Power transmission line diagnosis method and system based on physical information generation and causal analysis

Abstract

The invention discloses a power transmission line diagnosis method and system based on physical information generation and causal analysis, and belongs to the technical field of power transmission line component state diagnosis. Comprising the steps that multi-source data of a target power transmission line component are acquired, and the multi-source data comprise equipment attribute parameters, environment parameters and a real observation image acquired through image acquisition equipment; inputting the equipment attribute parameters and the environment parameters into the physical information generative model to generate a multi-modal reference image; calculating difference information between the real observation image and the multi-modal reference image; and based on a preset causal graph model, performing causal analysis on the difference information, deducing an equipment state reason causing generation of the difference information, and outputting a diagnosis result. A multi-modal reference image conforming to a physical rule is generated by combining a physical information generative model with equipment attribute parameters and environmental parameters, the problem of imaging degradation caused by environmental interference in real observation is solved, and the diagnosis robustness in a complex environment is remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power transmission line component state diagnosis, and particularly relates to a power transmission line diagnosis method and system based on physical information generation and causal analysis. BACKGROUND

[0002] As an important part of the power system, the power transmission line undertakes the key function of power transmission. Its operation state is directly related to the reliability and safety of the power grid. Due to long-term exposure to complex natural environments such as wind, rain, ice, snow, temperature changes, and ultraviolet radiation, defects such as mechanical fatigue, corrosion, loosening, and fracture are prone to occur in the power transmission line body and fittings such as pre-stranding wire, tension clamp, and suspension clamp. If these defects cannot be found and handled in time, serious accidents such as wire breakage and tower collapse may occur, causing large-scale power outages and social and economic losses.

[0003] To ensure the safety of the power transmission line, regular inspection is one of the core works of power operation and maintenance. The traditional inspection mainly relies on manual visual observation or ground telescope observation, which has the limitations of low efficiency, high risk, and great influence of subjective factors. In recent years, with the popularization of unmanned aerial vehicles, helicopters, and fixed cameras for inspection, automatic and semi-automatic inspection technologies based on images have developed rapidly, greatly improving the coverage and frequency of data collection. Correspondingly, image intelligent analysis technology has become the key support for identifying line defects and evaluating the operation state.

[0004] At present, the transmission line image analysis method mainly includes two types: the method based on traditional image processing mainly relies on edge detection, threshold segmentation, morphological operation, template matching and other algorithms to locate and simply identify defects of insulators, conductors, fittings and other components. This kind of method has certain effect under the condition of simple background and good illumination, but is significantly affected by image quality, sensitive to noise, occlusion, illumination change, complex background and other interference, and has poor robustness. The method based on deep learning, especially the convolutional neural network, has strong advantages in target detection, semantic segmentation and other tasks, can automatically learn features, and identify more complex defect modes such as insulator burst, bird nest and foreign object suspension. However, the deep learning method relies on a large number of high-quality labeled samples for training, and in actual engineering, the problem of few defect samples and high labeling cost is very prominent. In addition, the existing method is a “black box” model, and its identification logic is disconnected with the physical failure mechanism of the line components, so it is difficult to maintain stable performance when the sample is insufficient or the imaging condition is poor. More importantly, the existing method generally ignores a basic challenge: the imaging degradation problem in complex environments. The transmission line is mostly located in the wild, mountainous area, coastal area and other places, and the image is easily affected by fog, rain, snow, backlight, motion blur and other factors, resulting in loss of details of key components, decrease of contrast, increase of noise, and serious influence on the accuracy of subsequent analysis. Both traditional methods and deep learning methods lack a mechanism for targeted enhancement of image quality degradation, which limits the recognition accuracy and generalization ability in real scene.

[0005] Therefore, in the actual operation and maintenance environment with scarce samples and variable imaging conditions, an imaging enhancement and state diagnosis method combining the advantages of physical priori and data-driven is urgently needed to improve image quality, enhance defect recognition ability, and provide stable and reliable visual input for subsequent intelligent analysis. SUMMARY

[0006] The purpose of the present application is to overcome the above-mentioned deficiencies, and to provide a transmission line diagnosis method and system based on physical information generation and causal analysis.

