Detection and rating method for external damage prevention risk of power transmission channel

Abstract

The invention relates to the technical field of power risk assessment, in particular to a power transmission channel anti-external damage risk detection and rating method, which comprises the following steps: acquiring image data, distance data and meteorological data of a power transmission channel; preprocessing the monitoring data to obtain standardized data; obtaining the category, position and physical attribute information of the external damage target through deep learning target recognition; mapping machine account data to obtain a severity coefficient, and querying a target type coefficient, a distance coefficient, a speed coefficient, a direction coefficient, a size (or quality) coefficient and a weather coefficient; carrying out continuous multiplication operation on the seven coefficients and then carrying out normalization to generate risk scores; and performing grading early warning according to the risk score and triggering a differentiated operation and maintenance work order. According to the method, quantitative sorting of external damage risks is realized, the problem of uneven operation and maintenance resource allocation caused by existing binary warning is solved, the risk assessment accuracy is improved through a coefficient weight dynamic updating mechanism, and a high-risk external damage scene is effectively and preferentially handled.

Description

Technical Field

[0001] This invention relates to the field of power risk assessment technology, and in particular to a method for detecting and rating the risk of external damage to power transmission channels. Background Technology

[0002] The external damage monitoring technology for power transmission channels mainly involves deploying vibration sensors, fiber optic sensing devices, and image acquisition equipment to continuously collect environmental data in real time. When abnormal vibration or visual changes are detected, the system analyzes the data characteristics through signal processing algorithms to identify potential external damage sources such as foreign object intrusion or mechanical impact, thereby triggering an early warning mechanism to assist maintenance personnel in timely intervention, reduce fault risks, and maintain the reliability of the power grid.

[0003] Existing power transmission line external damage monitoring technologies suffer from the following technical pain points: monitoring methods relying on visible light monitoring or manual inspections can only output binary alarms indicating the presence or absence of external damage targets, failing to quantify and rank risks of different levels, such as crane contact with lines or excessively high-altitude mechanical construction. All alarms are treated the same, forcing maintenance personnel to apply the same response procedures to low-risk field pedestrian alarms and high-risk crane operation alarms near 220kV high-speed rail traction lines. In application scenarios where the average length of lines maintained per person doubles, limited human resources are consumed on verifying a large number of low-risk alarms, while high-risk scenarios that actually trigger tripping get out of control due to response delays, ultimately leading to a severe mismatch between maintenance resource allocation and actual risk. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for detecting and rating the risks of external damage to power transmission channels, solving the technical problem of uneven allocation of operation and maintenance resources in high-risk external damage scenarios of power transmission channels due to the lack of risk quantification and ranking capabilities.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows:

[0006] This invention provides a method for detecting and rating the risk of external damage to power transmission channels, comprising:

[0007] Step 1: Collect monitoring data of the power transmission channel, including image data, distance data, and meteorological data;

[0008] Step 2: Preprocess the monitoring data collected in Step 1 to obtain standardized data;

[0009] Step 3: Perform target identification on the standardized data obtained in Step 2 to obtain the category, location, and physical attribute information of the externally damaged target;

[0010] Step 4: Based on the category and location of the external damage target obtained in Step 3, obtain the severity coefficient from the mapping ledger data, query the target type coefficient based on the category of the external damage target, query the distance coefficient, velocity coefficient, direction coefficient and size (or mass) coefficient based on the location and physical attribute information of the external damage target, and query the weather coefficient based on the meteorological data obtained in Step 2.

[0011] Step 5: Calculate the risk score by taking the target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size (or mass) coefficient, and weather coefficient obtained in Step 4.

[0012] Step 6: Based on the risk score obtained in Step 5, classify the risks into different levels, generate warning levels, and trigger maintenance work orders corresponding to the warning levels to prioritize the allocation of maintenance resources to high-risk external damage scenarios.

[0013] Furthermore, the method for detecting and rating the risk of external damage to power transmission channels described in this invention includes step 1: collecting image data using a visible light camera and an infrared thermal imager simultaneously triggered; collecting distance data using a lidar to scan the spatial position between the external damage target and the conductor; and collecting meteorological data to record wind speed, rainfall, and visibility parameters. The visible light camera, infrared thermal imager, and lidar are installed on the power transmission tower, and the meteorological data acquisition equipment is installed on the top of the tower.

