Power transmission line fault detection method and inspection system

Abstract

The invention relates to a power system fault diagnosis technology, in particular to a power transmission line fault detection method and an inspection system, and aims to solve the problem that transient events cannot be captured in time due to overlong acquisition interval or system response delay adopted by a monitoring system in the prior art in a timing snapshot strategy or a conventional real-time monitoring mode. According to the invention, the microwave radar penetrates through severe weather to collect basic data, high-frequency collection of the visual module under the trigger of an instantaneous event is combined, multi-source data fusion is realized by using a feature constraint mechanism, and localization processing is carried out by relying on the edge calculation unit, so that the problem of data missing in a complex meteorological environment is effectively solved, and the accuracy of data fusion is improved. The method ensures timely capture and accurate identification of transient fault events, can effectively deal with severe weather conditions, timely capture transient fault events, improve fault identification accuracy, and reduce data transmission burden.

Description

Technical Field

[0001] This application relates to power system fault diagnosis technology, specifically to a method for detecting transmission line faults and an inspection system. Background Technology

[0002] In the practice of power transmission line inspection, complex meteorological environments pose a significant challenge. Power transmission lines are widely distributed in areas with complex geographical conditions such as mountainous and forested areas, and frequently encounter severe weather such as heavy fog, sandstorms, rain, and snow. This results in severely blurred image data acquired by traditional visual acquisition equipment, with a large amount of key detail information missing, making it difficult for image analysis-based intelligent recognition algorithms to accurately distinguish between real fault targets and environmental interference factors.

[0003] Meanwhile, transient fault events such as lightning strikes, bird activity, or brief contact of foreign objects with power lines are characterized by their suddenness and extremely short duration. Existing monitoring systems, employing timed capture strategies or conventional real-time monitoring modes, cannot capture these brief events in a timely manner due to excessively long acquisition intervals or system response delays. Furthermore, detection methods relying on single visual sensors are susceptible to interference from factors such as changes in natural lighting and cluttered backgrounds, resulting in insufficient stability of recognition results. Moreover, the transmission and processing of massive amounts of raw video data consumes excessive network resources, leading to low efficiency in fault tracing and delayed response, posing a significant threat to the safe and stable operation of transmission lines.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] (a) Technical problems to be solved The purpose of this application is to provide a method for detecting faults in power transmission lines and a system for inspecting faults in power transmission lines, which can effectively cope with severe weather conditions, capture instantaneous fault events in a timely manner, improve the accuracy of fault identification, and reduce the burden of data transmission.

[0006] (II) Technical Solution Firstly, this application discloses a method for detecting faults in transmission lines, the technical solution of which is as follows: The first type of data on the transmission line environment is collected through the first sensing module, which is a microwave radar capable of penetrating severe weather. The first type of data includes target motion parameters and conductor physical state parameters. The second type of data of the transmission line environment is collected through the second sensing module, which is a vision acquisition module with high-definition imaging capability. The second type of data is image or video data containing details of the fault target. The second sensing module performs time synchronization and spatial calibration with the first sensing module, and the synchronization accuracy and calibration deviation meet the requirements for collaborative operation. Based on the first type of data, determine whether there is a preset instantaneous event. An instantaneous event is a transmission line abnormality-related event with a duration not exceeding a preset duration. When a transient event is detected, the second sensing module is triggered to switch to high-frequency acquisition mode to obtain detailed image data of the transient event period; By fusing the first type of data and detailed image data through a feature constraint mechanism, fault identification is performed and fault judgment results are generated. The feature constraint mechanism is the process of injecting the target spatial information in the first type of data into the identification process of the second type of data. The fault determination results are processed by a lightweight identification model running on an edge computing unit deployed locally on the inspection device, and a simplified information containing the fault determination results is uploaded.

[0007] Furthermore, this application also proposes that the target motion parameters in the first type of data include the approach speed of the foreign object, and the physical state parameters of the conductor include the vibration amplitude of the conductor; the preset instantaneous event determination conditions are: the approach speed of the foreign object exceeds a preset speed threshold, or the vibration amplitude of the conductor exceeds a preset vibration threshold.

[0008] Furthermore, this application also proposes that the preset velocity threshold is 0.3m / s to 0.5m / s, the preset vibration threshold is 0.3mm to 0.8mm, and the preset duration of the instantaneous event is ≤2 seconds.

[0009] Furthermore, this application proposes that the feature constraint mechanism specifically involves injecting the target spatial trajectory and location information from the first type of data into the feature extraction stage of the visual recognition model corresponding to the second type of data through an attention mechanism or feature mapping method, thereby constraining the target recognition range.

[0010] Furthermore, this application also proposes that the method further includes: Receive feedback information based on the fault determination results; Based on the feedback information, the event judgment conditions of the first sensing module and / or the operating parameters of the second sensing module in high-frequency operating mode are adaptively adjusted.

[0011] Secondly, this application also discloses a transmission line fault inspection system, the technical solution of which is as follows: The first sensing module is a microwave radar capable of penetrating severe weather conditions, used to collect the first type of data of the power transmission line environment, including target motion parameters and conductor physical state parameters. The second sensing module is a vision acquisition module with high-definition imaging capabilities, used to collect the second type of data of the power transmission line environment, which is image or video data containing details of the fault target. The synchronization calibration module is used to achieve time synchronization and spatial calibration between the first and second sensing modules to ensure the accuracy of collaborative operation. The processing unit is configured as follows: Based on the first type of data, it is determined whether there is a preset instantaneous event. An instantaneous event is a transmission line abnormality-related event with a duration not exceeding the preset duration. When a transient event is detected, the second sensing module is triggered to switch to high-frequency acquisition mode; Fault identification is performed by fusing the first type of data and detailed image data collected by the second sensing module through a feature constraint mechanism; The edge computing unit, deployed locally on the inspection device, is used to run a lightweight identification model and generate fault determination results. The edge computing unit is also used to upload simplified information containing the fault determination results.

