Communication optical cable line intelligent inspection fault point rapid positioning method and device

Abstract

The invention relates to the technical field of communication engineering, in particular to a communication optical cable line intelligent inspection fault point rapid positioning method and device, and the method comprises the steps: obtaining optical cable line state data, constructing an intelligent fault detection model, recognizing an abnormal signal through wavelet transform and a threshold determination method, training an LSTM neural network in combination with historical fault data, and pre-judging a fault type. When a fault is detected, an unmanned aerial vehicle inspection unit is triggered, an inspection path is planned based on an improved Dijkstra algorithm according to the fault type and suspected fault area geographic information, an optical cable line is scanned and detected through a laser radar, physical fault features are recognized through a YOLOv5 algorithm, and geographic coordinates of a fault point are calibrated in combination with an inertial navigation system; and fusing sensor data and a detection result, evaluating a fault confidence level by adopting a DS evidence theory, and generating an alarm work order. Therefore, the problems of long positioning time consumption, high omission ratio, difficulty in distinguishing fault types and the like in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the field of communication engineering technology, and in particular to a method and device for quickly locating fault points during intelligent inspection of communication optical cable lines. Background Art

[0002] As communication networks evolve toward higher speeds and greater intelligence, the scale and complexity of optical fiber cables, the physical carriers of information transmission, are growing exponentially. These cables extend from dense urban pipelines to long-distance trunk lines in the wild, supporting multimodal communication needs such as voice, data, and video. In this context, real-time inspection of optical fiber cables and rapid fault location have become critical components for ensuring the stability of communication networks. Highly accurate, robust, and intelligent fault location technologies are urgently needed to address diverse fault scenarios such as fiber breaks, bends, and splice losses in complex environments.

[0003] However, current communication optical cable line inspection and fault location technologies have significant shortcomings. Traditional manual inspections are restricted by the geographical environment, with slow responses and high missed detection rates. Traditional OTDR technology uses a single wavelength for detection and is sensitive to optical fiber bending loss and ambient temperature, making it difficult to distinguish fault types and positioning accuracy exceeding 10 meters. The existing system has weak automated calibration capabilities and a high proportion of manual repeated calibration costs. When dealing with complex scenarios, it lacks multi-wavelength signal fusion and dynamic environmental compensation mechanisms. Positioning takes a long time and has a false alarm rate of 15%, making it difficult to meet real-time, high-precision and high-reliability operation and maintenance requirements. Summary of the Invention

[0004] The present application provides a method and device for quickly locating fault points in intelligent inspection of communication optical cable lines to solve the problems of long positioning time, high missed detection rate, and difficulty in distinguishing fault types in the prior art.

[0005] The first embodiment of the present application provides a method for quickly locating fault points of a communication optical cable line through intelligent inspection, comprising the following steps: Step A1: obtaining status data such as vibration frequency, temperature gradient, and optical power attenuation of the optical cable line; Step A2: constructing an intelligent fault detection model based on the status data, identifying abnormal signals through wavelet transform and threshold judgment method, training LSTM neural network in combination with historical fault data, and pre-judging fault types such as optical cable breakage, external force damage, and optical fiber aging; Step A3: when a fault warning is detected, triggering the drone inspection unit, and using the improved method to detect faults based on the fault type and geographical information of the suspected fault area. The Dijkstra algorithm is used to generate a target inspection path. The drone's onboard laser radar and AI vision module, combined with the target inspection path, scans and inspects the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. Combined with the inertial navigation system, the geographic coordinates of the fault point are quickly calibrated. Step A4: The edge computing node fuses the sensor data with the drone's detection results, analyzes the fault data using DS evidence theory, determines the confidence level of the fault point, generates an alarm work order containing the fault type, location coordinates, and severity, and pushes it to the operation and maintenance platform.

[0006] In a preferred embodiment, an intelligent fault detection model is constructed based on the state data, and abnormal signals are identified by wavelet transform and threshold judgment method. The specific steps are as follows: Step B1, the acquired vibration frequency signal Perform continuous wavelet transform to obtain wavelet coefficients , the formula is ,in, is a normalization factor that ensures that the transformation maintains energy consistency at different scales a, , is the original function, b is the translation parameter, is the mother wavelet function, is the differential symbol; Step B2, set the threshold ,in is the mean of the wavelet coefficients, is the standard deviation, k is the threshold coefficient (k ),like , it is determined to be an abnormal signal.

[0007] In a preferred embodiment, the method for determining the threshold coefficient k in step B2 includes: step B201, collecting historical abnormal signal data to build a training set, and calculating the false positive rate under different k values and missed detection rate ; Step B202, with As the goal, the optimal k value is determined by the golden section method.

[0008] In a preferred embodiment, the target inspection path is generated by using the improved Dijkstra algorithm. The specific steps are as follows: Step C1, abstract the geographical information of the suspected fault area into a weighted graph , where V is the node set (tower location), E is the edge set (cable segment), W is the edge weight matrix, and weight Calculated by the linear weighted combination formula , is the straight-line distance between nodes i and j, is the terrain complexity factor (value range [0,1]), , is the weight coefficient; step C2, taking the starting tower S as the source point, use the Dijkstra algorithm to find the minimum weight path to the target tower D.

[0009] In a preferred embodiment, an alarm work order containing the fault type, location coordinates, and severity is generated and pushed to the operation and maintenance platform. The specific steps are as follows: Step D1: Based on the fault type and coordinates output by the fault assessment model, combined with optical cable equipment parameters (such as fiber core type, fusion point location) and terrain characteristics, a multi-dimensional report containing the fault severity and repair suggestions is generated. Step D2: The PSO algorithm is used to optimize the repair path in the multi-dimensional report. The formula is: , ,in, is the velocity of particle i in generation t; w is the inertia weight; , is the learning factor; , is a random number; is the best historical position of particle i; is the global optimal position; The optimal fault repair reference path is obtained by taking the position of particle i in the tth generation. Step D3: Convert the fault location report and the optimal repair path into visual instructions, push them to the operation and maintenance platform in real time through the edge computing node, and support real-time viewing and navigation on the mobile app.

[0010] In a preferred embodiment, an LSTM neural network is trained in combination with historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging. The specific steps are as follows: Step E1: Extract feature vectors such as signal intensity fluctuation, temperature change rate, and optical power attenuation slope to construct a training data set; Step E2: Use five-fold cross-validation to optimize the LSTM network hyperparameters, including the number of hidden layers, the number of neurons, and the learning rate. The optimization goal is to minimize the classification loss function: , where M is the number of samples, C is the number of fault types, is the true label, is the predicted probability, is the loss value, i is the sample index, c is the category index, and ln is the natural logarithm; Step E3: Evaluate the model performance through accuracy and F1 score. The formula is: , ,in, is the accuracy rate, TP is the positive class, the number of samples predicted by the model as positive, TN is the negative class, the number of samples predicted by the model as negative, FP is the negative class, the number of samples incorrectly predicted as positive by the model, FN is the positive class, the number of samples incorrectly predicted as negative by the model, F1 is the F1 score, the value range is [0,1]. The closer to 1, the more balanced and excellent the model's performance in precision and recall is. Precision is the proportion of samples predicted as positive that are actually positive, and Recall is the proportion of samples actually positive that are correctly predicted as positive.