[0014] Furthermore, in the method for detecting and rating the risk of external damage to power transmission channels described in this invention, step 2 includes: performing defogging enhancement and distortion correction processing on image data, and performing timestamp alignment processing on distance data and meteorological data; the timestamp alignment processing synchronizes the image data, distance data, and meteorological data to a unified time reference to generate a time-aligned data set; and performing size standardization processing on the time-aligned image data to crop and generate image tiles of a uniform size.

[0015] Furthermore, in the method for detecting and rating the risk of external damage to power transmission channels described in this invention, step 3 includes: inputting the image tiles generated in step 2 into a deep learning target detection model to identify the subcategories of the external damage target and output pixel bounding boxes; combining the distance data collected in step 1, mapping the pixel bounding boxes to three-dimensional space through coordinate transformation, and calculating the actual physical size of the external damage target; and calculating the movement speed of the external damage target and its movement direction relative to the transmission line conductor based on the continuous time series distance data and image data.

[0016] Furthermore, in the method for detecting and rating the risk of external damage to power transmission channels described in this invention, step 4 includes: locating the corresponding transmission tower number based on the location information of the external damage target in the image obtained in step 3; querying the power production management system based on the tower number to obtain the tower's voltage level parameters, load characteristics indicators, and historical tripping records; calculating the severity coefficient based on the obtained ledger parameters; simultaneously querying the target type coefficient mapping table based on the subcategories of the external damage target identified in step 3, querying the size (or mass) coefficient mapping table based on the actual physical size value or mass calculated in step 3, querying the speed coefficient mapping table and direction coefficient mapping table based on the movement speed and movement direction relative to the transmission line conductor calculated in step 3, and querying the weather coefficient mapping table based on the meteorological data obtained in step 2.

[0017] Furthermore, in the method for detecting and rating the risk of external damage to power transmission channels described in this invention, step 4 further includes: the distance coefficient is queried using the safe clearance distance corresponding to the tower voltage level obtained in step 5 as the benchmark parameter, and the ratio of the actual measured distance collected in step 1 to the benchmark parameter as the query index, and the coefficient value is obtained from the distance coefficient segmentation mapping table; the target type coefficient is obtained from the target type coefficient mapping table using the external damage target sub-category identified in step 3 as the query key; the speed coefficient is obtained from the speed coefficient mapping table using the motion speed value calculated in step 3 as the query key; the direction coefficient is obtained from the direction coefficient mapping table using the motion direction angle of the external damage object relative to the transmission line conductor calculated in step 3 as the query key; the size (or mass) coefficient is obtained from the size (or mass) coefficient mapping table using the actual physical size value or mass calculated in step 3 as the query key; and the weather coefficient is obtained from the weather coefficient mapping table using the meteorological parameter value obtained in step 2 as the query key.

[0018] Furthermore, in the method for detecting and rating the risk of external damage to power transmission channels described in this invention, step 5 includes: multiplying the target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size (or mass) coefficient, and weather coefficient obtained in step 4 to obtain an initial risk value; performing normalization processing on the initial risk value and mapping the normalization result to a numerical range of 0-100; and performing an integer operation on the mapped value to generate a risk score.

[0019] Furthermore, in the method for detecting and rating external damage risks to power transmission channels described in this invention, step 6 includes: comparing the risk score generated in step 5 with a preset threshold range to determine the warning level; generating a maintenance work order based on the warning level: a red warning work order is generated for a risk score ≥ 60, an orange warning work order is generated for a risk score 35-59, a yellow warning work order is generated for a risk score 15-34, and a blue warning work order is generated for a risk score < 15; and sending the maintenance work order to the maintenance system.

[0020] Furthermore, the method for detecting and rating external damage risks to power transmission channels according to the present invention also includes step 7: periodically collecting records of actual tripping events and normal operation and maintenance records of power transmission channels to construct a training dataset; applying machine learning algorithms to analyze the training dataset and calculating the contribution weights of the seven coefficients involved in step 4 to the tripping events; recalculating the values ​​of target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size (or mass) coefficient, and weather coefficient based on the contribution weights; and updating the values ​​in the coefficient mapping table used in steps 4 to 6 when the coefficient values ​​change beyond a set threshold.

[0021] Furthermore, in the method for detecting and rating external damage risks to power transmission channels described in this invention, step 7 further includes: counting the current cumulative sample size; when the cumulative sample size is less than 500, calculating the initial weight value using the analytic hierarchy process (AHP); when the cumulative sample size reaches or exceeds 500, calculating the data-driven weight value using the SHAP value analysis method; during the transition period from less than 500 to reaching or exceeding 500, combining the initial weight value and the data-driven weight value using a weighted average algorithm, gradually increasing the proportion of the data-driven weight value and decreasing the proportion of the initial weight value as the cumulative sample size increases.

