A method for predicting the cleaning range of a power transmission line corridor by fusing tree species knowledge graph

Abstract

This invention relates to the field of power transmission line operation and maintenance technology, and discloses a method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs. The method includes the following steps: collecting point cloud data and multi-source data from meteorology, vegetation mechanisms, and electrical environment; performing time-series alignment, completion, and distortion removal preprocessing; constructing a tree species knowledge graph and a tree proxy model, outputting growth and risk prediction parameters; accurately extracting tree obstacle point clouds and matching tree species through an attention mechanism; calculating corrected safety distances and establishing a multi-dimensional hazard classification system; combining meteorological data to correct growth parameters, predicting future risks, dynamically delineating the clearance range, adjusting inspection cycles, and visualizing the results. This invention significantly improves prediction accuracy, overcomes the shortcomings of traditional inspections, saves annual operation and maintenance costs, reduces tripping losses, achieves precise control and mode transformation, considers multiple benefits, has strong adaptability, and broad prospects for promotion.

Description

Technical Field

[0001] This invention relates to the field of power transmission line operation and maintenance technology, and more specifically, to a method for predicting the clearance range of power transmission line corridors by integrating tree species knowledge graphs. Background Technology

[0002] Transmission lines are the core infrastructure for the safe and stable operation of power systems, and tree obstructions are one of the main causes of transmission line tripping. Traditional manual inspections are inefficient and risky, while excessive clearing significantly increases maintenance costs, and insufficient clearing directly leads to line outages and tripping. Currently, proactive maintenance is mainly carried out periodically using drone inspections and special manual inspections. However, periodic inspections are usually conducted on a fixed schedule, making it impossible to monitor the dynamic changes of tree obstructions in real time. Especially during seasons when trees grow rapidly, untimely clearing can easily lead to the accumulation of hidden dangers, causing tripping accidents and seriously threatening the safe and stable operation of the power grid.

[0003] Currently, the management of tree obstructions in power transmission line corridors mainly relies on traditional manual inspections and periodic clearing, supplemented by drone inspections in some scenarios. However, the overall technical level still has significant limitations: traditional manual inspection methods suffer from low efficiency, high cost, and insufficient judgment accuracy, making it difficult to meet the growing demand for precise management of tree obstruction clearing and wildfire prevention. Data collection methods are limited, mostly relying on point cloud data or manual observation data, failing to effectively integrate multi-source information such as meteorological, vegetation mechanisms, and electrical environment data. Furthermore, data from different sources exhibits temporal differences and format heterogeneity, resulting in weak data fusion and processing capabilities. There is a lack of precise management solutions based on tree species knowledge graphs. Safety distance calculations, tree growth predictions, clearing range settings, and inspection cycle settings are not dynamically adjusted based on the differentiated characteristics of tree species, relying on fixed standards or manual experience, leading to inaccurate risk assessments and a lack of targeted management measures. Therefore, there is an urgent need to provide a method for predicting the clearing range of power transmission line corridors that integrates tree species knowledge graphs to solve the above problems. Summary of the Invention

[0004] To address the aforementioned technical challenges, this invention provides a method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs. This significantly improves the accuracy of clearance range prediction and overcomes the problems of low efficiency and insufficient precision in traditional manual inspections. Dynamic management reduces annual maintenance costs and minimizes tripping losses, while multi-source data fusion provides reliable support. Relying on tree species knowledge graphs for precise management promotes the transformation of tree obstacle management models, balancing safety, economic, social, and ecological benefits. It is highly adaptable and has broad prospects for widespread application.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] A method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs includes the following steps:

[0007] S1. The UAV is equipped with a lidar to collect point cloud data of the power transmission line corridor. The supporting equipment simultaneously collects meteorological, vegetation mechanism and electrical environment data. All data are stored in association with the tower number and the collection timestamp. The point cloud data is aligned with the three types of time-series data, namely meteorological data, vegetation mechanism data and electrical environment data, through a dynamic time warping algorithm. The missing areas of the point cloud data are weighted and filled, and the distorted point cloud data is removed.

[0008] S2. Test the biomechanical and electrical characteristics of the main tree species in the test corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database, develop a configurable tree proxy model based on the graph, integrate it into the power transmission line tree obstacle risk assessment platform, and output tree risk and growth prediction parameters.

[0009] S3. Construct an attention mechanism point cloud classification model to locate high-risk areas of tree obstacles around the guide line. Combine point cloud image features with tree species knowledge graph to match tree species categories and remove interference items to complete accurate classification of tree obstacle point clouds.

[0010] S4. Calculate the safety distance correction coefficient based on tree resistivity to obtain the final safe distance between the conductor and the tree barrier; establish a tree barrier hazard classification system including biomechanical and electrical risk dimensions, and upgrade the tree barrier hazard ledger;

[0011] S5. Introduce meteorological data to correct tree growth parameters, combine the lodging probability output by the tree proxy model to predict the future risk of tree obstacles, adjust the core clearing area according to the biomechanical characteristics of tree species, adjust the inspection cycle according to the electrical risk level, and dynamically display the tree obstacle growth prediction and electrical risk heat map in the tree obstacle risk assessment platform.

[0012] As a preferred embodiment of the present invention, step S1 includes the following specific steps:

[0013] S11. The UAV, equipped with lidar equipment, will conduct operations on the power transmission line corridor, controlling the flight path altitude between 80 and 130 meters, ensuring that the collected point cloud density is not less than 20 points / m², and the collected point cloud data includes three-dimensional coordinates and collection timestamps.

[0014] S12. Multi-source auxiliary data is collected independently of the UAV's lidar using supporting equipment. Meteorological data is collected from automatic weather stations deployed every 500m along the power transmission line corridor, including ambient temperature, relative humidity, instantaneous wind speed, and corresponding collection timestamps. Vegetation mechanism data is obtained through on-site sampling combined with laboratory testing, including tree species, current growth height, tree age, diameter at breast height, and corresponding collection timestamps. Electrical environment data is collected on-site using a portable electrical parameter detector, including line operating voltage, operating current, environmental resistivity, and corresponding collection timestamps.

[0015] S13. Using the rule of combining tower number and collection timestamp as a unified identifier, an associated index is built for all collected data to achieve a one-to-one mapping between point cloud data and multi-source auxiliary data. All data is stored in a distributed database composed of MySQL and MongoDB databases. Structured parameters such as tower number and temperature are stored in the MySQL database, while unstructured point cloud data is stored in the MongoDB database, supporting fast data tracing and retrieval.

[0016] As a preferred embodiment of the present invention, S1 further includes the following steps:

[0017] S14. Based on the timestamp sequence of point cloud data, align the three types of time-series data: meteorological data, vegetation mechanism data, and electrical environment data. Eliminate the time difference caused by different devices. By backtracking the optimal alignment path, perform linear interpolation or downsampling on the time-series data to keep the timestamps of various types of data consistent, and obtain the aligned dataset.

[0018] S15. Verify the data alignment effect and calculate the alignment similarity. The similarity should be no less than 95%. If the requirement is not met, readjust the alignment path.

[0019] S16. For point cloud missing areas caused by complex terrain such as valleys and steep slopes, an attention mechanism weighted completion algorithm is adopted. First, the effective point cloud set around the missing area is divided. Then, weights are assigned according to the distance between each effective point cloud and the missing area. The weighted effective point cloud data is used to complete the coordinate information of the missing area and generate the completed point cloud dataset.

[0020] S17. Use the isolated forest algorithm to detect distorted points in the point cloud, construct 100 isolated trees, calculate the anomaly score for each point cloud, set the anomaly score threshold to 0.7, remove distorted points caused by severe weather with anomaly scores not lower than the threshold, and output the final preprocessed point cloud dataset.

[0021] As a preferred embodiment of the present invention, S2 includes the following specific steps:

[0022] S21. Select the main tree species that cover more than 95% of the common tree species in the transmission line corridor, and carry out biomechanical and electrical characteristic tests in combination in the laboratory and on site. Measure the bending strength, tensile strength, tree resistivity, and breakdown voltage under different humidity gradients of the tree species. Integrate all test data to build a basic database to provide data support for the construction of a tree species knowledge graph.

[0023] S22. Construct a multi-dimensional tree species knowledge graph containing nodes, relationships, and attributes. The graph is divided into four layers: basic attribute layer, biomechanical layer, electrical characteristic layer, and growth characteristic layer. The Neo4j graph database is used to store the knowledge graph. At the same time, a dedicated query interface is built to support the quick retrieval of parameters of the corresponding dimension by tree species.

[0024] S23. Develop a configurable tree proxy model based on tree species knowledge graph data, integrate the model into the power transmission line tree obstacle risk assessment platform, take tree species, relative humidity, temperature and tree age as model input conditions, output three core prediction parameters: tree growth height in the next six months, tree lodging probability and lightning breakdown risk, and store all output parameters synchronously in the assessment platform to provide data basis for subsequent tree obstacle risk calculation.

