the actual operating environment on life, thereby improving the accuracy and reliability of the prediction results and helping to optimize predictive maintenance strategies.

[0088] In some embodiments of the present application, the remaining service life satisfies the following formula (2): (2); In formula (2), represents the environmental compensation coefficient, Indicates the remaining service life, represents the average value of the temperature gradient change rate in the temperature anomaly feature, represents the scale parameter, represents the shape parameter, Represents the mean multiplier for the Weibull distribution model.

[0089] Formula (2) dynamically calculates the remaining useful life of the equipment by quantifying key temperature anomaly characteristics during equipment aging (such as the average rate of change of the temperature gradient), combining them with the life characteristic parameters of the equipment material (scale and shape parameters) and environmental compensation factors. This calculation mechanism incorporates the accelerated effect of temperature anomalies on equipment aging into the evaluation system and achieves the numerical conversion of life prediction through a statistical model (Weibull distribution mean multiplier). Its technical value lies in providing a precise time basis for equipment maintenance decisions. For example, in the case of large transformers, when an intensified temperature gradient anomaly is detected, the formula can shorten the remaining useful life estimate in real time, guiding operation and maintenance personnel to formulate maintenance or replacement plans in advance, and avoiding unexpected downtime accidents caused by equipment aging and failure.

[0090] In some embodiments of the present application, assuming that the aging risk level is a high risk level, the risk parameter mapping table is called to obtain the scale parameter of the Weibull distribution model. For 8000 hours, the shape parameters is 1.6; at this time, the mean temperature gradient change rate The environmental compensation coefficient is 0.22. The mean multiplier of the Weibull distribution model is calculated to be approximately 0.897, and the remaining service life is finally calculated to be approximately 4142 hours.

[0091] In the same situation, if we ignore the temperature gradient and environmental influence, that is, is 0, The mean multiplier of the Weibull distribution model is calculated to be 1, and the final calculated remaining service life is approximately 7176 hours. However, according to common sense, abnormal temperatures and high temperature environments will shorten the remaining service life of large transformers, highlighting the need for dynamic correction.

[0092] In some embodiments of the present application, the process of marking abnormal areas of a large transformer on a three-dimensional temperature field grid model and issuing an alarm message based on temperature anomaly characteristics, aging risk level, and remaining service life includes but is not limited to the following steps.

[0093] Step S510: Mark the abnormal area corresponding to the temperature abnormality feature in the three-dimensional temperature field grid model, and send its corresponding coordinates to the remote monitoring platform.

[0094] In step S510, the temperature anomaly signatures derived from the previous analysis are first used to precisely mark abnormal areas with potential problems within the three-dimensional temperature field grid model. This not only visually demonstrates which parts of the transformer may be at risk of aging or failure, but also provides a clear location reference for subsequent maintenance work. Furthermore, the specific coordinates of these marked areas are sent to a remote monitoring platform, enabling off-site technicians to gain real-time insights into equipment status, conduct remote diagnosis, and provide decision support, improving the efficiency and accuracy of troubleshooting.

[0095] Step S520: If the aging risk level is a high risk level or an emergency risk level, or the remaining service life is less than a preset life threshold, an alarm message is sent to the remote monitoring platform.

[0096] In step S520, based on the aging risk level assessment results and the remaining service life prediction, if the transformer is found to be at a high risk level or an emergency risk level, or its remaining service life is lower than the preset safety threshold, the system will automatically trigger the alarm mechanism and send an alarm message to the remote monitoring platform. This step is crucial because it is directly related to the safe and stable operation of the power system. Timely alarms can prompt operation and maintenance personnel to take quick measures, such as arranging emergency repairs, replacing components, or adjusting loads, to avoid potential accidents and reduce economic losses and safety hazards. At the same time, this data-driven risk warning mechanism is more efficient and accurate than traditional manual inspections, and helps to improve the intelligent management level of the entire power grid.

[0097] Secondly, refer to Figure 4 The embodiment of the present application also provides a non-contact large-scale transformer aging detection system, including a data acquisition module 610, a three-dimensional temperature field construction module 620, a temperature anomaly feature extraction module 630 and an aging risk prediction module 640.