[0022] Beneficial effects of this invention;

[0023] This invention synchronously collects image data, distance data, and meteorological data from power transmission channels. After preprocessing, standardized data is generated. Deep learning target recognition technology is used to obtain the category, location, and physical attribute information of external damage targets. Combined with ledger data, a severity coefficient is obtained. Target type coefficient, distance coefficient, speed coefficient, direction coefficient, size (or mass) coefficient, and weather coefficient are queried. Through multiplication and normalization, a quantified risk score is generated. Based on the risk score, tiered warnings are issued and differentiated operation and maintenance work orders are automatically triggered. This achieves accurate quantification and ranking of external damage risks, enabling operation and maintenance resources to be prioritized for high-risk scenarios. This effectively solves the problem of average resource allocation caused by the existing binary alarm mechanism. At the same time, by regularly collecting tripping event samples and normal operation and maintenance records, and using machine learning algorithms to dynamically update coefficient weights, the risk assessment model has self-evolution capabilities, can adapt to changes in the field environment, and continuously improves the accuracy of risk identification. This significantly reduces the occurrence rate of tripping accidents, optimizes operation and maintenance efficiency, and enhances the reliability of power grid operation. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0026] Figure 2 This is a schematic diagram of the distance coefficient of the present invention;

[0027] Figure 3 This is a schematic diagram of the velocity coefficient of the present invention;

[0028] Figure 4 This is a schematic diagram of the motion direction coefficient of the present invention;

[0029] Figure 5 This is a schematic diagram of the vehicle height / mass of the present invention;

[0030] Figure 6 This is a weather diagram for the present invention;

[0031] Figure 7 This is a schematic diagram illustrating the severity coefficient of the present invention. Detailed Implementation

[0032] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0033] This invention provides a method for detecting and rating the risk of external damage to power transmission channels, comprising:

[0034] Step 1: Collect monitoring data of the power transmission channel, including image data, distance data, and meteorological data;

[0035] Step 2: Preprocess the monitoring data collected in Step 1 to obtain standardized data;

[0036] Step 3: Perform target identification on the standardized data obtained in Step 2 to obtain the category, location, and physical attribute information of the externally damaged target;

[0037] Step 4: Based on the category and location of the external damage target obtained in Step 3, obtain the severity coefficient from the mapping ledger data, query the target type coefficient based on the category of the external damage target, query the distance coefficient, velocity coefficient, direction coefficient and size (or mass) coefficient based on the location and physical attribute information of the external damage target, and query the weather coefficient based on the meteorological data obtained in Step 2.

[0038] Step 5: Calculate the risk score by taking the target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size (or mass) coefficient, and weather coefficient obtained in Step 4.

[0039] Step 6: Based on the risk score obtained in Step 5, classify the risks into different levels, generate warning levels, and trigger maintenance work orders corresponding to the warning levels to prioritize the allocation of maintenance resources to high-risk external damage scenarios.

[0040] In the method for detecting and rating the risk of external damage to power transmission channels, step 1 involves the collection of multi-source monitoring data. Image data is collected simultaneously by visible light cameras and infrared thermal imagers deployed on transmission towers, covering different lighting conditions day and night. Distance data is obtained by using lidar scanning to acquire the three-dimensional spatial position between the external damage target and the conductor. Meteorological data, such as wind speed, rainfall, and visibility parameters, is recorded by sensors installed on the top of the towers. This multi-source data collection method provides a comprehensive input foundation for subsequent analysis. The simultaneous acquisition of visible light and infrared data can effectively cope with complex environments such as fog and night, while lidar point cloud data can accurately quantify the safe distance.

[0041] Step 2 involves standardizing and preprocessing the collected raw monitoring data. Image data undergoes dehazing enhancement and geometric distortion correction to eliminate environmental interference and lens distortion. Distance and meteorological data are synchronized to a millisecond-level timeline using a timestamp alignment algorithm, forming a unified spatiotemporal reference dataset. After time alignment, the image data is further standardized in size, cropped into fixed-size tile structures. This process ensures data consistency and provides a regular input for subsequent algorithm processing. The key to this preprocessing stage is establishing spatiotemporal correlations between multi-source data, laying the foundation for subsequent fusion analysis.