[0025] As a preferred embodiment of the present invention, step S3 includes the following specific steps:

[0026] S31. Extract conductor point clouds from the preprocessed point cloud dataset and construct a three-dimensional contour model of the conductor. The model defines the conductor range by the lowest elevation and the highest elevation of the conductor. At the same time, it identifies high-risk areas around the conductor. High-risk areas are point cloud areas whose distance from the three-dimensional contour model of the conductor is no greater than the basic safety distance. The basic safety distance is determined according to the line voltage level.

[0027] S32. Construct an attention mechanism point cloud classification model. Input the preprocessed point cloud dataset into the model and calculate the attention weight of each point cloud. The weight calculation logic is the ratio of the exponent of the mapping result of the feature matching function on the elevation and density features of the point cloud in the high-risk area to the sum of the exponents of the corresponding mapping results of all point clouds. Set a weight threshold and output the point clouds in the high-risk area with attention weights greater than or equal to the weight threshold as the candidate set of tree obstacle point clouds. At the same time, calculate the recall rate, which is the ratio of the number of correctly identified tree obstacle point clouds to the sum of the number of correctly identified tree obstacle point clouds and the number of missed tree obstacle point clouds. The recall rate should not be less than 98%. If it is not met, adjust the weight threshold and reclassify.

[0028] S33. Extract image and morphological features from the candidate set of tree barrier point clouds to form a feature set. The feature set includes three parameters: crown width, tree height, and diameter at breast height.

[0029] S34. Retrieve the morphological parameter set corresponding to all tree species in the tree species knowledge graph through the query interface. The parameter set includes the standard crown width, standard tree height and standard diameter at breast height of each tree species. Calculate the similarity between the feature set and the morphological parameter set of each tree species. The similarity calculation logic is 1 minus the ratio of the Euclidean distance between the two to the maximum Euclidean distance. Set the similarity threshold.

[0030] S35. Retain point clouds with similarity not lower than the similarity threshold and match them with the corresponding tree species. Remove interference items such as weeds and rocks with similarity lower than the similarity threshold. Finally, output the accurate tree obstacle point cloud dataset. The accurate tree obstacle point cloud dataset includes tree species, morphological feature set and actual distance from tree obstacle to guide line.

[0031] As a preferred embodiment of the present invention, the calculation of the safety distance correction coefficient in S4 is specifically as follows:

[0032] Obtain basic safety distances from line design specifications. From tree species knowledge graph Retrieve the resistivity of the corresponding tree species From the aligned meteorological data Get the current relative humidity ;

[0033] The safety distance correction factor is calculated using the following formula. , ,in These are the weighting coefficients. , The base value for tree species resistivity, with a safety distance correction factor. With relative humidity Increase resistivity Decrease and increase;

[0034] Final safe distance Calculate according to the following formula:

[0035]

[0036] Simultaneously determine potential hazards: if the actual distance of the tree obstruction... Less than the final safe distance If so, it will be included in the list of hidden danger trees;

[0037] The tree obstacle hazard classification system described in S4 is specifically established as follows:

[0038] A three-level, four-element hazard classification system is constructed. The four elements are spatial distance, biomechanical, electrical risk, and growth characteristics. Among them, the spatial distance element is... Biomechanical elements are Electrical risk factors are Growth characteristics are Scoring range for each element The higher the score, the greater the risk; The actual distance to the tree barrier. For the final safe distance, The probability of a tree falling over. To mitigate the risk of lightning strikes, For the next six months' growth height;

[0039] The overall score is calculated using the following formula: Weight , , , ;

[0040] Hazard levels are classified according to a comprehensive score. It is classified as a Level I major hidden danger, with a comprehensive score of The hazard is classified as Level II general hazard, with a comprehensive score of [missing information]. The hazard is classified as Level III, a minor safety concern.

[0041] The upgrade of the tree obstacle hazard log in S4 is as follows:

[0042] Add fields for tree species, resistivity, probability of lodging, breakdown risk, comprehensive level, and corrected safe distance, and enter all parameters of the hazard tree obstacle set to form a standardized hazard database.

[0043] As a preferred embodiment of the present invention, step S5 specifically includes:

[0044] S51. Introduce forecast meteorological data for the next six months, extract average temperature and extreme wind speed, and correct the tree growth height for the next six months.

[0045] S52. Combining the inhibitory effects of average temperature and extreme wind speed on tree growth in future meteorological conditions, the original predicted growth height is dynamically adjusted by setting meteorological correction weights and meteorological influence functions to obtain a corrected growth height that better reflects the actual growth situation.

[0046] S53. Calculate the change in tree height over the next six months. This change is the difference between the corrected growth height and the current height. Then, obtain the future distance from the tree barrier to the guide wire. This distance is the sum of the actual distance of the tree barrier and the change in tree height.

[0047] S54. Calculate the future comprehensive risk. Based on the current tree barrier comprehensive score, incorporate the influence weights of tree fall probability and lightning strike risk, quantify and superimpose the effects of the two core risks on future hidden dangers, and obtain the future tree barrier comprehensive hidden danger risk value.

[0048] S55. Delineate the core clearing area according to the level of tree obstacle hazard. The clearing area for Level I major hazard tree obstacles is the sum of the final safe distance and the corrected growth height, extending outward from the guide line. The clearing area for Level II general hazard tree obstacles is the final safe distance. The clearing area for Level III minor hazard tree obstacles is the future distance from the tree obstacle to the guide line, ensuring coverage of areas with future growth risks.

[0049] S56. Set the inspection cycle according to the classification of electrical risk factors. If the electrical risk factor score is not lower than eight points, it is high risk and the inspection cycle is seven days; if the electrical risk factor score is not lower than five points but lower than eight points, it is medium risk and the inspection cycle is thirty days; if the electrical risk factor score is lower than five points, it is low risk and the inspection cycle is ninety days. If the probability of tree falling is not lower than 0.7, the inspection cycle is halved.

[0050] S57. Synchronize the core cleaning scope, the set inspection cycle, the tree growth prediction curve, and the risk heat map to the tree obstacle risk assessment platform. In the risk heat map, Level I hazards are marked in red, Level II hazards are marked in yellow, and Level III hazards are marked in green, which supports operation and maintenance personnel to view and schedule intuitively.

[0051] As a preferred embodiment of the present invention, it also includes S6: verifying the tree species characteristic parameters and the fusion effect of multi-source data, and after meeting the preset error requirements, collecting new data at a fixed period to update and iterate the tree species knowledge graph and tree proxy model parameters to continuously improve the prediction accuracy.

[0052] S6 specifically refers to:

[0053] S61. Verify the characteristic parameters of tree species. Select no less than 30 typical hidden tree obstacles on site, measure characteristic parameters such as bending strength and tree resistivity, retrieve the corresponding parameters in the tree species knowledge graph, and calculate the relative error. The relative error is the absolute value of the difference between the measured parameter and the graph parameter divided by the measured parameter and then multiplied by 100%. The relative error is required to be no more than 5%. If it exceeds this range, the measured parameters are added to the knowledge graph and the corresponding node attributes are corrected.

[0054] S62. Verify the effect of multi-source data fusion. For the same transmission line corridor, prediction is carried out using point cloud data only and multi-source fusion data respectively. The multi-source fusion data includes point cloud data, meteorological data and tree species knowledge graph to obtain two sets of prediction results. Based on the actual on-site cleanup verification results, the prediction errors of the two sets are calculated respectively, and then the error reduction rate is calculated. The error reduction rate is the absolute value of the difference between the two sets of prediction errors divided by the prediction error of point cloud data only and then multiplied by 100%. The error reduction rate is required to be no less than 30%. If it is not met, the data preprocessing algorithm or tree proxy model parameters are backtracked and optimized.

[0055] S63. Set a fixed iteration cycle of six months, collect newly added tree species test data, operation and maintenance cleanup feedback data, meteorological impact parameter change data and prediction deviation data within the cycle, supplement new data according to the four-layer structure of the knowledge graph, and update node attributes and relationships.

[0056] S64. Gradient descent is used to correct the parameters of the tree surrogate model. The parameter correction logic is that the new parameter is equal to the old parameter minus the product of the learning rate and the gradient of the loss function. The learning rate is 0.01, and the loss function is the mean square error, which is the sum of the squares of the differences between the predicted value and the actual value divided by the number of samples. The iterated knowledge graph and the model are re-integrated into the evaluation platform to complete the full-process parameter update to improve the prediction accuracy.

[0057] As a preferred embodiment of the present invention, the formula for dynamically adjusting the original predicted growth height in S52 is as follows:

[0058] ;in, To correct the growth height, This is the originally predicted growth height. For meteorological correction weights, This is a meteorological influence function;

[0059] The formula is:

[0060] in, It is the average temperature extracted from forecast meteorological data for the next six months. This is an extreme wind speed.

