A power transmission line environment risk identification method based on multi-modal internet of things sensing

Abstract

The present application relates to a kind of power transmission line environmental risk identification method based on multi-modal internet of things sensing, it is related to power transmission line state monitoring and risk identification technical field.It includes the multi-modal internet of things sensing data of each line section is collected, and according to line section and same collection identification is merged, generates multi-modal observation set;According to the multi-modal observation set, the quality parameters corresponding to each mode of the mode are calculated, and generate a set of trusted weights;According to the set of trusted weights, each mode observation feature in the multi-modal observation set is weighted and fused, generates the initial risk identification result of each line section;According to the line section correlation between each line section, the initial risk identification result is revised, and the environmental risk grade of each line section is output.The present application improves the accuracy, stability and section continuity of power transmission line environmental risk identification.

Description

Technical Field

[0001] This invention belongs to the field of power transmission line condition monitoring and risk identification technology, specifically a method for identifying environmental risks of power transmission lines based on multimodal IoT sensing. Background Technology

[0002] With the development of online monitoring and status awareness technologies for transmission lines, a multi-source sensing system consisting of visible light imaging, infrared imaging, micro-meteorological monitoring, vibration monitoring, and tilt monitoring has gradually been formed along the lines. Related technologies have evolved from single-point acquisition and single-parameter alarms to multi-source access, edge analysis, and tiered early warning. While existing solutions can acquire information such as temperature, humidity, wind speed and direction, images, and the condition of the transmission line itself, they still commonly employ fixed-weight fusion, independent judgment of single sections, or single-modal-dominated judgment methods in scenarios involving complex weather, obstruction interference, local inaccuracies, and the coexistence of section propagation effects. This results in insufficient stability of the identification results when different modal quality fluctuates, and makes it difficult to reflect the risk transmission relationship between adjacent line sections, thus affecting the accuracy and continuity of environmental risk level determination.

[0003] CN114184232A discloses a multi-parameter integrated monitoring system for transmission lines. This scheme accesses transmission line monitoring data through a data acquisition module, edge terminal, and remote management terminal, and performs hierarchical analysis of abnormal data, which can improve monitoring and analysis efficiency. However, its focus is on multi-parameter integrated access and abnormal data processing, and it does not establish a correspondence between modal quality parameters and reliable weights based on the differences in data quality of different modes within the same line section. Therefore, it is difficult to differentiate and weight the multi-modal observation results under conditions of image occlusion, meteorological changes, or local sensor drift. CN113162232A discloses a risk assessment and defense decision-making system and method for transmission line equipment. This scheme accesses meteorological environmental data and constructs a risk assessment model, which can provide early warning and operation and maintenance decision support. However, its focus is on the risk assessment platform and defense decision application, and it does not establish a dynamic correction mechanism for fluctuations in the observation quality of each mode, nor does it perform hierarchical correction for the spatial correlation between adjacent line sections.

[0004] CN118627021A discloses a lightweight method for identifying potential hazards in power transmission terminals based on multimodal fusion. This scheme utilizes image data and environmental information for multimodal fusion and forms a lightweight hazard detection model through distillation and quantization, which can improve the efficiency of terminal inspection. However, it mainly focuses on the identification of potential hazards in power transmission terminals and the lightweight deployment of models, with an emphasis on image-text multimodal fusion. It does not establish a processing chain of "modal quality parameters - credible weight set - section correlation correction" for environmental risk identification at the line section level. As shown in CN117854235A, existing meteorological disaster forecasting and early warning technologies for dense power transmission channels can collect micro-weather station, radar, and image data to identify icing and galloping. However, this type of scheme is more biased towards disaster early warning and image identification, and still lacks a unified processing framework for simultaneous data acquisition merging, unified multimodal quality evaluation, and risk transmission correction between adjacent line sections.

[0005] Given the problems of inconsistent multimodal observation quality, insufficient adaptability of fixed fusion methods, and inadequate utilization of risk transmission relationships between sections in existing transmission line environmental monitoring technologies, this invention is proposed. Therefore, the problem this invention aims to solve is how to merge multimodal IoT sensor data collected in the same batch into line sections during the transmission line environmental risk identification process, quantify the observation quality of each mode and generate reliable weights, and then correct the initial identification results by combining the correlation relationships between line sections, thereby obtaining a more stable environmental risk level. As mentioned in the background section, existing technologies focus on multi-parameter acquisition, platform early warning, or terminal hidden danger identification. This invention, however, solves the problems of accuracy in identifying transmission line environmental risk levels and determining section continuity through multimodal observation set construction, modal quality parameter calculation, reliable weight weight fusion, and line section correlation correction. It belongs to the field of transmission line condition monitoring and risk identification technology. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a method for identifying environmental risks in power transmission lines based on multimodal IoT sensing, comprising: Multimodal IoT sensor data from each line section are collected and merged according to line section and the same collection identifier to generate a multimodal observation set; Based on the multimodal observation set, calculate the modal quality parameters corresponding to each mode and generate a set of reliable weights; Based on the set of credible weights, the features of each modality observation in the multimodal observation set are weighted and fused to generate the initial risk identification results for each line segment; The initial risk identification results are corrected based on the correlation between different line sections, and the environmental risk level of each line section is output.

[0008] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the generation of the multimodal observation set includes: Visible light acquisition devices, infrared acquisition devices, meteorological acquisition devices, conductor vibration acquisition devices, and tower tilt angle acquisition devices are deployed in each section of the line to collect visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data for the corresponding line sections, respectively. Based on the tower number, span number, and geographical location of each line section, the visible light image data, the infrared image data, the meteorological monitoring data, the conductor vibration monitoring data, and the tower tilt angle monitoring data are processed to correspond to the line section, generating zoned observation data corresponding to each line section; Based on the acquisition time corresponding to the partitioned observation data, perform same-time correspondence processing to generate same-time observation data groups for each line segment; The same observation data sets corresponding to each line segment are merged to generate a multimodal observation set corresponding to each line segment.