[0007] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a transmission line diagnosis method based on physical information generation and causal analysis, comprising the following steps: obtaining multi-source data of a target transmission line component, the multi-source data including device attribute parameters, environmental parameters and real observation images; inputting the equipment attribute parameters and the environment parameters into a pre-trained physical information generative model to generate a multi-modal reference image of the target power transmission line component under corresponding conditions; wherein the physical information generative model is trained by introducing a physical law constraint term, so that the generated multi-modal reference image conforms to the physical characteristics of the target power transmission line component in a healthy state; calculating difference information between the multi-modal reference image and the real observation image; performing causal analysis on most of the difference information based on a pre-set causal diagram model to deduce the equipment state causes leading to the difference information, and obtaining a diagnosis result.

[0008] In the step of inputting the equipment attribute parameters and the environment parameters into a pre-trained physical information generative model to generate a multi-modal reference image of the target power transmission line component under corresponding conditions, the training process of the physical information generative model is as follows: using a cGAN model as an initial generative model; pre-training the initial generative model by using a data set generated based on physical simulation, wherein the physical simulation simulates the electrical, thermal and mechanical behaviors of the target power transmission line component under different equipment attribute parameters and environment parameters; In the pre-training process, a loss function is constructed by a physical law constraint term to obtain a trained physical information generative model.

[0009] The physical law constraint term includes a thermodynamic constraint, and the multi-modal reference image is an infrared reference image. The thermodynamic constraint makes the pixel value distribution of the infrared reference image meet the temperature field distribution law of the heat conduction equation, and the specific formula is as follows:

[0010] wherein, represents a curl, is a material thermal conductivity, and T is a temperature, is a density, and c is a specific heat capacity, is an internal heat source term.

[0011] The physical law constraint term includes an electromagnetic field constraint, and the multi-modal reference image is an electromagnetic distribution reference image. The electromagnetic field constraint is used to make the pixel intensity distribution of the generated electromagnetic distribution reference image conform to the spatial variation law of the electromagnetic field described by the Maxwell equations.

[0012] The physical law constraint term includes an optical imaging constraint, and the multi-modal reference image is a visible light reference image. The optical imaging constraint introduces a physical model of a rendering equation to ensure that the lighting, shadow and reflection characteristics of the generated visible light reference image conform to the optical law of the actual scene.

[0013] The physical model of the rendering equation is as follows:

[0014] where, represents the exit luminance of point p along the exit direction , is the self-luminance, is the reflected light, represents the integral over the hemispherical space , describes the ability of the surface at point p to reflect incident light from direction to exit direction , is the actual incident radiance from the environment or light source, the term represents the influence of the incident angle on the effective received light energy.

[0015] In the step of calculating the difference information between the multi-modal reference image and the real observation image, the difference information includes pixel-level residual, structural similarity deviation, and spectral response difference.

[0016] In the step of performing causal analysis on the difference information based on the preset causal graph model, inferring the equipment state reason causing the difference information, and outputting a diagnosis result, the causal analysis includes intervention on environmental interference variables and equipment state variables for attribution judgment, and the intervention process includes: setting ideal conditions and performing simulated intervention based on the ideal conditions to obtain an intervention structure, analyzing based on the intervention result to obtain an intervention analysis result; adjusting one of the equipment state variables while keeping the other variables unchanged, and analyzing the change trend of the difference information to obtain an adjustment analysis result; outputting a diagnosis report based on the intervention analysis result and the adjustment analysis result.

[0017] Further comprising the following steps: receiving a counterfactual query request containing future settings of equipment attribute parameters and / or environmental parameters, generating a predicted image at a future time point through a physical information generative model based on the settings in the counterfactual query request, and outputting a state prediction report based on the predicted image.