[0061] This invention also provides a transmission line corridor clearing range prediction system that integrates tree species knowledge graph, including: a data acquisition and preprocessing module, a tree species knowledge graph construction and proxy model development module, a tree obstacle point cloud accurate extraction module, a safety distance calculation and hazard classification module, and a clearing planning and visualization module;

[0062] The data acquisition and preprocessing module is used to collect point cloud data of the power transmission line corridor by using a drone equipped with a lidar, and to control the supporting equipment to simultaneously collect meteorological, vegetation mechanism and electrical environment data. All data are stored in association with tower number and collection timestamp. The point cloud data is aligned with the three types of time-series data (meteorological, vegetation mechanism and electrical environment) through a dynamic time warping algorithm. The missing areas of the point cloud data are filled in with weighted data and the distorted point cloud data is removed.

[0063] The tree species knowledge graph construction and proxy model development module is used to test the biomechanical and electrical characteristics of major tree species in the transmission line corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database; based on the tree species knowledge graph, a configurable tree proxy model is developed, and the proxy model is integrated into the transmission line tree obstacle risk assessment platform to output tree risk and growth prediction parameters through the model.

[0064] The tree obstacle point cloud precise extraction module is used to construct an attention mechanism point cloud classification model, which is used to locate high-risk areas of tree obstacles around the guide wire; the point cloud image features are combined with the tree species knowledge graph to match tree species categories, and interference items are removed to complete the precise classification of tree obstacle point clouds;

[0065] The safety distance calculation and hazard classification module is used to calculate the safety distance correction coefficient based on the tree resistivity to obtain the final safety distance between the conductor and the tree barrier; to establish a tree barrier hazard classification system that includes biomechanical and electrical risk dimensions, and to upgrade the tree barrier hazard ledger;

[0066] The cleaning planning and visualization module is used to incorporate meteorological data to correct tree growth parameters, combine the probability of lodging to predict future risks of tree obstacles, adjust the core cleaning area based on the biomechanical characteristics of tree species, and adjust the inspection cycle according to the electrical risk level. The module also dynamically displays tree obstacle growth prediction data and electrical risk heat map in the power transmission line tree obstacle risk assessment platform.

[0067] The beneficial technical effects of this invention are:

[0068] By utilizing drone-based LiDAR data collection and an attention mechanism model, the accuracy of cleaning range prediction is significantly improved, addressing the shortcomings of traditional manual inspections, such as low efficiency and insufficient precision. Dynamic management reduces both over- and under-cleaning, decreasing the number of cleaning operations per year. Based on a cleaning range of 100 kilometers per operation and a maintenance cost of 6,500 yuan per kilometer, this translates to annual savings of 1.95 million yuan in maintenance costs, while also reducing losses from power outages.

[0069] Breaking through the limitations of traditional single data collection, this method associates and stores point cloud data with meteorological, vegetation mechanism, and electrical environment data. It uses a dynamic time warping algorithm to solve the problems of temporal differences and format heterogeneity, completes data alignment, completion, and distortion correction, and outputs a standardized dataset to provide comprehensive and reliable support for subsequent evaluation.

[0070] By constructing a multi-dimensional tree species knowledge graph and a configurable agent model, the biomechanical, electrical, and growth characteristics of tree species are integrated into the entire process of safety distance calculation, growth prediction, and cleaning and inspection plan formulation. This replaces fixed standards and manual experience, and solves the core problems of inaccurate risk assessment and lack of targeted control.

[0071] This will transform tree obstruction management from passive clearing and regular general clearing to proactive prediction and precise control, significantly improving work efficiency and helping power supply companies increase revenue; enhancing power supply reliability, reducing power outage complaint rates, and improving corporate image; reducing unnecessary logging, taking into account ecological protection, and achieving synergistic benefits in safety, economy, society, and ecology.

[0072] It is adaptable to transmission line corridors of various voltage levels and different terrains. Its core algorithm and knowledge graph can be flexibly adapted to regional tree species and meteorological characteristics, making it highly universal. It provides an effective solution for the intelligent upgrading of power transmission line operation and maintenance, and has broad prospects for promotion. Attached Figure Description

[0073] Figure 1 This is the overall flowchart of the present invention.

[0074] Figure 2 This is a schematic diagram of point cloud acquisition and modeling in this invention. Detailed Implementation

[0075] In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0076] Combination Figure 1-2 The present invention provides the following embodiments:

[0077] Example 1:

[0078] A method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs includes the following steps:

[0079] S1. The UAV is equipped with a lidar to collect point cloud data of the power transmission line corridor. The supporting equipment simultaneously collects meteorological, vegetation mechanism and electrical environment data. All data are stored in association with the tower number and the collection timestamp. The point cloud data is aligned with the three types of time-series data, namely meteorological data, vegetation mechanism data and electrical environment data, through a dynamic time warping algorithm. The missing areas of the point cloud data are weighted and filled, and the distorted point cloud data is removed.

[0080] S2. Test the biomechanical and electrical characteristics of the main tree species in the test corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database, develop a configurable tree proxy model based on the graph, integrate it into the power transmission line tree obstacle risk assessment platform, and output tree risk and growth prediction parameters.

[0081] S3. Construct an attention mechanism point cloud classification model to locate high-risk areas of tree obstacles around the guide line. Combine point cloud image features with tree species knowledge graph to match tree species categories and remove interference items to complete accurate classification of tree obstacle point clouds.

[0082] S4. Calculate the safety distance correction coefficient based on tree resistivity to obtain the final safe distance between the conductor and the tree barrier; establish a tree barrier hazard classification system including biomechanical and electrical risk dimensions, and upgrade the tree barrier hazard ledger;

[0083] S5. Introduce meteorological data to correct tree growth parameters, combine the lodging probability output by the tree proxy model to predict the future risk of tree obstacles, adjust the core clearing area according to the biomechanical characteristics of tree species, adjust the inspection cycle according to the electrical risk level, and dynamically display the tree obstacle growth prediction and electrical risk heat map in the tree obstacle risk assessment platform.

[0084] As a preferred embodiment of the present invention, step S1 includes the following specific steps:

[0085] S11. The UAV, equipped with lidar equipment, will conduct operations on the power transmission line corridor, controlling the flight path altitude between 80 and 130 meters, ensuring that the collected point cloud density is not less than 20 points / m², and the collected point cloud data includes three-dimensional coordinates and collection timestamps.

[0086] S12. Multi-source auxiliary data is collected independently of the UAV's lidar using supporting equipment. Meteorological data is collected from automatic weather stations deployed every 500m along the power transmission line corridor, including ambient temperature, relative humidity, instantaneous wind speed, and corresponding collection timestamps. Vegetation mechanism data is obtained through on-site sampling combined with laboratory testing, including tree species, current growth height, tree age, diameter at breast height, and corresponding collection timestamps. Electrical environment data is collected on-site using a portable electrical parameter detector, including line operating voltage, operating current, environmental resistivity, and corresponding collection timestamps.

[0087] S13. Using the rule of combining tower number and collection timestamp as a unified identifier, an associated index is built for all collected data to achieve a one-to-one mapping between point cloud data and multi-source auxiliary data. All data is stored in a distributed database composed of MySQL and MongoDB databases. Structured parameters such as tower number and temperature are stored in the MySQL database, while unstructured point cloud data is stored in the MongoDB database, supporting fast data tracing and retrieval.

[0088] By limiting the drone flight path altitude and point cloud density, the point cloud data can accurately capture the spatial relationship between the guide wire and tree obstacles; multi-source auxiliary data are collected at fixed intervals or in a standardized manner to ensure the spatiotemporal consistency of the data; a distributed database is used to store structured and unstructured data separately, which not only meets the efficient query requirements of structured data, but also adapts to the storage characteristics of unstructured point cloud data. A unified identifier is used to build an associated index, laying the foundation for data traceability and rapid retrieval for subsequent multi-source data fusion.

[0089] As a preferred embodiment of the present invention, S1 further includes the following steps:

[0090] S14. Based on the timestamp sequence of point cloud data, align the three types of time-series data: meteorological data, vegetation mechanism data, and electrical environment data. Eliminate the time difference caused by different devices. By backtracking the optimal alignment path, perform linear interpolation or downsampling on the time-series data to keep the timestamps of various types of data consistent, and obtain the aligned dataset.

[0091] S15. Verify the data alignment effect and calculate the alignment similarity. The similarity should be no less than 95%. If the requirement is not met, readjust the alignment path.

[0092] S16. For point cloud missing areas caused by complex terrain such as valleys and steep slopes, an attention mechanism weighted completion algorithm is adopted. First, the effective point cloud set around the missing area is divided. Then, weights are assigned according to the distance between each effective point cloud and the missing area. The weighted effective point cloud data is used to complete the coordinate information of the missing area and generate the completed point cloud dataset.

[0093] S17. Use the isolated forest algorithm to detect distorted points in the point cloud, construct 100 isolated trees, calculate the anomaly score for each point cloud, set the anomaly score threshold to 0.7, remove distorted points caused by severe weather with anomaly scores not lower than the threshold, and output the final preprocessed point cloud dataset.

