A power transmission line inspection method based on multi-source data and related equipment

Abstract

The embodiment of the application provides a power transmission line inspection method based on multi-source data and related equipment, and belongs to the technical field of power inspection. The embodiment of the application can obtain inspection data in multiple dimensions, including high-altitude inspection data, ground inspection data, manual inspection data and fixed monitoring data, and utilize a multi-source feature model to identify defect types from the high-altitude inspection data, the ground inspection data, the manual inspection data and the fixed monitoring data in multiple angles, arrange an inspection line based on the identified defect types, increase the probability of finding defects of the power transmission line, increase the identification accuracy of the defects of the power transmission line, and improve the response speed of the inspection personnel.

Description

Technical Field

[0001] This application relates to the field of power line inspection technology, and in particular to a method and related equipment for power transmission line inspection based on multi-source data. Background Technology

[0002] In the field of power transmission line inspection, traditional inspection methods mostly rely on a certain detection terminal to carry out the work. However, anomalies in power transmission lines may manifest in different forms, and the detection terminal may form blind spots due to limitations such as working angle and coverage. Single-modal data has obvious blind spots in the identification of complex defects, making it difficult to simultaneously and accurately capture multiple types of defect features such as geometric anomalies and dynamic electrical parameter anomalies. It is also impossible to comprehensively capture the status information of key components and channels of the line, resulting in the inability to detect line anomalies in a timely manner. This seriously restricts the utilization value of inspection data and fails to meet the needs of accurate identification of line defects.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this application is to propose a transmission line inspection method and related equipment based on multi-source data, so as to collect operation data of transmission lines over a wider range, accurately determine the operation status of the lines, and generate inspection routes that better meet actual needs.

[0005] To achieve the above objectives, one aspect of this application proposes a transmission line inspection method based on multi-source data, the method comprising:

[0006] Acquire multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data obtained when multiple terminals inspect transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The multi-source inspection data are respectively input into the multi-source feature model to determine the defect type corresponding to the multi-source inspection data. The multi-source feature model corresponds to the multi-source inspection data. Based on the defect type and the scene characteristics obtained from the multi-source inspection data, the inspection route of the transmission line is generated.

[0007] In some embodiments, the multi-source feature model includes at least two single-source feature models, and the step of inputting the multi-source inspection data into the multi-source feature model respectively to determine the defect type corresponding to the multi-source inspection data includes: Determine the data type of the multi-source inspection data; The multi-source inspection data is input into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model; The defect category is determined by fusing the reference defect and the defect confidence from the single-source feature model using a decision fusion method.

[0008] In some embodiments, generating the inspection route of the transmission line based on the defect type and scene features obtained from the multi-source inspection data includes: Determine the operating characteristics of the transmission line under the defect type; By utilizing the operational characteristics and scene characteristics obtained from the multi-source inspection data, the line priority of the transmission line is obtained through the inspection model. Based on the line priority of the transmission line, the initial inspection line of the transmission line is generated using the A* algorithm; By combining the reward function generated based on the line priority, the initial inspection path is optimized to obtain the inspection route of the transmission line.

[0009] In some embodiments, the single-source feature model includes: a time-series feature model, an image feature model, and a point cloud feature model; the step of inputting the multi-source inspection data into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model includes: Time-series data, image data, and point cloud data are obtained from the multi-source inspection data; The time-series data is input into the time-series feature model to obtain the first reference defect and the first defect confidence level output by the time-series feature model; the image data is input into the image feature model to obtain the second reference defect and the second defect confidence level output by the image feature model; the point cloud data is input into the point cloud feature model to obtain the third reference defect and the third defect confidence level output by the point cloud feature model.

[0010] In some embodiments, a decision fusion method is used to fuse the reference defect and the defect confidence from the single-source feature model to determine the defect category, including: The first defect confidence level, the second defect confidence level, and the third defect confidence level are respectively converted into basic probability assignment functions; Based on the probability allocation function, the defect consistency among the first reference defect, the second reference defect, and the third reference defect is calculated; If the defect consistency is less than or equal to the defect threshold, the basic probability allocation function is fused using the DS fusion algorithm to determine the defect category.

[0011] In some embodiments, after calculating the defect consistency among the first reference defect, the second reference defect, and the third reference defect according to the probability allocation function, the method further includes: If the defect consistency is greater than the defect threshold, obtain the historical accuracy of the time-series feature model, the image feature model, and the point cloud feature model; Using the historical accuracy, the weight factors of the temporal feature model, the image feature model, and the point cloud feature model are modified; The first reference defect, the second reference defect, the third reference defect, the first defect confidence, the second defect confidence, and the third defect confidence are obtained again using the temporal feature model after modifying the weight factors, the image feature model, and the point cloud feature model, until the defect consistency among the first defect confidence, the second defect confidence, and the third defect confidence is less than or equal to the defect threshold.

[0012] In some embodiments, optimizing the initial inspection path by combining a reward function generated based on the line priority to obtain the inspection route of the transmission line includes: The first reward score and first reward level for the route priority are determined; the second reward score and second reward level for inspection coverage are determined; the third reward score and third reward level for inspection resources are determined; and the fourth reward score and fourth reward level for collaborative inspection are determined. Based on the first reward score and the first reward level, the second reward score, the second reward level, the third reward score, the third reward level, the fourth reward score, and the fourth reward level, the initial inspection path is optimized to obtain the inspection path of the transmission line.

