A health monitoring system and method for tuning a water engineering optical fiber network
By combining multi-source sensor arrays and edge computing nodes, and integrating DS evidence theory and SVM regression model, real-time and accurate fault monitoring and location of the fiber optic network for water diversion projects were achieved, reducing deployment costs and improving monitoring efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YUNNAN COMM IND SERVICE CO LTD
- Filing Date
- 2025-08-06
- Publication Date
- 2026-07-21
AI Technical Summary
Existing optical cable monitoring technology has problems such as insufficient real-time performance, low positioning accuracy, limited coverage and high deployment cost in water diversion projects. It cannot effectively monitor sudden faults and quickly locate fault points, and is susceptible to environmental noise interference, resulting in a high false alarm rate.
By adopting a combination of multi-source sensor arrays, edge computing nodes, and cloud management platforms, the system synchronously collects fiber optic environmental parameters through sensing modules, constructs a LoRa star network and an industrial-grade PON fiber optic network, and combines the DS evidence theory fusion model and the SVM regression model for fault early warning, location, and classification, thereby achieving multi-dimensional sensing and lightweight deployment.
It enables real-time monitoring of fiber optic networks, improves fault identification accuracy to 98.7%, reduces false alarm rate to 1.3%, improves positioning accuracy to ≤50 meters, and reduces equipment cost per kilometer by 30%, solving the problems of high cost and complexity of traditional solutions.
Smart Images
Figure CN120915375B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of optical fiber fault detection technology, and specifically to a health monitoring system and method for optical fiber networks used in water diversion projects. Background Technology
[0002] In large-scale water diversion projects, optical fiber channels are the core infrastructure supporting the intelligent operation of the project, undertaking critical tasks such as transmitting water diversion dispatch instructions, monitoring equipment status, and transmitting video images back. Because the project lines are typically tens to hundreds of kilometers long, passing through geologically complex areas (such as tunnels, rivers, and high-intensity earthquake zones), optical fibers face multiple environmental threats: Geological disasters: soil subsidence, rock fractures, etc., cause mechanical stress concentration in the optical fiber, leading to breakage or signal attenuation; Climatic factors: extreme temperature changes (-40℃ to +70℃) cause aging of the optical fiber sheath material, and humidity corrosion (above 95% RH) causes fiber optic connector failure.
[0003] Traditional monitoring methods rely on manual inspections or single-sensor detection, which have the following problems: poor real-time performance (unable to monitor sudden faults around the clock); low positioning accuracy (difficult to quickly locate fault points); limited coverage (long-distance projects require the deployment of a large number of devices, resulting in high costs and complex management); and human interference (construction excavation, theft, and other activities directly damage the physical structure of the optical cable). If these risks are not monitored in a timely manner, they may lead to communication interruptions, which in turn may cause safety accidents such as uncontrolled water conveyance equipment and delayed disaster warnings. Therefore, building a full-line optical cable status monitoring system is an urgent need to ensure the safe operation of water diversion projects.
[0004] Current technologies in the field of optical cable monitoring can be mainly divided into three categories, and their core architecture and technical characteristics are as follows:
[0005] Category 1: OTDR-based fiber optic physical characteristic detection technology; Technical principle: Pulsed light is emitted into the optical cable using an optical time domain reflectometer (OTDR), and the backscattered light signal is received. The fiber break point or abnormal loss area is located based on the signal attenuation curve. The theoretical positioning accuracy is 1-5 meters (depending on the pulse width). Typical applications: Mainly used for periodic testing of fiber optic links (e.g., once a day), suitable for static loss monitoring.
[0006] The second category: Environmental monitoring technologies based on single sensors; Vibration monitoring systems: Accelerometers or fiber optic vibration sensors are deployed along the optical cable to identify human-caused damage such as digging and knocking by analyzing the vibration signal spectrum, with a positioning accuracy of approximately 100-200 meters (depending on sensor spacing). Temperature and humidity monitoring networks: Distributed temperature and humidity sensors (such as DHT11) are networked together, transmitting data via RS485 or LoRa to monitor environmental parameters inside optical cable junction boxes and provide early warnings of connector oxidation problems caused by excessive humidity.