[0094] The Dynamic Time Warping algorithm can flexibly handle the time difference of different time series data through optimal path backtracking, and linear interpolation or downsampling ensures the uniformity of data dimensions. Similarity verification ensures that the alignment effect meets the requirements of subsequent fusion. The attention mechanism weighted completion allocates weights based on the logic of more accurate representation of missing regions by near point clouds, improving the accuracy of missing data completion. The isolated forest algorithm can efficiently identify discrete distortion points caused by severe weather through integrated detection of multiple isolated trees, avoiding abnormal data from interfering with subsequent tree obstacle identification and distance calculation.

[0095] As a preferred embodiment of the present invention, S2 includes the following specific steps:

[0096] S21. Select the main tree species that cover more than 95% of the common tree species in the transmission line corridor, and carry out biomechanical and electrical characteristic tests in combination in the laboratory and on site. Measure the bending strength, tensile strength, tree resistivity, and breakdown voltage under different humidity gradients of the tree species. Integrate all test data to build a basic database to provide data support for the construction of a tree species knowledge graph.

[0097] S22. Construct a multi-dimensional tree species knowledge graph containing nodes, relationships, and attributes. The graph is divided into four layers: basic attribute layer, biomechanical layer, electrical characteristic layer, and growth characteristic layer. The Neo4j graph database is used to store the knowledge graph. At the same time, a dedicated query interface is built to support the quick retrieval of parameters of the corresponding dimension by tree species.

[0098] S23. Develop a configurable tree proxy model based on tree species knowledge graph data, integrate the model into the power transmission line tree obstacle risk assessment platform, take tree species, relative humidity, temperature and tree age as model input conditions, output three core prediction parameters: tree growth height in the next six months, tree lodging probability and lightning breakdown risk, and store all output parameters synchronously in the assessment platform to provide data basis for subsequent tree obstacle risk calculation.

[0099] Joint testing covering over 95% of common tree species ensures the representativeness and comprehensiveness of basic data, providing a reliable data source for the knowledge graph. The four-layer structure of the knowledge graph is logically layered according to basic attributes, core characteristics, and application parameters, aligning with the parameter retrieval needs of tree barrier risk assessment. The Neo4j graph database is adapted to the node association storage characteristics of the graph, and a dedicated query interface improves parameter retrieval efficiency. The tree proxy model uses tree species as the core input, combined with key influencing factors such as meteorology and tree age, to quantify tree growth and risk patterns through data-driven modeling. The core parameters output directly support subsequent safe distance calculations and risk assessments.

[0100] As a preferred embodiment of the present invention, step S3 includes the following specific steps:

[0101] S31. Extract conductor point clouds from the preprocessed point cloud dataset and construct a three-dimensional contour model of the conductor. The model defines the conductor range by the lowest elevation and the highest elevation of the conductor. At the same time, it identifies high-risk areas around the conductor. High-risk areas are point cloud areas whose distance from the three-dimensional contour model of the conductor is no greater than the basic safety distance. The basic safety distance is determined according to the line voltage level.

[0102] S32. Construct an attention mechanism point cloud classification model. Input the preprocessed point cloud dataset into the model and calculate the attention weight of each point cloud. The weight calculation logic is the ratio of the exponent of the mapping result of the feature matching function on the elevation and density features of the point cloud in the high-risk area to the sum of the exponents of the corresponding mapping results of all point clouds. Set a weight threshold and output the point clouds in the high-risk area with attention weights greater than or equal to the weight threshold as the candidate set of tree obstacle point clouds. At the same time, calculate the recall rate, which is the ratio of the number of correctly identified tree obstacle point clouds to the sum of the number of correctly identified tree obstacle point clouds and the number of missed tree obstacle point clouds. The recall rate should not be less than 98%. If it is not met, adjust the weight threshold and reclassify.

[0103] S33. Extract image and morphological features from the candidate set of tree barrier point clouds to form a feature set. The feature set includes three parameters: crown width, tree height, and diameter at breast height.

[0104] S34. Retrieve the morphological parameter set corresponding to all tree species in the tree species knowledge graph through the query interface. The parameter set includes the standard crown width, standard tree height and standard diameter at breast height of each tree species. Calculate the similarity between the feature set and the morphological parameter set of each tree species. The similarity calculation logic is 1 minus the ratio of the Euclidean distance between the two to the maximum Euclidean distance. Set the similarity threshold.

[0105] S35. Retain point clouds with similarity not lower than the similarity threshold and match them with the corresponding tree species. Remove interference items such as weeds and rocks with similarity lower than the similarity threshold. Finally, output the accurate tree obstacle point cloud dataset. The accurate tree obstacle point cloud dataset includes tree species, morphological feature set and actual distance from tree obstacle to guide line.

[0106] The basic safety distance is determined according to the voltage level, which complies with the industry standard for safe operation of transmission lines. The three-dimensional contour model of the conductor accurately defines the boundary of the risk area. The attention mechanism focuses on the core features of tree obstacles in high-risk areas through weight calculation, which improves the targeting of tree obstacle point cloud screening. The recall rate constraint ensures that no high-risk tree obstacles are missed. The morphological feature set selects the different core parameters of tree species such as crown width, tree height, and diameter at breast height. The similarity is calculated by Euclidean distance, which can effectively distinguish different tree species from non-tree obstacle interference items, and achieve accurate classification and tree species matching of tree obstacle point clouds.

[0107] As a preferred embodiment of the present invention, the calculation of the safety distance correction coefficient in S4 is specifically as follows:

[0108] Obtain basic safety distances from line design specifications. From tree species knowledge graph Retrieve the resistivity of the corresponding tree species From the aligned meteorological data Get the current relative humidity ;

[0109] The safety distance correction factor is calculated using the following formula. , ,in These are the weighting coefficients. , The base value for tree species resistivity, with a safety distance correction factor. With relative humidity Increase resistivity Decrease and increase;

[0110] Final safe distance Calculate according to the following formula:

[0111]

[0112] Simultaneously determine potential hazards: if the actual distance of the tree obstruction... Less than the final safe distance If so, it will be included in the list of hidden danger trees;

[0113] The tree obstacle hazard classification system described in S4 is specifically established as follows:

[0114] A three-level, four-element hazard classification system is constructed. The four elements are spatial distance, biomechanical, electrical risk, and growth characteristics. Among them, the spatial distance element is... Biomechanical elements are Electrical risk factors are Growth characteristics are Scoring range for each element The higher the score, the greater the risk; The actual distance to the tree barrier. For the final safe distance, The probability of a tree falling over. To mitigate the risk of lightning strikes, For the next six months' growth height;

[0115] The overall score is calculated using the following formula: Weight , , , ;

[0116] Hazard levels are classified according to a comprehensive score. It is classified as a Level I major hidden danger, with a comprehensive score of The hazard is classified as Level II general hazard, with a comprehensive score of [missing information]. The hazard is classified as Level III, a minor safety concern.

[0117] The upgrade of the tree obstacle hazard log in S4 is as follows:

[0118] Add fields for tree species, resistivity, probability of lodging, breakdown risk, comprehensive level, and corrected safe distance, and enter all parameters of the hazard tree obstacle set to form a standardized hazard database.

[0119] The safety distance correction coefficient incorporates relative humidity and tree species resistivity, with weight allocation tailored to the degree of influence of these two factors on the safety distance, making the corrected safety distance more consistent with the actual environment and tree species characteristics. The four-element grading system covers current spatial risks, physical stability, electrical safety, and future growth risks, with weight settings based on the priority of each element's impact on tree obstacle hazards. The comprehensive scoring achieves a quantitative assessment of hazard risks. The upgraded ledger supplements core technical parameters, forming a standardized database to provide data support for subsequent cleanup planning and maintenance traceability.

[0120] As a preferred embodiment of the present invention, step S5 specifically includes:

[0121] S51. Introduce forecast meteorological data for the next six months, extract average temperature and extreme wind speed, and correct the tree growth height for the next six months.

[0122] S52. Combining the inhibitory effects of average temperature and extreme wind speed on tree growth in future meteorological conditions, the original predicted growth height is dynamically adjusted by setting meteorological correction weights and meteorological influence functions to obtain a corrected growth height that better reflects the actual growth situation.

[0123] S53. Calculate the change in tree height over the next six months. This change is the difference between the corrected growth height and the current height. Then, obtain the future distance from the tree barrier to the guide wire. This distance is the sum of the actual distance of the tree barrier and the change in tree height.

[0124] S54. Calculate the future comprehensive risk. Based on the current tree barrier comprehensive score, incorporate the influence weights of tree fall probability and lightning strike risk, quantify and superimpose the effects of the two core risks on future hidden dangers, and obtain the future tree barrier comprehensive hidden danger risk value.

[0125] S55. Delineate the core clearing area according to the level of tree obstacle hazard. The clearing area for Level I major hazard tree obstacles is the sum of the final safe distance and the corrected growth height, extending outward from the guide line. The clearing area for Level II general hazard tree obstacles is the final safe distance. The clearing area for Level III minor hazard tree obstacles is the future distance from the tree obstacle to the guide line, ensuring coverage of areas with future growth risks.