[0018] In a second aspect, the present application provides a power transmission line diagnosis system based on physical information generation and causal analysis, comprising: a data acquisition module for acquiring multi-source data of target power transmission line components, the multi-source data including equipment attribute parameters, environmental parameters, and real observation images; A multimodal reference image generation module is used to input the device attribute parameters and environmental parameters into a pre-trained physical information generative model to generate a multimodal reference image of the target transmission line component under corresponding conditions; wherein, the physical information generative model is trained by introducing physical law constraint terms so that the generated multimodal reference image conforms to the physical characteristics of the target transmission line component in a healthy state; The difference information calculation module is used to calculate the difference information between the multimodal reference image and the real observed image; The causal analysis module is used to perform causal analysis on the difference information based on a preset causal graph model, infer the cause of the device status that led to the difference information, and obtain a diagnostic result.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a transmission line diagnostic method based on physical information generation and causal analysis. Instead of directly identifying defects in potentially degraded real-world observation images, it utilizes a physical information generation model to generate a multimodal reference image representing the "ideal healthy state" under corresponding conditions, based on known equipment attributes and environmental parameters. Anomalies are located by calculating the difference between the generated multimodal reference image and the real-world observation image. This method shifts the focus of analysis from interpreting degraded images to identifying deviations from the multimodal reference image, significantly reducing reliance on the quality of the original image and effectively overcoming the interference of complex environments such as fog, rain, and backlighting on diagnostic results, thus significantly improving the robustness of diagnosis in complex environments.

[0020] Furthermore, embedding physical laws as constraints into the training process of the physical information generative model means that the generated multimodal baseline images are not simply statistical extrapolations of data, but rather images that conform to physical principles. Subsequently, by using a causal graph model to analyze the discrepancies, the root causes of the differences can be inferred. This provides maintenance personnel with a clear chain of physical evidence, significantly improving the credibility and acceptability of the diagnostic results.

[0021] Furthermore, the multimodal benchmark images in this scheme are generated by a physical information generative model, rather than relying on a large number of labeled healthy / defective samples for training, thus fundamentally reducing the requirement for a large number of defective samples. For rare defects or entirely new types of parts, only the device attribute parameters need to be adjusted and input into the physical information generative model to generate the corresponding multimodal benchmark images, thereby enabling state diagnosis in cases with small or even zero samples, and solving the pain point of weak generalization ability of deep learning models in scenarios with scarce samples.

[0022] Furthermore, through causal analysis of the difference information, this solution can further infer the type, severity, and even development trend of the defect, achieving a deeper level of status diagnosis and risk assessment. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural diagram of the system composition of the present invention. Detailed Implementation

[0024] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0025] Example 1 like Figure 1 As shown, a transmission line diagnostic method based on physical information generation and causal analysis includes the following steps: S1: Acquire multi-source data of the target transmission line components, including equipment attribute parameters, environmental parameters, and real observation images; S2: Input the device attribute parameters and environmental parameters into a pre-trained physical information generative model to generate a multimodal reference image of the target transmission line component under corresponding conditions; wherein, the physical information generative model is trained by introducing physical law constraint terms so that the generated multimodal reference image conforms to the physical characteristics of the target transmission line component in a healthy state; S3: Calculate the difference information between the multimodal reference image and the actual observed image; S4: Based on a preset cause-effect graph model, perform causal analysis on the difference information, infer the cause of the device status that led to the difference information, and obtain a diagnostic result.

[0026] Specifically, in S1, multi-source data of the target transmission line components is acquired. This multi-source data includes at least equipment attribute parameters, environmental parameters, and real observation images obtained through image acquisition equipment.

[0027] During the inspection of power transmission line components, drones or other inspection equipment are typically used to collect data from the components via onboard acquisition devices. However, current technologies usually only collect image data and then optimize and analyze it. While this method can analyze power transmission line components to some extent, it relies heavily on labeled data for defect identification. For example, when using deep learning models for defect identification, a large number of labeled samples are required to ensure accuracy. However, in actual inspections, defect samples are scarce and labeling costs are high, which can easily lead to insufficient samples and inaccurate defect identification. Furthermore, over-reliance on image data makes it difficult to effectively address imaging degradation under complex weather conditions.