[0011] In a preferred embodiment, the laser radar and AI vision module carried by the drone are combined with the target inspection path to scan and detect the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. Combined with the inertial navigation system, the geographical coordinates of the fault point are quickly calibrated. The specific steps are as follows: Step F1: The laser radar point cloud data is mapped to the geographic coordinate system using the coordinate system conversion formula: , where R is the rotation matrix and T is the translation vector. is the sensor coordinate system coordinate, is the geographic coordinate; Step F2: The AI ​​vision module outputs the pixel coordinates of the fault target through the YOLOv5 algorithm , combined with the camera intrinsic parameter matrix K to inversely calculate the spatial coordinates: , where d is the target distance.

[0012] In a preferred embodiment, fault data decision-level fusion is performed through DS evidence theory. The specific steps are as follows: Step G1: Establish a fault identification framework Θ, which includes fault types such as optical cable breakage, external force damage, and optical fiber aging; Step G2: Calculate the basic probability distribution function m(A) of each fault type based on sensor data and drone detection results, where A is a subset of Θ; Step G3: Calculate the joint basic probability distribution function using the DS synthesis rule, which is , where C is the fused fault type, and are the basic probability distribution functions for sensor data and drone detection results, respectively. is the basic probability distribution function obtained after combination, A is the variable set The object of action, B is the variable set The object of action, is an empty set; Step G4: Determine the confidence level of the fault point according to the joint basic probability distribution function.

[0013] In a preferred embodiment, when determining the confidence level of the fault point, a conflict degree index is introduced. : ,when When >0.7, the confidence level of the fault point is high; when When <0.4, it is judged as low, among which, To identify the framework, and for two different basic probability distribution functions.

[0014] The third embodiment of the present application provides a construction device for a communication optical cable line intelligent inspection fault point rapid positioning system, including: an acquisition module for acquiring state data such as vibration frequency, temperature gradient, and optical power attenuation of the optical cable line; an identification module for constructing an intelligent fault detection model based on the state data, identifying abnormal signals through wavelet transform and threshold judgment method, training LSTM neural network in combination with historical fault data, and pre-judging fault types such as optical cable breakage, external force damage, and optical fiber aging; a calibration module for triggering a drone inspection unit when a fault warning is detected, and using the fault type and geographic information of the suspected fault area to detect the fault. An improved Dijkstra algorithm is used to generate a target inspection path. The laser radar and AI vision module carried by the drone are combined with the target inspection path to scan and inspect the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. Combined with the inertial navigation system, the geographic coordinates of the fault point are quickly calibrated. A generation module is used to fuse sensor data with the drone detection results at the edge computing node, and the fault data decision-level fusion is performed through the DS evidence theory to determine the confidence level of the fault point. An alarm work order containing the fault type, location coordinates, and severity is generated and pushed to the operation and maintenance platform.

[0015] The beneficial effects of the present invention are as follows: by acquiring status data such as the vibration frequency, temperature gradient, and optical power attenuation of optical cable lines, an intelligent fault detection model integrating wavelet transform and LSTM neural network is constructed to pre-judge the fault type. Upon detecting an early warning, a drone inspection unit is triggered, an improved Dijkstra algorithm is used to generate a target inspection path, and an AI vision module (YOLOv5 algorithm) is used to identify physical fault characteristics and calibrate coordinates using an inertial navigation system. The data is then fused through edge computing nodes using DS evidence theory to determine the fault confidence level and generate an alarm work order. This shortens the average time required for fault location, reduces the false alarm rate, and improves the real-time, accuracy, and automation level of communication optical cable line operation and maintenance. This solves the problems of long positioning time, high missed detection rate, and difficulty in distinguishing fault types in existing technologies.

[0016] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of a process for quickly locating fault points in intelligent inspection of communication optical cable lines according to an embodiment of the present application; Figure 2 A schematic diagram of a long-distance optical fiber communication system provided according to an embodiment of the present application; Figure 3 This is an example diagram of an optical cable intelligent inspection system for a provincial-level communication operator provided according to one embodiment of the present application; Figure 4 This is a diagram illustrating an example of a power inspection scenario system according to one embodiment of the present application; Figure 5 This is an example diagram of a drone inspection operation provided according to one embodiment of the present application; Figure 6 This is an example diagram of a power line inspection scenario provided according to one embodiment of the present application; Figure 7 This is an example diagram of a regional power grid fault repair task provided according to one embodiment of the present application; Figure 8 This is a flow chart of a method for quickly locating fault points in intelligent inspection of communication optical cable lines according to one embodiment of the present application; Figure 9 A schematic diagram of a structure of a construction device for a communication optical cable line intelligent inspection fault point rapid location system provided in accordance with an embodiment of the present application; DETAILED DESCRIPTION

[0018] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0019] The following describes, with reference to the accompanying drawings, a method and apparatus for rapidly locating fault points in intelligent inspections of communication optical cable lines according to an embodiment of the present application. To address the difficulty in distinguishing fault types mentioned in the background art, the present application provides a method for rapidly locating fault points in intelligent inspections of communication optical cable lines. This method obtains state data such as the vibration frequency, temperature gradient, and optical power attenuation of the optical cable line, constructs an intelligent fault detection model that integrates wavelet transform and LSTM neural network to pre-determine the fault type. Upon detecting an early warning, a drone inspection unit is triggered. An improved Dijkstra algorithm is used to generate a target inspection path, and an AI vision module (YOLOv5 algorithm) is used to identify physical fault characteristics and an inertial navigation system is used to calibrate coordinates. The data is then fused by edge computing nodes using DS evidence theory to determine the fault confidence level and generate an alarm work order. This method shortens the average time required for fault location, reduces the false alarm rate, and improves the real-time, accuracy, and automation level of communication optical cable line operation and maintenance. This method solves the problems of long positioning time, high missed detection rate, and difficulty distinguishing fault types in the prior art.

[0020] Specifically, Figure 1 A flow chart of a method for quickly locating fault points in intelligent inspection of communication optical cable lines provided in an embodiment of the present application.

[0021] like Figure 1 As shown, the method for quickly locating fault points in intelligent inspection of communication optical cable lines includes the following steps: In step A1, status data such as vibration frequency, temperature gradient, and optical power attenuation of the optical cable line are obtained.

[0022] Among them, optical power attenuation refers to the phenomenon that when an optical signal is transmitted in a transmission medium such as an optical fiber, the optical power intensity gradually decreases as the transmission distance increases due to factors such as medium absorption, scattering, bending loss and connector loss.

[0023] It can be understood that the embodiments of the present application quantify the degree of decrease in power intensity of optical signals due to medium loss during transmission, identify abnormal loss sections in optical fiber links, provide key data for fault prediction of optical cable lines, and provide early warning of transmission performance degradation, so that operation and maintenance personnel can quickly locate sections with excessive loss and carry out targeted repairs, thereby improving inspection efficiency and the accuracy of fault handling.