[0126] S56. Set the inspection cycle according to the classification of electrical risk factors. If the electrical risk factor score is not lower than eight points, it is high risk and the inspection cycle is seven days; if the electrical risk factor score is not lower than five points but lower than eight points, it is medium risk and the inspection cycle is thirty days; if the electrical risk factor score is lower than five points, it is low risk and the inspection cycle is ninety days. If the probability of tree falling is not lower than 0.7, the inspection cycle is halved.

[0127] S57. Synchronize the core cleaning scope, the set inspection cycle, the tree growth prediction curve, and the risk heat map to the tree obstacle risk assessment platform. In the risk heat map, Level I hazards are marked in red, Level II hazards are marked in yellow, and Level III hazards are marked in green, which supports operation and maintenance personnel to view and schedule intuitively.

[0128] Meteorological data is used to correct growth height, and the accuracy of growth prediction is improved based on the plant growth patterns of temperature affecting growth rate and extreme wind speed inhibiting growth. Future comprehensive risk calculations overlay two key risks: lodging and lightning breakdown, to quantify the future evolution trend of potential hazards. The cleanup scope is dynamically delineated according to the hazard level, ensuring full coverage of high-risk areas while avoiding excessive cleanup of low-risk areas. The inspection cycle is linked to electrical risks and the probability of lodging, enabling risk-based hierarchical management, while visualization enhances the intuitiveness and efficiency of operation and maintenance decisions.

[0129] As a preferred embodiment of the present invention, it also includes S6: verifying the tree species characteristic parameters and the fusion effect of multi-source data, and after meeting the preset error requirements, collecting new data at a fixed period to update and iterate the tree species knowledge graph and tree proxy model parameters to continuously improve the prediction accuracy.

[0130] S6 specifically refers to:

[0131] S61. Verify the characteristic parameters of tree species. Select no less than 30 typical hidden tree obstacles on site, measure characteristic parameters such as bending strength and tree resistivity, retrieve the corresponding parameters in the tree species knowledge graph, and calculate the relative error. The relative error is the absolute value of the difference between the measured parameter and the graph parameter divided by the measured parameter and then multiplied by 100%. The relative error is required to be no more than 5%. If it exceeds this range, the measured parameters are added to the knowledge graph and the corresponding node attributes are corrected.

[0132] S62. Verify the effect of multi-source data fusion. For the same transmission line corridor, prediction is carried out using point cloud data only and multi-source fusion data respectively. The multi-source fusion data includes point cloud data, meteorological data and tree species knowledge graph to obtain two sets of prediction results. Based on the actual on-site cleanup verification results, the prediction errors of the two sets are calculated respectively, and then the error reduction rate is calculated. The error reduction rate is the absolute value of the difference between the two sets of prediction errors divided by the prediction error of point cloud data only and then multiplied by 100%. The error reduction rate is required to be no less than 30%. If it is not met, the data preprocessing algorithm or tree proxy model parameters are backtracked and optimized.

[0133] S63. Set a fixed iteration cycle of six months, collect newly added tree species test data, operation and maintenance cleanup feedback data, meteorological impact parameter change data and prediction deviation data within the cycle, supplement new data according to the four-layer structure of the knowledge graph, and update node attributes and relationships.

[0134] S64. Gradient descent is used to correct the parameters of the tree surrogate model. The parameter correction logic is that the new parameter is equal to the old parameter minus the product of the learning rate and the gradient of the loss function. The learning rate is 0.01, and the loss function is the mean square error, which is the sum of the squares of the differences between the predicted value and the actual value divided by the number of samples. The iterated knowledge graph and the model are re-integrated into the evaluation platform to complete the full-process parameter update to improve the prediction accuracy.

[0135] At least 30 typical trees were selected for field testing to ensure statistical representativeness of the sample. The 5% relative error threshold met the engineering accuracy requirements, achieving accurate calibration of the knowledge graph parameters. The multi-source data fusion effect was verified through comparative experiments, quantifying the advantages of the fusion scheme compared to single data. A 30% error reduction rate ensured the value of the fusion. The six-month iteration cycle aligned with the tree growth pattern and the pace of maintenance data accumulation. The gradient descent method guided parameter optimization through the gradient of the loss function. The mean squared error effectively measured the prediction deviation. Continuous iteration enabled the knowledge graph and model to adapt to environmental changes and maintain long-term prediction accuracy.

[0136] As a preferred embodiment of the present invention, the formula for dynamically adjusting the original predicted growth height in S52 is as follows:

[0137] ;in, To correct the growth height, This is the originally predicted growth height. For meteorological correction weights, This is a meteorological influence function;

[0138] The formula is as follows:

[0139]

[0140] in, It is the average temperature extracted from forecast meteorological data for the next six months. This is an extreme wind speed.

[0141] The meteorological correction weight of 0.02 balances the influence of meteorological factors and the inherent growth characteristics of tree species on growth height, avoiding excessive deviation of meteorological correction from the basic growth law. In the meteorological influence function, 15℃ is used as the suitable growth baseline temperature. The linear coefficient quantifies the promoting or inhibiting effect of temperature on growth, and extreme wind speed reflects its inhibitory effect on tree growth through a negative coefficient. The overall formula realizes the quantitative modeling of the influence of meteorological factors on growth height, making the correction results more consistent with the actual growth scenario.

[0142] Example 2:

[0143] This invention also provides a transmission line corridor clearing range prediction system that integrates tree species knowledge graph, including: a data acquisition and preprocessing module, a tree species knowledge graph construction and proxy model development module, a tree obstacle point cloud accurate extraction module, a safety distance calculation and hazard classification module, and a clearing planning and visualization module;

[0144] The data acquisition and preprocessing module is used to collect point cloud data of the power transmission line corridor by using a drone equipped with a lidar, and to control the supporting equipment to simultaneously collect meteorological, vegetation mechanism and electrical environment data. All data are stored in association with tower number and collection timestamp. The point cloud data is aligned with the three types of time-series data (meteorological, vegetation mechanism and electrical environment) through a dynamic time warping algorithm. The missing areas of the point cloud data are filled in with weighted data and the distorted point cloud data is removed.

[0145] The tree species knowledge graph construction and proxy model development module is used to test the biomechanical and electrical characteristics of major tree species in the transmission line corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database; based on the tree species knowledge graph, a configurable tree proxy model is developed, and the proxy model is integrated into the transmission line tree obstacle risk assessment platform to output tree risk and growth prediction parameters through the model.

[0146] The tree obstacle point cloud precise extraction module is used to construct an attention mechanism point cloud classification model, which is used to locate high-risk areas of tree obstacles around the guide wire; the point cloud image features are combined with the tree species knowledge graph to match tree species categories, and interference items are removed to complete the precise classification of tree obstacle point clouds;

[0147] The safety distance calculation and hazard classification module is used to calculate the safety distance correction coefficient based on the tree resistivity to obtain the final safety distance between the conductor and the tree barrier; to establish a tree barrier hazard classification system that includes biomechanical and electrical risk dimensions, and to upgrade the tree barrier hazard ledger;

[0148] The cleaning planning and visualization module is used to incorporate meteorological data to correct tree growth parameters, combine the probability of lodging to predict future risks of tree obstacles, adjust the core cleaning area based on the biomechanical characteristics of tree species, and adjust the inspection cycle according to the electrical risk level. The module also dynamically displays tree obstacle growth prediction data and electrical risk heat map in the power transmission line tree obstacle risk assessment platform.

[0149] The following are some applications of this invention:

[0150] 1. Acquisition and preprocessing of lidar point cloud and multi-source data

[0151] Unmanned aerial vehicles (UAVs) equipped with lidar devices conduct operations along power transmission line corridors to control flight path altitude. Ensure the density of the collected point cloud The collected point cloud dataset is denoted as ,in In three-dimensional coordinates, The data is collected using timestamps in the format YYYY-MM-DDHH:MM:SS, and stored according to the LAS1.4 standard for easy subsequent processing.

[0152] Independently separate from drone lidar data collection, multi-source auxiliary data is collected synchronously through dedicated equipment: meteorological data. Data was collected by automatic weather stations deployed every 500 meters along the corridor and recorded as follows:

[0153] ,in For ambient temperature, Relative humidity, Instantaneous wind speed; vegetation mechanism data Obtained through on-site sampling combined with laboratory testing, denoted as ,in As a tree species, At its current height, Tree age Diameter at breast height; electrical environment data Data were collected on-site using a portable electrical parameter tester and recorded as follows: ,in This is the operating voltage of the line. For operating current, For environmental resistivity. All collected data are categorized as follows: The rules are uniformly indexed to build related indexes. It achieves a one-to-one mapping between point cloud data and multi-source auxiliary data. The data storage adopts a distributed database architecture of MySQL and MongoDB. MySQL is used to store structured parameters such as tower number and temperature, while MongoDB is used to store unstructured point cloud data, supporting fast data tracing and retrieval.