[0013] To achieve the above objectives, another aspect of this application proposes a transmission line inspection device based on multi-source data, the device comprising: The acquisition module is used to acquire multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data acquired when multiple terminals inspect transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The determination module is used to input the multi-source inspection data into the multi-source feature model respectively, and determine the defect type corresponding to the multi-source inspection data, wherein the multi-source feature model corresponds to the multi-source inspection data; The generation module is used to generate the inspection route of the transmission line based on the defect type and the scene characteristics obtained from the multi-source inspection data.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the methods described above.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0016] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the methods described above. The embodiments of this application include at least the following beneficial effects: This application provides a method, device, electronic device, storage medium, and program product for power transmission line inspection based on multi-source data. This solution acquires multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data acquired during multi-terminal inspections of power transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The multi-source inspection data is input into a multi-source feature model to determine the defect type corresponding to the multi-source inspection data. Based on the defect type and scene features obtained from the multi-source inspection data, an inspection route for the power transmission line is generated. This application acquires inspection data from multiple dimensions, including high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data, and uses a multi-source feature model to identify defect types from multiple angles in the high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The inspection route is arranged based on the identified defect types, which increases the probability of discovering power transmission line defects, increases the accuracy of defect identification, and improves the response speed of inspection personnel. Attached Figure Description

[0017] Figure 1 This is a flowchart of the transmission line inspection method based on multi-source data provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart of step S102 in the document; Figure 3 yes Figure 1 The flowchart of step S103 in the process; Figure 4 This is a schematic diagram of the structure of the transmission line inspection device based on multi-source data provided in the embodiments of this application; Figure 5 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0021] 1) Transmission line inspection: This is an operation and maintenance method that involves regular monitoring of transmission equipment through manual observation and instrument testing. It aims to discover equipment defects and potential line hazards, and ensure the safe operation of the power grid.

[0022] 2) Multi-source inspection data: refers to the operational data obtained by inspecting transmission lines through multiple terminals and in multiple ways.

[0023] 3) A-star algorithm: A-star algorithm is a heuristic search algorithm used to find the shortest path from the starting point to the ending point on a graph plane.

[0024] 4) Decision fusion algorithm: This is an algorithm that combines the results of multiple classifiers to evaluate the final decision.

[0025] 5) The Dempster-Shafer Fusion Algorithm is an algorithm used to fuse multiple sources of evidence or information. It is based on the Dempster-Shafer theory and aims to draw more accurate conclusions by integrating various pieces of evidence.

[0026] In related technologies, although automated equipment can be used to assist in the inspection of power transmission lines, the defect data collected is relatively simple and cannot fully capture the status information of key components and channels of the line, resulting in the inability to detect line anomalies in a timely manner.

[0027] In view of this, this application provides a transmission line inspection method and related equipment based on multi-source data. This method acquires inspection data from multiple dimensions, including high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. It also uses a multi-source feature model to identify defect types from multiple angles in the high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. Based on the identified defect types, the inspection line is arranged, which can increase the probability of discovering transmission line defects, increase the accuracy of identifying transmission line defects, and improve the response speed of inspection personnel.

[0028] The transmission line inspection method based on multi-source data provided in this application relates to the field of power inspection technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle-mounted terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the transmission line inspection method based on multi-source data, but is not limited to the above forms.

[0029] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0030] The following is a detailed description and explanation of the solutions in the embodiments of the present invention, using specific application examples: Figure 1This is an optional flowchart of the transmission line inspection method based on multi-source data provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S106.

[0031] Step S101: Obtain multi-source inspection data from multiple terminals. Multi-source inspection data refers to real-time data obtained when multiple terminals inspect transmission lines. Multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data.

[0032] The multi-source inspection data in this embodiment is acquired in real time by various terminal devices. These terminals allow for multi-dimensional inspection of transmission lines. The multi-source inspection data designed in this embodiment inspects transmission lines from four dimensions: a high-altitude inspection layer, a ground inspection layer, a near-body operation layer, and a fixed detection layer.

[0033] High-altitude inspection involves surveying power transmission lines from a perspective higher than the horizontal plane. This can be achieved by deploying industrial-grade multi-rotor drones and fixed-wing drones in coordinated formations. The multi-rotor drones (with a payload of ≥5kg, endurance of ≥40min, and equipped with a 4K visible light camera, infrared thermal imager, and lidar) are responsible for close-range, detailed scanning of key components such as tension towers, insulator strings, and conductor joints. The fixed-wing drones (with an endurance of ≥120min, cruising speed of ≥60km / h, and equipped with a high-resolution wide-field camera) are responsible for rapid, wide-area coverage of long-distance transmission line corridors (such as sections crossing mountains and rivers). The two types of drones are connected by pre-set flight paths to achieve high-altitude coordination of "wide-area inspection + detailed scanning of key areas".

[0034] Ground-level inspections involve examining power transmission lines from a ground perspective. This can be achieved by deploying wheeled / tracked power transmission line inspection robots, adaptable to different terrains (flatlands, mountains, hills). Equipped with high-definition cameras, partial discharge sensors, and temperature and humidity sensors, the robots move along the bases of towers, grounding grids, and vegetation areas along the lines to inspect for defects in ground equipment (such as cracks in tower foundations and corrosion of grounding grids) and potential safety hazards along the passageways (such as tree obstructions and debris accumulation). Simultaneously, they complement the "air-ground" perspective of high-altitude drones, avoiding missed defects due to obstructed views.

[0035] Performing near-site operational layer inspections involves inspecting transmission lines from a manual inspection perspective. Maintenance personnel can be equipped with handheld smart terminals (such as industrial tablets or AR glasses). These terminals have built-in AI models for identifying transmission line defects, supporting offline image capture and real-time analysis of insulators, hardware, and other components. They can also receive suspected defect data transmitted from high-altitude / ground terminals, assisting maintenance personnel in on-site verification and confirmation, thus solving the problem of "remote identification being questionable and on-site verification lacking evidence."

[0036] The implementation of fixed monitoring layer inspection involves collecting data on transmission lines in real time through fixed devices. Fixed monitoring terminals are deployed in key sections of the transmission lines (such as heavy icing areas, areas prone to lightning, and areas with severe pollution). These terminals include tower tilt sensors, conductor temperature sensors, insulator leakage current sensors, and video surveillance cameras. The terminals have the ability to collect data 24 hours a day and upload line operation status data in real time, providing basic data support for the priority scheduling of inspections by mobile terminals (drones, robots).