[0024] For example, Figure 2As shown in the figure, in long-distance fiber-optic communication systems, after an optical signal originates from the transmitter, its optical power gradually decreases with increasing transmission distance due to factors such as absorption of specific wavelengths by the fiber (such as infrared absorption by hydroxyl ions in quartz glass), Rayleigh scattering (scattering losses caused by the interaction of light with the fiber's internal molecular structure), mode leakage caused by excessive bending (such as bending losses when the fiber's curvature radius is less than a critical value), and contact mismatch at splices or connectors (such as coupling losses caused by end-face contamination and misalignment). For example, after 100 kilometers of transmission in a cross-regional backbone network, the optical power attenuates by a cumulative 3dB due to these losses, equivalent to only 50% of the initial signal energy. Without timely compensation using erbium-doped fiber amplifiers (EDFAs) or repeaters, the signal-to-noise ratio at the receiver will continue to deteriorate, ultimately leading to increased bit error rates or even interruption of data transmission. This phenomenon clearly demonstrates the direct impact of optical power attenuation on communication quality and highlights the importance of accurately measuring the attenuation coefficient and properly deploying power compensation equipment in fiber-optic network planning to ensure stable signal transmission.

[0025] In step A2, an intelligent fault detection model is constructed based on the status data. Abnormal signals are identified through wavelet transform and threshold judgment method. The LSTM neural network is trained with historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging.

[0026] Among them, the threshold judgment method is a method of setting a specific critical value, comparing the input data or indicators with the threshold, and performing classification, status judgment or decision triggering.

[0027] It can be understood that the embodiment of the present application sets a specific critical value, compares the input status data with the threshold in real time, identifies abnormal signals, and performs preliminary classification and status judgment on faults such as optical cable breakage, external force damage, and optical fiber aging, thereby providing an abnormal trigger mechanism for the intelligent fault detection model. By combining the signal feature extraction capability of wavelet transform and the time series modeling capability of LSTM neural network for historical fault data, the accuracy of fault detection and the ability to predict multiple types of faults are improved, and the risk of missed detection and false alarms is reduced.

[0028] In the embodiment of the present application, an intelligent fault detection model is constructed based on the state data, and abnormal signals are identified through wavelet transform and threshold judgment method. The specific steps are as follows: Step B1: Get the vibration frequency signal Perform continuous wavelet transform to obtain wavelet coefficients , the formula is ,in, is a normalization factor that ensures that the transformation maintains energy consistency at different scales a, , is the original function, b is the translation parameter, is the mother wavelet function, is the differential symbol; Step B2: Set the threshold ,in is the mean of the wavelet coefficients, is the standard deviation, k is the threshold coefficient (k ),like , it is determined to be an abnormal signal.

[0029] Among them, continuous wavelet transform is a mathematical tool for time-frequency analysis of signals through continuously changing scale and displacement parameters. The continuous wavelet transform formula is: ,in, is a normalization factor that ensures that the transformation maintains energy consistency at different scales a, , is the original function, b is the translation parameter, is the mother wavelet function, is the differential symbol.

[0030] It can be understood that the embodiments of the present application perform multi-resolution time-frequency analysis on the vibration frequency signal to capture the local features of the signal at different time and frequency scales, decompose the original signal into wavelet coefficients of different scales and positions through the translation and scaling operations of the mother wavelet function, retain the time domain details of the signal, use the normalization factor to ensure energy consistency at each scale, provide feature data for threshold determination, improve the sensitive capture capability of transient abnormal components in non-stationary signals, locate abnormal changes in vibration signals caused by equipment failures, and provide multi-dimensional feature information for intelligent fault detection.

[0031] For example, Figure 2 As shown in the figure, in a distributed fiber-optic vibration sensing system, the continuous wavelet transform (CWT) is used to analyze fiber-optic vibration signals. Multi-scale time-frequency decomposition captures transient features in the signal (such as 0.1-second pulses generated by external force impacts), converting the one-dimensional vibration signal into a two-dimensional time-frequency image. Combined with the ResNet50 neural network, this method accurately identifies intrusion events (such as impact, shearing, and climbing), achieving an average accuracy of 98.91% and a single-case detection time of only 0.376 seconds. For example, when construction machinery accidentally touches an optical cable, the CWT can quickly locate the vibration source in the time-frequency domain (with an accuracy of ±2 meters). Feature extraction distinguishes between human damage and environmental interference (such as background noise caused by wind and rain), significantly reducing false alarm rates. This method has been successfully applied to optical cable security monitoring, enabling real-time dynamic monitoring of long-distance optical cables and providing an efficient and reliable early warning method for communication line operation and maintenance.

[0032] In the embodiment of the present application, the method for determining the threshold coefficient k in step B2 includes: Step B201: Collect historical abnormal signal data to build a training set and calculate the false positive rate under different k values and missed detection rate ; Step B202: As the goal, the optimal k value is determined by the golden section method.

[0033] Among them, the golden section method is a one-dimensional optimization algorithm based on the golden section ratio.

[0034] It can be understood that in the embodiment of the present application, in the optimization process of iteratively narrowing the interval with the goal of minimizing the false detection rate and the missed detection rate, there is no need to calculate the derivative to approximate the extreme point of the unimodal function, and the optimal k value can be quickly determined with a smaller number of iterations, thereby improving the efficiency of threshold optimization, reducing the computational cost, and balancing the risks of false detection and missed detection, thereby improving the accuracy of abnormal signal detection.

[0035] For example, in the anomaly detection of vibration signals of communication optical cables, the golden section method is used to optimize the threshold coefficient k to balance the false detection rate and the missed detection rate: the initial interval [k_min=0.2, k_max=0.8] is set, with the goal of minimizing the total error rate. The interval is iteratively narrowed by the golden section ratio (0.618). The first calculation is k1=0.2+0.618×(0.8-0.2)=0.5708, k2=0.8-0.618×(0.8-0.2)=0.4292. After comparing the error rates of the two points, the interval is narrowed to [k_min=0.2, k_max=0.5708]. After 5 iterations, the optimal k=0.482 is determined, which reduces the total error rate from the initial 22% to 6.8%, reducing the computational complexity by 70% compared with the traditional grid search method, significantly improving the threshold optimization efficiency and detection accuracy.