[0154] Point cloud dataset timestamp sequence Based on meteorological data Vegetation mechanism data Electrical environment data The three types of time-series data were aligned, and the Dynamic Time Warping (DTW) algorithm was used to eliminate time differences caused by different devices: first, the distance matrix was calculated, and then the meteorological data was aligned. For example, its timestamp sequence is Construct the Euclidean distance matrix Matrix elements This is used to measure the time difference between two points; then, a cumulative distance matrix is ​​constructed. According to the formula Perform iterative calculations, with boundary conditions set as follows: ; then from Backtracking to obtain the optimal alignment path The time-series data is linearly interpolated or downsampled according to the path to obtain the aligned dataset. Finally, calculate the alignment similarity. ,Require If this requirement is not met, the alignment path will be readjusted to ensure the reliability of data fusion.

[0155] Addressing point cloud gaps caused by complex terrain such as valleys and steep slopes. The attention-based weighted completion algorithm is used for processing: first, the effective point cloud of the neighborhood of the missing region is divided. Calculate the attention weights of each neighboring point. ,in The Euclidean distance between two points is then calculated using the formula. Calculate the completed coordinates and generate the completed dataset. .

[0156] The Isolation Forest algorithm is used to detect distorted points in a point cloud, constructing 100 isolated trees and calculating anomaly scores for each point cloud. ,in For point Average path length across all isolated trees Sample size The average path length constant, ,in For the harmonic function; set the anomaly scoring threshold. Remove The severe weather distortion points are used to output the final preprocessed point cloud dataset. .

[0157] 2. Construction of Tree Species Knowledge Graph and Development of Tree Proxy Model

[0158] Select the main tree species in the transmission line corridor This means covering over 95% of common tree species and conducting joint laboratory and field tests.

[0159] The bending strength of tree species was determined using a universal testing machine. ,tensile strength Recorded as ; The resistivity of trees was measured using a high resistivity meter At relative humidity A gradient is set every 10% within the range, and the breakdown voltage is tested using a withstand voltage testing machine. The humidity-resistivity-breakdown characteristic curve was fitted using the least squares method, and the fitting formula is as follows: ,in These are the fitting coefficients, obtained from the regression of experimental data, and the required goodness of fit is... Integrate the above test data to form a basic database. This provides data support for the construction of a tree species knowledge graph.

[0160] Constructing a multi-dimensional tree species knowledge graph , where nodes Includes tree species nodes, parameter nodes, and environment nodes; relationships This refers to the relationships between nodes, such as poplar trees having bending strength; attributes. These are the quantization parameters corresponding to the nodes.

[0161] The knowledge graph is structured in four layers: basic attributes, biomechanics, electrical properties, and growth properties. The basic attributes layer includes tree species, age, diameter at breast height (DBH), and growth region; the biomechanics layer includes bending strength. ,tensile strength and the corresponding reference values; the electrical characteristics layer includes the tree resistivity. Breakdown voltage relative humidity Related data; the growth characteristic layer includes annual growth, growth cycle, and crown width; the knowledge graph is stored using the Neo4j graph database, and a dedicated query interface is built, categorized by tree species. and hierarchy As input, output parameters for the corresponding dimension. It supports quickly retrieving parameters of the corresponding dimension by tree type, and the interface response time is fast. .

[0162] Based on tree species knowledge graph Data development configurable tree agent model The model was integrated into the power transmission line tree obstacle risk assessment platform, with the model based on tree species. relative humidity ,temperature Tree age Given the input conditions, the output consists of three core prediction parameters:

[0163] Growth height in the next 6 months: ,in The current height of the tree; The growth coefficient is retrieved from the growth characteristic layer of the knowledge graph for different tree species. Different values, such as poplar ; , The average humidity and average temperature for the next 6 months;

[0164] Probability of tree falling: ,in This is a terrain correction factor for mountainous areas. Plains , The baseline value for the bending strength of this tree species was retrieved from the knowledge graph. The maximum wind speed along the corridor over the next 6 months is predicted from meteorological data.

[0165] Lightning breakdown risk: ,in This is the line voltage correction factor. hour ,otherwise , The resistivity reference value for this tree species was retrieved from the knowledge graph. This refers to the line operating voltage;

[0166] Model output set The data is simultaneously stored in the power transmission line tree obstacle risk assessment platform to provide data for subsequent tree obstacle risk calculations.

[0167] 3. Precise extraction of attention mechanism obstacle point cloud

[0168] From the preprocessed point cloud dataset Extract the point cloud of the conductor and construct a 3D contour model of the conductor. ,in , The elevations of the lowest and highest points of the conductor are identified, and high-risk areas around the conductor are also determined. ,in The safety distance for conductor foundations is determined according to the line voltage level, such as for a 220kV line. .

[0169] Constructing an attention-based point cloud classification model The preprocessed point cloud dataset After inputting the data into the model, the attention weights for each point cloud are calculated. ,in The feature matching function focuses on high-risk areas. Elevation and density characteristics of interior point clouds;

[0170] Set weight threshold The weight threshold can be dynamically adjusted, and point clouds with attention weights greater than or equal to the threshold in high-risk areas are output as candidate tree obstacle point clouds. Simultaneously calculate recall rate ,in To correctly identify the number of tree obstacle point clouds, To determine the number of tree obstruction point clouds that were missed in the initial assessment, the following is required: If the conditions are not met, the weight threshold will be adjusted and the classification will be redefined.

[0171] From Tree Barrier Point Cloud Candidate Set Extract image and morphological features to form a feature set. ,in For the crown, For the height of the tree, Diameter at breast height;

[0172] Retrieve the morphological parameter sets corresponding to all tree species in the tree species knowledge graph through the query interface. This includes the standard crown width, standard tree height, and standard diameter at breast height for each tree species, and calculates the similarity between the feature set and the morphological parameter set for each tree species. Set a similarity threshold Point clouds with similarity at or above the threshold are retained and matched with the corresponding tree species. Interference items such as weeds and rocks with similarity below the threshold are removed, and the final output is an accurate tree obstacle point cloud dataset. ,in This represents the actual distance from the tree barrier to the guide wire.

[0173] 4. Comprehensive calculation of safe distance and risk between power lines and tree obstacles

[0174] Obtain basic safety distances from line design specifications. From tree species knowledge graph Retrieve the resistivity of the corresponding tree species From the aligned meteorological data Get the current relative humidity .

[0175] According to the formula Calculate the safety distance correction factor, where The weighting coefficients were determined through engineering experiments. , ,and ; The resistivity benchmark value for this tree species is derived from the knowledge graph. Retrieve from the middle; safety distance correction factor With relative humidity Increase the resistivity of trees Decrease and increase.

[0176] According to the formula Calculate the final safe distance and simultaneously assess potential hazards. If the actual distance to the tree obstruction... Then the tree barrier will be included in the list of potential tree barriers. .

[0177] A three-level, four-element tree barrier hazard classification system is constructed, with the four elements being spatial distance, biomechanical, electrical risk, and growth characteristics.

[0178] Spatial distance elements: ;

[0179] Biomechanical elements: ;

[0180] Electrical risk factors: ;

[0181] Growth characteristics: ;

[0182] The scoring range for each element is [missing information]. The higher the score, the greater the risk. The actual distance to the tree barrier. For the final safe distance, The probability of a tree falling over. To mitigate the risk of lightning strikes, This indicates the height to be measured over the next six months.

[0183] According to the formula Calculate the overall score, where the weights are... , , , It was determined through expert scoring combined with engineering verification, and .

[0184] Hazard levels are classified according to a comprehensive score. Classified as Level I major hidden danger, Classified as Level II general hazard, It is classified as a Level III minor hazard.

[0185] Upgrade the tree obstacle hazard log, adding fields for tree species, resistivity, probability of lodging, breakdown risk, overall level, and corrected safe distance, and inputting the set of hazard tree obstacles. All parameters are used to form a standardized hidden danger database.

[0186] 5. Prediction of tree growth and detailed planning for clearing tree obstacles

[0187] Incorporating forecast meteorological data for the next 6 months Extract the average temperature from it and extreme wind speeds The tree's height for the next 6 months is adjusted using the following formula:

[0188]

[0189] in Meteorological adjustment weights, meteorological influence function Extreme weather can inhibit tree growth, therefore The result is negative;

[0190] Calculate the change in tree height over the next 6 months This allows us to obtain the future distance from the tree barrier to the conductor. ;

[0191] According to the formula Calculate the overall future risks and quantify the risk value of future tree-related hazards.

[0192] The core cleanup area is delineated according to the level of tree hazard: the cleanup area for Level I major hazard tree obstacles is as follows: Extending outwards from the guide line, the clearing area for Class II general hidden danger tree obstacles is defined as follows: The area for clearing Level III minor hazard tree obstructions is defined as follows: This ensures coverage of areas at risk of future growth.

[0193] Based on electrical risk factors Graded inspection cycles and electrical risk factor scoring Divided into high-risk categories, inspection cycle ; Divided into medium risk, inspection cycle ; Divided into low-risk categories, inspection cycle If the probability of trees falling down The inspection cycle is halved.