[0098] The data acquisition module 610 is used to collect three-dimensional images, surface temperature distribution data and environmental data of large transformers in a non-contact manner, and perform time alignment using GNSS synchronized timestamps.

[0099] The three-dimensional temperature field construction module 620 is used to construct a three-dimensional temperature field grid model of a large transformer based on the three-dimensional image and surface temperature distribution data, and perform grid encryption processing on the winding area, core area and heat dissipation area.

[0100] Temperature anomaly feature extraction module 630 is used to calculate an environmental compensation coefficient based on environmental data and historical environmental data. Based on the environmental compensation coefficient, surface temperature distribution data, and a preset time sliding window, a prediction model is used to calculate a dynamic threshold. Based on the dynamic threshold, temperature anomaly features are extracted from the three-dimensional temperature field grid model.

[0101] The aging risk prediction module 640 is used to input the temperature anomaly characteristics into the machine learning model and output the aging risk level of the large transformer.

[0102] Furthermore, refer to Figure 5 , an embodiment of the present application provides a non-contact large transformer aging detection device, which is movable and deployed on unmanned inspection equipment.

[0103] As a mobile and intelligent monitoring device, the device is deployed on unmanned inspection equipment and can be flexibly moved to different locations around the transformer to achieve real-time, non-destructive testing of the transformer's operating status. The device does not require physical contact with the transformer, avoiding the potential safety hazards and operational complexity of traditional detection methods, while improving detection efficiency and adaptability. By deploying the device outside the transformer, key parameters such as its surface temperature distribution, environmental data, and three-dimensional images can be continuously obtained for real-time analysis, thereby accurately assessing the transformer's aging and health status. This mobile feature makes the device suitable for a variety of substation environments and transformers of different models, enhancing the system's versatility and convenience in on-site application, and providing strong support for the implementation of intelligent inspections and condition-based maintenance in power systems.

[0104] The device includes a GNSS unit 710 , a thermal imager 720 , a 3D depth camera 730 , an environmental sensor group 740 , an edge computing unit 750 , and a communication unit 760 .

[0105] GNSS unit 710 provides synchronized timestamps, aligning surface temperature distribution data, 3D images, and environmental data. This unit provides precise time synchronization, generating synchronized timestamps to ensure accurate temporal alignment of collected surface temperature distribution data, 3D images, and environmental data. This is crucial for subsequent data analysis, as it ensures the continuity and consistency of all collected information across the time dimension, providing a reliable time reference for non-contact large transformer aging testing.

[0106] The Thermal Imager 720 is used to collect surface temperature distribution data for large transformers. By accurately measuring the surface temperature of large transformers, potential hot spots or areas of abnormal temperature rise can be identified, which are often early signs of internal equipment failure or degradation. Thermal imaging technology enables rapid scanning of large areas without direct contact with the equipment, improving inspection efficiency and safety.

[0107] The 3D depth camera 730 is used to capture 3D images of large transformers. Based on these high-precision 3D images, a 3D model of the transformer can be constructed, which is crucial for subsequent analysis of physical structural changes and assessment of aging. Furthermore, the 3D images enable mesh refinement of key components (such as windings and cores) to improve the accuracy of local analysis.

[0108] Environmental sensor group 740 collects environmental data from large transformers. Composed of multiple sensors, it monitors the transformer's surrounding environmental parameters in real time, including temperature, humidity, and wind speed. This environmental data is crucial for understanding the impact of external conditions on transformer performance and forms the basis for calculating environmental compensation coefficients. This helps eliminate measurement bias caused by environmental factors and improves the accuracy of aging assessments.

[0109] Edge computing unit 750 includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned non-contact large-scale transformer aging detection method. The advantage of edge computing is that it can perform preliminary processing close to the data source, reducing latency and lowering the load on remote servers.

[0110] Communication unit 760 is used to communicate with the remote monitoring platform. Serving as a bridge between on-site detection devices and the remote monitoring platform, it ensures that collected data and analysis results are promptly transmitted to back-end management personnel. This not only facilitates real-time monitoring but also enables remote experts to quickly respond to any potential risk warnings and implement necessary maintenance measures, greatly improving emergency response capabilities and management efficiency.