[0028] In this embodiment, in addition to acquiring image data, the device attribute parameters and real-time environmental parameters of the transmission line components are also acquired simultaneously. The actual observation images acquired by the image acquisition device are used as multi-source data, and then analysis is performed based on this multi-source data.

[0029] The equipment attribute parameters include, but are not limited to, the geometric dimensions, material properties, electrical parameters, and spatial information of transmission line components, such as the insulator disc diameter, disc thickness, creepage distance, etc., the conductor diameter, cross-sectional area, etc., and the standard dimensions of fittings, etc.; material properties include the insulator material, conductive material, emissivity, etc.; spatial information refers to the coordinate position, installation angle, and relative distance of the transmission line components in three-dimensional space, etc.; environmental parameters refer to the dynamically changing parameters of the external conditions of the transmission line components, including but not limited to real-time ambient temperature, humidity, wind speed, light intensity, electrical load, background interference, etc., where electrical load refers to the real-time load current passing through the conductor, and background interference refers to the temperature of background objects (displayed as an infrared image) and the presence or absence of obstructions; To improve the efficiency of subsequent analysis, the real observation images obtained by the image acquisition equipment can be restricted by conditions including but not limited to size and resolution, such as uniformly adjusting to a fixed resolution and pixel format. The real observation images can also be preprocessed, and the preprocessing process can refer to the methods of denoising, enhancing and registering images in the existing technology to obtain the preprocessed real observation images.

[0030] Specifically, in S2, the acquired equipment attribute parameters and real-time environmental parameters are input into a pre-trained physical generative model to generate a multimodal reference image of the target transmission line component under corresponding conditions. The physical information generative model is trained by introducing physical law constraint terms to make the generated multimodal reference image conform to the physical characteristics of the target transmission line component in a healthy state.

[0031] Furthermore, the physical information generative model is a generative artificial intelligence model based on deep learning and trained by introducing physical law constraints. In the existing technology, the physical information generative model can combine physical laws and data-driven methods to ensure that its output data conforms to physical laws, such as heat conduction equations, electromagnetic field distribution laws and optical imaging models, thereby generating sample data that conforms to real physical characteristics under specific environmental and device parameters. In a preferred embodiment, the training process of the physical information generative model includes at least the following steps: 1) Use the cGAN model as the initial generative model; Preferably, the initial generative model can be one of the deep generative models such as Generative Adversarial Network (GAN), Variational Autoencoder (VAE), or Diffusion model. Conditional Generative Adversarial Network (cGAN) is preferred, which uses device attribute parameters and environmental parameters as conditional inputs to guide the generation of multimodal reference images.

[0032] 2) The initial generative model was pre-trained using a dataset generated based on physical simulation, where the physical simulation simulated the electrical, thermal, and mechanical behavior of the target transmission line components under different equipment attribute parameters and environmental parameters.

[0033] In the training process of the physical information generative model, pre-training is required first. This pre-training includes using a dataset generated based on physical simulation to pre-train the initial generative model so that the initial generative model can initially learn the physical response characteristics of transmission line components under different operating conditions. This physical simulation needs to be based on the electrical, thermal, and mechanical behavior of the target transmission line components under different equipment attribute parameters and environmental parameters. For example, finite element analysis software is used to simulate the electric field distribution and leakage current of insulators under specific voltage levels and ambient humidity; computational fluid dynamics and thermodynamics simulation is used to calculate the steady-state temperature field of conductors under specific load current, ambient temperature, and wind speed; and an optical rendering engine is used to generate visible light images with correct lighting effects based on the optical properties of materials (such as reflectivity) and real-time lighting conditions.

[0034] Based on the simulation data above, a paired sample set containing multimodal inputs (device attribute parameters, environmental parameters) and corresponding physical response images is constructed for supervised training, thereby forming a training sample mapping relationship between parameters and images.

[0035] 3) During the pre-training process, a loss function is constructed using physical law constraints. These physical law constraints are used to ensure that the output of the physical information generative model conforms to the preset physical laws.

[0036] Furthermore, the physical constraints include thermodynamic constraints, electromagnetic field constraints, and optical imaging constraints.