[0007] The third category combines manual inspection with offline testing. Manual inspection involves regularly (e.g., weekly) dispatching maintenance personnel to inspect along the fiber optic cable route, using visual checks and portable OTDR testers to detect obvious fault points. Offline analysis involves aggregating test data to a management platform, relying on human experience to determine the fault type, lacking real-time capability.
[0008] The core problems and shortcomings of existing technologies:
[0009] 1. Insufficient real-time capability, unable to cope with sudden failures: OTDR detection requires manual triggering or timed execution (intervals of several hours to several days), and cannot capture instantaneous fracture events (such as sudden fractures caused by geological disasters); a single vibration sensor is only sensitive to specific types of interference (such as mechanical vibration), and has no monitoring capability for progressive failures (accounting for more than 60%) such as sudden temperature changes and slow stress accumulation.
[0010] 2. Limited positioning accuracy and fault identification capabilities: OTDR has a "blind zone" problem (5-10 meters near-end blind zone, 1-2 meters event blind zone), which cannot accurately identify the fault location in short-distance dense joint scenarios (such as one joint every 50 meters in a tunnel); a single sensor relies on threshold judgment (such as alarm when the vibration amplitude exceeds 5g), which is easily affected by environmental noise (such as vibration from wind and rain), and the false alarm rate is as high as 15%-20%, leading to frequent misjudgments by maintenance personnel.
[0011] 3. Long-distance deployment is costly and management is complex: Traditional solutions require independent deployment of sensor networks for each monitoring function (strain, vibration, temperature and humidity), resulting in high equipment costs per kilometer; multiple systems store data independently (e.g., vibration data on platform A, OTDR data on platform B), forming "data silos", lacking cross-dimensional correlation analysis capabilities, and fault diagnosis relies on manual comprehensive judgment, which is inefficient. Summary of the Invention
[0012] One of the objectives of this invention is to provide a health monitoring system for fiber optic networks used in water diversion projects, thereby solving the aforementioned problems.
[0013] To achieve the above objectives, a health monitoring system for fiber optic networks in water diversion projects is provided, comprising:
[0014] Sensing module: used to synchronously acquire fiber optic environmental parameters through edge computing nodes and a multi-source sensor array; the multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors, and tensile sensors;
[0015] Transmission network construction module: used to form a star network of adjacent edge computing nodes through LoRa for short-distance networking; adopts industrial-grade PON fiber optic network, enabling edge nodes to access optical line terminal (OLT) through optical splitters to build a point-to-multipoint backbone transmission network.
[0016] The cloud management platform is used to preprocess the data acquired by the transmission network construction module to obtain vibration signals, strain data, and temperature and humidity data. The preprocessed data is then input into a preset DS evidence theory fusion model to obtain a warning trigger signal indicating whether a fault warning is triggered. In the DS evidence theory fusion model, vibration signals, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculations and confidence scores are performed based on the BPA function of each evidence body. When the final confidence score reaches a set threshold, a warning trigger signal is generated. The platform is also used to locate and classify faults using a trained fault classification model after an abnormal signal is detected at an edge node, and to output the fault location and fault type.
[0017] Furthermore, the vibration signal: uses a 50-500H fiber optic fault detection bandpass filter to remove environmental noise, and extracts several frequency components as feature values through fast Fourier transform; the strain data: uses the moving average method to filter out high-frequency interference, and calculates the strain change rate per minute as a feature value; the temperature and humidity data: performs polynomial fitting, removes the trend term, and calculates the deviation between the real-time value and the historical mean as a feature value; the strain data: uses the moving average method to filter out high-frequency interference, and calculates the strain change rate per minute as a feature value.
[0018] Furthermore, the cloud management platform includes a fault location module:
[0019] The fault location module is used to trigger the synchronous acquisition of optical signals by the two OTDR devices after an abnormal signal is detected at the edge node, and to obtain the backscattering curves S1(x) and S2(x); to perform wavelet denoising on the curves, identify the loss abrupt change points x1 and x2, and extract the loss gradient features; and to establish the loss gradient-physical distance mapping relationship using an SVM regression model, with the calculation formula as follows:
[0020]
[0021] Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber.