[0037] Since multi-source inspection data is obtained from different terminals, in order to utilize it, it is necessary to ensure the synchronous acquisition of multi-source inspection data. To determine the spatiotemporal data of the acquired multi-source inspection data, the following elements are required: Spatial Positioning Synchronization: All mobile terminals (drones, robots, handheld terminals) integrate a BeiDou high-precision positioning module (supporting BeiDou-3 B1I / B2I bands, positioning accuracy ≤1m) and are equipped with an IMU (Inertial Measurement Unit) for auxiliary positioning. To address obstructions during drone flight (such as trees or towers blocking BeiDou signals), a fusion algorithm of "BeiDou positioning + IMU inertial navigation" is adopted, using Kalman filtering to correct positioning errors and ensure spatial positioning accuracy is stable within ±0.5m. Fixed monitoring terminals use pre-calibrated latitude and longitude coordinates (error ≤0.3m) as a spatial reference. When mobile terminals collect data, they simultaneously record positioning information (latitude, longitude, and elevation) and link it to the power transmission line ledger (such as tower number, span, and line mileage) to achieve precise mapping between "data and physical location."

[0038] Time Synchronization: A two-layer time synchronization mechanism combining PTP (Precise Time Protocol, IEEE 1588) and NTP (Network Time Protocol) is adopted. A time server (time accuracy ≤10ns) is deployed at the transmission line inspection command center, and a time reference is distributed to each terminal via 5G / fiber optic network. Mobile terminals (drones, robots) support the PTP protocol, and the time synchronization error with the time server is controlled within ±1ms. Fixed monitoring terminals use the NTP protocol for time calibration, with an error ≤10ms. At the same time, all data collected by all terminals are timestamped (accurate to the millisecond level) to ensure that line data collected by different terminals at the same time can be directly correlated, avoiding errors in defect timing judgment due to time deviation (such as the time matching of partial discharge signals and infrared temperature anomalies).

[0039] In addition to determining the spatiotemporal data, this embodiment also describes the preprocessing procedure for multi-source inspection data, specifically including: For visible light / infrared images acquired by drones, inspection robots, or handheld terminals, the following steps are first performed: distortion correction (correcting lens distortion based on the camera intrinsic parameter matrix), illumination normalization (using histogram equalization or Retinex algorithm to eliminate image brightness differences under different lighting conditions), and denoising (using Gaussian filtering to remove random noise and bilateral filtering to preserve defect edge features). For point cloud data acquired by LiDAR, statistical filtering is used to remove outliers (such as dust interference points in the air), voxel grid downsampling is used to reduce point cloud density (reducing computation while ensuring the integrity of defect features), and ground segmentation algorithms are used to separate the line equipment point cloud (such as towers and conductors) from the background point cloud (such as trees and ground).

[0040] For electrical parameters (such as partial discharge and leakage current) and environmental parameters (such as temperature, humidity, and wind speed) collected by fixed monitoring terminals and robots, a sliding window filter (the window size is dynamically adjusted according to the data sampling frequency, such as using a window of 5 sampling points for 10Hz sampling data) is used to remove high-frequency noise. Interpolation algorithms (such as linear interpolation and cubic spline interpolation) are used to fill in missing data values ​​(such as sampling interruptions caused by temporary sensor failures). At the same time, the time series data is normalized (Min-Max normalization is used to map the data to the [0,1] interval) to eliminate the impact of differences in the magnitude of different parameters on the AI ​​model.

[0041] Step S102: Input the multi-source inspection data into the multi-source feature model respectively to determine the defect type corresponding to the multi-source inspection data.

[0042] In some embodiments, step S102 involves inputting multi-source inspection data into a multi-source feature model to determine the defect type corresponding to the multi-source inspection data, such as... Figure 2 As shown, it includes: Step S201: Determine the data type of the multi-source inspection data.

[0043] After acquiring multi-source inspection data, preprocessing is performed on the data. This preprocessing method depends on the data type. For example, if the multi-source inspection data is image data, image preprocessing is performed, specifically including: For visible light / infrared images acquired by UAVs and handheld terminals, the following steps are first performed: distortion correction (correcting lens distortion based on camera intrinsic parameter matrix), illumination normalization (using histogram equalization or Retinex algorithm to eliminate image brightness differences under different lighting conditions), and denoising (using Gaussian filtering to remove random noise and bilateral filtering to preserve defect edge features). For point cloud data acquired by LiDAR, statistical filtering is used to remove outliers (such as dust interference points in the air), voxel grid downsampling is used to reduce point cloud density (reducing computational load while ensuring the integrity of defect features), and ground segmentation algorithm is used to separate line equipment point clouds (such as towers and conductors) from background point clouds (such as trees and ground).

[0044] When the multi-source inspection data consists of electrical parameters (such as partial discharge and leakage current) and environmental parameters (such as temperature, humidity, and wind speed) collected by fixed monitoring terminals and robots, a sliding window filter is used (the window size is dynamically adjusted according to the data sampling frequency, such as using a window of 5 sampling points for 10Hz sampling data) to remove high-frequency noise. Interpolation algorithms (such as linear interpolation and cubic spline interpolation) are used to fill in missing data values ​​(such as sampling interruptions caused by temporary sensor failures). At the same time, the time-series data is normalized (using Min-Max normalization to map the data to the [0,1] interval) to eliminate the impact of differences in the magnitude of different parameters on the AI ​​model.

[0045] After preprocessing the multi-source inspection data, feature extraction is performed on multi-source inspection data of the same type based on the data type, obtaining the features of each type of terminal data. The final terminal data features are obtained by fusing multi-source inspection data of the same type.

[0046] Step 202: Input the multi-source inspection data into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model.

[0047] Multi-source inspection data are input into a multi-source feature model to extract defect features. The multi-source feature model contains multiple single-source feature models. Each single-source feature model can output the defect type of the transmission line and the confidence level of the defect type for the input multi-source inspection data.