[0036] In the embodiment of the present application, the LSTM neural network is trained in combination with historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging. The specific steps are as follows: Step E1: Extract feature vectors such as signal intensity fluctuation, temperature change rate, and optical power attenuation slope to construct a training data set; Step E2: Use five-fold cross-validation to optimize the LSTM network hyperparameters, including the number of hidden layers, number of neurons, and learning rate. The optimization goal is to minimize the classification loss function: , where M is the number of samples, C is the number of fault types, is the true label, is the predicted probability, is the loss value, i is the sample index, c is the category index, and ln is the natural logarithm; Step E3: Evaluate model performance through accuracy and F1 score. The formula is: , ,in, is the accuracy rate, TP is the positive class, the number of samples predicted by the model as positive, TN is the negative class, the number of samples predicted by the model as negative, FP is the negative class, the number of samples incorrectly predicted as positive by the model, FN is the positive class, the number of samples incorrectly predicted as negative by the model, F1 is the F1 score, the value range is [0,1]. The closer to 1, the more balanced and excellent the model's performance in precision and recall is. Precision is the proportion of samples predicted as positive that are actually positive, and Recall is the proportion of samples actually positive that are correctly predicted as positive.

[0037] Among them, five-fold cross-validation is to divide the data set evenly into five parts, use one of them as the validation set and the other four as the training set for model training and evaluation each time, and take the average value of the results after repeating five times to more reliably measure the performance of the model.

[0038] It can be understood that the embodiment of the present application optimizes the LSTM network hyperparameters by dividing the training data set into five parts, using one part as the validation set and the remaining four parts as the training set, thereby preventing the deviation of a single data division, using data to comprehensively evaluate the performance of hyperparameter combinations on different subsets, reducing the risk of overfitting, and making the hyperparameter optimization results have generalization capabilities, thereby improving the stability of the model in practical applications.

[0039] For example, in the inspection of communication optical cables, five-fold cross-validation is used. This involves dividing sensor-collected time-series data, such as signal strength fluctuations and temperature change rates, into five equal parts. Four of these parts are used in turn to train the LSTM network and optimize hyperparameters such as the number of hidden layers and neurons. The remaining part is used to verify model performance. Ultimately, the stability is assessed by taking the average of the five accuracy rates (e.g., 98.5%) and F1 scores (e.g., 0.93). This method significantly improves model generalization, for example, reducing the false alarm rate from 15% to 8%. Combined with OTDR technology, it increases fault location accuracy from ±50 meters to ±10 meters, effectively reducing manual inspection workload and shortening repair time.

[0040] In step A3, when a fault warning is detected, the drone inspection unit is triggered. Based on the fault type and the geographic information of the suspected fault area, the improved Dijkstra algorithm is used to generate a target inspection path. The drone's onboard lidar and AI vision module, combined with the target inspection path, scans and inspects the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress to the junction box. Combined with the inertial navigation system, the geographic coordinates of the fault point are quickly calibrated.

[0041] Among them, the YOLOv5 algorithm is a single-stage target detection algorithm based on PyTorch.

[0042] It can be understood that the embodiment of the present application uses a lightweight network architecture and multi-scale feature fusion technology to perform fault identification, and combines the improved Dijkstra path planning algorithm to perform real-time detection of 15 types of physical faults such as tower tilt, cable damage, and junction box deformation, thereby improving the average accuracy mean. It adopts an adaptive anchor frame mechanism and data enhancement strategy to cope with complex environments such as backlight and haze, and combines the lidar point cloud and inertial navigation system to control the calibration error of the three-dimensional coordinates of the fault point and improve inspection efficiency.

[0043] For example, Figure 3 As shown, in a provincial telecom operator's intelligent optical cable inspection system, the YOLOv5 algorithm is deployed on the AI ​​vision module of a DJI Matrice 300RTK drone. When abnormal optical power fluctuations are detected, the system triggers the drone to scan the suspected fault area along a spiral inspection path (covering a 2-kilometer radius) generated by a modified Dijkstra algorithm. At a cruising altitude of 100 meters, the YOLOv5s model (7.2M parameters) processes 4K aerial imagery at 38 FPS. Using a multi-scale feature fusion network, it accurately identifies potential hazards such as cable damage (pixel-level detection accuracy of 0.8mm), tower tilt (angular error ≤1.5°), and splice box deformation (volume change rate ≥5%). Combining LiDAR point cloud data with an RTK positioning system, the fault point's geographic coordinates are synchronized to the operation and maintenance platform in real time (positioning error ±0.12 meters). During an emergency inspection after a rainstorm, the system completed the work that would have taken 8 hours for a regular manual inspection in just 30 minutes, successfully discovered three hidden cable break points, and achieved a fault identification accuracy rate of 98.7%, significantly improving the response speed and refinement of optical cable operation and maintenance.

[0044] In the embodiment of the present application, the target inspection path is generated using the improved Dijkstra algorithm, and the specific steps are as follows: Step C1: Abstract the geographic information of the suspected fault area into a weighted graph , where V is the node set (tower location), E is the edge set (cable segment), W is the edge weight matrix, and weight Calculated by the linear weighted combination formula , is the straight-line distance between nodes i and j, is the terrain complexity factor (value range [0,1]), , is the weight coefficient; Step C2: Using the starting tower S as the source point, use the Dijkstra algorithm to find the minimum weight path to the target tower D.

[0045] Among them, the linear weighted combination formula is a mathematical expression that multiplies multiple variables by their corresponding weight coefficients and then linearly sums them to comprehensively consider the influence of different variables and obtain a new comprehensive result. , is the straight-line distance between nodes i and j, is the terrain complexity factor (value range [0,1]), , is the weight coefficient.

[0046] It can be understood that the embodiment of the present application organically integrates the straight-line distance between nodes and the terrain complexity factor by using a linear weighted combination formula, adjusts the weight coefficient, and makes the weight of the edge comprehensively reflect the impact of distance and terrain conditions on the inspection cost during the actual inspection process, determines the one-sidedness of the weight, and makes the edge weight in the weighted graph fit the actual geographical environment, providing accurate data for the Dijkstra algorithm, so that the target inspection path can adapt to the difference in travel difficulty caused by the terrain, and improve the practicality of inspection path planning.

[0047] For example, Figure 4 As shown, in the power inspection scenario, the linear weighted combination formula can comprehensively measure the influencing factors. I will enrich the example content through specific numerical values ​​and comparisons of different scenarios, and show the practical application of the formula in path planning. In the mountainous power inspection path planning, assume that the straight-line distance between nodes X and Y is 800 meters, the area is a steep mountain, and the terrain complexity factor is evaluated as 0.8; the straight-line distance between nodes M and N is 1200 meters, and it is in a plain area, and the terrain complexity factor is only 0.2. If the straight-line distance weight coefficient is set to 0.3 and the terrain complexity factor weight coefficient is 0.7, the weight of the edge between nodes X and Y is calculated by the linear weighted combination formula. =0.3×800+0.7×0.8×800=240+448=688; the weight of the edge between nodes M and N =0.3×1200+0.7×0.2×1200=360+168=528. Although the straight-line distance between X and Y is short, the complex terrain gives it a higher weight. Substituting these weights into the improved Dijkstra algorithm, the generated path avoids high-weighted XY segments and selects lower-weighted paths, such as the MN segment. This prevents inspectors from getting bogged down in mountainous terrain, saving both energy and time while reducing the risk of getting lost or falling. This makes inspections more efficient and safer, fully demonstrating the value of the linear weighted combination formula in comprehensively considering multiple factors to optimize path planning.