[0042] Step 3 utilizes preprocessed standardized data for target recognition and feature extraction. Standardized image tiles are input into a deep learning-based target detection model to identify subcategories of externally damaged targets and generate pixel-level bounding boxes. Combined with distance data acquired by LiDAR, coordinate transformation is performed using camera calibration parameters to map two-dimensional pixel coordinates to three-dimensional space, calculating the actual physical size or mass of the externally damaged target. Mass calculation involves estimating mass using a fixed density table, e.g., a truck weighing 1 ton, a kite weighing 0.2 kg, etc. Based on continuous time-series data, the velocity and direction of motion relative to the power transmission line conductors are calculated using differential operations. This step achieves the transformation from pixel-level recognition to physical quantity calculation, providing rich feature information for risk quantification.

[0043] Step 4 quantifies risk factors through multi-dimensional coefficient queries. Based on the location information obtained from target identification, it is mapped to the tower number in the power grid asset management system. The voltage level, load importance, and historical tripping records of the corresponding line are then queried to calculate the severity coefficient reflecting the line's vulnerability. Simultaneously, a pre-set target type coefficient mapping table is queried based on the identified external damage target subcategories. Size (or mass) coefficient, speed coefficient, and direction coefficient mapping tables are queried based on the calculated physical size or object mass, speed, and direction of movement, respectively. Finally, a weather coefficient mapping table is queried in conjunction with pre-processed meteorological data. This coefficientization process transforms continuous physical quantities into standardized risk contribution values.

[0044] Step 5 involves a multi-factor fusion risk assessment. The seven coefficients obtained in Step 4 are multiplied together to obtain an initial risk value. This initial value is then mapped to the 0-100 range using a linear normalization algorithm. The normalization result is rounded to the nearest integer to generate the final risk score. This multiplicative model considers the coupling effect of each risk factor, while the normalization process ensures the comparability of risk scores across different scenarios.

[0045] Step 6 implements tiered early warning and resource scheduling based on risk scores. Risk scores are compared with preset thresholds to categorize into four warning levels: red, orange, yellow, and blue. A red warning corresponds to a risk score of 60 or higher, triggering an immediate action work order; an orange warning corresponds to a score of 35-59, generating a daily inspection work order; a yellow warning corresponds to a score of 15-34, scheduling weekly inspections; and a blue warning corresponds to a score below 15, performing monthly routine inspections. Warning work orders are automatically distributed to the operations and maintenance management system, enabling priority handling of high-risk scenarios and precise allocation of operations and maintenance resources.

[0046] In the data acquisition phase, the visible light camera and infrared thermal imager employ a hardware synchronization triggering mechanism to ensure spatiotemporal consistency between the visible light and infrared images captured simultaneously. The lidar generates 3D point cloud data by scanning the relative position between the damaged target and the power line, while meteorological sensors continuously record wind speed, rainfall, and visibility parameters. All acquisition equipment is fixed at specific locations on the transmission tower, with the meteorological sensor mounted on the top of the tower to avoid obstruction. This deployment method enables the simultaneous acquisition of multi-dimensional field data, providing a foundation for subsequent fusion analysis.

[0047] The preprocessing stage first performs dehazing enhancement and distortion correction on the image data. An adaptive histogram equalization algorithm is used to improve image quality under low-light conditions, and lens distortion is eliminated through camera calibration parameters. Distance data and meteorological data are synchronized to a unified time base using a timestamp alignment algorithm, forming a time-aligned multi-source data set. After alignment, the image data is further standardized in size and cropped into a fixed-size tile structure. This processing not only eliminates data heterogeneity caused by equipment differences but also provides regular input for the target recognition algorithm.

[0048] The target recognition process inputs standardized image tiles into a deep learning model, outputting detailed subcategories and pixel-level bounding boxes for the damaged target. Combining distance data acquired by LiDAR, a perspective transformation matrix maps the two-dimensional pixel coordinates to three-dimensional space, calculating the actual physical dimensions of the damaged target. Based on continuous time-series data, an inter-frame difference algorithm calculates motion velocity and direction vectors, for example, to track the real-time trajectory of a crane boom, thereby dynamically updating its spatial pose information.

[0049] The coefficient query phase locates the corresponding tower number based on the target identification results, obtains the line voltage level, load characteristics, and historical tripping frequency by accessing the power production management system, and comprehensively calculates a coefficient reflecting the severity of the line's critical state. Simultaneously, a pre-set coefficient mapping table is invoked to query type coefficients based on target subcategories, size (or mass) coefficients based on physical dimensions or mass values, speed and direction coefficients based on motion vectors, and weather coefficients based on meteorological data, resulting in standardized values ​​for seven risk factors.