[0048] Specifically, time-series data, image data, and point cloud data are obtained from multi-source inspection data; Time series data is input into the time series feature model to obtain the first reference defect and the first defect confidence level output by the time series feature model; image data is input into the image feature model to obtain the second reference defect and the second defect confidence level output by the image feature model; point cloud data is input into the point cloud feature model to obtain the third reference defect and the third defect confidence level output by the point cloud feature model.

[0049] For image data (visible light, infrared), single-source feature models can use a "CNN+Transformer" hybrid architecture to extract defect features. CNN (such as ResNet50, EfficientNet) is responsible for extracting local features (such as the edges of broken insulators and the texture of broken strands in conductors), while Transformer (such as the attention mechanism of ViT) is responsible for capturing global features (such as the overall posture anomaly of the insulator string and the sag change of the conductor). By concatenating the features of the two, an image feature vector (with a dimension of 1024) is formed.

[0050] For time-series data (partial discharge, temperature), LSTM (Long Short-Term Memory) or Transformer Encoder structures are used to extract time-series features (such as pulse features of partial discharge and trend change features of temperature), and the time-series features are mapped into 512-dimensional feature vectors through fully connected layers.

[0051] For point cloud data (LiDAR): PointNet++ network is used to extract local geometric features of the point cloud (such as the crack depth of the tower foundation and the cross-sectional deformation of the conductor) through hierarchical sampling and grouping operations, and convert them into 512-dimensional point cloud feature vectors.

[0052] The single-source feature model provided in this embodiment can identify defect categories based on multi-source inspection data of the same type. Based on rich terminal data features, it can effectively improve the accuracy of defect identification.

[0053] After obtaining different types of terminal data features, feature fusion is further performed on the terminal data features to obtain a fused feature vector.

[0054] A feature fusion module based on cross-attention is employed to achieve adaptive fusion of different types of features. First, image features, temporal features, and point cloud features are input into a feature alignment layer, and mapped to the same feature space (uniformly 1024 dimensions) through linear transformation. Then, a cross-attention mechanism is introduced to calculate the correlation weights between different features (e.g., the correlation between "abnormal insulator temperature" in image features and "increased leakage current" in temporal features), and the weight values ​​are normalized using a softmax function. Finally, the different features are weighted and summed according to the correlation weights to obtain a fused feature vector (1024 dimensions). This vector simultaneously contains visual features of the image, dynamic features of the temporal sequence, and geometric features of the point cloud, providing comprehensive feature support for subsequent defect identification. The cross-modal feature recognition provided in this embodiment can identify defect categories at the same location, such as line components like towers or conductors, based on inspection data from different terminals. Based on rich terminal data features, it can effectively improve the accuracy of defect identification.

[0055] In some embodiments, this application also includes a process for establishing a single-source feature model.

[0056] For example, this embodiment provides a process for building a single-source feature model for identifying insulator defects using image inspection data, including: Dataset Construction and Annotation: Insulator images were collected under various scenarios (covering different lighting, weather, and angles, with a sample size of ≥100,000 images). Rare defects such as "damage" and "spontaneous explosion" were expanded to ≥5,000 images / class through data augmentation (such as rotation, scaling, and noise addition). A combination of manual annotation and AI pre-annotation was adopted. The LabelImg tool was used to annotate the defect regions and categories, and PascalVOC format annotation files were generated. The dataset was divided into training set (70%), validation set (20%), and test set (10%).

[0057] Image preprocessing and feature enhancement: The original image is standardized to a uniform size of 512×512 pixels, and Z-score normalization (mean 0, standard deviation 1) is used to eliminate differences in pixel value magnitude; for the temperature features of the infrared image, the gradient features of the temperature heatmap (such as the edge gradient of the high temperature area) are extracted and fused with the texture features of the visible light image (such as the gray-level co-occurrence matrix) as supplementary input.

[0058] Model Architecture Design: Basic Backbone Network: An improved ResNet50 is used, removing the last 3 fully connected layers and retaining the first 4 convolutional modules to output a 512×16×16 local feature map (capturing details such as insulator damage edges and dirt textures); Global Feature Extraction: A Transformer encoder (6 layers, 8 self-attention heads) is connected after the backbone network to flatten the local feature map into a sequence (16×16=256 tokens), and the self-attention mechanism is used to capture the overall morphological association of the insulator string (such as the positional offset of adjacent insulators after self-explosion); Classification Head: The [CLS] token (representing global features) output by the Transformer is connected to 2 fully connected layers (hidden dimension 256), and finally the probability distribution (confidence) of 4 types of defects is output through softmax.

[0059] Training strategy and optimization: Loss function: Weighted cross-entropy loss is used (the classes with small sample sizes, such as "self-destruction" and "damage", are assigned a weight of 1.5 to balance the imbalance of samples); Optimizer: AdamW (initial learning rate 1e-4, decaying by 10% every 5 epochs), 50 training epochs, and an early stopping strategy is adopted (if the accuracy on the validation set does not improve for 5 consecutive epochs, the training is stopped); Regularization: Dropout (probability 0.3) and L2 regularization (weight decay of 1e-5) are added to suppress overfitting.

[0060] Model evaluation and iteration: Core metrics: accuracy (≥98%), F1 score for each type of defect ("damaged", "self-explosion" ≥95%, "dirty" ≥97%); Error analysis: For misclassified samples (e.g., "severely dirty" misclassified as "damaged"), supplement edge cases of this type of sample (e.g., dirt covering the damaged area), and retrain until the metrics meet the standards.

[0061] For example, this embodiment provides a process for building a single-source feature model for identifying line defects using time-series data, including: Time-series dataset construction and preprocessing: Time-series data collection: including conductor temperature (sampling frequency 1Hz, duration ≥100,000 hours) and partial discharge pulses (sampling frequency 10kHz, single segment duration 10s, ≥50,000 segments), divided into "normal / abnormal" labels, with "icing" data supplemented through low-temperature environment simulation experiments; Data cleaning: outliers (such as jumps caused by sensor failures) were removed using the 3σ criterion, and missing values ​​were filled using linear interpolation; Feature engineering: sliding window segmentation was performed on the time-series data (window size 100 sampling points, step size 50), and time-domain features (mean, variance, peak value, kurtosis) and frequency-domain features (main frequency components were extracted through FFT) of each window were extracted to form a 20-dimensional / window feature vector.