[0009] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the generation of a reliable weight set includes: Visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data are separated from the multimodal observation sets corresponding to each line section, and the quality detection results of the corresponding modes are generated respectively. Based on the quality detection results, the modal quality parameters corresponding to each modality are calculated. The modal quality parameters include image sharpness, image occlusion rate, temperature distribution continuity, data integrity, signal fluctuation stability, and acquisition time difference. The modal quality parameters corresponding to each modality are normalized, and a set of reliable weights is generated.

[0010] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the normalization processing of the modal quality parameters corresponding to each mode and the generation of a reliable weight set include: The modal quality parameters corresponding to each mode within the same line segment are subjected to directional unification processing. Among them, image clarity, temperature distribution continuity, data integrity and signal fluctuation stability retain the original direction of change, while image occlusion rate and acquisition time difference are converted into inverse quality parameters. The modal quality parameters corresponding to each mode within the same line section are compressed according to the minimum and maximum values ​​of the corresponding parameters to generate standard quality values ​​between zero and one. Based on the proportion of the standard quality value corresponding to each mode to the sum of all standard quality values ​​in the same line segment, a reliable weight value corresponding to each mode is generated, and the reliable weight values ​​corresponding to each mode are merged into a reliable weight value set.

[0011] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the generation of initial risk identification results for each line section includes: Visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features are extracted from the multimodal observation sets corresponding to each line section to generate observation features corresponding to each mode; Based on the credible weights corresponding to each mode, the observation features corresponding to each mode are weighted and processed, and then combined according to the credible weights and observation features corresponding to each mode within the same line segment to generate fused observation features; The fused observation features corresponding to each line segment are input into the risk identification model to generate the initial risk identification results for each line segment.

[0012] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the step of inputting the fused observation features corresponding to each line segment into the risk identification model to generate initial risk identification results corresponding to each line segment includes: The fused observation features corresponding to each line segment are input into the risk identification model, and the risk identification model generates the risk identification score corresponding to each line segment. Based on the risk identification scores corresponding to each line segment, a risk level label is generated for each line segment. The risk level label corresponding to each line segment is determined as the initial risk identification result for each line segment.

[0013] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the step of correcting the initial risk identification result according to the correlation between each line segment and outputting the environmental risk level of each line segment includes: Based on the tower adjacency relationship, span continuity relationship and wind direction transmission relationship between each line segment, the corresponding line segment association relationship is generated for each line segment; The initial risk identification results corresponding to each line segment are correlated with the initial risk identification results corresponding to adjacent line segments according to the line segment correlation relationship to generate the correlation correction results corresponding to each line segment; Based on the correlation correction results corresponding to each line segment, the initial risk identification results corresponding to each line segment are adjusted to generate the environmental risk level corresponding to each line segment.

[0014] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the generation of line segment association relationships corresponding to each line segment includes: Based on the tower number and span number corresponding to each line segment, determine the tower adjacency relationship and span continuity relationship between each line segment, and generate a set of adjacent line segments; Based on the wind direction monitoring data corresponding to each line section, determine the wind direction transmission relationship between adjacent line sections; The tower adjacency relationship, the span continuity relationship, and the wind direction transmission relationship are merged to generate the line segment association relationship corresponding to each line segment.

[0015] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the generation of correlation correction results corresponding to each line section includes: Based on the line segment association relationship corresponding to each line segment, extract the initial risk identification result set of adjacent line segments corresponding to the current line segment; The initial risk identification result corresponding to the current line segment is compared with each initial risk identification result in the set of initial risk identification results of the adjacent line segments to generate the risk difference result corresponding to the current line segment. Based on the risk difference results and the correlation between the line segments, the correlation correction results corresponding to the current line segment are generated.

[0016] As a preferred technical solution for a transmission line environmental risk identification method based on multimodal IoT sensing, the step of adjusting the initial risk identification results corresponding to each line segment according to the correlation correction results to generate the environmental risk level corresponding to each line segment includes: Based on the correlation correction results corresponding to each line segment, the risk correction direction corresponding to each line segment is determined; When the risk correction direction is upward, the initial risk identification result corresponding to the current line segment is increased by one level; when the risk correction direction is downward, the initial risk identification result corresponding to the current line segment is decreased by one level; when the risk correction direction is to remain unchanged, the initial risk identification result corresponding to the current line segment is maintained.

[0017] The beneficial effects of this invention are as follows: By merging multimodal IoT sensor data from various line sections into a multimodal observation set, this invention ensures the consistency of observation results from different sources within the same section, reducing interference from asynchronous acquisition and mismatched risk identification. By calculating the modal quality parameters corresponding to each modality and generating a set of reliable weights, it achieves differentiated allocation of low-quality and high-quality modes, improving identification stability under complex weather, occlusion, and local inaccuracy conditions. By weighted fusion of the observation features of each modality and generating initial risk identification results, it improves the targeting of environmental risk assessment. Furthermore, by combining the correlation between line sections to correct the initial risk identification results, it enhances the ability to determine risk transmission between adjacent sections, ultimately improving the accuracy and continuity of environmental risk level output. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth. Example 1

[0023] Reference Figure 1 This embodiment provides a method for identifying environmental risks of power transmission lines based on multimodal IoT sensing, including: S1. Collect multimodal IoT sensor data from each line section, and merge them according to line section and the identifier of the same collection to generate a multimodal observation set. This step requires specific explanation: S1.1. Install visible light acquisition devices, infrared acquisition devices, meteorological acquisition devices, conductor vibration acquisition devices and tower tilt angle acquisition devices in each line section to collect visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data and tower tilt angle monitoring data for the corresponding line sections respectively. S1.2. Based on the tower number, span number and geographical location of each line section, the visible light image data, the infrared image data, the meteorological monitoring data, the conductor vibration monitoring data and the tower tilt angle monitoring data are processed to correspond to the line section, generating the zoned observation data corresponding to each line section; S1.3. Based on the acquisition time corresponding to the partitioned observation data, perform same-time correspondence processing to generate the same-time observation data group corresponding to each line section. S1.4 Merge the same observation data sets corresponding to each line segment to generate a multimodal observation set corresponding to each line segment.