[0037] The physical constraint term includes a thermodynamic constraint. The multimodal reference image can be an infrared reference image. The thermodynamic constraint is used to ensure that the pixel value distribution of the generated infrared reference image satisfies the temperature field distribution law described by the heat conduction equation, and to ensure that the temperature gradient of each region in the infrared image is consistent with the actual physical conditions.

[0038] The temperature field distribution described by the heat conduction equation is determined by both Fourier's law and the energy conservation equation, specifically expressed as heat divergence:

[0039] in: Indicates curl, T represents the thermal conductivity of the material, and T represents the temperature. Where c is density and c is specific heat capacity. For internal heat source terms; under steady-state conditions, The equation simplifies to It is used to constrain the rationality of the spatial distribution of temperature in the generated image, ensure that the high-temperature area matches the input parameters such as current load and heat dissipation conditions, and avoid the phenomenon of heat backflow that violates the second law of thermodynamics.

[0040] The physical law constraint includes electromagnetic field constraint. The multimodal reference image can be an electromagnetic distribution reference image. This electromagnetic field constraint is used to make the pixel intensity distribution of the generated electromagnetic distribution reference image conform to the electromagnetic field spatial variation law described by Maxwell's equations, thereby ensuring that the electromagnetic field hot spot location is highly consistent with the actual corona discharge area. The spatial variation law of electromagnetic field described by Maxwell's equations ensures that the generated electromagnetic field distribution is not only accurate in spatial shape, but also truly reflects the influence of voltage, current and material dielectric properties on the electric field strength.

[0041] The physical constraints include optical imaging constraints. The multimodal reference image can be a visible light reference image. By introducing a physical model into the rendering equation, this constraint ensures that the lighting, shadow, and reflection characteristics of the generated visible light reference image conform to the optical laws of the actual scene, especially maintaining the realism of the surface texture and three-dimensional deformation of the parts under complex lighting conditions.

[0042] The physical model that introduces rendering equations simulates various processes of light. These rendering equations can be referenced from existing rendering processes that simulate the interaction between light and materials based on physical laws. The detailed formula can be expressed as follows:

[0043] This equation describes the situation at point p along the exit direction. Outgoing light brightness Self-luminescent It consists of two parts: the reflected light and the integral term. Indicates from all incident directions incident light Surface bidirectional reflection distribution function The contribution after the action, and is weighted by the cosine of the angle between the normal n and the incident direction; where, Represents the hemispherical space The integral ensures that the light contribution from all possible incident directions is fully considered; the formula contains... Determined by material properties, it describes how a surface directs incident light from a direction at point p. Reflected in the direction of emission ability, The actual incident radiance from the environment or light source. This item reflects the impact of the incident angle on the effective received light energy, ensuring that the transition between shadows and light and dark conforms to the real physical lighting model. Thus, it serves as a geometric attenuation factor to reflect the impact of the incident angle on the effective lighting area. When the angle between the incident light and the surface normal increases, the energy received per unit area decreases accordingly, which conforms to the basic physical principle of Lambert's law.

[0044] Loss functions established using physical constraint terms can be used to guide the generator to output multimodal reference images that conform to physical laws during generative adversarial networks (GANs). This loss function calculates the deviation between the generated multimodal reference images and the generative model based on real physical information, applying the aforementioned physical constraint terms to various multimodal reference images. For example, using discharge physics constraints, electromagnetic field distribution laws, and optical imaging models as regularization terms, a multiphysics coupling loss function can be constructed. Using this loss function to train the initial generative model is a common deep learning optimization technique. The formula for the loss function can be found by referring to... ,in It represents the physical deviation loss in the heat conduction process, used to constrain the rationality of the temperature field distribution during the discharge process; This corresponds to the electromagnetic field distribution regularization term derived from Maxwell's equations. For optical imaging loss based on rendering equations, each physical term is weighted by a coefficient. , , Balance the data to ensure that the generated multimodal reference images still satisfy the true physical laws under various conditions.

[0045] The loss function constructed using physical law constraints can train the physical information generative model, enabling it to not only rely on data-driven statistical laws when generating multimodal reference images, but also follow various physical laws. This ensures that the generated multimodal reference images maintain physical consistency under different acquisition environments, reducing the risk of physical distortion caused by environmental parameter disturbances or changes in equipment status.