[0048] In the embodiment of the present application, the laser radar and AI vision module carried by the drone are combined with the target inspection path to scan and detect the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress of the joint box. The geographic coordinates of the fault point are quickly calibrated in combination with the inertial navigation system. The specific steps are as follows: Step F1: The LiDAR point cloud data is mapped to the geographic coordinate system using the coordinate system conversion formula: , where R is the rotation matrix and T is the translation vector. is the sensor coordinate system coordinate, is the geographic coordinate; Step F2: The AI ​​vision module outputs the pixel coordinates of the fault target through the YOLOv5 algorithm , combined with the camera intrinsic parameter matrix K to inversely calculate the spatial coordinates: , where d is the target distance.

[0049] Among them, the coordinate system conversion formula is a regular expression for converting coordinate values ​​between different coordinate systems through mathematical operations. The formula is , where R is the rotation matrix and T is the translation vector. is the sensor coordinate system coordinate, is the geographic coordinate.

[0050] It can be understood that the embodiment of the present application uses a coordinate system conversion formula to uniformly map the sensor coordinate system data of the lidar and the pixel coordinates of the AI ​​vision module to the geographic coordinate system through parameters such as the rotation matrix, translation vector and camera intrinsic parameter matrix, so that multi-source heterogeneous data can be spatially aligned, so that the terrain information scanned by the lidar during the drone inspection process and the fault target position identified by vision can be accurately matched. Combined with the inertial navigation system, the pixel coordinates or sensor coordinates of the fault point are converted into actual geographic coordinates, providing accurate spatial position information for the rapid positioning, precise calibration and maintenance scheduling of the fault point, thereby improving the automation level and accuracy of fault detection and positioning in inspection operations.

[0051] For example, Figure 5As shown in the figure, during the drone inspection operation, the laser radar records the coordinates of the tilt feature point of a certain tower as (2.8 meters, -1.5 meters, 15 meters) in its own sensor coordinate system. The coordinate system uses the current position of the drone as the origin. At the same time, the AI ​​vision module captures an image of water entering the cable junction box. The pixel position of the fault target in the photo is (300 pixels, 250 pixels). The actual distance between the drone and the target is measured by the ranging module as 80 meters. By using pre-calibrated device parameters (such as the rotation angle of the drone sensor relative to geographic north and the height difference between the drone's origin and the ground coordinate system), the system first "translates" the lidar sensor coordinates into familiar geographic coordinates. For example, (2.8, -1.5, 15) is converted to 30.25°N, 119.87°E, and 120 meters above sea level. The system then combines the pixel coordinates (300, 250) identified by the vision module with a distance of 80 meters to calculate the actual location of the fault in three-dimensional space. This is then unified into the aforementioned geographic coordinate system, ultimately determining the precise location of the water ingress at the junction box: 30.25°N, 119.87°E, and 118 meters above sea level. This process allows the drone to automatically integrate the "local location information" collected by various devices into "global geographic coordinates" without manual intervention. Maintenance personnel can then navigate directly to the location using the data, saving over 70% of the time required for location analysis compared to traditional manual troubleshooting, significantly improving troubleshooting efficiency.

[0052] In step A4, the edge computing node integrates the sensor data with the drone detection results, analyzes the fault data using DS evidence theory, determines the confidence level of the fault point, generates an alarm work order containing the fault type, location coordinates, and severity, and pushes it to the operation and maintenance platform.

[0053] Among them, DS evidence theory is a mathematical theory that integrates the trust function and likelihood function of multi-source evidence to reason and synthesize the uncertainty of propositions.

[0054] It can be understood that the embodiments of the present application process the uncertainty and conflicting information in the data through the mathematical synthesis of the trust function and the likelihood function, accurately calculate the confidence level of the fault point based on the fused evidence, generate comprehensive alarm work orders, prevent misjudgment or missed judgment caused by noise or detection blind spots in a single data source, provide the operation and maintenance platform with accurate fault decision-making basis, and improve the automation level of fault diagnosis in power inspections and the operation and maintenance response efficiency.

[0055] For example, Figure 6As shown in the figure, in a power line inspection scenario, edge computing nodes collect data from multiple sources: a sensor detects a sudden temperature rise on a tower, determining a 70% probability of failure; a drone's visual inspection reveals a slight tilt in the tower, assigning a 65% probability of failure; and another vibration sensor's feedback indicates normal data, indicating only a 20% probability of failure. Faced with this conflicting and uncertain information, the DS evidence theory is applied, treating each data source as independent evidence. Through a fusion calculation using trust and likelihood functions, the system ultimately determines an 85% confidence level that the tower is faulty. Based on this information, the system generates an alarm work order containing the fault type (structural anomaly), location coordinates, and severity (moderate fault), and pushes it to the operation and maintenance platform. This allows operation and maintenance personnel to respond quickly and accurately address potential faults, avoiding oversight or excessive maintenance caused by misjudgment of a single data point.

[0056] In the embodiment of the present application, an alarm work order including the fault type, location coordinates, and severity is generated and pushed to the operation and maintenance platform. The specific steps are as follows: Step D1: Based on the fault type and coordinates output by the fault assessment model, combined with optical cable equipment parameters (such as fiber core type and splice location) and terrain characteristics, a multi-dimensional report including fault severity and repair suggestions is generated.

[0057] Step D2: Use the PSO algorithm to optimize the repair path in the multi-dimensional report. The formula is: , ,in, is the velocity of particle i in generation t; w is the inertia weight; , is the learning factor; , is a random number; is the best historical position of particle i; is the global optimal position; For the position of particle i in the tth generation, the optimal fault repair reference path is obtained.

[0058] Step D3: Convert the fault location report and optimal repair path into visual instructions, push them to the operation and maintenance platform in real time through the edge computing node, and support real-time viewing and navigation on the mobile app.

[0059] Among them, the PSO algorithm is an optimization algorithm that simulates the collaborative behavior of biological groups such as bird flocks and fish schools. It is based on swarm intelligence and searches for the optimal solution by iteratively updating the position and velocity of particles. The PSO algorithm formula is: , ,in, is the velocity of particle i in generation t; w is the inertia weight; , is the learning factor; , is a random number; is the best historical position of particle i; is the global optimal position; is the position of particle i in the tth generation.

[0060] It can be understood that the embodiments of the present application iteratively update the particle speed and position, and use parameters such as inertia weight and learning factor to guide the particles to search for the historical best and global best positions, efficiently find the best in the repair path, quickly determine the optimal fault repair reference path, avoid ineffective exploration, reduce resource consumption, make the repair plan obtained by the operation and maintenance platform more scientific and reasonable, enable operation and maintenance personnel to efficiently perform repair tasks, and improve the efficiency and quality of fault handling.