[0024] In a preferred embodiment, the method for identifying environmental risks of transmission lines based on multimodal IoT sensing is deployed for environmental risk identification scenarios of overhead transmission lines in mountainous areas, overhead transmission lines crossing rivers, and transmission lines in windy and humid areas. Taking a 220 kV transmission line as an example, the transmission line is divided into multiple line sections. Each line section is divided by the span between two adjacent towers, and is uniquely corresponding to the tower number, span number, and geographical location.

[0025] In a preferred embodiment, step S1.1, which involves deploying visible light acquisition devices, infrared acquisition devices, meteorological acquisition devices, conductor vibration acquisition devices, and tower tilt angle acquisition devices in each line section to collect visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data for the corresponding line sections, can be performed as follows: For each line section, visible light acquisition devices and infrared acquisition devices are installed at the crossarm positions of the towers at both ends of the span, meteorological acquisition devices are installed in the upper middle part of the towers, conductor vibration acquisition devices are installed near the conductor suspension points, and tower tilt angle acquisition devices are installed at the main structure positions of the towers. The visible light acquisition device collects visible light image data covering the conductors, insulator strings, fittings, and adjacent vegetation areas of the line section. The infrared acquisition device collects infrared image data of the corresponding area. The meteorological acquisition device collects meteorological monitoring data corresponding to wind speed, wind direction, ambient temperature, ambient humidity, and rainfall status. The conductor vibration acquisition device collects conductor vibration monitoring data corresponding to conductor vibration frequency, vibration amplitude, and vibration duration. The tower tilt angle acquisition device collects tower tilt angle monitoring data corresponding to the tilt angle values ​​in the transverse and longitudinal directions of the line. Taking the section of the line between towers T105 and T106 as an example, the visible light acquisition device collected a set of visible light image data with a resolution of 1920×1080; the infrared acquisition device collected a set of infrared image data with a temperature range of -20 degrees Celsius to 120 degrees Celsius; the meteorological acquisition device recorded a wind speed of 12.4 meters per second, a wind direction of north-northwest, an ambient temperature of 6.8 degrees Celsius, an ambient humidity of 84%, and a rainfall condition of light rain; the conductor vibration acquisition device recorded a vibration frequency of 3.6 Hz, a vibration amplitude of 18 mm, and a vibration duration of 46 seconds; and the tower tilt angle acquisition device recorded a tilt angle of 0.42 degrees in the transverse direction and a tilt angle of 0.31 degrees in the longitudinal direction. Using the above acquisition methods, on the one hand, the image-based monitoring data covers surface icing, foreign object suspension, conductor heating, and external obstruction; on the other hand, the numerical monitoring data covers wind load, vibration status, and tower attitude changes, thus providing a complete data foundation for subsequent environmental risk identification.

[0026] In a preferred embodiment, step S1.2, which involves processing the visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data according to the tower number, span number, and geographical location of each line segment to generate zoned observation data corresponding to each line segment, can be performed as follows: First, the span between two adjacent towers is used as the boundary unit of the line segment, with the end with the smaller tower number as the starting end and the end with the larger tower number as the ending end, forming a fixed line segment numbering rule. Second, for various monitoring data collected at the same time, the tower number, span number, and geographical location corresponding to their installation positions are checked, and monitoring data from the same line segment are grouped into the same line segment. For image data covering two adjacent line segments, unique assignment is made based on the span position corresponding to the center of the image field of view; when the center of the image field of view falls within the T105-T106 span range, the image data is assigned to the T105-T106 line segment. For meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data, the corresponding line sections are determined based on the fixed installation locations of the data acquisition devices. Taking the T105-T106 line section as an example, its geographical range is from 118.8123°E to 118.8168°E and from 32.0451°N to 32.0497°N. Visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data collected within this range are uniformly classified into the T105-T106 line section after location verification, thus forming the zoned observation data for this line section. Through the above line section correspondence processing, cross-mixing of monitoring data from different line sections is avoided, ensuring that the risk identification targets for each subsequent line section are unique and clear.

[0027] In a preferred embodiment, step S1.3, which involves performing same-time correspondence processing based on the acquisition time corresponding to the partitioned observation data to generate same-time observation data groups for each line segment, can be performed as follows: For partitioned observation data within the same line segment, the acquisition times corresponding to visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data are extracted respectively, and same-time correspondence is performed using minute-level time windows. Specifically, the earliest arriving acquisition time is taken as the current merging starting point, and the remaining modal data within 30 seconds before and after the starting point are merged into the same observation. If a modal data exceeds this time range, the modal data is merged into the next observation. Taking the T105-T106 line section as an example, the visible light image data was acquired at 10:15:08, the infrared image data at 10:15:12, the meteorological monitoring data at 10:15:00, the conductor vibration monitoring data at 10:15:21, and the tower tilt angle monitoring data at 10:15:17. All five types of data fall within the range of 10:15:00 to 10:15:30, therefore they are classified as the same observation data group. If another set of infrared image data was acquired at 10:16:03, it would not be included in this same observation data group but would be assigned to the next same observation data group. This same-time correspondence processing allows for joint characterization of different modal data under the same environmental conditions, reducing state distortion caused by the dispersion of acquisition times.

[0028] In a preferred embodiment, the merging of the same observation data sets corresponding to each line segment in step S1.4 to generate a multimodal observation set corresponding to each line segment can be performed as follows: For each line segment, multiple same observation data sets arranged in the order of acquisition time are sequentially merged into the multimodal observation set corresponding to that line segment, wherein each same observation data set serves as an observation unit in the multimodal observation set. Each observation unit includes at least one set of visible light image data, one set of infrared image data, one set of meteorological monitoring data, one set of conductor vibration monitoring data, and one set of tower tilt angle monitoring data; when a certain mode is missing in a certain observation, a missing mark is retained at the corresponding position, and this missing situation is included in the data completeness in subsequent modal quality detection. Taking the T105-T106 line section as an example, three sets of observation data were generated at 10:15, 10:20, and 10:25 respectively. These three sets of observation data were then merged into a multimodal observation set corresponding to the T105-T106 line section in chronological order. Through this merging method, each line section forms a multimodal observation set organized in chronological order, providing a unified data organization format for subsequent modal quality detection, observation feature extraction, and risk identification model input.