[0046] Specifically, in S3, based on the real observation data obtained by the image acquisition equipment, the difference information is calculated with the multimodal reference image. This difference information includes pixel-level residuals, structural similarity deviations, and spectral response differences.

[0047] After generating the multimodal reference image, the difference information between it and the real observation image can be calculated. This difference information includes at least pixel-level residuals, structural similarity deviations, and spectral response differences.

[0048] By comparing the infrared reference image with the measured thermal image pixel by pixel, the residual distribution of the temperature field is obtained. The deviation of the thermal distribution morphology on the surface of the device is evaluated by combining the structural similarity index. The color fidelity of the visible light image under different lighting conditions is quantified by using the spectral angle mapping method.

[0049] It should be noted that the various types of difference information obtained from the above difference information calculation are all existing technologies. Their calculation methods and application logic have clear technical implementation paths in this field and can be flexibly selected according to specific detection needs. They are not limited or described in detail in this embodiment.

[0050] Specifically, in S4, a causal graph model is preset, and causal analysis is performed on the obtained difference information based on the causal graph model. The cause of the equipment state that caused the difference information is inferred, and a diagnostic result is obtained. The causal analysis includes at least intervening in environmental disturbance variables and equipment state variables to make attribution judgments.

[0051] A causal graphical model is a probabilistic graphical model used to represent causal relationships between variables. It describes the transmission path of environmental disturbance variables (such as temperature, humidity, and light intensity) and equipment state variables (such as insulation aging and poor contact) to the observed differences through a directed acyclic graph. The application process of the causal graphical model can refer to existing technologies, such as the causal graphical model processing method and apparatus disclosed in Chinese patent application CN117521816A. In this embodiment, the detailed usage method is not explained, but only the intervention process is described. Interventions on environmental disturbance variables and equipment state variables should at least include: Set ideal conditions, conduct simulated interventions based on these ideal conditions, obtain intervention results, analyze the intervention results, and obtain intervention analysis results. Ideal conditions could be as follows: ambient temperature constant at 25℃, relative humidity maintained at 40%RH, light intensity stable at 500 lux and no external electromagnetic interference. If the difference information is significantly reduced under these conditions, it indicates that the actual difference is mainly caused by environmental interference variables. This is because under these ideal conditions, the environmental variables have been intervened to ensure that there are no extreme or adverse disturbance factors. This means that most of the differences observed in the actual inspection environment are due to environmental interference such as temperature fluctuations, humidity changes, and uneven lighting, rather than abnormal equipment condition. If the discrepancy information still exists significantly or increases under ideal conditions, it indicates that the discrepancy information is likely caused by the inherent abnormality of the equipment itself. In this case, further intervention of the equipment state variables is required for precise positioning. The intervention of equipment state variables includes at least: adjusting one of the equipment state variables while keeping other variables unchanged, analyzing the trend of the discrepancy information, and obtaining the adjustment analysis results; for example, simulating the degradation process of the dielectric constant of the insulating material and observing its impact on the consistency of infrared radiation characteristics and visible light texture. If the discrepancy information increases significantly, it can be determined that insulation aging is the main cause of the discrepancy information. The above example of the intervention process is only one possible implementation method. In practical applications, similar intervention analysis can be performed on other equipment state variables based on the same principle. Based on the results of the intervention analysis and the adjustment analysis, the specific equipment state causes that produce the difference information can be determined, and a diagnostic result can be obtained. It should be noted that a significant increase in the above-mentioned difference information can be judged by a threshold. This threshold can be obtained based on historical data statistical analysis. For example, if the range of difference information under normal operating conditions is set to within ±15%, then when the difference information exceeds this range, it is judged as a significant increase. This threshold can be dynamically adjusted according to the equipment type, operating years and operating conditions, and is not limited by a specific value. A diagnostic report is generated based on the results of intervention analysis and regulation analysis.