[0050] The distance coefficient is calculated based on the safe clearance corresponding to the line voltage level. It is segmented and mapped using the ratio of the actual measured distance to the benchmark value. For example, when the benchmark clearance for a 220kV line is 3.5 meters, a measured distance of 1.75 meters corresponds to a coefficient of 0.85. Other coefficients are obtained directly from tables using key parameters. For instance, a crane target corresponds to a type coefficient of 1.0, and a wind speed of 12 m / s corresponds to a weather coefficient of 0.8. These mapping relationships are generated through training with historical accident data, reflecting the contribution of different factors to risk.

[0051] The risk score is calculated using a multiplicative model that integrates seven types of coefficients. The initial risk value is linearly normalized and mapped to the 0-100 range, and finally rounded to generate the final risk score. The multiplicative model takes into account the coupling effect between factors. For example, when a crane target is too close to the conductor in thunderstorms, the superposition of various coefficients will produce a significantly high risk score, while a pedestrian target is still considered low-risk even if the distance is relatively close under clear weather conditions.

[0052] The tiered early warning mechanism classifies risks into four threshold levels: a red alert triggers an immediate response process, an orange alert generates a daily inspection task, a yellow alert schedules a weekly inspection, and a blue alert executes a monthly routine inspection. Work orders are automatically pushed to the maintenance mobile terminal, guiding on-site personnel to prioritize high-risk scenarios such as crane operations according to their risk level, avoiding response delays caused by the even distribution of resources.

[0053] The weight update process periodically collects tripping event samples and normal operation and maintenance records, trains a binary classification model using the LightGBM algorithm, and analyzes the contribution of each coefficient to the tripping event using SHAP values. When the coefficient weight changes exceed a threshold, the coefficient mapping table is dynamically updated. For example, after a new kite entanglement accident is added under windy and rainy weather, the system automatically increases the weight ratio of the weather coefficient and the type coefficient.

[0054] When the sample size is less than 500, the analytic hierarchy process (AHP) is used to calculate expert weights. Once the sample size is sufficient, the SHAP data-driven weighting method is switched to. During the transition phase, a weighted blending approach is used to achieve a smooth transition. For example, when the sample size is 300, the expert weight accounts for 40% and the data weight accounts for 60%. As the sample size accumulates, the system gradually transitions to a fully data-driven mode to avoid evaluation bias during the system's cold start phase.

[0055] This invention relates to the practical application of a method for detecting and rating the risk of external damage to power transmission channels. Visible light cameras and infrared thermal imagers are deployed on the transmission towers, both acquiring image data through a hardware-synchronized triggering mechanism. LiDAR scans the spatial position between the external damage target and the conductor to generate distance data, and meteorological sensors record wind speed, rainfall, and visibility parameters. All data acquisition equipment is fixed at specific locations on the towers, with the meteorological sensors installed at the top to avoid obstruction. This multi-source data acquisition method provides comprehensive input for subsequent analysis.

[0056] The acquired image data undergoes dehazing enhancement and distortion correction. Distance data and meteorological data are synchronized to a unified time base using a timestamp alignment algorithm, forming a time-aligned dataset. The aligned image data is then cropped into uniform-sized tile structures to eliminate data heterogeneity caused by device differences. The preprocessed, standardized image tiles are input into a deep learning target detection model to identify subcategories of externally damaged targets and output pixel bounding boxes. Combined with LiDAR point cloud data, the actual physical dimensions are calculated through coordinate transformation. Based on continuous time-series data, the motion speed and direction are calculated using an inter-frame difference algorithm.

[0057] The target identification results are used to locate the corresponding tower number, access the power production management system to obtain the voltage level, load type, and historical trip records, and calculate the severity coefficient. Simultaneously, a preset coefficient mapping table is queried to obtain the type coefficient based on the target subcategory, the size (or mass) coefficient based on the physical dimensions, and the speed coefficient, direction coefficient, and weather coefficient based on the motion vector and meteorological parameters, respectively. The distance coefficient is calculated based on the safe clearance corresponding to the voltage level, and the ratio of the actual measured distance to the benchmark value serves as the query index.