[0194] The core cleaning scope, the set inspection cycle, the tree growth prediction curve, and the risk heat map are synchronized to the power transmission line tree obstacle risk assessment platform. In the risk heat map, Level I hazards are marked in red, Level II hazards are marked in yellow, and Level III hazards are marked in green, which supports operation and maintenance personnel to view and dispatch intuitively.

[0195] 5. Solution Validation and Knowledge Graph Iteration

[0196] At least 30 typical trees with potential hazards were selected on-site, and their bending strength was measured. Tree resistivity Characteristic parameters, from tree species knowledge graph Retrieve corresponding parameters , According to the relative error formula Calculate the relative error, requiring If the values ​​exceed this range, the measured parameters will be added to the knowledge graph, and the corresponding node attributes will be corrected.

[0197] For the same transmission line corridor, predictions were made using point cloud data alone and multi-source fusion data, which included point cloud data, meteorological data, and tree species knowledge graphs. Two sets of prediction results were obtained. , Verification results based on actual on-site cleanup. Using this as a baseline, calculate the prediction errors for the two sets of data respectively. , According to the formula Calculate the error reduction rate, requiring If the conditions are not met, the algorithm will backtrack to the first step of optimizing the data preprocessing algorithm, or to the second step of correcting the tree proxy model parameters.

[0198] A fixed iteration cycle of 6 months is set, during which newly added tree species test data, operation and maintenance cleanup feedback data, meteorological impact parameter change data, and prediction deviation data are collected; the new data is then supplemented into the tree species knowledge graph according to its four-layer structure. In this process, new tree species nodes are added, parameter attributes are updated, and node relationships are optimized. Gradient descent is used to correct the tree proxy model parameters; the parameter correction logic is as follows:

[0199]

[0200] in For model coefficients, including wait; The learning rate; The gradient of the loss function is given, where the loss function uses the mean squared error. ,in For predicted values, This is the actual value. The number of samples;

[0201] The iterative knowledge graph and tree proxy model were reintegrated into the transmission line tree obstacle risk assessment platform to complete the full-process parameter update and continuously improve prediction accuracy.

[0202] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting a cleaning range of a transmission line corridor by fusing tree species knowledge graph, characterized in that, Includes the following steps: S1. The UAV is equipped with a lidar to collect point cloud data of the power transmission line corridor. The supporting equipment simultaneously collects meteorological, vegetation mechanism and electrical environment data. All data are stored in association with the tower number and the collection timestamp. The point cloud data is aligned with the three types of time-series data, namely meteorological data, vegetation mechanism data and electrical environment data, through a dynamic time warping algorithm. The missing areas of the point cloud data are weighted and filled, and the distorted point cloud data is removed. S2. Test the biomechanical and electrical characteristics of the main tree species in the test corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database, develop a configurable tree proxy model based on the graph, integrate it into the power transmission line tree obstacle risk assessment platform, and output tree risk and growth prediction parameters. S3. Construct an attention mechanism point cloud classification model to locate high-risk areas of tree obstacles around the guide line. Combine point cloud image features with tree species knowledge graph to match tree species categories and remove interference items to complete accurate classification of tree obstacle point clouds. S4. Calculate the safety distance correction coefficient based on tree resistivity to obtain the final safe distance between the conductor and the tree barrier; establish a tree barrier hazard classification system including biomechanical and electrical risk dimensions, and upgrade the tree barrier hazard ledger; S5. Introduce meteorological data to correct tree growth parameters, combine the lodging probability output by the tree proxy model to predict the future risk of tree obstacles, adjust the core clearing area according to the biomechanical characteristics of tree species, adjust the inspection cycle according to the electrical risk level, and dynamically display the tree obstacle growth prediction and electrical risk heat map in the tree obstacle risk assessment platform. 2.The method of claim 1, wherein, S1 includes the following specific steps: S11. The UAV, equipped with lidar equipment, will conduct operations on the power transmission line corridor, controlling the flight path altitude between 80 and 130 meters, ensuring that the collected point cloud density is not less than 20 points / m², and the collected point cloud data includes three-dimensional coordinates and collection timestamps. S12. Multi-source auxiliary data is collected independently of the UAV's lidar using supporting equipment. Meteorological data is collected from automatic weather stations deployed every 500m along the power transmission line corridor, including ambient temperature, relative humidity, instantaneous wind speed, and corresponding collection timestamps. Vegetation mechanism data is obtained through on-site sampling combined with laboratory testing, including tree species, current growth height, tree age, diameter at breast height, and corresponding collection timestamps. Electrical environment data is collected on-site using a portable electrical parameter detector, including line operating voltage, operating current, environmental resistivity, and corresponding collection timestamps. S13. Using the rule of combining tower number and collection timestamp as a unified identifier, an associated index is built for all collected data to achieve a one-to-one mapping between point cloud data and multi-source auxiliary data. All data is stored in a distributed database composed of MySQL and MongoDB databases. Structured parameters such as tower number and temperature are stored in the MySQL database, while unstructured point cloud data is stored in the MongoDB database, supporting fast data tracing and retrieval. 3.The method of claim 2, wherein, S1 also includes the following steps: S14. Based on the timestamp sequence of point cloud data, align the three types of time-series data: meteorological data, vegetation mechanism data, and electrical environment data. Eliminate the time difference caused by different devices. By backtracking the optimal alignment path, perform linear interpolation or downsampling on the time-series data to keep the timestamps of various types of data consistent, and obtain the aligned dataset. S15. Verify the data alignment effect and calculate the alignment similarity. The similarity should be no less than 95%. If the requirement is not met, readjust the alignment path. S16. For point cloud missing areas caused by complex terrain such as valleys and steep slopes, an attention mechanism weighted completion algorithm is adopted. First, the effective point cloud set around the missing area is divided. Then, weights are assigned according to the distance between each effective point cloud and the missing area. The weighted effective point cloud data is used to complete the coordinate information of the missing area and generate the completed point cloud dataset. S17. Use the isolated forest algorithm to detect distorted points in the point cloud, construct 100 isolated trees, calculate the anomaly score for each point cloud, set the anomaly score threshold to 0.7, remove distorted points caused by severe weather with anomaly scores not lower than the threshold, and output the final preprocessed point cloud dataset.

4. The method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs according to claim 1, characterized in that, S2 includes the following specific steps: S21. Select the main tree species that cover more than 95% of the common tree species in the transmission line corridor, and carry out biomechanical and electrical characteristic tests in combination in the laboratory and on site. Measure the bending strength, tensile strength, tree resistivity, and breakdown voltage under different humidity gradients of the tree species. Integrate all test data to build a basic database to provide data support for the construction of a tree species knowledge graph. S22. Construct a multi-dimensional tree species knowledge graph containing nodes, relationships, and attributes. The graph is divided into four layers: basic attribute layer, biomechanical layer, electrical characteristic layer, and growth characteristic layer. The Neo4j graph database is used to store the knowledge graph. At the same time, a dedicated query interface is built to support the quick retrieval of parameters of the corresponding dimension by tree species. S23. Develop a configurable tree proxy model based on tree species knowledge graph data, integrate the model into the power transmission line tree obstacle risk assessment platform, take tree species, relative humidity, temperature and tree age as model input conditions, output three core prediction parameters: tree growth height in the next six months, tree lodging probability and lightning breakdown risk, and store all output parameters synchronously in the assessment platform to provide data basis for subsequent tree obstacle risk calculation.

5. The method for predicting the clearance range of transmission line corridors by integrating tree species knowledge graphs according to claim 1, characterized in that, Step S3 includes the following specific steps: S31. Extract conductor point clouds from the preprocessed point cloud dataset and construct a three-dimensional contour model of the conductor. The model defines the conductor range by the lowest elevation and the highest elevation of the conductor. At the same time, it identifies high-risk areas around the conductor. High-risk areas are point cloud areas whose distance from the three-dimensional contour model of the conductor is no greater than the basic safety distance. The basic safety distance is determined according to the line voltage level. S32. Construct an attention mechanism point cloud classification model. Input the preprocessed point cloud dataset into the model and calculate the attention weight of each point cloud. The weight calculation logic is the ratio of the index of the mapping result of the feature matching function on the elevation and density features of the point cloud in the high-risk area to the sum of the indices of the corresponding mapping results of all point clouds. Set a weight threshold, output the point cloud with attention weight greater than or equal to the weight threshold in the high-risk area as the candidate set of tree obstacle point cloud, and calculate the recall rate at the same time. The recall rate is the ratio of the number of correctly identified tree obstacle point clouds to the sum of the number of correctly identified tree obstacle point clouds and the number of missed tree obstacle point clouds. The recall rate is required to be no less than 98%. If it is not met, the weight threshold is adjusted and the classification is carried out. S33. Extract image and morphological features from the candidate set of tree barrier point clouds to form a feature set. The feature set includes three parameters: crown width, tree height, and diameter at breast height. S34. Retrieve the morphological parameter set corresponding to all tree species in the tree species knowledge graph through the query interface. The parameter set includes the standard crown width, standard tree height and standard diameter at breast height of each tree species. Calculate the similarity between the feature set and the morphological parameter set of each tree species. The similarity calculation logic is 1 minus the ratio of the Euclidean distance between the two to the maximum Euclidean distance. Set the similarity threshold. S35. Retain point clouds with similarity not lower than the similarity threshold and match them with the corresponding tree species. Remove interference items such as weeds and rocks with similarity lower than the similarity threshold. Finally, output the accurate tree obstacle point cloud dataset. The accurate tree obstacle point cloud dataset includes tree species, morphological feature set and actual distance from tree obstacle to guide line.