[0029] The purpose of this step is to unify and merge the multi-source heterogeneous monitoring data scattered across the transmission line site according to the line section and the relationship of the same acquisition, forming a multi-modal observation set that can be directly used in subsequent risk analysis. The technical problem it solves is that in existing transmission line environmental monitoring schemes, the acquisition locations, acquisition times, and line sections of different modal data are inconsistent, leading to confusion in subsequent risk identification, distorted state representation, and unclear result correspondences. Through the above step S1, visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data can be uniformly classified into a single line section and unified to the same observation level, thus bringing the beneficial effects of clear data sources, accurate state correspondence, and a complete foundation for subsequent processing.

[0030] S2. Calculate the modal quality parameters corresponding to each mode based on the multimodal observation set, and generate a set of reliable weights. This step requires specific explanation: S2.1 Separate visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data from the multi-modal observation set corresponding to each line section, and generate quality detection results for the corresponding modes respectively; S2.2. Based on the quality detection results, calculate the modal quality parameters corresponding to each modality. The modal quality parameters include image sharpness, image occlusion rate, temperature distribution continuity, data integrity, signal fluctuation stability, and acquisition time difference. S2.3 Normalize the modal quality parameters corresponding to each modality and generate a set of reliable weights.

[0031] Furthermore, S2.3.1, the modal quality parameters corresponding to each mode within the same line segment are subjected to directional unification processing, wherein image clarity, temperature distribution continuity, data integrity and signal fluctuation stability retain the original direction of change, and image occlusion rate and acquisition time difference are converted into reverse quality parameters; S2.3.2. The modal quality parameters corresponding to each mode within the same line section are subjected to interval compression processing according to the minimum and maximum values ​​of the corresponding parameters to generate standard quality values ​​between zero and one. S2.3.3. Generate a trusted weight value for each mode according to the proportion of the standard quality value corresponding to each mode to the sum of all standard quality values ​​in the same line segment, and merge the trusted weight values ​​corresponding to each mode into a trusted weight value set.

[0032] In a preferred embodiment, step S2.1, which involves separating visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data from the multimodal observation set corresponding to each line section, and generating corresponding mode quality detection results, can be performed as follows: For each line section corresponding to the multimodal observation set, visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data are sequentially extracted from each observation unit, and corresponding mode quality detection items are established. For visible light image data, quality inspection items include the sharpness of image edges, the proportion of obscured areas, and the difference in acquisition time. For infrared image data, quality inspection items include the continuity of temperature distribution, whether abnormally high-temperature areas are broken, and the difference in acquisition time. For meteorological monitoring data, quality inspection items include whether wind speed, wind direction, ambient temperature, ambient humidity, and rainfall status are complete, and the difference in acquisition time. For conductor vibration monitoring data, quality inspection items include whether vibration frequency sequence, vibration amplitude sequence, and vibration duration are complete, whether the changes in adjacent sampled values ​​are stable, and the difference in acquisition time. For tower tilt angle monitoring data, quality inspection items include whether the tilt angle values ​​in the transverse and longitudinal directions are complete, whether the tilt angle changes are stable, and the difference in acquisition time. Taking the observation unit corresponding to the T105-T106 line section at 10:15 as an example, the visible light image data showed a clear outline of the conductor edge, with some foliage obscuring the upper part of the image, the obscured area accounting for approximately 12% of the entire image; the infrared image data showed a continuous temperature transition on the conductor and insulator surfaces, with no obvious temperature block breakage; the meteorological monitoring data showed the presence of wind speed, wind direction, ambient temperature, ambient humidity, and rainfall; the conductor vibration monitoring data showed a complete 10-second sampling sequence, with stable changes in adjacent vibration amplitudes; the tower tilt angle monitoring data showed both transverse and longitudinal tilt angles, with minimal changes between adjacent samplings. These detection results were used as the quality detection results for the corresponding modes.

[0033] In a preferred embodiment, step S2.2 involves calculating the modal quality parameters corresponding to each mode based on the quality detection results. These modal quality parameters include image sharpness, image occlusion rate, temperature distribution continuity, data integrity, signal fluctuation stability, and acquisition time difference. Specifically, this can be performed as follows: For visible light image data, image sharpness is determined by statistically analyzing the detail levels of the conductor edge region and the insulator outline region. Images with rich edge details and clear outline transitions are considered high-resolution. The image occlusion rate is determined by statistically analyzing the proportion of the area where the conductor body, insulator string, and fittings are obscured by tree branches, bird nests, fog droplets, or raindrops. For infrared image data, temperature distribution continuity is determined by comparing the continuity of temperature changes on the conductor body, fitting connection positions, and insulator surface. A higher temperature distribution continuity corresponds to a smooth temperature difference change between adjacent areas and a complete hot spot outline. For meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data, data completeness is determined by the correspondence between the data items that should be present in the observation unit and the actual data items collected. When wind speed, wind direction, ambient temperature, ambient humidity, and rainfall conditions are all present, the meteorological monitoring data is considered complete. When vibration frequency, vibration amplitude, and vibration duration are all present, the conductor vibration monitoring data is considered complete. When both transverse and longitudinal tilt angle values ​​are present, the tower tilt angle monitoring data is considered complete. Signal fluctuation stability is determined by the smoothness of the change in adjacent sampled values. Higher signal fluctuation stability corresponds to the absence of sudden jumps or discrete spikes in continuous sampling of conductor vibration monitoring data or tower tilt angle monitoring data. Acquisition time difference is determined by the time difference between the acquisition time of each modal data and the reference acquisition time in the same observation. Taking the observation unit corresponding to 10:15 on the T105-T106 line section as an example, the image clarity of the visible light image data is judged to be 0.88, and the image occlusion rate is judged to be 0.12; the temperature distribution continuity of the infrared image data is judged to be 0.91; the data integrity of the meteorological monitoring data is judged to be 1.00; the data integrity of the conductor vibration monitoring data is judged to be 1.00, and the signal fluctuation stability is judged to be 0.84; the data integrity of the tower tilt angle monitoring data is judged to be 1.00, and the signal fluctuation stability is judged to be 0.93; the acquisition time difference between the visible light image data and the reference acquisition time is 8 seconds, the infrared image data is 12 seconds, the meteorological monitoring data is 0 seconds, the conductor vibration monitoring data is 21 seconds, and the tower tilt angle monitoring data is 17 seconds. The above modal quality parameters constitute the basis for the subsequent generation of reliable weights.