[0012] Furthermore, this application also proposes that the processing unit is specifically configured to: determine whether the speed at which the foreign object approaches and / or the vibration amplitude of the conductor in the first type of data exceed the corresponding preset threshold, wherein the preset threshold is a configurable speed threshold and vibration threshold.

[0013] Furthermore, this application also proposes that the feature constraint mechanism executed by the processing unit is as follows: the target spatial trajectory and location information in the first type of data are integrated into the visual recognition process of the second type of data through feature layer injection, thereby constraining the target box prediction range to reduce interference.

[0014] Furthermore, this application proposes that the fault identification model is a lightweight model whose computational complexity and size are suitable for running on edge computing units.

[0015] Furthermore, this application also proposes that the system further includes a dynamic optimization module, configured as follows: Receive feedback information based on the fault determination results; The event determination conditions of the first sensing module and / or the high-frequency acquisition parameters of the second sensing module are adaptively adjusted based on the feedback information.

[0016] (III) Beneficial Effects Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes microwave-type radar to penetrate severe weather conditions and collect basic data. It combines this with a vision module for high-frequency data acquisition triggered by instantaneous events, and employs a feature constraint mechanism to achieve multi-source data fusion. Simultaneously, it relies on edge computing units for localized processing, effectively solving the problem of data loss in complex weather environments. This ensures the timely capture and accurate identification of instantaneous fault events, effectively responding to severe weather conditions, capturing instantaneous fault events in a timely manner, improving fault identification accuracy, and reducing data transmission burden. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the logic structure of a power transmission line fault detection method. Detailed Implementation

[0019] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Example 1

[0022] like Figure 1 As shown in the figure, this application proposes a method for detecting faults in transmission lines, including the following steps: S100. Collect first type of data of the transmission line environment through the first sensing module. The first sensing module is a microwave radar that can penetrate severe weather. The first type of data includes target motion parameters and conductor physical state parameters. Among them, microwave radars specifically select operating frequencies of 24GHz~77GHz, detection ranges of 0~500m, and response times of ≤100μs, suitable for 110kV~500kV transmission line inspection scenarios. Target motion parameters, such as the speed at which foreign objects approach, and conductor physical state parameters, such as the amplitude of conductor vibration, are used.

[0023] S200: Collect a second type of data about the transmission line environment through a second sensing module. The second sensing module is a vision acquisition module with high-definition imaging capabilities. The second type of data is image or video data containing details of the fault target. The visual acquisition module can be a visible light camera, an infrared thermal imager, or a combination of both. The visible light camera has a resolution of ≥16 million pixels, and the infrared thermal imager can detect temperatures ranging from -40℃ to 80℃.

[0024] S300, the second sensing module and the first sensing module perform time synchronization and spatial calibration, and the synchronization accuracy and calibration deviation meet the requirements of collaborative work. Time synchronization is achieved through a GPS timing module or a high-precision clock synchronization protocol; spatial calibration is achieved through hardware coaxial mounting and software coordinate mapping to ensure that the pixel coordinates of radar spatial information and visual images correspond precisely.

[0025] S400. Based on the first type of data, determine whether there is a preset instantaneous event. The instantaneous event is a transmission line abnormality-related event with a duration not exceeding a preset duration. The judgment is based on transmission line safety standards (such as DL / T283-2018) and historical fault data, with a preset duration of ≤2 seconds. The judgment condition is that the speed at which the foreign object approaches exceeds a preset speed threshold or the amplitude of conductor vibration exceeds a preset vibration threshold. S500: When a transient event is detected, the second sensing module is triggered to switch to high-frequency acquisition mode to acquire detailed image data of the transient event period. S600: By fusing the first type of data and detailed image data through a feature constraint mechanism, fault identification is performed and fault judgment results are generated. The feature constraint mechanism is the process of injecting the target spatial information in the first type of data into the identification process of the second type of data. Specifically, one or a combination of the following two methods may be used: I. Attention Mechanism: Dynamically track the target's spatial trajectory captured by radar, amplify the visual feature weight of the trajectory area, and suppress background interference; II. Feature Mapping Method: The target area located by radar is transformed into a region of interest (ROI) in the visual image, constraining the visual model to extract features only from this region; both methods inject radar spatial information into the intermediate feature extraction layer (such as Conv3~Conv4 layer) of the visual recognition model. The S700 uses an edge computing unit deployed locally on the inspection device to run a lightweight identification model to process fault determination results and uploads simplified information containing the fault determination results.

[0026] In practical applications, the first sensing module can be understood as a device capable of target detection using electromagnetic wave characteristics. Its main function is to penetrate complex weather conditions such as rain, snow, and fog to acquire target motion parameters and conductor physical state parameters. For example, this module can use continuous wave radar technology or pulse radar technology to measure the target's velocity, position, and conductor vibration amplitude.

[0027] Furthermore, the target motion parameters of the first type of data can be extracted through the Doppler effect, while the physical state parameters of the conductor can be obtained by analyzing the phase change of the reflected signal.

[0028] The second sensing module is a device with high-resolution imaging capabilities, whose main function is to capture detailed information about faulty targets.

[0029] This module can acquire image or video data using visible light cameras, infrared thermal imagers, or other devices with high-definition imaging capabilities. For example, in nighttime or low-light conditions, infrared imaging technology can be used to obtain the thermal radiation information of the target, thereby compensating for the shortcomings of visible light imaging.