[0061] For example, Figure 7 As shown in the figure, in a regional power grid fault repair task, five faulty substations are located at different locations, with coordinates A(1,1), B(3,4), C(5,2), D(2,6), and E(7,5). The operations team needs to plan a repair route with the shortest total distance. When using the PSO algorithm for optimization, each possible repair sequence is considered a "particle," and 10 different route plans are initially generated (for example, the first plan is A→C→B→E→D, with a total distance of approximately 14.7 kilometers). During the algorithm's operation, each "particle" adjusts its search direction based on its own previously found shortest route (individual optimality) and the shortest route currently found by all particles (global optimality). For example, if a particle originally had a longer route, it will gradually move toward a more optimal route sequence, such as prioritizing repairs at nearby substations. After multiple iterations of optimization, the optimal sequence, A→B→C→D→E, is finally "selected" from the chaotic route combinations, shortening the total distance to approximately 12.4 kilometers, a reduction of approximately 15% compared to the initial plan. This method allows operation and maintenance personnel to directly follow the optimal route generated by the algorithm without manual trial and error, greatly reducing time spent on the road and vehicle damage, and significantly improving fault repair efficiency.

[0062] In the embodiment of the present application, the fault data decision-level fusion is performed using the DS evidence theory. The specific steps are as follows: Step G1: Establish a fault identification framework Θ, including fault types such as optical cable breakage, external force damage, and optical fiber aging; Step G2: Based on the sensor data and the UAV detection results, calculate the basic probability distribution function m(A) of each fault type, where A is a subset of Θ; Step G3: Calculate the joint basic probability distribution function using the DS synthesis rule, the formula is: , where C is the fused fault type, and are the basic probability distribution functions for sensor data and drone detection results, respectively. is the basic probability distribution function obtained after combination, A is the variable set The object of action, B is the variable set The object of action, is an empty set; Step G4: Determine the confidence level of the fault point according to the joint basic probability distribution function.

[0063] The basic probability distribution function is a function used in DS evidence theory to distribute probability mass to each subset of the domain power set.

[0064] It can be understood that the embodiment of the present application distributes the uncertainty information in the sensor data and drone detection results to each subset of the fault identification framework in the form of probability mass by calculating the basic probability distribution function, describes the degree of trust in different fault propositions, provides a data expression basis for multi-source evidence fusion, allows the retention of fuzzy information that cannot determine the specific fault type, and fuses evidence conflicts from different sources through DS synthesis rules, so that the confidence level calculation is more in line with the complexity of actual fault scenarios, thereby improving the accuracy of data fusion in power inspections.

[0065] In the embodiment of the present application, when determining the confidence level of the fault point, a conflict degree index is introduced. : ,when When >0.7, the confidence level of the fault point is high; when When <0.4, it is judged as low, among which, To identify the framework, and for two different basic probability distribution functions.

[0066] Among them, the conflict index is a quantitative parameter used to measure the degree of contradiction or disagreement between different evidences in multi-source information fusion scenarios such as DS evidence theory. The formula for calculating the conflict index is: ,when When >0.7, the confidence level of the fault point is high; when When <0.4, it is judged as low, among which, To identify the framework, and for two different basic probability distribution functions.

[0067] It can be understood that the embodiment of the present application quantifies the degree of contradiction between different basic probability distribution functions by calculating the conflict degree index, reflects the degree of disagreement in fault judgment between multiple sources of evidence such as sensor data and drone detection results, and grades the confidence level of the fault point by setting a clear threshold. When the conflict degree is high (>0.7), it warns that there may be data contradictions or missing key information; when the conflict degree is moderate (0.4-0.7) or low (<0.4), it provides a basis for the confidence level, so that fault location and risk assessment are in line with the actual situation, so that operation and maintenance personnel give priority to high-conflict and high-risk faults, and improve fault handling efficiency.

[0068] For example, in optical cable fault detection, sensor data indicates a basic probability of 0.6 for "cable break" and 0.4 for "unknown fault." Drone detection results assign a basic probability of 0.5 for "external damage" and 0.5 for "unknown fault." Calculating the conflict index between the two reveals that the sensor's identification of "cable break" and the drone's identification of "external damage" are mutually exclusive events (the intersection is empty). The sum of the products of their probabilities (i.e., the conflict index) is 0.6 × 0.5 = 0.3. Based on preset thresholds (>0.7 for high, 0.4-0.7 for medium, and <0.4 for low), a conflict index of 0.3 indicates minimal disagreement between the two sources of evidence. The system then assigns a "low" confidence level to the fault point, suggesting possible insufficient data coverage or atypical fault characteristics. Based on this, operation and maintenance personnel prioritize checking whether there are blind spots in the detection range of sensors and drones to avoid underestimation and omission of potential risks due to conflicting evidence. This not only improves the scientific nature of fault confidence grading, but also provides data conflict warning for accurate troubleshooting and optimizes the allocation efficiency of maintenance resources.

[0069] According to the embodiment of the present application, a method for rapidly locating fault points in intelligent inspections of communication optical cable lines is proposed. By acquiring state data such as the vibration frequency, temperature gradient, and optical power attenuation of the optical cable line, an intelligent fault detection model integrating wavelet transform and LSTM neural network is constructed to pre-judge the fault type. When an early warning is detected, the drone inspection unit is triggered, an improved Dijkstra algorithm is used to generate the target inspection path, and an AI vision module (YOLOv5 algorithm) is used to identify physical fault characteristics and calibrate the coordinates of the inertial navigation system. The edge computing node then uses the DS evidence theory to fuse the data to determine the fault confidence level and generate an alarm work order. This shortens the average time taken to locate the fault, reduces the false alarm rate, and improves the real-time, accuracy, and automation level of the operation and maintenance of communication optical cable lines. This solves the problems of long positioning time, high missed detection rate, and difficulty in distinguishing fault types in the existing technology.

[0070] The following will describe a method for quickly locating fault points in intelligent inspection of communication optical cable lines through a specific embodiment. Figure 8 Shown, including: Vibration, temperature, and optical power monitoring equipment are deployed every 2 kilometers along the optical cable. Vibration frequency is acquired at a sampling frequency of 100 Hz, temperature gradient is acquired with an accuracy of ±0.5°C, and optical power attenuation and other status data are acquired with a resolution of 0.01 dB. The continuous wavelet transform formula is used to (Morlet wavelet is selected and the normalization factor is used to ensure energy consistency) to process the vibration frequency signal, and the threshold coefficient k=3 is determined by the golden section method. Determine abnormal signals; extract feature vectors such as signal intensity fluctuations to construct a training set of 5,000 data sets, and use 5-fold cross-validation to optimize the LSTM neural network hyperparameters (2 hidden layers, 64 and 32 neurons, and a learning rate of 0.001) to minimize the classification loss function. The trained model was evaluated with an accuracy of 92% and an F1 score of 0.88, enabling the pre-judgment of various types of optical cable faults.