[0062] Model Architecture Design: Temporal Feature Extraction: A bidirectional LSTM network (2 layers, 128 hidden units per layer) is used to capture the dependencies between time series data (such as the slow temperature rise trend before an anomaly, and the pulse periodicity of partial discharge); Attention Mechanism: A temporal attention layer is connected after the LSTM output to assign weights to features at different times within the window (such as higher weight at the peak of the partial discharge pulse); Classification Head: The attention-weighted features are connected to a fully connected layer (64 hidden dimensions), and the probability distribution (confidence) of 4 types of anomalies is output through softmax.

[0063] Training strategy and optimization: Loss function: FocalLoss (focuses on difficult-to-classify samples, such as distinguishing between "early partial discharge" and "normal"); Optimizer: RMSprop (learning rate 5e-5, decay coefficient 0.9), 30 training epochs, using time series cross-validation (to avoid data leakage); Data augmentation: Gaussian noise (signal-to-noise ratio 30dB) and time stretching (±10%) are added to the time series data to improve the robustness of the model.

[0064] Model Evaluation and Iteration: Core metrics: Time series accuracy (≥96%), anomaly detection latency (≤5s, for slow-developing defects such as icing); Iterative optimization: For confused samples of "temperature anomaly" and "icing" (both are accompanied by temperature decrease), add environmental humidity features (icing requires high humidity), increase the humidity feature dimension in the input layer (from 20 to 21 dimensions), and retrain until the confusion rate is <3%.

[0065] For example, this embodiment provides a process for building a single-source feature model for identifying structural defects in towers using point cloud data, including: Point cloud dataset construction and preprocessing: Point cloud data collection: including laser point clouds of towers, fittings, and conductors (point cloud quantity ≥ 1 million points per scene, sample size ≥ 5000 scenes), where "fitting deformation" and "conductor strand breakage" are supplemented by simulation samples generated through 3D modeling; Point cloud cleaning: outliers are removed by statistical filtering (neighborhood points < 5 are considered noise), and the point cloud density is unified to 100,000 points / scene by voxel mesh downsampling (voxel size 5mm); Segmentation and annotation: the point clouds of towers, fittings, and conductors are separated by a RANSAC-based planar segmentation algorithm, and the defect categories (such as tower tilt angle and fitting deformation) are manually annotated.

[0066] Model Architecture Design: Geometric Feature Extraction: PointNet++ network is adopted. Key point sets are selected through sampling layer (FPS algorithm), local regions are constructed through grouping layer (BallQuery), and local geometric features (such as point cloud normal vectors, curvature, and distance distribution) are extracted through MLP layer; Hierarchical Feature Fusion: Local features are hierarchically aggregated (from 100,000 points → 10,000 points → 1,000 points) through SetAbstraction module to capture multi-scale features from details (point cloud missing of broken conductor strands) to the global (overall tilt of towers); Classification Head: The final global features (1024 dimensions) are connected to a fully connected layer (hidden dimension 256), and the probability distribution (confidence) of 4 types of structural defects is output through softmax.

[0067] Training strategy and optimization: Loss function: Combine cross-entropy loss (classification) and mean squared error loss (auxiliary regression, such as predicting tower tilt angle) to improve defect localization accuracy; Optimizer: SGD (momentum 0.9, learning rate 1e-3, cosine annealing scheduling), 40 training rounds, using random point order augmentation (shuffling the input order of point cloud to adapt to the disorder of point cloud); Regularization: Add Dropout (probability 0.2) and random point drop (5% probability of deleting non-critical point cloud) to simulate LiDAR occlusion scenarios.

[0068] Model Evaluation and Iteration: Core Indicators: Classification accuracy (≥97%), tower tilt angle prediction error (≤0.5°), fitting deformation recognition IoU (≥0.85); Iterative Optimization: To address the issue of missed detection of conductor strand breaks, a "point cloud density mutation" feature (point cloud density drops sharply at the break point) was added to the local feature layer of PointNet++. After retraining, the missed detection rate decreased from 5% to 1%.

[0069] Step 203: Using a decision fusion method, reference defects and defect confidence from the single-source feature model are fused to determine the defect category.

[0070] After each single-source feature model outputs the defect type and confidence level of the transmission line, the output results of multiple unit feature models can be combined to determine the final defect type. Specifically, the confidence level of the single-source feature model is converted into a basic probability assignment (BPA) function. For example, if the confidence level of the single-source feature model identifying "insulator damage" in image data is 0.8, it is converted into a BPA value m(damaged) = 0.8 and m(other) = 0.2. Then, the conflict coefficient between different pieces of evidence is calculated (to determine the consistency of the output results of each single-source feature model; the smaller the conflict coefficient, the higher the consistency). If the conflict coefficient is ≤0.5, the evidence is directly fused using the DS synthesis rule. If the conflict coefficient is >0.5, a weighting factor is introduced (dynamically adjusted according to the historical accuracy of the single-source feature model, such as 0.6 for an image model with 95% accuracy and 0.4 for a time series model with 90% accuracy) to reduce the impact of conflicting evidence before fusion. Finally, based on the fused BPA function, the defect type with the highest confidence level is selected as the final diagnostic result. If the highest confidence level is <0.6, it is marked as a "suspected defect," triggering other verification processes.

[0071] Step S103: Based on the defect type and scene characteristics obtained from multi-source inspection data, generate the inspection route of the transmission line.

[0072] In some embodiments, the identified defect types of the transmission lines can be used to generate inspection routes for the transmission lines, instructing users to reach the defect locations quickly and address the defects in a timely manner.

[0073] Step S103: Based on the defect type and scene characteristics obtained from multi-source inspection data, generate the inspection route of the transmission line, such as... Figure 3 As shown, it includes: Step S301: Determine the operating characteristics of the transmission line under the defect type.