[0034] In a preferred embodiment, the normalization processing of the modal quality parameters corresponding to each mode in step S2.3, and the generation of a set of reliable weights, can be performed as follows. First, in step S2.3.1, the modal quality parameters corresponding to each mode within the same line segment are processed to unify their direction. Specifically, image clarity, temperature distribution continuity, data integrity, and signal fluctuation stability are directly used as positive quality parameters in subsequent processing; image occlusion rate and acquisition time difference are converted into inverse quality parameters, i.e., the less occlusion, the higher the quality; the smaller the acquisition time difference, the higher the quality. Taking the observation unit corresponding to 10:15 in the T105-T106 line segment as an example, the image occlusion rate is 0.12, which corresponds to a higher occlusion quality after inverse processing; the acquisition time difference of infrared image data is 12 seconds, and the acquisition time difference of conductor vibration monitoring data is 21 seconds, so the time difference quality corresponding to the infrared image data is higher than that of the conductor vibration monitoring data. Secondly, in step S2.3.2, the modal quality parameters corresponding to each mode within the same line section are subjected to interval compression processing based on the minimum and maximum values ​​of the corresponding parameters, so that the standard quality values ​​corresponding to each mode all fall between zero and one. Taking this observation unit as an example, the standard quality values ​​obtained from visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data are 0.82, 0.86, 0.93, 0.74, and 0.88, respectively. Finally, in step S2.3.3, the confidence weights corresponding to each mode are generated according to the proportion of the standard quality value corresponding to each mode to the sum of all standard quality values ​​within the same line section, and the confidence weights corresponding to each mode are merged into a confidence weight set. Taking the aforementioned standard quality values ​​as an example, the sum of the standard quality values ​​for the five modes is 4.23. Therefore, the confidence weight for visible light image data is 0.194, for infrared image data it is 0.203, for meteorological monitoring data it is 0.220, for conductor vibration monitoring data it is 0.175, and for tower tilt monitoring data it is 0.208. This forms the confidence weight set for the observation unit corresponding to 10:15 on the T105-T106 line section. Using this processing method, each mode can participate in the analysis according to its current data quality proportion in subsequent feature fusion, avoiding excessive interference from low-quality modes on the risk identification results.

[0035] The purpose of this step is to assess the quality of multimodal observation sets for each line section and transform the assessment results into a set of reliable weights that can be directly used in feature fusion. The technical problem it addresses is that the field acquisition environment of transmission lines is complex; images may be obstructed or blurred, and numerical monitoring data may be missing, drifting, or have discrepancies in acquisition time. Directly applying equal weights to multimodal data can easily amplify the adverse effects of low-quality data. Through step S2, the image clarity, image obstruction rate, temperature distribution continuity, data integrity, signal fluctuation stability, and acquisition time difference of different modes can be transformed into unified standard quality values ​​and reliable weights. This results in the beneficial effects of consistent modal participation with data quality, limited impact of low-quality modes, and a more stable fusion foundation.

[0036] S3. Based on the trusted weight set, weighted fusion of the modal observation features in the multimodal observation set is performed to generate initial risk identification results for each line segment. This step requires specific explanation: S3.1 Extract visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features from the multi-modal observation set corresponding to each line section, and generate observation features corresponding to each mode; S3.2. Based on the trusted weights corresponding to each mode, the observation features corresponding to each mode are weighted and processed, and the trusted weights and observation features corresponding to each mode within the same line segment are combined to generate fused observation features. S3.3 Input the fused observation features corresponding to each line segment into the risk identification model to generate the initial risk identification results corresponding to each line segment.

[0037] S3.3.1 Input the fused observation features corresponding to each line segment into the risk identification model, and generate the risk identification score corresponding to each line segment from the risk identification model; S3.3.2. Based on the risk identification scores corresponding to each line segment, the risk level is determined and a risk level label corresponding to each line segment is generated. S3.3.3. The risk level label corresponding to each line segment is determined as the initial risk identification result corresponding to each line segment.

[0038] In a preferred embodiment, step S3.1, which involves extracting visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features from the multi-modal observation sets corresponding to each line section, and generating observation features corresponding to each mode, can be performed as follows: For visible light image data, visible light image features are extracted from the conductor area, insulator string area, hardware connection area, and adjacent vegetation area. These visible light image features include the proportion of foreign object area, the proportion of visible icing area on the conductor, the number of insulator contamination patches, the intrusion distance of adjacent tree barriers, and the proportion of areas with abnormal surface color. For infrared image data, infrared image features are extracted from the conductor body, tension clamp, suspension clamp, and insulator string area. These infrared image features include the proportion of the highest temperature area, the number of areas with concentrated temperature differences, the smoothness of the temperature gradient, and the number of local abnormal hot spots. For meteorological monitoring data, meteorological monitoring features corresponding to wind speed ranges, wind direction categories, ambient temperature ranges, ambient humidity ranges, and rainfall status categories are extracted. For conductor vibration monitoring data, conductor vibration features corresponding to vibration frequency level, vibration amplitude level, vibration duration level, and vibration change trend are extracted. For tower tilt monitoring data, the tower tilt characteristics corresponding to the tilt level in the transverse direction, the tilt level in the longitudinal direction, and the tilt trend are extracted. Taking the observation unit corresponding to 10:15 on the T105-T106 line section as an example, the visible light image characteristics include a foreign object area ratio of 4%, a conductor icing visible area ratio of 16%, and a distance of 1.3 meters from the intrusion of nearby trees; the infrared image characteristics include a maximum temperature area ratio of 3%, two local abnormal hot spots, and a maximum temperature difference of 11 degrees Celsius between the conductor and the fittings; the meteorological monitoring characteristics include a wind speed range of 10 to 15 meters per second, a wind direction of north-northwest, an ambient humidity range of 80% to 90%, and a rainfall type of light rain; the conductor vibration characteristics include a vibration frequency level of moderate, a vibration amplitude level of relatively high, and a vibration duration level of moderate to high; the tower tilt characteristics include a low tilt level in the transverse direction and a low tilt level in the longitudinal direction. The above items constitute the observation characteristics corresponding to each mode.