[0071] When the LSTM neural network issues a suspected optical cable break fault warning, the geographical information of the mountain fault area is abstracted into a weighted graph (the node set V contains 20 tower locations, and the edge set E is the optical cable segment). The linear weighted formula is used to calculate the distance between the straight line and the terrain complexity factor (valued at 0.7). (Weight coefficient ) calculates the edge weights, takes the starting tower S as the source point, and uses the improved Dijkstra algorithm to generate the target inspection path; dispatches a drone equipped with a high-precision lidar (measurement accuracy of ±1.5cm) and a 50-megapixel AI vision module for inspection. The lidar point cloud data is mapped to the geographic coordinate system using the coordinate system conversion formula. The AI ​​vision module uses the YOLOv5 algorithm to identify the tower tilt fault. Combined with the camera's intrinsic parameter matrix K and the 15-meter target distance, the spatial coordinates are inversely calculated through the formula, and the geographic coordinates of the fault point are calibrated to 31.23° north latitude and 118.76° east longitude.

[0072] A fault identification framework Θ including types such as "cable break" is established. According to the sensor data, the basic probability distribution of "cable break" is set to 0.6 and "unknown fault" is set to 0.4. The basic probability distribution of "tower tilt (which may cause cable break)" by drone detection results is set to 0.7 and "unknown fault" is set to 0.3. The conflict index between the basic probability distribution functions of different data sources is calculated to be 0.42, and the confidence level of the fault point is determined to be medium. The DS synthesis rule formula is used. After integrating the evidence, it was determined that the fault type was "optical cable break risk", the location coordinates were 31.23° north latitude and 118.76° east longitude, and the severity was moderate.

[0073] Based on the fault assessment results, combined with the G.652D fiber core type, the location of the splicing point 2 kilometers away from the fault point, and the mountainous terrain characteristics, a multi-dimensional report containing the fault type, location, severity, and repair suggestions (carry climbing and splicing tools) is generated; the PSO algorithm is used to calculate the fault type and location. , Optimize the repair path, initialize 20 particles, and set the inertia weight , learning factor = =1.6, and iteratively determined the optimal repair sequence for the three fault points (fault point 1 → fault point 3 → fault point 2), reducing the distance by 25%. The fault location report and the optimal path were converted into visual instructions, which were pushed to the operation and maintenance platform and mobile app through the edge computing node. The app's real-time navigation function helped operation and maintenance personnel quickly reach the fault point for repair.

[0074] In summary, the present invention acquires multi-dimensional status data in real time by deploying multiple monitoring devices along the optical cable and setting sampling parameters, and captures abnormal signals by combining continuous wavelet transform with optimized threshold judgment method; utilizes LSTM neural network training model to pre-judge faults with high accuracy and F1 score, thereby improving handling time; when the early warning is triggered, the improved Dijkstra algorithm is combined with factors such as terrain to plan the inspection route; drones equipped with high-precision equipment and advanced algorithms quickly locate the coordinates of the fault point; DS evidence theory integrates multi-source data, and scientifically determines the fault confidence level through conflict degree indicators, thereby improving the reliability of fault diagnosis; optimizes the repair path through PSO algorithm, and combines visual command push with mobile navigation to enable operation and maintenance personnel to efficiently arrive at and repair the fault, thereby improving the timeliness of fault detection, the accuracy of positioning and the efficiency of repair, reducing the cost of manual inspection and the risk of missed detection, and ensuring the stable operation of communication optical cable lines.

[0075] Next, a construction device for a communication optical cable line intelligent inspection fault point rapid location system proposed in an embodiment of the present application will be described with reference to the accompanying drawings.

[0076] Specifically, Figure 9 A schematic diagram of the composition of a construction device for a communication optical cable line intelligent inspection fault point rapid location system provided in an embodiment of the present application.

[0077] like Figure 9 As shown, the construction device 10 of the communication optical cable line intelligent inspection fault point rapid positioning system includes: an acquisition module 100, an identification module 200, a calibration module 300, and a generation module 400.

[0078] The acquisition module 100 is used to obtain status data such as the vibration frequency, temperature gradient, and optical power attenuation of the optical cable line. The identification module 200 is used to build an intelligent fault detection model based on the status data, identify abnormal signals through wavelet transform and threshold judgment, and train an LSTM neural network based on historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging. The calibration module 300 is used to trigger the drone inspection unit when a fault warning is detected. Based on the fault type and geographic information of the suspected fault area, the improved Dijkstra algorithm is used to generate a target inspection path. The laser radar and AI vision module on the drone are combined with the target inspection path to scan and inspect the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. The geographic coordinates of the fault point are quickly calibrated in combination with the inertial navigation system. The generation module 400 is used to integrate sensor data with drone detection results at the edge computing node, perform decision-level fusion of fault data through DS evidence theory, determine the confidence level of the fault point, generate an alarm work order containing the fault type, location coordinates, and severity, and push it to the operation and maintenance platform.

[0079] It should be noted that the above explanation of an embodiment of a method for rapid fault location of a communication optical cable line intelligent inspection is also applicable to a construction device of a communication optical cable line intelligent inspection fault location rapid location system of this embodiment, and will not be repeated here.

[0080] According to an embodiment of the present application, a device for constructing a system for rapidly locating fault points in intelligent inspections of communication optical cable lines is proposed. By acquiring state data such as the vibration frequency, temperature gradient, and optical power attenuation of the optical cable line, an intelligent fault detection model integrating wavelet transform and LSTM neural network is constructed to pre-judge the fault type. Upon detecting an early warning, a drone inspection unit is triggered. An improved Dijkstra algorithm is used to generate a target inspection path, and an AI vision module (YOLOv5 algorithm) is used to identify physical fault characteristics and calibrate coordinates using an inertial navigation system. The data is then fused through the DS evidence theory at an edge computing node to determine the fault confidence level and generate an alarm work order. This shortens the average time taken to locate faults, reduces the false alarm rate, and improves the real-time, accuracy, and automation level of communication optical cable line operation and maintenance. This solves the problems of long positioning time, high missed detection rate, and difficulty in distinguishing fault types in the prior art.

[0081] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0082] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0083] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0084] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0085] Those skilled in the art will appreciate that all or part of the steps in the method for implementing the above-mentioned embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

Claims

1. A method for quickly locating fault points in intelligent inspection of communication optical cable lines, characterized in that: The following steps are involved: Step A1: Obtain status data such as vibration frequency, temperature gradient, and optical power attenuation of the optical cable line; Step A2: Building an intelligent fault detection model based on the status data, identifying abnormal signals through wavelet transform and threshold judgment method, and training an LSTM neural network based on historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging; Step A3: When a fault warning is detected, the drone inspection unit is triggered. Based on the fault type and the geographic information of the suspected fault area, an improved Dijkstra algorithm is used to generate a target inspection path. The drone's onboard laser radar and AI vision module, combined with the target inspection path, scans and inspects the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. The geographic coordinates of the fault point are quickly calibrated in combination with the inertial navigation system. Step A4: The edge computing node integrates the sensor data with the drone detection results, analyzes the fault data using DS evidence theory, determines the confidence level of the fault point, generates an alarm work order containing the fault type, location coordinates, and severity, and pushes it to the operation and maintenance platform.