6. The method of claim 1, wherein the method further comprises: The specific calculation of the safety distance correction factor in S4 is as follows: Obtain basic safety distances from line design specifications. From tree species knowledge graph Retrieve the resistivity of the corresponding tree species From the aligned meteorological data Get the current relative humidity ; The safety distance correction factor is calculated using the following formula. , ,in These are the weighting coefficients. , The base value for tree species resistivity, with a safety distance correction factor. With relative humidity Increase resistivity Decrease and increase; Final safety distance This is calculated as follows: Simultaneously determine potential hazards: if the actual distance of the tree obstruction... Less than the final safe distance If so, it will be included in the list of hidden danger trees; The tree obstacle hazard classification system described in S4 is specifically established as follows: A three-level, four-element hazard classification system is constructed. The four elements are spatial distance, biomechanical, electrical risk, and growth characteristics. Among them, the spatial distance element is... Biomechanical elements are Electrical risk factors are Growth characteristics are Scoring range for each element The higher the score, the greater the risk; The actual distance to the tree barrier. For the final safe distance, The probability of a tree falling over. To mitigate the risk of lightning strikes, For the next six months' growth height; The overall score is calculated using the following formula: Weight , , , ; Hazard levels are classified according to a comprehensive score. It is classified as a Level I major hidden danger, with a comprehensive score of The hazard is classified as Level II general hazard, with a comprehensive score of [missing information]. The hazard is classified as Level III, a minor safety concern. The upgrade of the tree obstacle hazard log in S4 is as follows: Add fields for tree species, resistivity, probability of lodging, breakdown risk, comprehensive level, and corrected safe distance, and enter all parameters of the hazard tree obstacle set to form a standardized hazard database.

7. The method of claim 1, wherein the method further comprises: The specific steps of S5 include: S51. Introduce forecast meteorological data for the next six months, extract average temperature and extreme wind speed, and correct the tree growth height for the next six months. S52. Combining the inhibitory effects of average temperature and extreme wind speed on tree growth in future meteorological conditions, the original predicted growth height is dynamically adjusted by setting meteorological correction weights and meteorological influence functions to obtain a corrected growth height that better reflects the actual growth situation. S53. Calculate the change in tree height over the next six months. This change is the difference between the corrected growth height and the current height. Then, obtain the future distance from the tree barrier to the guide wire. This distance is the sum of the actual distance of the tree barrier and the change in tree height. S54. Calculate the future comprehensive risk. Based on the current tree barrier comprehensive score, incorporate the influence weights of tree fall probability and lightning strike risk, quantify and superimpose the effects of the two core risks on future hidden dangers, and obtain the future tree barrier comprehensive hidden danger risk value. S55. Delineate the core clearing area according to the level of tree obstacle hazard. The clearing area for Level I major hazard tree obstacles is the sum of the final safe distance and the corrected growth height, extending outward from the guide line. The clearing area for Level II general hazard tree obstacles is the final safe distance. The clearing area for Level III minor hazard tree obstacles is the future distance from the tree obstacle to the guide line, ensuring coverage of areas with future growth risks. S56. Set the inspection cycle according to the classification of electrical risk factors. If the electrical risk factor score is not lower than eight points, it is high risk and the inspection cycle is seven days; if the electrical risk factor score is not lower than five points but lower than eight points, it is medium risk and the inspection cycle is thirty days; if the electrical risk factor score is lower than five points, it is low risk and the inspection cycle is ninety days. If the probability of tree falling is not lower than 0.7, the inspection cycle is halved. S57. Synchronize the core cleaning scope, the set inspection cycle, the tree growth prediction curve, and the risk heat map to the tree obstacle risk assessment platform. In the risk heat map, Level I hazards are marked in red, Level II hazards are marked in yellow, and Level III hazards are marked in green, which supports operation and maintenance personnel to view and schedule intuitively. 8.The method of claim 1, wherein, It also includes S6, verifying the tree species characteristic parameters and the fusion effect of multi-source data. After meeting the preset error requirements, new data is collected at fixed intervals to update and iterate the tree species knowledge graph and tree proxy model parameters, continuously improving the prediction accuracy. S6 specifically refers to: S61. Verify the characteristic parameters of tree species. Select no less than 30 typical hidden tree obstacles on site, measure characteristic parameters such as bending strength and tree resistivity, retrieve the corresponding parameters in the tree species knowledge graph, and calculate the relative error. The relative error is the absolute value of the difference between the measured parameter and the graph parameter divided by the measured parameter and then multiplied by 100%. The relative error is required to be no more than 5%. If it exceeds this range, the measured parameters are added to the knowledge graph and the corresponding node attributes are corrected. S62. Verify the effect of multi-source data fusion. For the same transmission line corridor, prediction is carried out using point cloud data only and multi-source fusion data respectively. The multi-source fusion data includes point cloud data, meteorological data and tree species knowledge graph to obtain two sets of prediction results. Based on the actual on-site cleanup verification results, the prediction errors of the two sets are calculated respectively, and then the error reduction rate is calculated. The error reduction rate is the absolute value of the difference between the two sets of prediction errors divided by the prediction error of point cloud data only and then multiplied by 100%. The error reduction rate is required to be no less than 30%. If it is not met, the data preprocessing algorithm or tree proxy model parameters are backtracked and optimized. S63. Set a fixed iteration cycle of six months, collect newly added tree species test data, operation and maintenance cleanup feedback data, meteorological impact parameter change data and prediction deviation data within the cycle, supplement new data according to the four-layer structure of the knowledge graph, and update node attributes and relationships. S64. Gradient descent is used to correct the parameters of the tree surrogate model. The parameter correction logic is that the new parameter is equal to the old parameter minus the product of the learning rate and the gradient of the loss function. The learning rate is 0.01, and the loss function is the mean square error, which is the sum of the squares of the differences between the predicted value and the actual value divided by the number of samples. The iterated knowledge graph and the model are re-integrated into the evaluation platform to complete the full-process parameter update to improve the prediction accuracy. 9.The method of claim 7, wherein, The formula for dynamically adjusting the original predicted growth height in S52 is as follows: ;in, To correct the growth height, This is the originally predicted growth height. For meteorological correction weights, This is a meteorological influence function; The formula is as follows: in, It is the average temperature extracted from forecast meteorological data for the next six months. This is an extreme wind speed. 10.A system for predicting a cleaning range of a transmission line corridor by fusing tree species knowledge graph, characterized in that, include: The module includes: data acquisition and preprocessing, tree species knowledge graph construction and proxy model development, tree obstacle point cloud precise extraction, safe distance calculation and hazard classification, and cleanup planning and visualization. The data acquisition and preprocessing module is used to collect point cloud data of the power transmission line corridor by using a drone equipped with a lidar, and to control the supporting equipment to simultaneously collect meteorological, vegetation mechanism and electrical environment data. All data are stored in association with tower number and collection timestamp. The point cloud data is aligned with the three types of time-series data (meteorological, vegetation mechanism and electrical environment) through a dynamic time warping algorithm. The missing areas of the point cloud data are filled in with weighted data and the distorted point cloud data is removed. The tree species knowledge graph construction and proxy model development module is used to test the biomechanical and electrical characteristics of major tree species in the transmission line corridor, construct a multi-dimensional tree species knowledge graph and store it in a dedicated database; based on the tree species knowledge graph, a configurable tree proxy model is developed, and the proxy model is integrated into the transmission line tree obstacle risk assessment platform to output tree risk and growth prediction parameters through the model. The tree obstacle point cloud precise extraction module is used to construct an attention mechanism point cloud classification model, which is used to locate high-risk areas of tree obstacles around the guide wire; the point cloud image features are combined with the tree species knowledge graph to match tree species categories, and interference items are removed to complete the precise classification of tree obstacle point clouds; The safety distance calculation and hazard classification module is used to calculate the safety distance correction coefficient based on the tree resistivity to obtain the final safety distance between the conductor and the tree barrier; to establish a tree barrier hazard classification system that includes biomechanical and electrical risk dimensions, and to upgrade the tree barrier hazard ledger; The cleaning planning and visualization module is used to incorporate meteorological data to correct tree growth parameters, combine the probability of lodging to predict future risks of tree obstacles, adjust the scope of the core cleaning area according to the biomechanical characteristics of tree species, and adjust the inspection cycle according to the electrical risk level. The power transmission line tree obstacle risk assessment platform dynamically displays tree obstacle growth prediction data and electrical risk heat map.

Images