[0052] Furthermore, the diagnostic method also includes step S5: receiving a counterfactual query request, the counterfactual query request containing future settings for device attribute parameters and / or environmental parameters, generating a predicted image of future time points through a physical information generative model based on the settings in the counterfactual query request, and outputting a state prediction report based on the predicted image.

[0053] To determine the future state of transmission line components based on existing data, counterfactual reasoning can be used to simulate image predictions under different parameter settings, and the predicted images can be analyzed. The counterfactual query request sets hypothetical modifications to the equipment attribute parameters and / or environmental parameters at a future point in time. For example, it is assumed that the slight contamination level of the currently discovered transmission line components will increase by 20% in the next three months. Based on this hypothetical modification, the physical information generative model generates a new multimodal baseline image.

[0054] Based on the newly generated multimodal reference image, it is compared with the current actual observation image, the difference information is extracted, and the analysis results of the difference information and the state prediction report based on the analysis results are output.

[0055] Preferably, the analysis results of the difference information include at least the trend, magnitude and spatial distribution characteristics of the difference changes, and the state prediction report based on the analysis results includes at least the potential risk areas, key influencing factors and possible failure modes of future state changes. Based on the above embodiments, one possible scenario is: if the generated multimodal reference image deviates from the current real observation image in the insulator string area in terms of infrared and visible light characteristics, and the difference is increased by a certain percentage compared to the current state, then it is determined that there is a risk of hot spot deterioration in the insulator string area. The key influencing factors are the coupling effect of pollution accumulation and partial discharge, which may cause flashover faults. It is recommended to carry out cleaning in advance or install anti-pollution flashover coating, and dynamically adjust the maintenance cycle according to the equipment operating environment. If the thermal characteristics of the conductor splice area are significantly enhanced in the generated multimodal reference image, combined with changes in environmental parameters such as wind speed and load current, it can be inferred that there is a risk of poor contact or metal fatigue, which may develop into strand breakage or fracture fault. It is necessary to arrange live fastening or reinforcement measures in a timely manner. If the generated multimodal reference image shows an abnormal temperature field distribution in the tower foundation area, accompanied by changes in humidity and vibration parameters, there may be a risk of settlement or tilting. The key influencing factors are the superposition of geological loosening and external loads, which can easily lead to structural instability. It is recommended to verify the data by combining deformation monitoring data and implement reinforcement treatment. Based on the above embodiments, the present invention has at least the following technical effects: by integrating physical information with a data-driven generative model, it can predict the future state of transmission lines; by using counterfactual reasoning and image difference analysis, it can improve the predictive ability of transmission lines and effectively identify potential fault risk areas and key influencing factors.

[0056] Example 2 This invention provides a power transmission line diagnostic system based on physical information generation and causal analysis, comprising: The data acquisition module is used to acquire multi-source data of the target transmission line components, including equipment attribute parameters, environmental parameters, and real observation images; A multimodal reference image generation module is used to input the device attribute parameters and environmental parameters into a pre-trained physical information generative model to generate a multimodal reference image of the target transmission line component under corresponding conditions; wherein, the physical information generative model is trained by introducing physical law constraint terms so that the generated multimodal reference image conforms to the physical characteristics of the target transmission line component in a healthy state; The difference information calculation module is used to calculate the difference information between the multimodal reference image and the real observed image; The causal analysis module is used to perform causal analysis on most discrepancies based on a preset causal graph model, infer the cause of the device status that leads to the discrepancies, and obtain diagnostic results.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A power transmission line diagnosis method based on physical information generation and causal analysis, characterized by, The method comprises the following steps: Obtaining multi-source data of a target power transmission line component, wherein the multi-source data comprises equipment attribute parameters, environmental parameters and real observation images; Inputting the equipment attribute parameters and the environmental parameters into a pre-trained physical information generative model to generate multi-modal reference images of the target power transmission line component under corresponding conditions; wherein the physical information generative model is trained by introducing a physical law constraint term; Calculating difference information between the multi-modal reference images and the real observation images; Performing causal analysis on the difference information based on a preset causal diagram model to deduce equipment state causes leading to the difference information, and obtaining a diagnosis result.