[0111] In addition, an embodiment of the present application provides a computer-readable storage medium storing a program executable by a processor. When the program is executed by the processor, it is used to implement the aforementioned non-contact large transformer aging detection method.

[0112] In summary, the non-contact large transformer aging detection method, system, device and medium provided by the embodiments of the present application have the following technical effects.

[0113] First, the embodiments of the present application avoid interference with the operating status of the equipment by adopting a non-contact data collection method, including using a thermal imager to obtain surface temperature distribution data, a 3D depth camera to capture three-dimensional images, and an environmental sensor group to collect real-time environmental parameters, while improving the security and efficiency of data collection.

[0114] Secondly, the embodiment of the present application utilizes the time synchronization service provided by the GNSS unit to ensure the accurate alignment of all collected data in the time dimension, providing a reliable time reference for subsequent data analysis. The use of edge computing units not only enables preliminary processing close to the data source, reducing data transmission delays, but also reduces the load on remote servers, improving the response speed and processing power of the entire system.

[0115] Furthermore, by constructing a three-dimensional temperature field grid model and combining it with environmental compensation coefficients and dynamic threshold calculations, this method can accurately identify transformer aging characteristics and abnormal areas. A machine learning model is used to predict aging risk levels, and the Weibull distribution model and expected value method are combined to assess remaining life, further enhancing the accuracy and reliability of aging status assessments.

[0116] Furthermore, the device is designed to be mobile and deployed in unmanned inspection facilities, making it suitable for a variety of substation environments and transformer types, enhancing the system's versatility and ease of field application. The application of transfer learning technology enables the machine learning model to adapt to different transformer types, improving the system's flexibility and accuracy. This overall solution not only overcomes the limitations of traditional detection methods but also effectively reduces unnecessary maintenance costs and power outages, providing strong support for intelligent inspection and condition-based maintenance in power systems. It has significant practical application value and socioeconomic benefits.

[0117] In some optional embodiments, the functions / operations mentioned in the block diagram may not occur in the order mentioned in the operation schematic diagram. For example, depending on the functions / operations involved, the two boxes shown in succession may actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiments presented and described in the flow chart of the present application are provided in an exemplary manner for the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logical flows presented herein. Optional embodiments are contemplated in which the order of the various operations is changed and the sub-operations described as a part of a larger operation are performed independently.

[0118] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present application. More specifically, given the properties, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the module will be understood within the routine skills of an engineer. Therefore, a person skilled in the art will be able to implement the present application as set forth in the claims using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the appended claims and their equivalents.

[0119] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several programs that enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0120] The logic and / or steps represented in a flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable programs for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can retrieve and execute a program from a program execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, a program execution system, apparatus, or device.

[0121] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, for example, the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing it in a suitable manner as necessary, and then storing it in a computer memory.

[0122] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable program execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0123] In the above description of this specification, reference to the terms "one embodiment / implementation," "another embodiment / implementation," or "certain embodiments / implementations" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in the embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0124] Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

[0125] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A non-contact large transformer aging detection method, characterized in that: The following steps are involved: Collect 3D images, surface temperature distribution data, and environmental data of large transformers in a non-contact manner, and use GNSS synchronized timestamps for time alignment. Constructing a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and performing grid encryption processing on the winding area, the core area, and the heat dissipation area therein; Calculating an environmental compensation coefficient based on the environmental data and historical environmental data; Calculating a dynamic threshold value through a prediction model according to the environmental compensation coefficient, the surface temperature distribution data and a preset time sliding window; extracting temperature anomaly features from the three-dimensional temperature field grid model according to the dynamic threshold; The temperature anomaly characteristics are input into a machine learning model to output the aging risk level of the large transformer.

2. The non-contact large transformer aging detection method according to claim 1, characterized in that: The method of constructing a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and performing grid encryption processing on the winding area, the core area, and the heat dissipation area thereof, comprises the following steps: Generating an initial grid structure model of the large transformer using a three-dimensional reconstruction algorithm according to the three-dimensional image; Using an image recognition algorithm, identifying and marking the winding area, the core area, and the heat dissipation area in the initial grid structure model; Each data point in the surface temperature distribution data is matched and mapped to the initial grid structure model to construct a three-dimensional temperature field grid model of the large transformer, and local grid encryption processing is performed on the winding area, core area and heat dissipation area therein.