[0022] Furthermore, the cloud management platform includes a fault classification module:
[0023] Fault classification module: used to collect a fault sample library and build a fault classification model based on a BP neural network structure; it is also used to input abnormal signals detected by edge nodes into the trained fault classification model and output the fault type.
[0024] Furthermore, the edge computing node adopts a lightweight structural design, including a processor, a LoRa / PON dual communication module architecture, a solar panel, and a lithium battery.
[0025] Furthermore, the cloud management platform also includes:
[0026] Visualization module: Based on the fault type output by the fault classification module and the GIS map, the module loads the corresponding fault points onto the GIS map to display the real-time status of the optical cable and allows users to view historical data curves within a set time period through the fault points.
[0027] The second objective of this invention is to provide a health monitoring method for fiber optic networks used in water diversion projects, comprising:
[0028] Sensing steps: Simultaneously acquire fiber optic environmental parameters through edge computing nodes and a multi-source sensor array; the multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors, and tensile sensors;
[0029] Transmission network construction steps: Connect adjacent edge computing nodes to form a star network via LoRa for short-distance networking; use an industrial-grade PON fiber optic network to enable edge nodes to access the optical line terminal (OLT) via a splitter to build a point-to-multipoint backbone transmission network.
[0030] Health monitoring steps: The data acquired by the transmission network construction module is preprocessed to obtain vibration signals, strain data, and temperature and humidity data; the preprocessed data is input into a preset DS evidence theory fusion model to obtain a warning trigger signal for whether a fault warning is triggered. In the DS evidence theory fusion model, vibration signals, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculation and confidence calculation are performed according to the BPA function of a single evidence body. When the final confidence reaches a set threshold, a warning trigger signal is generated; it is also used to locate and classify faults through a trained fault classification model after an abnormal signal is detected at the edge node, and output the fault location and fault type.
[0031] Furthermore, the vibration signal: uses a 50-500H fiber optic fault detection bandpass filter to remove environmental noise, and extracts several frequency components as feature values through fast Fourier transform; the strain data: uses the moving average method to filter out high-frequency interference, and calculates the strain change rate per minute as a feature value; the temperature and humidity data: performs polynomial fitting, removes the trend term, and calculates the deviation between the real-time value and the historical mean as a feature value; the strain data: uses the moving average method to filter out high-frequency interference, and calculates the strain change rate per minute as a feature value.
[0032] Furthermore, the health monitoring steps specifically include the following steps:
[0033] Fault localization steps: After detecting an abnormal signal at the edge node, trigger the OTDR devices at both ends to synchronously acquire optical signals and obtain backscattering curves S1(x) and S2(x); perform wavelet denoising on the curves, identify loss abrupt change points x1 and x2, and extract loss gradient features; use an SVM regression model to establish the loss gradient-physical distance mapping relationship, the calculation formula is as follows:
[0034]
[0035] Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber.
[0036] Furthermore, the health monitoring steps specifically include the following steps:
[0037] Fault classification steps: Collect a fault sample library and build a fault classification model based on the BP neural network structure; input the abnormal signals detected by the edge nodes into the trained fault classification model and output the fault type.
[0038] Principles and advantages:
[0039] 1. Hardware Dimension: Multi-dimensional Perception and Lightweight Deployment
[0040] 1) Multi-source sensor array integration technology
[0041] By physically integrating fiber optic strain sensors, vibration acceleration sensors, temperature and humidity composite sensors, and tensile sensors, a three-dimensional sensing system with 12 core monitoring parameters (strain, vibration frequency, temperature and humidity, tensile force, etc.) is constructed, covering three major categories of fault causes in optical cables: mechanical stress, environmental erosion, and human interference.
[0042] Technical benefits: Compared with traditional single sensor solutions, the integrity of monitoring parameters is improved by 300%, and eight typical fault modes can be identified (such as strain abrupt changes caused by soil settlement and connector oxidation caused by high temperature and humidity).
[0043] 2) Lightweight design of edge computing nodes
[0044] It adopts a processor + LoRa / PON dual communication module architecture, integrates Modbus / TCP protocol conversion function, realizes real-time data acquisition and local preprocessing for 500H fiber optic fault detection, and supports hybrid power supply of solar energy + lithium battery.