2. A method for quickly locating fault points in intelligent inspection of communication optical cable lines according to claim 1, characterized in that: An intelligent fault detection model is constructed based on the state data, and abnormal signals are identified through wavelet transform and threshold judgment method. The specific steps are as follows: Step B1: Get the vibration frequency signal Perform continuous wavelet transform to obtain wavelet coefficients , the formula is ,in, is a normalization factor that ensures that the transformation maintains energy consistency at different scales a, , is the original function, b is the translation parameter, is the mother wavelet function, is the differential symbol; Step B2: Set the threshold ,in is the mean of the wavelet coefficients, is the standard deviation, k is the threshold coefficient (k ),like , it is determined to be an abnormal signal.

3. A method for quickly locating fault points in intelligent inspection of communication optical cable lines according to claim 2, characterized in that: The method for determining the threshold coefficient k in step B2 includes: Step B201: Collect historical abnormal signal data to build a training set and calculate the false positive rate under different k values and missed detection rate ; Step B202: As the goal, the optimal k value is determined by the golden section method.

4. A method for quickly locating fault points in intelligent inspection of communication optical cable lines according to claim 1, characterized in that: The target inspection path is generated using the improved Dijkstra algorithm. The specific steps are as follows: Step C1: Abstract the geographic information of the suspected fault area into a weighted graph , where V is the node set (tower location), E is the edge set (cable segment), W is the edge weight matrix, and weight Calculated by the linear weighted combination formula , is the straight-line distance between nodes i and j, is the terrain complexity factor (value range [0,1]), , is the weight coefficient; Step C2: Using the starting tower S as the source point, use the Dijkstra algorithm to find the minimum weight path to the target tower D.

5. The method for quickly locating fault points in intelligent inspection of communication optical cable lines according to claim 1, characterized in that: Generate an alarm ticket containing the fault type, location coordinates, and severity, and push it to the operation and maintenance platform. The specific steps are as follows: Step D1: Based on the fault type and coordinates output by the fault assessment model, combined with optical cable equipment parameters (such as fiber core type and splice location) and terrain characteristics, a multi-dimensional report including fault severity and repair suggestions is generated; Step D2: Use the PSO algorithm to optimize the repair path in the multi-dimensional report. The formula is: , ,in, is the velocity of particle i in generation t; w is the inertia weight; , is the learning factor; , is a random number; is the best historical position of particle i; is the global optimal position; For the position of particle i in generation t, the optimal fault repair reference path is obtained; Step D3: Convert the fault location report and optimal repair path into visual instructions, push them to the operation and maintenance platform in real time through the edge computing node, and support real-time viewing and navigation on the mobile app.

6. A method for rapid fault location during intelligent inspection of a communication optical cable line according to claim 1, characterized in that: Combined with historical fault data, the LSTM neural network is trained to pre-diagnose fault types such as optical cable breakage, external force damage, and optical fiber aging. The specific steps are as follows: Step E1: Extract feature vectors such as signal intensity fluctuation, temperature change rate, and optical power attenuation slope to construct a training data set; Step E2: Use five-fold cross-validation to optimize the LSTM network hyperparameters, including the number of hidden layers, number of neurons, and learning rate. The optimization goal is to minimize the classification loss function: , where M is the number of samples, C is the number of fault types, is the true label, is the predicted probability, is the loss value, i is the sample index, c is the category index, and ln is the natural logarithm; Step E3: Evaluate model performance through accuracy and F1 score. The formula is: , ,in, is the accuracy rate, TP is the positive class, the number of samples predicted by the model as positive, TN is the negative class, the number of samples predicted by the model as negative, FP is the negative class, the number of samples incorrectly predicted as positive by the model, FN is the positive class, the number of samples incorrectly predicted as negative by the model, F1 is the F1 score, the value range is [0,1]. The closer to 1, the more balanced and excellent the model's performance in precision and recall is. Precision is the proportion of samples predicted as positive that are actually positive, and Recall is the proportion of samples actually positive that are correctly predicted as positive.

7. The method for quickly locating fault points in intelligent inspection of communication optical cable lines according to claim 1, characterized in that: The drone uses the laser radar and AI vision module, combined with the target inspection path, to scan and inspect the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. Combined with the inertial navigation system, the geographic coordinates of the fault point are quickly calibrated. The specific steps are as follows: Step F1: The LiDAR point cloud data is mapped to the geographic coordinate system using the coordinate system conversion formula: , where R is the rotation matrix and T is the translation vector. is the sensor coordinate system coordinate, is the geographic coordinate; Step F2: The AI ​​vision module outputs the pixel coordinates of the fault target through the YOLOv5 algorithm , combined with the camera intrinsic parameter matrix K to inversely calculate the spatial coordinates: , where d is the target distance.

8. The method for rapid fault location during intelligent inspection of a communication optical cable line according to claim 1, characterized in that: The DS evidence theory is used to perform decision-level fusion of fault data. The specific steps are as follows: Step G1: Establish a fault identification framework Θ, including fault types such as optical cable breakage, external force damage, and optical fiber aging; Step G2: Based on the sensor data and the UAV detection results, calculate the basic probability distribution function m(A) of each fault type, where A is a subset of Θ; Step G3: Calculate the joint basic probability distribution function using the DS synthesis rule, the formula is: , where C is the fused fault type, and are the basic probability distribution functions for sensor data and drone detection results, respectively. is the basic probability distribution function obtained after combination, A is the variable set The object of action, B is the variable set The object of action, is an empty set; Step G4: Determine the confidence level of the fault point according to the joint basic probability distribution function.

9. A method for rapid fault location during intelligent inspection of a communication optical cable line according to claim 8, characterized in that: When determining the confidence level of the fault point, the conflict degree index is introduced : ,when When >0.7, the confidence level of the fault point is high; when When <0.4, it is judged as low, among which, To identify the framework, and for two different basic probability distribution functions.

10. A device for constructing a communication optical cable line intelligent inspection fault point rapid positioning system, characterized in that: include: An acquisition module is used to obtain status data such as vibration frequency, temperature gradient, and optical power attenuation of the optical cable line; An identification module is used to build an intelligent fault detection model based on the status data, identify abnormal signals through wavelet transform and threshold judgment method, and train an LSTM neural network based on historical fault data to pre-judge fault types such as optical cable breakage, external force damage, and optical fiber aging; The calibration module is used to trigger the drone inspection unit when a fault warning is detected. Based on the fault type and the geographic information of the suspected fault area, the improved Dijkstra algorithm is used to generate a target inspection path. The drone's onboard laser radar and AI vision module, combined with the target inspection path, scans and inspects the optical cable line. The YOLOv5 algorithm is used to identify physical fault characteristics such as tower tilt, cable wear, and water ingress into the junction box. The inertial navigation system is used to quickly calibrate the geographic coordinates of the fault point. A generation module is used for edge computing nodes to fuse sensor data with the drone detection results, perform fault data decision-level fusion through DS evidence theory, determine the confidence level of the fault point, generate an alarm work order containing the fault type, location coordinates, and severity, and push it to the operation and maintenance platform.

Images