3. The non-contact large transformer aging detection method according to claim 1, characterized in that: The environmental data includes real-time temperature, real-time humidity and real-time wind speed, and the historical environmental data includes historical average temperature, historical average humidity and historical average wind speed; The environmental compensation coefficient satisfies the following formula: ； in, represents the environmental compensation coefficient; represents the real-time temperature, represents the real-time humidity, Indicates the real-time wind speed; represents the historical average temperature, represents the historical average humidity, represents the historical average wind speed; represents the temperature weight coefficient, represents the humidity weight coefficient, represents the wind speed weight coefficient; .

4. The non-contact large transformer aging detection method according to claim 1, characterized in that: The dynamic thresholds include a temperature gradient amplitude threshold and a temperature standard deviation threshold; The step of extracting temperature anomaly features from the three-dimensional temperature field grid model according to the dynamic threshold comprises the following steps: Calculating in real time the temperature gradient amplitude of each grid cell and the temperature standard deviation of adjacent grid cells in the three-dimensional temperature field grid model; Marking grid cells whose temperature gradient amplitude exceeds the temperature gradient amplitude threshold as first-level abnormal areas; Marking grid cells whose temperature standard deviation exceeds the temperature standard deviation threshold as secondary abnormal areas; The temperature gradient change rate of the primary abnormal area and the temperature fluctuation range of the secondary abnormal area are extracted as the temperature abnormality features.

5. The non-contact large transformer aging detection method according to claim 1, characterized in that: The machine learning model includes a support vector machine model and a random forest model; the machine learning model adapts to different transformers through transfer learning.

6. The non-contact large transformer aging detection method according to claim 1, characterized in that: The length of the time sliding window is 6 hours, and the sliding step is 30 minutes.

7. The non-contact large transformer aging detection method according to claim 1, characterized in that: The prediction model includes a temporal convolutional network model; the temporal convolutional network model includes three dilated convolutional layers; wherein the dilation coefficient of the first dilated convolutional layer is 2, the dilation coefficient of the second dilated convolutional layer is 4, and the dilation coefficient of the third dilated convolutional layer is 8.

8. Non-contact large transformer aging detection system, characterized by: It includes data acquisition module, three-dimensional temperature field construction module, temperature anomaly feature extraction module and aging risk prediction module; The data acquisition module is used to collect three-dimensional images, surface temperature distribution data and environmental data of large transformers in a non-contact manner, and perform time alignment using GNSS synchronized timestamps; The three-dimensional temperature field construction module is used to construct a three-dimensional temperature field grid model of the large transformer based on the three-dimensional image and the surface temperature distribution data, and perform grid encryption processing on the winding area, the core area and the heat dissipation area therein; The temperature anomaly feature extraction module is used to calculate an environmental compensation coefficient based on the environmental data and historical environmental data; calculate a dynamic threshold value through a prediction model based on the environmental compensation coefficient, the surface temperature distribution data and a preset time sliding window; and extract temperature anomaly features from the three-dimensional temperature field grid model based on the dynamic threshold value; The aging risk prediction module is used to input the temperature anomaly characteristics into a machine learning model and output the aging risk level of the large transformer.

9. Non-contact large transformer aging detection device, characterized in that: The device is mobile and deployed on unmanned inspection equipment, and includes a GNSS unit, a thermal imager, a 3D depth camera, an environmental sensor group, an edge computing unit, and a communication unit; The thermal imager is used to collect surface temperature distribution data of large transformers; The 3D depth camera is used to collect a three-dimensional image of the large transformer; The environmental sensor group is used to collect environmental data of the large transformer; The GNSS unit is used to provide a synchronization time stamp to perform time sequence alignment on the surface temperature distribution data, the three-dimensional image, and the environmental data; The edge computing unit includes a memory and a processor, the memory stores a computer program, and the processor implements the non-contact large transformer aging detection method according to any one of claims 1 to 7 when executing the computer program; The communication unit is used to communicate with the remote monitoring platform.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the non-contact large transformer aging detection method according to any one of claims 1 to 7 when executed by the processor.

Images