[0030] Time synchronization and spatial calibration can be achieved through hardware or software methods. For example, hardware methods can use high-precision clock and positioning modules to ensure that the time bases of the two sensing modules are consistent and their spatial coordinates are aligned; software methods can use algorithms to post-process the acquired data to eliminate errors caused by time delays or spatial deviations.

[0031] The identification of transient events can be achieved by setting thresholds. For example, a threshold can be set for the rate of change of velocity for target motion parameters; a threshold can be set for the vibration frequency or amplitude for conductor physical state parameters. When the monitored data exceeds the set range, a transient event is determined to exist.

[0032] The specific implementation methods of the feature constraint mechanism may include, but are not limited to, the following: mapping the target location information in the first type of data to the image coordinate system of the second type of data through geometric constraints, thereby limiting the recognition area; or dynamically adjusting the input weights of the visual recognition model so that the model prioritizes the feature information of the target area.

[0033] Lightweight recognition models for edge computing units can be compressed using techniques such as pruning, quantization, or knowledge distillation, thereby reducing computational complexity and storage requirements. For example, lightweight neural network architectures such as MobileNet or ShuffleNet can be employed to ensure that the model can run efficiently on resource-constrained devices.

[0034] The innovation of this application lies in solving the problem of accurate identification and tracing of transient faults under complex weather conditions through the collaborative operation of multimodal sensors and an intelligent triggering mechanism. Specifically, the anti-interference characteristics of the first sensing module provide a foundation for rapid detection of transient events; the high-frequency acquisition mode of the second sensing module ensures accurate capture of fault details; the feature constraint mechanism improves identification accuracy through the injection of spatial information; and the edge computing unit optimizes data transmission efficiency through local processing and streamlined information uploading. This end-to-end design effectively overcomes the problems of data failure, missed detection of transient faults, and high bandwidth consumption in traditional solutions under complex weather conditions, providing a reliable guarantee for the safe operation of transmission lines.

[0035] The working principle of this application embodiment is as follows: The first sensing module adopts a microwave radar that can penetrate severe weather conditions to collect the first type of data of the power transmission line environment. The first type of data includes target motion parameters and conductor physical state parameters, thereby ensuring that stable basic monitoring information can still be obtained under complex weather conditions.

[0036] The second sensing module is a vision acquisition module with high-definition imaging capabilities. Its function is to acquire image or video data containing details of the faulty target, which serves as an important basis for subsequent fault identification. To ensure the coordination of multi-source data, the second sensing module and the first sensing module achieve high-precision cooperation through time synchronization and spatial calibration. The synchronization accuracy and calibration deviation are strictly controlled to meet the requirements of collaborative operation.

[0037] Based on the first type of data, the system analyzes the target motion parameters and conductor physical state parameters to determine whether a preset instantaneous event exists. An instantaneous event is defined as a transmission line anomaly-related event with a duration not exceeding a preset time. When an instantaneous event is detected, the second sensing module is triggered to switch to a high-frequency acquisition mode, thereby acquiring detailed image data specifically for the instantaneous event period. This mechanism effectively avoids data redundancy issues associated with conventional continuous monitoring, while ensuring accurate capture of detailed information during critical periods.

[0038] Furthermore, by employing a feature constraint mechanism, target spatial information from the first type of data is injected into the recognition process of the second type of data, achieving deep fusion of multi-source data. Specifically, the target spatial trajectory and location information from the first type of data are used as prior constraints for the visual recognition model, thereby limiting the recognition range and suppressing misjudgments caused by environmental interference. The edge computing unit is deployed locally on the inspection device, running a lightweight recognition model to process the fault determination results and uploading only a simplified information containing the fault determination results. This design significantly reduces the bandwidth requirements for data transmission while improving fault response speed.

[0039] The overall technical solution forms a complete "monitoring-triggering-fusion-decision" closed loop through multimodal sensing collaboration, intelligent triggering mechanisms, and optimization of data fusion and edge computing. Microwave radar continuously monitors and rapidly identifies instantaneous anomalies, the vision module provides high-precision detail acquisition during critical periods, spatiotemporal calibration ensures data consistency, feature constraints improve recognition accuracy, and edge computing enables efficient processing. These technologies work closely together to solve the problem of accurate identification and tracing of instantaneous faults under complex weather conditions.

[0040] This application further proposes that the target motion parameters in the first type of data include the approach speed of the foreign object, and the physical state parameters of the conductor include the vibration amplitude of the conductor; the preset instantaneous event judgment conditions are: the approach speed of the foreign object exceeds the preset speed threshold, or the vibration amplitude of the conductor exceeds the preset vibration threshold.

[0041] Specifically, the foreign object approach speed refers to the change in speed at which an external object approaches the power transmission line conductor. It can be measured using the Doppler effect of microwave radar, with the aim of accurately detecting rapid intrusion behaviors that may cause instantaneous faults.

[0042] Conductor vibration amplitude refers to the change in amplitude of a transmission conductor under external force. It can be achieved by analyzing the echo signal intensity fluctuation of microwave radar, aiming to sensitively reflect the impact of abnormal conditions such as lightning strikes and bird pecking on the conductor. Preset speed threshold and preset vibration threshold are trigger conditions set according to actual scenario requirements. They can be flexibly configured according to the transmission line level and environmental characteristics, aiming to ensure the objectivity and adaptability of the judgment logic.

[0043] In detail, the above technical solution solves the accuracy problem of instantaneous fault identification under complex weather conditions by clarifying the core parameter types and their judgment logic. First, based on the first type of data collected by microwave radar, the system can stably acquire two highly interference-resistant physical parameters: the approach speed of foreign objects and the amplitude of conductor vibration, avoiding the shortcomings of traditional visual parameters that are easily affected by weather.