[0058] The initial risk value is obtained by multiplying seven coefficients together. After linear normalization, it is mapped to the range of 0 to 100, and then rounded to generate the final risk score. The risk score is compared with a preset threshold to determine four warning levels: a red warning triggers an immediate action work order, an orange warning generates a daily inspection task, a yellow warning arranges a weekly inspection, and a blue warning executes a monthly routine inspection. Work orders are automatically pushed to the operations and maintenance system to guide on-site personnel to prioritize high-risk scenarios.

[0059] The system periodically collects tripping event samples and normal operation and maintenance records, and uses machine learning algorithms to analyze the contribution of each coefficient to the tripping event. When the weight changes exceed a set threshold, the coefficient mapping table is updated. When the sample size is small, the analytic hierarchy process (AHP) is used to calculate the initial weights. Once the sample size is sufficient, the system switches to the SHAP value analysis method, and a weighted blending method is used to achieve a smooth transition during the transition phase. This dynamic update mechanism enables the system to adapt to changes in the field environment and continuously optimize the accuracy of risk assessment.

[0060] This invention achieves quantitative ranking of external damage risks to power transmission channels and precise allocation of operation and maintenance resources through multi-source data fusion, coefficient-based risk assessment, and adaptive weight updates, effectively solving the problem of response delay in high-risk scenarios.

[0061] Specific embodiments of the present invention are as follows:

[0062] Hardware deployment: The tower integrates a visible light / infrared dual-light gimbal, a laser ranging unit, and a mini weather station; the edge computing box has a built-in fanless low-power processor to execute the entire rating process; the central platform is responsible for ledger management, weight rolling updates, and work order distribution.

[0063] Data synchronization and preprocessing: Visible light images, infrared images, laser ranging values ​​and meteorological grid data are aligned with timestamps, and distortion correction and size standardization are performed to generate standard tiles and attach meteorological and coordinate information.

[0064] Target recognition and physical quantity conversion: Standard tile input is used to improve the YOLO model, and the output includes the outer break subclass, pixel box, and confidence score. Through pixel-world coordinate transformation, the actual size, velocity, direction of motion, and mass estimate of the target are obtained.

[0065] Ledger and Weather Mapping: Map pixel coordinates to tower numbers, read voltage level, historical trip count, and load characteristics to generate severity coefficients; at the same time, output weather coefficients based on meteorological codes.

[0066] Coefficient lookup and multiplication: Look up distance, speed, direction, size, weather, and sub-category coefficients in sequence, and normalize them all to the 0-1 range; multiply the seven coefficients to get the risk score R from 0 to 100.

[0067] Four-color grading and closed-loop: R is fixedly divided into four grades: red, orange, yellow, and blue. Corresponding inspection work orders are automatically generated and pushed to the operation and maintenance APP. After the on-site handling results are returned, the system automatically archives and updates the historical database.

[0068] Weight self-evolution: The central platform periodically pulls new tripped samples and recalculates the data weights. If the change of any dimension coefficient exceeds the set threshold, the edge box JSON file is hot-updated via MQTT to achieve weight self-evolution and zero-downtime maintenance.

[0069] Please also see Figures 2 to 6 This embodiment provides a risk detection and rating method for power transmission channel protection against external damage. For power transmission channels: First, data is obtained using a visible light 4K and infrared dual-light gimbal and lidar. Then, data synchronization and preprocessing are performed, aligning timestamps, and using NTP time synchronization for visible light / infrared / laser / meteorological data with an error <10ms. Next, image correction is performed, including dehazing (CLAHE), distortion correction (OpenCV undistort), and cropping to 1280×1280 tiles while retaining EXIF ​​coordinates. Then, target recognition and physical quantity conversion are performed. An improved YOLOv11 layer with a new 160×160 small target layer is added, CBAM attention is inserted, and 19 subclasses, pixel boxes, and confidence scores are output. The vehicle height is calculated using laser ranging (d) and the camera intrinsic parameter matrix. ,speed The direction θ = the angle between the displacement vector and the conductor. For mass estimation and testing, consult a fixed density table, such as: truck 1t, kite 0.2kg, etc. Based on voltage level, historical trip count, and load characteristics, take the highest score from the 10 / 7 / 4 / 1 table to obtain Ksev ∈ [1,10]. Based on wind speed + rainfall + thunderstorm, obtain Kweather ∈ {1.0,0.8,0.65,0.5,0.4}. Consult the tables sequentially: Ktype (19 constants), Kdist (5 levels), Kspeed (5 levels), Kdir (4 levels), Ksize (5 levels), Kweather (5 levels);

[0070] Round to the nearest integer between 0 and 100. Calculate the R value: R ≥ 60 for red, 35-59 for orange, 15-34 for yellow, and < 15 for blue. Implement appropriate countermeasures based on the color. Finally, the central platform periodically retrieves newly added tripped samples, recalculates the data weights, and repeats the above process for detection.