[0039] In a preferred embodiment, step S3.2, which involves assigning weights to the observation features corresponding to each mode based on the confidence weights, and combining the confidence weights and observation features of each mode within the same line segment to generate fused observation features, can be performed as follows: First, the observation features corresponding to each mode within the same observation unit are arranged in a unified order, namely, visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features. Second, the confidence weights corresponding to each mode are assigned to all observation features of that mode, ensuring that each observation feature within the same mode has the same quality proportion in subsequent combinations. Then, the observation features of each mode after weight assignment are combined in the aforementioned unified order to generate fused observation features corresponding to that observation of that line segment. Taking the observation unit corresponding to 10:15 on the T105-T106 line section as an example, the confidence weight of visible light image data is 0.194, the confidence weight of infrared image data is 0.203, the confidence weight of meteorological monitoring data is 0.220, the confidence weight of conductor vibration monitoring data is 0.175, and the confidence weight of tower tilt monitoring data is 0.208. Therefore, in the fused observation features, the influence of meteorological monitoring features and tower tilt features is higher than that of conductor vibration features, corresponding to the situation where the quality of meteorological monitoring data and tower tilt monitoring data is relatively high, while the quality of conductor vibration monitoring data is relatively low in this observation. Through the above weight allocation and combination processing, the fused observation features retain the content of each modality's features while also reflecting the current observation quality differences of each mode.

[0040] In a preferred embodiment, step S3.3, which involves inputting the fused observation features corresponding to each line segment into the risk identification model to generate initial risk identification results for each line segment, can be performed as follows: The risk identification model can be a multi-layer classification model, which includes an input layer, a first feature mapping layer, a second feature mapping layer, and an output layer. The input layer receives the fused observation features. The first feature mapping layer aggregates the combination relationships between visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features. The second feature mapping layer further discriminates the aggregated risk-related features. The output layer outputs the risk identification score. In step S3.3.1, the fused observation features corresponding to each line segment are input into the risk identification model, and the risk identification model generates the risk identification score for each line segment. Taking the T105-T106 line segment as an example, its risk identification score is 82 points. In step S3.3.2, the risk level is determined based on the risk identification score for each line segment, generating a risk level label for each line segment. Specifically, the risk identification can be determined according to the following risk level ranges: a risk identification score below 25 points corresponds to low risk; a risk identification score between 25 and 50 points corresponds to relatively low risk; a risk identification score above 50 points but not exceeding 75 points corresponds to relatively high risk; and a risk identification score above 75 points corresponds to high risk. Therefore, the risk level label for the T105-T106 line segment is high risk. In step S3.3.3, the risk level label corresponding to each line segment is determined as the initial risk identification result for that line segment. For example, the risk identification score for the T106-T107 line segment is 68 points, corresponding to relatively high risk; the risk identification score for the T107-T108 line segment is 34 points, corresponding to relatively low risk. Through the above method, the initial risk identification result for each line segment can be obtained.

[0041] The purpose of this step is to perform differentiated weighted fusion of the observation features of each modality based on the current data quality of each modality, and to provide the initial risk identification results for each line section through a risk identification model. The technical problem it addresses is that the environmental risk of transmission lines is often formed by the combined effects of image information, thermal state information, meteorological conditions, conductor dynamics, and tower attitude changes. A single modality is insufficient to fully reflect the risk state, while direct splicing of multiple modalities is susceptible to interference from low-quality data. Through step S3, the orderly fusion of multimodal observation features can be completed under credible weight constraints, generating risk identification scores and risk level labels. This results in a more comprehensive risk state representation, risk identification results that better match the on-site conditions, and clearer differences in initial risks among different line sections.

[0042] S4. Correct the initial risk identification results based on the correlation between each line segment, and output the environmental risk level of each line segment. This step requires specific explanation: S4.1. Based on the tower adjacency relationship, span continuity relationship and wind direction transmission relationship between each line section, generate the line section association relationship corresponding to each line section; S4.2. The initial risk identification results corresponding to each line segment are correlated with the initial risk identification results corresponding to adjacent line segments according to the correlation relationship of the line segments, and the correlation correction results corresponding to each line segment are generated. S4.3. Adjust the level of the initial risk identification result corresponding to each line segment according to the correlation correction result corresponding to each line segment, and generate the environmental risk level corresponding to each line segment.

[0043] S4.1.1. Based on the tower number and span number corresponding to each line section, determine the tower adjacency relationship and span continuity relationship between each line section, and generate a set of adjacent line sections; S4.1.2. Based on the wind direction monitoring data corresponding to each line section, determine the wind direction transmission relationship between adjacent line sections; S4.1.3 Merge the tower adjacency relationship, the span continuity relationship and the wind direction transmission relationship to generate the line segment association relationship corresponding to each line segment.

[0044] S4.2.1 Based on the line segment association relationship corresponding to each line segment, extract the initial risk identification result set of adjacent line segments corresponding to the current line segment; S4.2.2. Compare the initial risk identification result corresponding to the current line segment with each initial risk identification result in the set of initial risk identification results of the adjacent line segments to generate the risk difference result corresponding to the current line segment. S4.2.3. Based on the risk difference results and the line segment association relationship, generate the association correction result corresponding to the current line segment.

[0045] S4.3.1. Based on the correlation correction results corresponding to each line segment, determine the risk correction direction corresponding to each line segment; S4.3.2 When the risk correction direction is upward, the initial risk identification result corresponding to the current line segment is increased by one level; when the risk correction direction is downward, the initial risk identification result corresponding to the current line segment is decreased by one level; when the risk correction direction is to remain, the initial risk identification result corresponding to the current line segment is maintained.