[0044] By comparing the approach speed of a foreign object with a preset speed threshold, and the vibration amplitude of the conductor with a preset vibration threshold, the system can effectively distinguish between environmental interference and actual fault risks. For example, when the approach speed of a foreign object significantly exceeds the limit, the system determines it as an external intrusion event and triggers a high-frequency acquisition mode; when the vibration amplitude of the conductor abnormally increases, the system determines it as an internal abnormal event and initiates detailed image acquisition.

[0045] This "OR" logic design covers two typical transient fault scenarios: external intrusion and internal anomalies. It ensures comprehensive coverage of fault scenarios while avoiding missed detections caused by "AND" logic. Furthermore, the above solution is deeply integrated with the penetrating and severe weather-resistant characteristics of microwave radar, jointly constructing a complete link from monitoring to judgment to targeted data collection, significantly improving the reliability and efficiency of transient fault detection under complex weather conditions.

[0046] Through the above technical solutions, the system can accurately identify momentary faults under severe weather conditions, reduce false alarm and missed alarm rates, optimize resource utilization, and ensure that high-frequency acquisition mode is only activated when necessary, thereby improving overall inspection efficiency and safety.

[0047] This application further proposes a preset velocity threshold of 0.3m / s to 0.5m / s, a preset vibration threshold of 0.3mm to 0.8mm, and a preset duration of ≤2 seconds for instantaneous events.

[0048] Specifically, the preset speed threshold refers to the critical speed value used to determine whether a foreign object approaching a power transmission line poses a potential threat. It can be achieved by setting a speed range based on the accuracy of microwave radar acquisition. The purpose is to accurately capture the instantaneous approach behavior of dangerous foreign objects by adapting to the characteristics of foreign object intrusion under complex weather conditions.

[0049] The preset vibration threshold is the amplitude threshold used to distinguish between normal and abnormal vibration of a conductor. It can be achieved by setting intervals based on the conductor material properties and environmental vibration benchmarks. The purpose is to filter out daily wind load interference while effectively identifying abnormal vibration signals caused by lightning strikes or mechanical damage.

[0050] The preset duration of a transient event refers to the time window used to define whether an abnormal event related to a transmission line belongs to a transient fault. It can be achieved by setting an upper limit based on the statistical duration of typical transient faults. The purpose is to ensure that the system can efficiently lock onto short-term sudden abnormal events and trigger subsequent high-precision acquisition.

[0051] Specifically, the above solution addresses the reliability issues caused by ambiguity in threshold and duration settings in fault detection by precisely defining the key parameter range for instantaneous event determination. The preset speed threshold is set between 0.3 m / s and 0.5 m / s. This range is determined based on actual observation data of foreign object approach behavior in the transmission line environment. It effectively captures typical speed characteristics such as kite string bridging or bird activity, avoiding the overestimation of rapidly intruding dangerous foreign objects due to an excessively high threshold, or the misjudgment of normal wind swaying as a fault due to an excessively low threshold, thereby improving the accuracy of event triggering under complex weather conditions.

[0052] The preset vibration threshold is limited to the range of 0.3mm to 0.8mm. This range is derived from the difference in the physical state of the conductor under normal operation and fault conditions. It can accurately cover the abnormal vibration amplitude caused by lightning strikes or mechanical damage. It can identify small but critical damage signals of the conductor and filter out weak vibration interference caused by daily wind loads, ensuring that the vibration parameter judgment is strongly correlated with the actual fault.

[0053] The preset duration of instantaneous events is set to no more than 2 seconds. This duration is determined based on the statistical analysis of the duration of typical events such as lightning strikes and bird pecking. It can fully encompass fault processes shorter than 1 second, while avoiding the introduction of irrelevant environmental noise due to excessive duration or the omission of key stages of events due to excessively short duration. This enables the system to efficiently lock onto instantaneous anomalies and trigger subsequent high-precision data acquisition.

[0054] Furthermore, the aforementioned solution is closely related to the collaborative operation of the first and second sensing modules. The stable acquisition accuracy of microwave radar ensures precise judgment of speed and vibration parameters, while the high-frequency acquisition mode of the visual acquisition module significantly enhances the timeliness and robustness of fault tracing. This design not only improves the detection rate of instantaneous faults but also effectively reduces the false alarm rate, providing reliable technical support for transmission line inspection under complex weather conditions.

[0055] This application further proposes a feature constraint mechanism, which is as follows: by using an attention mechanism or feature mapping method, the target spatial trajectory and position information in the first type of data are injected into the feature extraction stage of the visual recognition model corresponding to the second type of data, thereby constraining the target recognition range.

[0056] Among them, the attention mechanism refers to a technique for dynamically adjusting the region of interest of the visual model. It can be implemented using methods such as channel attention, spatial attention, or hybrid attention, with the aim of flexibly focusing on key areas based on the dynamic characteristics of the target. Feature mapping refers to a technique for converting radar spatial information into regions of interest for the visual model. It can be implemented through methods such as coordinate mapping, region segmentation, or feature alignment, with the aim of providing a fixed region of interest for static or slowly changing targets.

[0057] Specifically, this solution addresses the issue of inaccurate target recognition range constraints under complex weather conditions through two differentiated implementation methods. For moving faulty targets, the attention mechanism dynamically tracks the spatial trajectory captured by radar, ensuring the visual model remains focused on the target's movement path. Even when dense fog blurs the visual image, it effectively filters out background light and shadow, trees, and other interfering factors. For static faulty targets, the feature mapping method transforms the spatial location determined by radar into a fixed region of interest, restricting the visual model to extract features only within that region and avoiding false activation of irrelevant backgrounds. This scenario-based adaptation mechanism ensures the accuracy of identifying various transient faults.