[0071] The 19 subcategories in target identification and physical quantity conversion are clustered, merged, and refined based on the external power transmission failures that have occurred to obtain a "minimum complete set". After undergoing the "technical featureization" process required by patent examination, they are divided into passenger vehicles such as buses, minibuses, cars, and SUVs; freight vehicles such as trucks, small trucks, and tractors; foreign objects such as kites, balloons, bird nests, and garbage; and also pedestrians, cattle, sheep, open flames, and smoke.

[0072] Parameter calculations for each component: based on tripping incidents and normal incidents in recent years;

[0073] Each segment completes 6 dimensions of features. (Tripped = 1, not tripped = 0), LightGBM binary classification, 10-fold crossover, AUC > 0.85 is retained, TreeSHAP outputs the 6-dimensional φi of each sample, and the mean of |φᵢ| is taken as the original contribution vector. , softmax normalization;

[0074] The sum is 1, thus obtaining the weight vector. The continuous value of w is divided into 5 segments according to 0-20%-40%-60%-80%-100%, and fixed coefficients are assigned: extremely high 1.00 / high 0.85 / medium 0.60 / low 0.30 / extremely low 0.05, forming a constant table of 5 levels (one for distance, speed, direction, size, and weather). The 19 subclasses are divided into [0.2, 1.0] from OneHot to SHAP and then through softmax, forming the Ktype constant table (19 rows).

[0075] For the subclass Ktype, all tripping events are relabeled, with each record being a 19-dimensional OneHot vector (only the corresponding subclass = 1, the rest = 0). Using the pre-trained LightGBM tripping model, TreeSHAP is run on each record to obtain a 19-dimensional φⱼ (the contribution value of the subclass to the "tripping probability"). The average of all |φⱼ| for the same subclass is taken to obtain 19 "subclass average contributions" Cⱼ (the larger the value, the more dangerous the situation). The sum is 1, and the value range naturally falls in [0.2, 1.0] (minimum approximately 0.2, maximum approximately 1.0). Write βⱼ into the JSON "Ktype table".

Claims

1. A method for detecting and rating the risk of external damage to power transmission channels, characterized in that, include: Step 1: Collect monitoring data of the power transmission channel, including image data, distance data, and meteorological data; Step 2: Preprocess the monitoring data collected in Step 1 to obtain standardized data; Step 3: Perform target identification on the standardized data obtained in Step 2 to obtain the category, location, and physical attribute information of the externally damaged target; Step 4: Based on the category and location of the external damage target obtained in Step 3, obtain the severity coefficient from the mapping ledger data, query the target type coefficient based on the category of the external damage target, query the distance coefficient, velocity coefficient, direction coefficient and size or mass coefficient based on the location and physical attribute information of the external damage target, and query the weather coefficient based on the meteorological data obtained in Step 2. Step 5: Calculate the risk score by combining the target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size or mass coefficient, and weather coefficient obtained in Step 4. Step 6: Based on the risk score obtained in Step 5, classify the risks into different levels, generate warning levels, and trigger maintenance work orders corresponding to the warning levels to prioritize the allocation of maintenance resources to high-risk external damage scenarios.

2. The method for detecting and rating the risk of external damage to power transmission channels according to claim 1, characterized in that, Step 1 includes: acquiring image data using a visible light camera and an infrared thermal imager simultaneously triggered; acquiring distance data using a lidar to scan the spatial position between the external target and the power line; and acquiring meteorological data to record wind speed, rainfall, and visibility parameters. The visible light camera, infrared thermal imager, and lidar are installed on the power transmission tower, and the meteorological data acquisition equipment is installed on the top of the tower.

3. The method for detecting and rating the risk of external damage to power transmission channels according to claim 2, characterized in that, Step 2 includes: performing dehazing enhancement and distortion correction on the image data, and performing timestamp alignment on the distance data and meteorological data; the timestamp alignment process synchronizes the image data, distance data and meteorological data to a unified time base to generate a time-aligned data set; and performing size normalization on the time-aligned image data and cropping it to generate image tiles of uniform size.