[0074] Step S302: Using operational characteristics and scene characteristics obtained from multi-source inspection data, the line priority of the transmission line is obtained through the inspection model.

[0075] Step 303: Based on the line priority of the transmission line, use the A* algorithm to generate the initial inspection line of the transmission line; Step 304: Combine the reward function generated based on line priority to optimize the initial inspection path and obtain the inspection line of the transmission line.

[0076] By utilizing various scenario data related to power transmission lines, environmental data, line operation data, and terminal status data along the transmission line are determined. Environmental data represents the scenario characteristics of the transmission line, while line operation data and terminal status data represent the operational characteristics of the transmission line under different defect types. Environmental data includes, for example, weather conditions such as wind speed, precipitation, and visibility; and terrain conditions such as tower locations and the distribution of obstacles along the line. Line operation data includes conductor load rate, insulator pollution level, and historical defect frequency. Terminal status data includes drone battery life, robot battery level, and equipment health status.

[0077] Since there are many types of data from various scenarios, in order to reduce data redundancy, this embodiment can also filter key data from the scenario data. Specifically, a feature selection algorithm based on gradient boosting tree (XGBoost) can be used to filter key scenario features (such as wind speed >10m / s, load rate >80%, historical defect frequency >3 times / year, and drone endurance <20min) from the collected multi-dimensional scenario data to eliminate the interference of redundant features.

[0078] After obtaining the key scene features, this embodiment can output the inspection route through the inspection model.

[0079] This embodiment provides a process for building an inspection model, specifically including: (1) Data preparation and labeling Data Acquisition: Collect multi-dimensional data of each section of the transmission line (based on towers), with a sample size of ≥8000 sets, covering three key scenario characteristics: environmental data characteristics (such as wind speed, visibility, temperature and humidity, precipitation), line operation data characteristics (such as conductor load rate, insulator pollution level, span), and historical defect characteristics (such as defect frequency, defect type, and handling time in the past 2 years), totaling 18 input features.

[0080] Labeling: According to the operation and maintenance specifications, the priority labeling standards are defined as high priority (e.g., wind speed > 18m / s + historical icing defects, load rate > 90% + partial discharge records), medium priority (e.g., wind speed 10-18m / s, medium pollution level), and low priority (e.g., clear weather + load rate < 60% + no historical defects). The labels are cross-labeled by 3 senior operation and maintenance personnel. Inconsistent samples are confirmed through on-site verification. Finally, the training set, validation set, and test set are divided into 7:2:1.

[0081] Data balancing: To address the issue of the low proportion of high-priority samples (approximately 15%), the SMOTE algorithm is used to generate synthetic samples, expanding the high-priority samples to 25% of the total samples, thus preventing the model from being biased towards the majority class.

[0082] (2) Feature data processing Data cleaning: Outliers (such as wind speed jumps caused by sensor failures) are removed using the 3σ criterion, and missing values ​​are filled using K-nearest neighbor interpolation (K=5); daily average values ​​are used for time-series features (such as load rate) to reduce noise interference.

[0083] Feature standardization: Numerical features (such as wind speed and load rate) are normalized using Z-score (mean 0, standard deviation 1) to eliminate differences in magnitude; categorical features (such as defect type) are converted into vector format using one-hot encoding.

[0084] Feature selection: XGBoost is used to calculate feature importance, retain the top 12 features (such as wind speed, load rate, historical defect frequency, and pollution level), and remove redundant features (such as air pressure, importance < 0.02) to reduce the computational load of the model.

[0085] (3) Architecture design It adopts an "MLP + attention mechanism" architecture to adapt to structured scene features: Input layer: 12 neurons, corresponding to the 12 key features selected.

[0086] Attention layer: 32 hidden units, with each feature weight calculated using Softmax (e.g., historical defect frequency weight 0.25, wind speed weight 0.2), dynamically enhancing the role of high-impact features.

[0087] Hidden layers: There are 3 fully connected layers with 128, 64 and 32 neurons respectively, all using the ReLU activation function; Dropout (probability 0.3) and L2 regularization (weight decay 1e-5) are added to each layer to suppress overfitting.

[0088] Output layer: 3 neurons, using the Softmax activation function, outputting high, medium, and low priority probability distributions.

[0089] (4) Model training and optimization Loss function: Weighted cross-entropy loss is used, with high, medium and low priority weights set to 1.5, 1.0 and 0.8 respectively to balance the sample distribution.

[0090] Optimizer and Scheduling: The AdamW optimizer (initial learning rate 1e-4) is used, combined with cosine annealing learning rate scheduling (decreasing by 10% every 5 rounds); the training run consists of 50 rounds, and an early stopping strategy is enabled (the training stops if the accuracy on the validation set does not improve for 5 consecutive rounds).

[0091] Evaluation and iteration: With overall accuracy (≥93%) and high-priority recall (≥96%) as the core indicators, after the test set meets the standards, 1,000 sets of real-time data are added every quarter for incremental training (freezing the underlying parameters and fine-tuning the attention layer and output layer) to adapt to changes in the route scenario.

[0092] When iteratively training the patrol model, constraints are important parameters that serve as boundary limits for path optimization, including: Equipment constraints: Drone endurance (single flight ≤ 40 min, with 10 min of battery charge reserved for return), robot movement speed (flat ground ≤ 3 km / h, mountainous ground ≤ 1.5 km / h), handheld terminal operating range (maintenance personnel's single inspection radius ≤ 5 km). Environmental constraints: Drones must cease flight when visibility is <500m; robots must cease movement when wind speed is >20m / s; all outdoor terminals must suspend operations during thunderstorms. Task constraints: When multiple terminals need to conduct collaborative inspections in the same section (such as drones + robots), the path must meet the following requirements: time connection (after the drone completes the high-altitude scan, the robot arrives at the ground work point within 30 minutes) and complementary coverage (the drone covers the upper part of the tower, and the robot covers the lower part of the tower, with no overlap and no omissions).