[0058] Meanwhile, the logic of injecting spatial information into the feature extraction stage changes the traditional passive mode of "identification first, verification later." By injecting precise radar information, after time synchronization and spatial calibration, into the early feature learning stage of the visual model, the model is constrained by radar spatial information from the source, fundamentally solving the problem of environmental interference. For example, in rainy or snowy weather, when visual images are filled with noise due to raindrops, the model extracts features only within the radar trajectory range, directly shielding the raindrop noise area and ensuring that fault identification is not affected by weather conditions. This fusion method, together with microwave radar anti-interference acquisition and instantaneous event-triggered visual acquisition, organically supports the accurate identification of instantaneous faults under complex weather conditions.

[0059] This application further proposes a method that includes: receiving feedback information based on the fault determination result; and adaptively adjusting the event judgment conditions of the first sensing module and / or the operating parameters of the second sensing module in the high-frequency operating mode according to the feedback information.

[0060] Specifically, feedback information can be understood as precise data with clear direction, mainly including three core categories: verification feedback from the main station system regarding fault determination results, such as "determined as foreign object intrusion, confirmed as a real fault after on-site verification" or "determined as conductor vibration, actually caused by strong wind interference"; manual error correction information from maintenance personnel, such as marking "false alarm" or "missed alarm" and their corresponding scenarios; and recognition performance data statistically compiled by the system itself, such as false alarm rate, missed alarm rate, and fault identification accuracy under specific environments. This specific feedback provides a reliable basis for parameter optimization, avoiding blind adjustments.

[0061] Event judgment conditions refer to the specific rules or standards used to trigger instantaneous event detection. These can be adapted by adjusting speed thresholds, vibration thresholds, event judgment sensitivity, and instantaneous event duration thresholds. Meanwhile, high-frequency operating parameters include acquisition frequency, acquisition duration, image resolution, and exposure parameters. Dynamic adjustment of these parameters ensures the system maintains efficient operation under various weather conditions.

[0062] The purpose of introducing the above mechanism is to solve the problem of insufficient adaptability of fixed parameters in complex environments, thereby improving the robustness and accuracy of the system.

[0063] In detail, the solution achieves self-learning and self-optimization capabilities in dynamic environments by constructing a closed-loop feedback mechanism.

[0064] First, the system receives verification results from the main station, manual error correction information, and performance statistics, and analyzes the adaptation deviation between parameters and the environment in real time. For example, in windy mountainous areas, if the feedback shows "the conductor vibration threshold is too low, causing frequent false alarms," ​​the vibration threshold is automatically increased; in foggy areas, if the feedback shows "the foreign object approach speed threshold is too high, causing missed alarms," ​​the speed threshold is lowered; in areas with strong sunlight, if the feedback shows "overexposed images acquired at high frequencies affect recognition," the exposure parameters of the vision module are automatically adjusted.

[0065] This dynamic adjustment not only allows for rapid adaptation to short-term weather changes but also enables long-term adaptation to environmental differences across different line sections, significantly reducing the false alarm rate in complex scenarios while maintaining a high fault identification accuracy. Furthermore, cross-module parameter adjustments achieve synergistic optimization of trigger accuracy and acquisition efficiency. For example, adjusting radar judgment sensitivity reduces invalid triggers, while simultaneously optimizing the acquisition duration of the vision module to avoid data redundancy, thereby improving recognition accuracy while maintaining low bandwidth and low power consumption characteristics.

[0066] This mechanism is deeply integrated with the microwave radar anti-interference acquisition and feature constraint fusion identification in the preceding scheme. The edge computing unit quickly outputs accurate fault judgment results, providing real-time data support for feedback. In turn, the feedback mechanism optimizes the core parameters of the first and second sensing modules, further enhancing the effect of multimodal fusion and forming a complete self-optimizing closed loop.

[0067] Example 2

[0068] In another embodiment, this application also discloses a transmission line fault inspection system, comprising: The first sensing module is a microwave radar capable of penetrating severe weather conditions, used to collect the first type of data of the power transmission line environment, including target motion parameters and conductor physical state parameters. The second sensing module is a vision acquisition module with high-definition imaging capabilities, used to collect the second type of data of the power transmission line environment, which is image or video data containing details of the fault target. The synchronization calibration module is used to achieve time synchronization and spatial calibration between the first and second sensing modules to ensure the accuracy of collaborative operation. The processing unit is configured to: determine whether a preset instantaneous event exists based on the first type of data, wherein the instantaneous event is a transmission line abnormality-related event with a duration not exceeding a preset duration; when it is determined that an instantaneous event exists, trigger the second sensing module to switch to high-frequency acquisition mode; and perform fault identification by fusing the first type of data and the detailed image data acquired by the second sensing module through a feature constraint mechanism. The edge computing unit, deployed locally on the inspection device, is used to run a lightweight identification model and generate fault determination results. The edge computing unit is also used to upload simplified information containing the fault determination results.

[0069] The core innovation of this embodiment lies in combining the anti-interference characteristics of microwave radar of the first sensing module with the high-definition imaging capability of the second sensing module through an intelligent triggering mechanism, and introducing a synchronous calibration module and a feature constraint mechanism, thereby solving the problem of difficulty in accurately identifying and tracing the source of instantaneous faults under complex meteorological conditions.

[0070] Specifically, the first sensing module can stably acquire target motion parameters and conductor physical state parameters under adverse weather conditions such as rain, snow, and heavy fog, providing a reliable foundation for instantaneous event detection; the second sensing module switches to a high-frequency acquisition mode after triggering to specifically capture fault details and avoid data redundancy in conventional monitoring; the synchronization calibration module achieves time synchronization error ≤5ms and spatial calibration deviation ≤0.5° through hardware and software collaboration, ensuring consistency of multi-source data; the feature constraint mechanism injects target spatial trajectory information from the first type of data into the visual recognition process, filtering environmental interference from the source; the edge computing unit supports lightweight model operation through localized computing power, significantly reducing bandwidth consumption and improving response speed, achieving efficient monitoring, accurate identification, and rapid source tracing.