[0046] In a preferred embodiment, step S4.1, which generates the line segment association relationship corresponding to each line segment based on the tower adjacency relationship, span continuity relationship, and wind direction transmission relationship between each line segment, can be performed as follows: In step S4.1.1, based on the tower number and span number corresponding to each line segment, the tower adjacency relationship and span continuity relationship between each line segment are determined, generating a set of adjacent line segments. Specifically, when two line segments share the same base tower, or when the span numbers are consecutive, they are determined to be adjacent line segments. For example, line segment T105-T106 and line segment T106-T107 share tower T106, therefore they have a tower adjacency relationship; line segment T106-T107 and line segment T107-T108 have consecutive span numbers, therefore they have a span continuity relationship. In step S4.1.2, the wind direction transmission relationship between adjacent line sections is determined based on the wind direction monitoring data corresponding to each line section. Specifically, combining the prevailing wind direction and line orientation at a certain moment, when there is a possibility that foreign objects, rain, fog, icing, or high humidity in the upwind line section may extend to the downwind line section, a wind direction transmission relationship is determined between the upwind and downwind line sections. Taking a certain observation as an example, the prevailing wind direction is north-northwest, and the T105-T106 line section is located on the upwind side of the T106-T107 line section. Therefore, there is a wind direction transmission relationship between the T105-T106 line section and the T106-T107 line section, with the former pointing towards the latter. In step S4.1.3, the tower adjacency relationship, span continuity relationship, and wind direction transmission relationship are merged to generate the line section association relationship corresponding to each line section. After consolidation, the line segment relationships of the T106-T107 section can simultaneously include the tower adjacency, span continuity, and wind direction transmission relationships with the T105-T106 section, as well as the tower adjacency and span continuity relationships with the T107-T108 section. Through the above processing, each line segment can obtain line segment relationships that reflect its spatial adjacency and environmental propagation.

[0047] In a preferred embodiment, step S4.2, which involves performing correlation calculations between the initial risk identification results corresponding to each line segment and the initial risk identification results corresponding to adjacent line segments according to the line segment association relationship, to generate the correlation correction results corresponding to each line segment, can be performed as follows: In step S4.2.1, based on the line segment association relationship corresponding to each line segment, the set of initial risk identification results for adjacent line segments corresponding to the current line segment is extracted. Taking line segment T106-T107 as an example, its associated adjacent line segments are line segments T105-T106 and T107-T108. Therefore, the set of initial risk identification results for its adjacent line segments includes the high risk of line segment T105-T106 and the lower risk of line segment T107-T108. In step S4.2.2, the initial risk identification result corresponding to the current line segment is compared with the initial risk identification results in the set of initial risk identification results for adjacent line segments to generate the risk difference result corresponding to the current line segment. Specifically, the risk levels are arranged in ascending order as low risk, lower risk, higher risk, and high risk, and the level difference between adjacent line segments and the current line segment is judged one by one. Taking the above example, the initial risk identification result of the current T106-T107 line segment is higher risk, which is one level lower than the high risk of the T105-T106 line segment and one level lower than the lower risk of the T107-T108 line segment. In step S4.2.3, based on the risk difference result and the correlation between line segments, the correlation correction result corresponding to the current line segment is generated. Specifically, when there is a wind direction transmission relationship between the adjacent upwind line segment and the current line segment, and the risk level of the adjacent upwind line segment is one level higher than that of the current line segment, the correlation correction result of the current line segment is determined to be upward; when the risk levels of the adjacent line segments on both sides of the current line segment are both one level lower than that of the current line segment, and there is no wind direction transmission relationship pointing towards the current line segment, the correlation correction result of the current line segment is determined to be downward; in other cases, it is determined to remain unchanged. Taking the above example, since the T105-T106 line segment is located on the upwind side, and its initial risk identification result is high risk, which is higher than the higher risk level of the T106-T107 line segment, and there is a wind direction transmission relationship between the two, the correlation correction result of the T106-T107 line segment is determined to be upward. For example, if the initial risk identification result for the T107-T108 line segment is low risk, while the initial risk identification result for its adjacent upwind line segment T106-T107 is high risk, and there is a continuous span relationship between the two, then the correlation correction result for the T107-T108 line segment can also be determined as upward adjustment. Using the above correlation calculation method, spatial adjacency and environmental propagation can be incorporated into the risk correction process.

[0048] In a preferred embodiment, step S4.3, which involves adjusting the initial risk identification result of each line segment based on the associated correction result to generate an environmental risk level for each line segment, can be performed as follows: Step S4.3.1 determines the risk correction direction for each line segment based on the associated correction result. Specifically, if the associated correction result is upward, the risk correction direction is upward; if the associated correction result is downward, the risk correction direction is downward; and if the associated correction result is unchanged, the risk correction direction is unchanged. Step S4.3.2, when the risk correction direction is upward, the initial risk identification result for the current line segment is increased by one level; when the risk correction direction is downward, the initial risk identification result for the current line segment is decreased by one level; and when the risk correction direction is unchanged, the initial risk identification result for the current line segment is retained. Taking the T106-T107 line section as an example, its initial risk identification result is relatively high risk, and the correlation correction result is upward, so the corresponding environmental risk level is adjusted to high risk. Taking the T107-T108 line section as another example, its initial risk identification result is relatively low risk, and the correlation correction result is upward, so the corresponding environmental risk level is adjusted to relatively high risk. Furthermore, if the initial risk identification result for a certain line section is high risk, even if its adjacent line sections are all relatively low risk, because this line section itself has significant conductor heating and large vibrations, its correlation correction result remains unchanged, and the corresponding environmental risk level remains high risk. Through the above-mentioned level adjustment methods, an environmental risk level that more closely reflects the spatial continuity of the line can be further formed based on the initial risk identification results.

[0049] The purpose of this step is to correlate and correct the initial risk identification results of each line segment by combining the spatial adjacency and environmental propagation relationships between line segments, thereby outputting an environmental risk level that better reflects the continuous distribution characteristics on site. The technical problem it addresses is that transmission line environmental risks exhibit propagation along the line and expansion to neighboring areas. If only the independent observation results of a single segment are used as the final conclusion, local misjudgments or distortions in the risk connection between adjacent segments are likely to occur. Through step S4, which incorporates tower adjacency relationships, span continuity relationships, and wind direction transmission relationships into the correlation calculation and level adjustment process, the beneficial effects of more consistent risk identification results with the actual spatial state of the line, smoother risk transitions between adjacent segments, and environmental risk levels that better reflect the on-site propagation patterns can be achieved.