2. The method of claim 1, wherein the method is characterized by, In the step of inputting the equipment attribute parameters and the environmental parameters into a pre-trained physical information generative model to generate multi-modal reference images of the target power transmission line component under corresponding conditions, the training process of the physical information generative model is as follows: Taking a cGAN model as an initial generative model; Pre-training the initial generative model by using a data set generated based on physical simulation, wherein the physical simulation simulates electrical, thermal and mechanical behaviors of the target power transmission line component under different equipment attribute parameters and environmental parameters; In the pre-training process, a loss function is constructed by a physical law constraint term to obtain a trained physical information generative model.

3. The method of claim 2, wherein the method is characterized by, The physical law constraint term comprises a thermodynamic constraint, the multi-modal reference images are infrared reference images, and the thermodynamic constraint is used to make pixel value distribution of the infrared reference images meet temperature field distribution rules of a heat conduction equation, and the specific formula is as follows: wherein, denotes the curl, is the thermal conductivity of the material, T is the temperature, is the density, c is the specific heat capacity, is the internal heat source term.

4. The method of claim 2, wherein the method is characterized by, The physical law constraint term comprises an electromagnetic field constraint, the multi-modal reference images are electromagnetic distribution reference images, and the electromagnetic field constraint is used to make pixel intensity distribution of the generated electromagnetic distribution reference images comply with spatial variation rules of electromagnetic fields described by Maxwell equations.

5. The method of claim 2, wherein the method is characterized by, The physical law constraint term comprises an optical imaging constraint, the multi-modal reference images are visible light reference images, and the optical imaging constraint is used to ensure that illumination, shadow and reflection characteristics of the generated visible light reference images comply with optical rules of an actual scene by introducing a physical model of a rendering equation.

6. The method of claim 5, wherein the method is characterized by, The physical model of the rendering equation is as follows: wherein, represents the luminous exitance of the exit light at point p in the exit direction , is the self-luminous, is the reflected light, represents the integration over the hemispherical space , describes the ability of the surface at point p to reflect incident light from direction to the exit direction , is the actual incident radiance from the environment or light source, the term represents the influence of the angle of incidence on the effective received light energy.

7. The method of claim 1, wherein the method is characterized by, In the step of calculating difference information between the multi-modal reference images and the real observation images, the difference information comprises pixel-level residuals, structural similarity deviations and spectral response differences.

8. The method of claim 1, wherein the method is characterized by, In the step of performing causal analysis on the difference information based on a preset causal diagram model to deduce equipment state causes leading to the difference information, and obtaining a diagnosis result, the causal analysis comprises intervention on environmental interference variables and equipment state variables to make attribution judgments, and the intervention process comprises: Setting ideal conditions, performing simulation intervention based on the ideal conditions to obtain intervention results, analyzing the intervention results to obtain intervention analysis results; Adjusting one of the equipment state variables while keeping other variables unchanged, analyzing the change trend of the difference information to obtain adjustment analysis results; Outputting a diagnosis report based on the intervention analysis results and the adjustment analysis results.

9. The method of claim 1, wherein the method is characterized by, Further comprising the following steps: A counterfactual query request is received, the counterfactual query request including future settings of device attribute parameters and / or environmental parameters, a predicted image at a future time point is generated by a physical information generative model based on the settings in the counterfactual query request, and a state prediction report based on the predicted image is output.

10. A power line diagnostic system based on physical information generation and causal analysis, characterized by, Comprise: a data acquisition module for acquiring multi-source data of a target power transmission line component, the multi-source data including device attribute parameters, environmental parameters, and real observation images; a multi-modal reference image generation module for inputting the device attribute parameters and environmental parameters into a pre-trained physical information generative model to generate multi-modal reference images of the target power transmission line component under corresponding conditions; wherein the physical information generative model is trained by introducing a physical law constraint term; a difference information calculation module for calculating difference information between the multi-modal reference images and the real observation images; a causal analysis module for performing causal analysis on the difference information based on a pre-set causal graph model to infer the device state causes leading to the difference information and obtain a diagnosis result.

Images