[0071] In practical applications, the first sensing module can be understood as a device capable of target detection using electromagnetic wave characteristics. Its main function is to penetrate complex weather conditions such as rain, snow, and fog to acquire target motion parameters and conductor physical state parameters. For example, this module can measure target velocity, position, and conductor vibration amplitude using continuous wave radar or pulse radar technology. Furthermore, the target motion parameters of the first type of data can be extracted using the Doppler effect, while the conductor physical state parameters can be obtained by analyzing the phase changes of the reflected signal.

[0072] The second sensing module is a device with high-resolution imaging capabilities, whose main function is to capture detailed information about faulty targets. Specifically, this module can acquire image or video data through visible light cameras, infrared thermal imagers, or other devices with high-definition imaging capabilities. For example, in nighttime or low-light conditions, infrared imaging technology can be used to obtain the thermal radiation information of the target, thereby compensating for the shortcomings of visible light imaging.

[0073] Time synchronization and spatial calibration can be achieved through hardware or software methods. For example, hardware methods can use high-precision clock and positioning modules to ensure that the time bases of the two sensing modules are consistent and their spatial coordinates are aligned; software methods can use algorithms to post-process the acquired data to eliminate errors caused by time delays or spatial deviations.

[0074] The identification of transient events can be achieved by setting thresholds. For example, a threshold can be set for the rate of change of velocity for target motion parameters; a threshold can be set for the vibration frequency or amplitude for conductor physical state parameters. When the monitored data exceeds the set range, a transient event is determined to exist.

[0075] The specific implementation methods of the feature constraint mechanism may include, but are not limited to, the following: mapping the target location information in the first type of data to the image coordinate system of the second type of data through geometric constraints, thereby limiting the recognition area; or dynamically adjusting the input weights of the visual recognition model so that the model prioritizes the feature information of the target area.

[0076] Lightweight recognition models for edge computing units can be compressed using techniques such as pruning, quantization, or knowledge distillation, thereby reducing computational complexity and storage requirements. For example, lightweight neural network architectures such as MobileNet or ShuffleNet can be employed to ensure that the model can run efficiently on resource-constrained devices.

[0077] The overall technical solution forms a complete "monitoring-triggering-fusion-decision" closed loop through multimodal sensing collaboration, intelligent triggering mechanisms, and optimization of data fusion and edge computing. Microwave radar continuously monitors and rapidly identifies instantaneous anomalies, the vision module provides high-precision detail acquisition during critical periods, spatiotemporal calibration ensures data consistency, feature constraints improve recognition accuracy, and edge computing enables efficient processing. These technologies work closely together to solve the problem of accurate identification and tracing of instantaneous faults under complex weather conditions.

[0078] This application further proposes that the processing unit is specifically configured to: determine whether the speed at which the foreign object approaches and / or the vibration amplitude of the conductor in the first type of data exceed the corresponding preset threshold, wherein the preset threshold is a configurable speed threshold and vibration threshold.

[0079] Specifically, the foreign object approach speed refers to the dynamic rate at which an external intruding object approaches a power transmission line. This can be achieved by using microwave radar to collect target motion parameters, with the aim of penetrating severe weather conditions to accurately capture the initial stage of an instantaneous event.

[0080] Among them, conductor vibration amplitude refers to the change in the physical state of a conductor under external forces. It can be detected by microwave radar to detect minute vibrations in the conductor, with the aim of consistently identifying potential conductor hazards. In addition, preset thresholds refer to the judgment criteria set according to the actual needs of the scenario. They can be configured remotely, manually set locally, or preset based on historical data templates. Remote configuration is completed by issuing commands through the main station system, while manual setting is achieved through on-site terminal operation. Historical data templates include threshold configuration schemes for typical scenarios such as mountainous areas, plains, and suburban areas, with the aim of adapting to the complex differences in scenarios with different regions, voltage levels, and line types.

[0081] In detail, this solution focuses on two core parameters: the speed at which foreign objects approach and the amplitude of conductor vibration. It uses "AND / OR" logic for flexible judgment, ensuring comprehensive coverage of external intrusion faults, conductor-related faults, and combined fault scenarios. For example, in mountainous areas with strong winds, the conductor vibration threshold can be remotely configured to increase from 0.3 mm to 0.6 mm, effectively reducing the false alarm rate of wind-induced vibration. In areas with high foreign object incidence, such as suburban areas, the foreign object approach speed threshold can be locally configured to decrease from 0.5 m / s to 0.3 m / s, significantly reducing missed instantaneous intrusions.

[0082] Simultaneously, this configuration logic works in conjunction with the synchronous calibration module to ensure the reliability of the first type of data acquisition, improving the accuracy of instantaneous event determination by more than 25% compared to the fixed threshold scheme. Furthermore, the configurable threshold design of the processing unit provides a basis for parameter adjustment in the subsequent dynamic optimization module, further enhancing the system's adaptability and robustness.

[0083] This application further proposes a feature constraint mechanism for the processing unit: the target spatial trajectory and location information in the first type of data are incorporated into the visual recognition process of the second type of data through feature layer injection, thereby constraining the target box prediction range to reduce interference.

[0084] Specifically, target spatial trajectory and location information refers to precise data after time synchronization and spatial calibration by the synchronization calibration module. This data can be acquired using microwave radar and generated as a binary mask or weight matrix through coordinate mapping. Feature layer injection refers to element-wise addition or channel-by-channel fusion operations performed in the intermediate feature extraction layer of the visual recognition model. For example, feature maps converted from radar spatial information are fused with visual feature maps at the Conv3 layer channel-by-channel. The purpose is to guide the visual model to focus on the actual location region of the target using the spatial information of the radar data.