[0050] It should be understood that numerous specific implementation decisions can be made during the development of any practical implementation, such as in any engineering or design project. Such development efforts may be complex and time-consuming, but for those skilled in the art who benefit from this disclosure, the development effort will be a routine work of design, manufacturing, and production without requiring much experimentation.

[0051] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying environmental risks of power transmission lines based on multimodal IoT sensing, characterized in that, include: Multimodal IoT sensor data from each line section are collected and merged according to line section and the same collection identifier to generate a multimodal observation set; Based on the multimodal observation set, calculate the modal quality parameters corresponding to each mode and generate a set of reliable weights; Based on the set of credible weights, the features of each modality observation in the multimodal observation set are weighted and fused to generate the initial risk identification results for each line segment; The initial risk identification results are corrected based on the correlation between different line sections, and the environmental risk level of each line section is output.

2. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 1, characterized in that, The generation of the multimodal observation set includes: Visible light acquisition devices, infrared acquisition devices, meteorological acquisition devices, conductor vibration acquisition devices, and tower tilt angle acquisition devices are deployed in each section of the line to collect visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data for the corresponding line sections, respectively. Based on the tower number, span number, and geographical location of each line section, the visible light image data, the infrared image data, the meteorological monitoring data, the conductor vibration monitoring data, and the tower tilt angle monitoring data are processed to correspond to the line section, generating zoned observation data corresponding to each line section; Based on the acquisition time corresponding to the partitioned observation data, perform same-time correspondence processing to generate same-time observation data groups for each line segment; The same observation data sets corresponding to each line segment are merged to generate a multimodal observation set corresponding to each line segment.

3. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 2, characterized in that, The generation of the trusted weight set includes: Visible light image data, infrared image data, meteorological monitoring data, conductor vibration monitoring data, and tower tilt angle monitoring data are separated from the multimodal observation sets corresponding to each line section, and the quality detection results of the corresponding modes are generated respectively. Based on the quality detection results, the modal quality parameters corresponding to each modality are calculated. The modal quality parameters include image sharpness, image occlusion rate, temperature distribution continuity, data integrity, signal fluctuation stability, and acquisition time difference. The modal quality parameters corresponding to each modality are normalized, and a set of reliable weights is generated.

4. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 3, characterized in that, The normalization process for the modal quality parameters corresponding to each modality, and the generation of a set of reliable weights, includes: The modal quality parameters corresponding to each mode within the same line segment are subjected to directional unification processing. Among them, image clarity, temperature distribution continuity, data integrity and signal fluctuation stability retain the original direction of change, while image occlusion rate and acquisition time difference are converted into inverse quality parameters. The modal quality parameters corresponding to each mode within the same line section are compressed according to the minimum and maximum values ​​of the corresponding parameters to generate standard quality values ​​between zero and one. Based on the proportion of the standard quality value corresponding to each mode to the sum of all standard quality values ​​in the same line segment, a reliable weight value corresponding to each mode is generated, and the reliable weight values ​​corresponding to each mode are merged into a reliable weight value set.

5. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 4, characterized in that, The initial risk identification results for each line segment are generated, including: Visible light image features, infrared image features, meteorological monitoring features, conductor vibration features, and tower tilt angle features are extracted from the multimodal observation sets corresponding to each line section to generate observation features corresponding to each mode; Based on the credible weights corresponding to each mode, the observation features corresponding to each mode are weighted and processed, and then combined according to the credible weights and observation features corresponding to each mode within the same line segment to generate fused observation features; The fused observation features corresponding to each line segment are input into the risk identification model to generate the initial risk identification results for each line segment.

6. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 5, characterized in that, The step of inputting the fused observation features corresponding to each line segment into the risk identification model to generate the initial risk identification results corresponding to each line segment includes: The fused observation features corresponding to each line segment are input into the risk identification model, and the risk identification model generates the risk identification score corresponding to each line segment. Based on the risk identification scores corresponding to each line segment, a risk level label is generated for each line segment. The risk level label corresponding to each line segment is determined as the initial risk identification result for each line segment.

7. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 6, characterized in that, The process of correcting the initial risk identification results based on the correlation between each line segment and outputting the environmental risk level of each line segment includes: Based on the tower adjacency relationship, span continuity relationship and wind direction transmission relationship between each line segment, the corresponding line segment association relationship is generated for each line segment; The initial risk identification results corresponding to each line segment are correlated with the initial risk identification results corresponding to adjacent line segments according to the line segment correlation relationship to generate the correlation correction results corresponding to each line segment; Based on the correlation correction results corresponding to each line segment, the initial risk identification results corresponding to each line segment are adjusted to generate the environmental risk level corresponding to each line segment.

8. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 7, characterized in that, The generation of the line segment association relationship corresponding to each line segment includes: Based on the tower number and span number corresponding to each line segment, determine the tower adjacency relationship and span continuity relationship between each line segment, and generate a set of adjacent line segments; Based on the wind direction monitoring data corresponding to each line section, determine the wind direction transmission relationship between adjacent line sections; The tower adjacency relationship, the span continuity relationship, and the wind direction transmission relationship are merged to generate the line segment association relationship corresponding to each line segment.

9. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 8, characterized in that, The generation of the associated correction results for each line segment includes: Based on the line segment association relationship corresponding to each line segment, extract the initial risk identification result set of adjacent line segments corresponding to the current line segment; The initial risk identification result corresponding to the current line segment is compared with each initial risk identification result in the set of initial risk identification results of the adjacent line segments to generate the risk difference result corresponding to the current line segment. Based on the risk difference results and the correlation between the line segments, the correlation correction results corresponding to the current line segment are generated.

10. The method for identifying environmental risks of transmission lines based on multimodal IoT sensing according to claim 9, characterized in that, The step of adjusting the initial risk identification results for each line segment based on the correlation correction results for each line segment to generate an environmental risk level for each line segment includes: Based on the correlation correction results corresponding to each line segment, the risk correction direction corresponding to each line segment is determined; When the risk correction direction is upward, the initial risk identification result corresponding to the current line segment is increased by one level; when the risk correction direction is downward, the initial risk identification result corresponding to the current line segment is decreased by one level; when the risk correction direction is to remain unchanged, the initial risk identification result corresponding to the current line segment is maintained.

Images