[0045] Technical benefits: The coverage radius of a single node reaches 2km, reducing the sensor deployment density by 30% compared to traditional solutions, and reducing the equipment cost per kilometer from 50,000 yuan to below 35,000 yuan, thus solving the problem of high deployment costs for long-distance projects.
[0046] 2. Algorithm Dimension: Intelligent Fusion and Precise Diagnosis
[0047] 1) DS Evidence Theory Multi-Source Data Fusion Algorithm
[0048] Differential weights are defined for vibration signal (weight ω1), strain data (ω2), and temperature and humidity (ω3). A fusion model containing 12 basic probability assignment (BPA) rules is constructed, and an early warning is triggered when the confidence level reaches a threshold.
[0049] Technical effects: The fault identification accuracy has been improved from 80% with traditional single sensors to 98.7%, and the false alarm rate has been reduced from 18% to 1.3%, solving the problem of misjudgment caused by environmental noise interference.
[0050] 2) Two-ended OTDR-SVM regression localization algorithm
[0051] By synchronously acquiring optical signals from both ends, an SVM regression model is used to establish a mapping relationship between "loss gradient and physical distance".
[0052] Technical benefits: The positioning accuracy is improved from 100 meters for traditional single-ended OTDRs to ≤50 meters, eliminating near-end blind spots (reduced from 10 meters to 2 meters), and enabling rapid and accurate location of fault points. Attached Figure Description
[0053] Figure 1 This is an architecture diagram of a health monitoring system for a fiber optic network used in a water diversion project, according to an embodiment of the present invention.
[0054] Figure 2 Hardware schematic of an edge computing node;
[0055] Figure 3 A flowchart for multi-source data fusion;
[0056] Figure 4 A flowchart for fault location;
[0057] Figure 5 This is a schematic diagram of a BP neural network structure. Detailed Implementation
[0058] The following detailed description illustrates the specific implementation method:
[0059] Example
[0060] A health monitoring system for fiber optic networks used in water diversion projects, basically as follows: Figure 1 , Figure 2 , Figure 3 As shown, it includes:
[0061] Sensing module: used to synchronously collect fiber optic environmental parameters through edge computing nodes and multi-source sensor arrays; the multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors and tensile sensors. The multi-source sensor array is a composite sensor group, with one group deployed every 50 meters, and important nodes are densified to 20 meters.
[0062] The strain sensor is attached to the surface of the optical cable armor layer to monitor axial stress. In this embodiment, the strain sensor is an optical fiber strain sensor (model: FBG-ST01): it adopts the principle of Bragg fiber grating, is attached to the surface of the optical cable armor layer, monitors axial strain (accuracy ±0.001με), and is rigidly connected to the optical cable through 3M epoxy resin adhesive.
[0063] The vibration sensor is fixed to the optical cable bracket and is set close to the strain sensor. In this embodiment, the vibration sensor is a vibration acceleration sensor (model: ADXL375): with a range of ±16g, a built-in 200H fiber optic fault detection low-pass filter, and is fixed to the optical cable bracket by a magnetic clamp, with a distance of ≤10cm between it and the strain sensor.
[0064] The temperature and humidity sensor is a temperature and humidity composite sensor (model: SHT30): it integrates temperature (accuracy ±0.3℃) and humidity (accuracy ±2% RH) sensing modules, is encapsulated in an IP68 protective housing, and is connected to the edge computing node via an RS485 bus.
[0065] The tension sensor measures the axial tension of the optical cable. In this embodiment, the tension sensor (model: HBM U2A) is connected in series between the optical cable fixing hardware and the tower to measure the axial tension of the optical cable (accuracy 0.1% FS) and outputs a 4-20mA analog signal.
[0066] The edge computing node adopts a lightweight structure design, including an ARM Cortex-A72 processor (1.8GHz fiber optic fault detection, integrated with 2GB DDR4 memory and 16GB eMMC storage, supporting Python / C++ algorithm deployment), a LoRa / PON dual communication module architecture (simultaneously configured with a LoRa wireless module (470MHz frequency band fiber optic fault detection, 20dBm transmission power, 2km coverage radius) and an industrial-grade PON fiber optic module (supporting 1.25Gbps rate, 200km repeaterless transmission), connected to a multi-source sensor array via an RJ45 network port), and a power system. The power system includes a solar panel (10W) and a lithium battery (12V / 20Ah), and is also equipped with an energy management chip, with a sleep current of <10μA, supporting wide temperature operation from -40℃ to +70℃.