[0085] The constrained target bounding box prediction range refers to the range of coordinates of the region of interest in the visual image converted from radar spatial trajectory information, and the feature response of non-target areas is masked by a masking mechanism. For example, if the radar locates the trajectory of a foreign object in the range of (100,200)-(300,400) pixels in the image, after injection, the model will only generate anchor boxes and calculate the target confidence within this range, thereby effectively reducing the interference of environmental noise on fault identification.

[0086] In detail, this scheme ensures that the time synchronization error between the radar and the vision module is ≤5ms and the coaxial deviation is ≤0.5° through a synchronous calibration module, laying the foundation for accurate injection of spatial information. Based on this, the processing unit transforms the target spatial trajectory and position information acquired by the radar into feature maps recognizable by the visual model, and performs deep fusion in the intermediate feature extraction layer of the convolutional neural network. This deep binding implementation allows the visual model to focus on the effective area indicated by the radar during the initial feature generation stage, avoiding irrelevant background information from entering the deep network and significantly reducing the accumulation of interference features. Simultaneously, the lightweight model running on the edge computing unit can efficiently process the fused feature data, ensuring that the system can quickly respond to transient faults even under complex weather conditions.

[0087] Through the above technical solutions, the system can achieve accurate fault identification in severe weather conditions, reducing the false alarm rate by more than 35% compared to existing technologies. The target box prediction deviation is controlled within 5 pixels, and the recall rate for instantaneous fault identification is increased to over 96%. Furthermore, this mechanism works organically with preceding modules, enabling accurate location of instantaneous faults such as kite string splices and conductor strand breaks even in dense fog with visibility ≤200m, significantly improving the reliability and efficiency of power transmission line inspection.

[0088] This application further proposes that the fault identification model is a lightweight model, whose computational complexity and size are suitable for running on edge computing units.

[0089] Specifically, lightweight processing refers to optimizing a model through a series of techniques to enable it to run efficiently in resource-constrained environments. In practical applications, lightweight processing can be achieved through model pruning, knowledge distillation, or parameter quantization. Model pruning reduces model complexity by removing redundant convolutional kernels and fully connected layer neurons; knowledge distillation uses a high-precision large model as a teacher model to train smaller student models to inherit its recognition capabilities; parameter quantization converts 32-bit floating-point parameters into 8-bit integers or 16-bit half-precision, thereby reducing storage and computational overhead. These techniques can be used individually or in combination, aiming to ensure that the model maintains high recognition accuracy while significantly reducing the number of parameters. The purpose of introducing lightweight processing is to solve the operational bottleneck problem caused by computational intensity on edge devices, ensuring the real-time performance of fault detection and system efficiency.

[0090] In detail, this solution utilizes a lightweight fault identification model that accurately matches the hardware architecture characteristics of edge computing units. Edge computing units typically employ low-power FPGAs, embedded CPUs (such as the ARM Cortex-A series), or neural network processors (NPUs). Therefore, the lightweight model must meet requirements such as supporting hardware parallel computing, avoiding complex branching structures, and ensuring continuous memory access. For example, the convolutional kernel size is adapted to the FPGA's DSP array to support parallel computing, reducing instruction jump overhead in the embedded CPU, and lowering data transfer latency through continuous memory access. Furthermore, the lightweight model works in close conjunction with the preceding feature constraint mechanism. Since the feature constraint mechanism filters environmental interference using radar spatial information, enabling visual recognition to focus on the effective area, the model does not need to adapt to complex backgrounds with redundant parameters, providing a prerequisite for lightweighting. Based on this, the low-latency inference of the lightweight model ensures that fault determination after feature fusion can be completed quickly, supporting the real-time response requirements for instantaneous faults.

[0091] Through the above technical solutions, the lightweight model not only achieves specific metrics such as a size ≤100MB, single-frame inference latency ≤500ms, and inference power consumption ≤2W, but also maintains a fault identification accuracy rate of over 95%, solving the problem of decreased accuracy in traditional lightweight models. Furthermore, this solution overcomes the limitations of existing technologies that separate "model design" from "edge hardware." By combining targeted lightweight technology with hardware architecture adaptation, it achieves a balance between high-precision identification, low-latency inference, and low-power operation, effectively supporting the stable operation of the power transmission line inspection system in low-power, low-computing-power outdoor scenarios.

[0092] This application further proposes that the above-mentioned transmission line fault inspection system also includes a dynamic optimization module, which is configured to: receive feedback information based on the fault determination result; and adaptively adjust the event determination conditions of the first sensing module and / or the high-frequency acquisition parameters of the second sensing module according to the feedback information.

[0093] Specifically, a dynamic optimization module refers to a functional unit that can automatically adjust system parameters based on actual operating data. This can be achieved using rule-based expert systems, machine learning models, or pre-defined mapping tables. Feedback information refers to the dataset used to guide parameter adjustments, which can include automatic verification results from the main station system, manual error correction annotations from maintenance personnel, system performance statistics, and data from external environmental linkages, among other sources.

[0094] Event determination conditions refer to the set of key parameters that affect the first sensing module's judgment of instantaneous events. These can be implemented in various forms, such as a foreign object approach speed threshold, a wire vibration amplitude threshold, an instantaneous event duration threshold, or a judgment sensitivity level. High-frequency acquisition parameters refer to the set of parameters that affect the operating characteristics of the second sensing module in high-frequency mode. These can be implemented using diverse settings such as image acquisition frequency, resolution, acquisition duration, exposure parameters, or frame rate.

[0095] In detail, this solution achieves dynamic optimization of system parameters by constructing a complete feedback closed-loop mechanism. The dynamic optimization module receives fault judgment results uploaded from the edge computing unit and generates adjustment instructions by combining multi-dimensional feedback information.