[0093] When iteratively training the inspection model, it is also necessary to define the state space and action space, where: State space S: includes the inspection priority of the line segment, the current location of the terminal, the remaining resources of the terminal (drone endurance, robot battery power), the set of inspected segments, and environmental constraints (such as weather and wind speed). It is represented in vector form, and the dimensions are dynamically adjusted according to the line scale (e.g., for a line with 100 poles, the state dimension is approximately 200). Action Space A: The actions of each terminal include "going to the next segment", "staying in the current segment for fine cleaning", "returning to the base station for power replenishment", and "cooperating with other terminals". For example, the drone's action A1 = {going to tower 10, staying at tower 9 for fine cleaning, returning to the base station, guiding the robot to tower 8}, and the robot's action A2 = {going to tower 8, staying at tower 7 for fine cleaning, returning to the base station, cooperating with the drone to clean tower 9}.

[0094] A reward mechanism will be introduced, with reward types including: Priority Reward R1: When the terminal patrols a high-priority segment, it receives a positive reward (R1=5), a reward of R1=2 when patrolling a medium-priority segment, a reward of R1=1 when patrolling a low-priority segment, and a negative reward (R1=-3) when patrolling without following priority (e.g., patrolling a low-priority segment first and then a high-priority segment). Coverage Reward R2: A positive reward (R2=10) is given when the terminal completes full coverage of the segment, a negative reward (R2=-5) is given when there is a coverage omission, and a reward of R2=-2 is given when the coverage is repeated; Resource Reward R3: A positive reward (R3=3) is given when the remaining resources of the terminal (such as the drone's battery life) meet the task requirements, and a negative reward (R3=-10) is given when the task is interrupted due to the depletion of resources. Collaboration Reward R4: A positive reward (R4=4) is given when multiple terminals collaborate to complete a section inspection (e.g., drones and robots have no time conflict and complementary coverage), and a negative reward (R4=-6) is given when there is a collaboration conflict (e.g., drones and robots occupy the same airspace at the same time).

[0095] Path balance reward function R=α R1+β R2+γ R3+δ R4, where α, β, γ, and δ are weighting coefficients (α=0.4, β=0.3, γ=0.2, and δ=0.1 obtained through grid search optimization. The weighting coefficient values ​​are based on the "core target priority" and "technical constraint reality" of transmission line inspection. The optimal balance scheme is determined after verification through grid search. In essence, it transforms the business logic of "safety first, coverage is king, resource controllability, and collaborative efficiency" in the operation and maintenance scenario into quantifiable weights that can be executed by the algorithm).

[0096] An improved A* algorithm is used to generate the initial path for each terminal: starting from the "segment with the highest patrol priority" and ending at the "base station location", the shortest path is planned in combination with constraints (such as drone endurance). At the same time, the path is adjusted through a conflict detection mechanism (such as spatiotemporal window detection to determine whether different terminals occupy the same space at the same time) to avoid conflicts between multiple terminals.

[0097] Please see Figure 4 This application also provides a transmission line inspection device based on multi-source data, which can implement the above-mentioned method. The device includes: The acquisition module 41 is used to acquire multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data acquired when multiple terminals inspect transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The determination module 42 is used to input the multi-source inspection data into the multi-source feature model respectively, and determine the defect type corresponding to the multi-source inspection data. The multi-source feature model corresponds to the multi-source inspection data. The generation module 43 is used to generate the inspection route of the transmission line based on the defect type and the scene features obtained from the multi-source inspection data.

[0098] In some embodiments, the multi-source feature model includes at least two single-source feature models, and the determining module 42 is configured to: Determine the data type of the multi-source inspection data; The multi-source inspection data is input into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model; The defect category is determined by fusing the reference defect and the defect confidence from the single-source feature model using a decision fusion method.

[0099] In some embodiments, the generation module 43 is configured to: Determine the operating characteristics of the transmission line under the defect type; By utilizing the operational characteristics and scene characteristics obtained from the multi-source inspection data, the line priority of the transmission line is obtained through the inspection model. Based on the line priority of the transmission line, the initial inspection line of the transmission line is generated using the A* algorithm; By combining the reward function generated based on the line priority, the initial inspection path is optimized to obtain the inspection route of the transmission line.

[0100] In some embodiments, the single-source feature model includes: a temporal feature model, an image feature model, and a point cloud feature model; the determination module 42 is used for: Time-series data, image data, and point cloud data are obtained from the multi-source inspection data; The time-series data is input into the time-series feature model to obtain the first reference defect and the first defect confidence level output by the time-series feature model; the image data is input into the image feature model to obtain the second reference defect and the second defect confidence level output by the image feature model; the point cloud data is input into the point cloud feature model to obtain the third reference defect and the third defect confidence level output by the point cloud feature model.

[0101] In some embodiments, the determining module 42 is configured to: The first defect confidence level, the second defect confidence level, and the third defect confidence level are respectively converted into basic probability assignment functions; Based on the probability allocation function, the defect consistency among the first reference defect, the second reference defect, and the third reference defect is calculated; If the defect consistency is less than or equal to the defect threshold, the basic probability allocation function is fused using the DS fusion algorithm to determine the defect category.

[0102] In some embodiments, the determining module 42 is configured to: If the defect consistency is greater than the defect threshold, obtain the historical accuracy of the time-series feature model, the image feature model, and the point cloud feature model; Using the historical accuracy, the weight factors of the temporal feature model, the image feature model, and the point cloud feature model are modified; The first reference defect, the second reference defect, the third reference defect, the first defect confidence, the second defect confidence, and the third defect confidence are obtained again using the temporal feature model after modifying the weight factors, the image feature model, and the point cloud feature model, until the defect consistency among the first defect confidence, the second defect confidence, and the third defect confidence is less than or equal to the defect threshold.