4. The method for detecting and rating the risk of external damage to power transmission channels according to claim 3, characterized in that, Step 3 includes: inputting the image tiles generated in Step 2 into a deep learning target detection model to identify the subcategories of the externally damaged target and output pixel bounding boxes; combining the distance data collected in Step 1, mapping the pixel bounding boxes to three-dimensional space through coordinate transformation, and calculating the actual physical size of the externally damaged target; and calculating the movement speed of the externally damaged target and its movement direction relative to the transmission line conductor based on the continuous time series distance data and image data.

5. The method for detecting and rating the risk of external damage to power transmission channels according to claim 4, characterized in that, Step 4 includes: locating the corresponding transmission tower number based on the location information of the externally damaged target in the image obtained in Step 3; querying the power production management system based on the tower number to obtain the tower's voltage level parameters, load characteristics indicators, and historical tripping records; calculating the severity coefficient based on the obtained ledger parameters; simultaneously querying the target type coefficient mapping table based on the subcategories of the externally damaged target identified in Step 3, querying the size or mass coefficient mapping table based on the actual physical size or mass calculated in Step 3, querying the speed coefficient mapping table and direction coefficient mapping table based on the movement speed and movement direction relative to the transmission line conductor calculated in Step 3, and querying the weather coefficient mapping table based on the meteorological data obtained in Step 2.

6. The method for detecting and rating the risk of external damage to power transmission channels according to claim 5, characterized in that, Step 4 also includes: The distance coefficient is retrieved using the safe clearance distance corresponding to the tower voltage level obtained in Step 5 as the reference parameter, and the ratio of the actual measured distance collected in Step 1 to the reference parameter as the query index, obtaining the coefficient value from the distance coefficient segmentation mapping table; the target type coefficient is retrieved from the target type coefficient mapping table using the external damage target sub-category identified in Step 3 as the query key; the velocity coefficient is retrieved from the velocity coefficient mapping table using the motion velocity value calculated in Step 3 as the query key; the direction coefficient is retrieved from the direction coefficient mapping table using the motion direction angle of the external damage object relative to the transmission line conductor calculated in Step 3 as the query key; the size (or mass) coefficient is retrieved from the size (or mass) coefficient mapping table using the actual physical size value or mass calculated in Step 3 as the query key; and the weather coefficient is retrieved from the weather coefficient mapping table using the meteorological parameter value obtained in Step 2 as the query key.

7. The method for detecting and rating the risk of external damage to power transmission channels according to claim 6, characterized in that, Step 5 includes: multiplying the target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size or mass coefficient and weather coefficient obtained in Step 4 to obtain an initial risk value; performing normalization processing on the initial risk value and mapping the normalization result to a numerical range of 0-100; and performing an integer operation on the mapped value to generate a risk score.

8. The method for detecting and rating the risk of external damage to power transmission channels according to claim 7, characterized in that, Step 6 includes: comparing the risk score generated in step 5 with the preset threshold range to determine the warning level; generating an operation and maintenance work order based on the warning level: a red warning work order is generated for a risk score ≥ 60, an orange warning work order is generated for a risk score 35-59, a yellow warning work order is generated for a risk score 15-34, and a blue warning work order is generated for a risk score < 15; and sending the operation and maintenance work order to the operation and maintenance system.

9. The method for detecting and rating the risk of external damage to power transmission channels according to claim 8, characterized in that, It also includes step 7, which involves periodically collecting records of actual tripping events and normal operation and maintenance records that occur in the transmission channel to build a training dataset; Machine learning algorithms are applied to analyze the training dataset, and the contribution weights of the seven coefficients involved in step 4 to the tripping event are calculated. The values ​​of target type coefficient, severity coefficient, distance coefficient, speed coefficient, direction coefficient, size or mass coefficient, and weather coefficient are recalculated based on the contribution weight. When the coefficient values ​​change beyond the set threshold, the values ​​in the coefficient mapping table used in steps 4 to 6 are updated.

10. The method for detecting and rating the risk of external damage to power transmission channels according to claim 9, characterized in that, Step 7 also includes: counting the current cumulative sample size; when the cumulative sample size is less than 500, using the analytic hierarchy process (AHP) to calculate the initial weight value; when the cumulative sample size reaches or exceeds 500, using the SHAP value analysis method to calculate the data-driven weight value; during the transition period from less than 500 to reaching or exceeding 500, using a weighted average algorithm to combine the initial weight value and the data-driven weight value, gradually increasing the proportion of the data-driven weight value and decreasing the proportion of the initial weight value as the cumulative sample size increases.

Images