[0096] For the first sensing module, the dynamic optimization module adjusts event judgment conditions such as the foreign object approach speed threshold or the conductor vibration amplitude threshold according to the false alarm rate or missed alarm rate in specific scenarios. For example, when false alarms increase due to strong winds, the system will automatically raise the vibration threshold to suppress interference; when missed alarms occur in sensitive areas, the speed threshold will be lowered to enhance detection sensitivity.

[0097] For the second sensing module, the dynamic optimization module comprehensively considers the balance between recognition accuracy and resource consumption, dynamically adjusting parameters such as acquisition frequency and resolution. For example, in scenarios with insufficient lighting but sufficient bandwidth, exposure parameters and resolution are prioritized to ensure detail; in scenarios with limited bandwidth but simple scenes, the frame rate is appropriately reduced while maintaining the acquisition of core details. This parameter adjustment mechanism forms a complete working chain with the preceding modules: the adjusted event judgment conditions are fed back to the processing unit to guide subsequent instantaneous event judgment, the optimized high-frequency acquisition parameters are sent to the second sensing module to improve the quality of fault detail acquisition, and the adjustment effect is continuously improved through the cyclical verification of "acquisition-judgment-feedback-optimization".

[0098] Through the above technical solutions, the system has achieved an upgrade from static configuration to dynamic optimization, significantly improving the accuracy of fault identification and operational stability in complex environments.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method of transmission line fault detection, characterized by, The method comprises: collecting first type data of the power transmission line environment through a first sensing module, the first sensing module being a microwave radar capable of penetrating severe weather, and the first type data including target motion parameters and conductor physical state parameters; collecting second type data of the power transmission line environment through a second sensing module, the second sensing module being a visual acquisition module with high-definition imaging capability, and the second type data being image or video data containing details of a fault target; the second sensing module and the first sensing module are time-synchronized and spatially calibrated, and the synchronization accuracy and calibration deviation meet the collaborative work requirements; judging whether a preset instantaneous event exists based on the first type data, the instantaneous event being a power transmission line abnormal event with a duration not exceeding a preset time length; when it is determined that the instantaneous event exists, triggering the second sensing module to switch to a high-frequency acquisition mode to obtain detailed image data of the instantaneous event period; fusing the first type data and the detailed image data through a feature constraint mechanism to perform fault identification and generate a fault determination result, the feature constraint mechanism being a process of injecting target spatial information in the first type data into the identification process of the second type data; processing the fault determination result through a light-weight identification model running on an edge computing unit deployed locally on the inspection device, and uploading simplified information containing the fault determination result.

2. The power transmission line fault detection method according to claim 1, characterized by, The target motion parameters in the first type data include foreign object approaching speed, and the conductor physical state parameters include conductor vibration amplitude; the preset instantaneous event determination condition is that the foreign object approaching speed exceeds a preset speed threshold or the conductor vibration amplitude exceeds a preset vibration threshold.

3. The method of claim 2, wherein The preset speed threshold is 0.3 m / s to 0.5 m / s, and the preset vibration threshold is 0.3 mm to 0.8 mm; the preset time length of the instantaneous event is ≤2 seconds.

4. The method of claim 1, wherein The feature constraint mechanism specifically injects target spatial trajectory and position information in the first type data into the feature extraction link of the corresponding visual recognition model of the second type data through an attention mechanism or a feature mapping method to constrain the target recognition range.

5. The method of claim 1, wherein The method further comprises: receiving feedback information based on the fault determination result; adaptively adjusting the event judgment condition of the first sensing module and / or the working parameter of the second sensing module in the high-frequency working mode according to the feedback information.

6. A power line fault patrol system characterized by comprising: The method comprises: a first sensing module, which is a microwave radar capable of penetrating severe weather, is used to collect first type data of the power transmission line environment, and the first type data includes target motion parameters and conductor physical state parameters; a second sensing module, which is a visual acquisition module with high-definition imaging capability, is used to collect second type data of the power transmission line environment, and the second type data is image or video data containing details of a fault target; a synchronization and calibration module is used to realize time synchronization and spatial calibration of the first sensing module and the second sensing module to ensure collaborative work accuracy; a processing unit is configured to: judge whether a preset instantaneous event exists based on the first type data, the instantaneous event being a power transmission line abnormal event with a duration not exceeding a preset time length; When it is determined that the transient event exists, triggering the second sensing module to switch to a high-frequency acquisition mode; Fusing the first type of data and detailed image data collected by the second sensing module through a feature constraint mechanism to perform fault recognition; An edge computing unit deployed locally on the inspection device, configured to run a lightweight recognition model and generate a fault determination result, and further configured to upload simplified information containing the fault determination result.

7. The power line fault patrol system according to claim 6, characterized by, The processing unit is specifically configured to determine whether a foreign object approaching speed and / or a conductor vibration amplitude in the first type of data exceeds a corresponding preset threshold, the preset threshold being a configurable speed threshold and a vibration threshold.

8. The transmission line fault patrol system of claim 6, wherein, The feature constraint mechanism executed by the processing unit is to inject the target space trajectory and position information in the first type of data into a visual recognition process of the second type of data through a feature layer to constrain a target frame prediction range and reduce interference.

9. The transmission line fault patrol system of claim 6, wherein, The fault recognition model is a model that has been subjected to lightweight processing, and its computational complexity and volume are suitable for running on the edge computing unit.

10. The power line fault patrol system according to claim 6, characterized by, The system further includes a dynamic optimization module configured to: receive feedback information based on the fault determination result; adaptively adjust event determination conditions of the first sensing module and / or high-frequency acquisition parameters of the second sensing module according to the feedback information.

Images