[0103] In some embodiments, the generation module 43 is configured to: The first reward score and first reward level for the route priority are determined; the second reward score and second reward level for inspection coverage are determined; the third reward score and third reward level for inspection resources are determined; and the fourth reward score and fourth reward level for collaborative inspection are determined. Based on the first reward score and the first reward level, the second reward score, the second reward level, the third reward score, the third reward level, the fourth reward score, and the fourth reward level, the initial inspection path is optimized to obtain the inspection path of the transmission line.

[0104] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0105] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0106] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0107] Please see Figure 5 , Figure 5 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 501 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 502 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 502 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501 using the methods described in the embodiments of this application. The input / output interface 503 is used to implement information input and output; The communication interface 504 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 505 transmits information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504); The processor 501, memory 502, input / output interface 503, and communication interface 504 are connected to each other within the device via bus 505.

[0108] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0109] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0110] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0111] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0112] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0113] The transmission line inspection method, apparatus, electronic device, storage medium, and program product based on multi-source data provided in this application acquire multi-source inspection data from multiple terminals. This multi-source inspection data consists of real-time data obtained during multi-terminal inspections of transmission lines, including high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The multi-source inspection data is input into a multi-source feature model to determine the corresponding defect types. Based on the defect types and scene features obtained from the multi-source inspection data, an inspection route for the transmission line is generated. This application acquires inspection data from multiple dimensions, including high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data, and uses a multi-source feature model to identify defect types from multiple angles within these data. Arranging inspection routes based on the identified defect types increases the probability of discovering transmission line defects, improves the accuracy of defect identification, and enhances the response speed of inspection personnel.

[0114] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0115] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0118] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0119] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0121] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0122] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0124] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for inspecting transmission lines based on multi-source data, characterized in that, The method includes: Acquire multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data obtained when multiple terminals inspect transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The multi-source inspection data are input into the multi-source feature model to determine the defect type corresponding to the multi-source inspection data; Based on the defect type and the scene characteristics obtained from the multi-source inspection data, the inspection route of the transmission line is generated.

2. The method according to claim 1, characterized in that, The multi-source feature model includes at least two single-source feature models. The step of inputting the multi-source inspection data into the multi-source feature model respectively to determine the defect type corresponding to the multi-source inspection data includes: Determine the data type of the multi-source inspection data; The multi-source inspection data is input into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model; The defect category is determined by fusing the reference defect and the defect confidence from the single-source feature model using a decision fusion method.

3. The method according to claim 1, characterized in that, The step of generating the inspection route of the transmission line based on the defect type and scene features obtained from the multi-source inspection data includes: Determine the operating characteristics of the transmission line under the defect type; By utilizing the operational characteristics and scene characteristics obtained from the multi-source inspection data, the line priority of the transmission line is obtained through the inspection model. Based on the line priority of the transmission line, the initial inspection line of the transmission line is generated using the A* algorithm; By combining the reward function generated based on the line priority, the initial inspection path is optimized to obtain the inspection route of the transmission line.

4. The method according to claim 2, characterized in that, The single-source feature model includes: a time-series feature model, an image feature model, and a point cloud feature model; the step of inputting the multi-source inspection data into the single-source feature model according to the data type to obtain the reference defect and defect confidence corresponding to the single-source feature model includes: Time-series data, image data, and point cloud data are obtained from the multi-source inspection data; The time-series data is input into the time-series feature model to obtain the first reference defect and the first defect confidence level output by the time-series feature model; the image data is input into the image feature model to obtain the second reference defect and the second defect confidence level output by the image feature model; the point cloud data is input into the point cloud feature model to obtain the third reference defect and the third defect confidence level output by the point cloud feature model.

5. The method according to claim 4, characterized in that, Using a decision fusion method, the reference defect and the defect confidence from the single-source feature model are fused to determine the defect category, including: The first defect confidence level, the second defect confidence level, and the third defect confidence level are respectively converted into basic probability assignment functions; Based on the probability allocation function, the defect consistency among the first reference defect, the second reference defect, and the third reference defect is calculated; If the defect consistency is less than or equal to the defect threshold, the basic probability allocation function is fused using the DS fusion algorithm to determine the defect category.

6. The method according to claim 5, after calculating the defect consistency among the first reference defect, the second reference defect, and the third reference defect according to the probability allocation function, the method further includes: If the defect consistency is greater than the defect threshold, obtain the historical accuracy of the time-series feature model, the image feature model, and the point cloud feature model; Using the historical accuracy, the weight factors of the temporal feature model, the image feature model, and the point cloud feature model are modified; The first reference defect, the second reference defect, the third reference defect, the first defect confidence, the second defect confidence, and the third defect confidence are obtained again using the temporal feature model after modifying the weight factors, the image feature model, and the point cloud feature model, until the defect consistency among the first defect confidence, the second defect confidence, and the third defect confidence is less than or equal to the defect threshold.

7. The method according to claim 3, characterized in that, The step of optimizing the initial inspection path by combining the reward function generated based on the line priority to obtain the inspection route of the transmission line includes: The first reward score and first reward level for the route priority are determined; the second reward score and second reward level for inspection coverage are determined; the third reward score and third reward level for inspection resources are determined; and the fourth reward score and fourth reward level for collaborative inspection are determined. Based on the first reward score and the first reward level, the second reward score, the second reward level, the third reward score, the third reward level, the fourth reward score, and the fourth reward level, the initial inspection path is optimized to obtain the inspection path of the transmission line.

8. A transmission line inspection device based on multi-source data, characterized in that, The device includes: The acquisition module is used to acquire multi-source inspection data from multiple terminals. The multi-source inspection data is real-time data acquired when multiple terminals inspect transmission lines. The multi-source inspection data includes high-altitude inspection data, ground inspection data, manual inspection data, and fixed monitoring data. The determination module is used to input the multi-source inspection data into the multi-source feature model respectively, and determine the defect type corresponding to the multi-source inspection data, wherein the multi-source feature model corresponds to the multi-source inspection data; The generation module is used to generate the inspection route of the transmission line based on the defect type and the scene characteristics obtained from the multi-source inspection data.

9. An electronic device / computer apparatus, characterized in that, The electronic device / computer apparatus includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images