[0067] Transmission network construction module: Used to connect adjacent edge computing nodes into a star network via LoRa for short-distance networking; a single gateway can connect 200 sensor nodes, achieving low-power wide-area connectivity (power consumption reduced by 80% compared to traditional WiFi). Employing an industrial-grade PON fiber optic network, edge nodes connect to the Optical Line Terminal (OLT) via a splitter (1:32), constructing a point-to-multipoint backbone transmission network; supporting IP-based data encapsulation and QoS (Quality of Service) management to ensure real-time transmission of monitoring data.
[0068] Cloud management platform: includes data platform, AI analysis engine and visualization module;
[0069] Data platform: Based on Hadoop distributed architecture, it includes HDFS storage cluster (single cluster capacity ≥10PB) and Kafka message queue, supports concurrent writing of 100,000 data entries per second, and storage period ≥5 years.
[0070] AI Analysis Engine: Integrates TensorFlow / PyTorch framework, deploys DS evidence theory fusion model, fault classification model based on SVM regression localization algorithm and BP neural network classifier, and supports online training of DS evidence theory fusion model and fault classification model (weight parameters are automatically updated once per hour).
[0071] Visualization Module: This module loads the fault points from the fault classification module onto a GIS map to display the real-time status of the optical cable. Users can view historical data curves within a set timeframe by clicking on the fault point. In this embodiment, a 0.1-meter-accuracy GIS map is developed to display the real-time status of the optical cable (green - normal / yellow - warning / red - fault). Clicking on a fault point allows viewing historical data curves (strain, vibration, temperature, humidity, etc.) over the past 30 days.
[0072] The AI analysis engine is used to preprocess the data acquired by the transmission network construction module to obtain vibration signals, strain data, and temperature and humidity data. The preprocessed data is then input into a preset DS evidence theory fusion model to obtain a warning trigger signal indicating whether a fault warning is triggered. In this model, vibration signals, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculations and confidence scores are performed based on the BPA function of each evidence body. When the final confidence score reaches a set threshold, a warning trigger signal is generated. Furthermore, after an abnormal signal is detected at an edge node, a trained fault classification model is used to locate and classify the fault, outputting the fault location and fault type.
[0073] The preprocessing includes: for the vibration signal, a 50-500H fiber optic fault detection bandpass filter is used to remove environmental noise, and several frequency components (50H fiber optic fault detection, 100H fiber optic fault detection, 200H fiber optic fault detection, etc.) are extracted as feature values through fast Fourier transform; for the strain data, a moving average method (window size 100 points) is used to filter out high-frequency interference, and the strain change rate per minute (Δε / min) is calculated as a feature value; for the temperature and humidity data, a polynomial fitting (3rd order polynomial) is performed, and after removing the trend term, the deviation between the real-time value and the historical mean (ΔT / ΔRH) is calculated as a feature value.
[0074] The computational analysis of the DS evidence theory fusion model includes the following modules:
[0075] Basic Parameter Analysis Module: Used to analyze basic parameters; the basic parameters include:
[0076] Fault type set: θ={F1,F1,...,F8}; including mechanical fracture, stress caused by soil settlement, damage caused by human excavation, aging of optical cable sheath, oxidation of connector, wind vibration fatigue, damage caused by sudden temperature change, and abnormal tensile force caused by loose hardware;
[0077] Evidence includes: vibration signal E1, strain data E2, and temperature and humidity data E3;
[0078] BPA function for a single evidence body: m i (F j ), representing the support level of the i-th piece of evidence for the j-th type of fault;
[0079] Uncertainty:
[0080] The preliminary fusion calculation module is used to randomly select two pieces of evidence from vibration signal E1, strain data E2, and temperature and humidity data E3, and calculate the first joint support sum according to the joint support sum calculation formula; it also calculates the first conflict coefficient of the two pieces of evidence, reflecting the degree of contradiction between the two pieces of evidence; and calculates the intermediate support based on the first joint support sum and the first conflict coefficient. The joint support sum calculation formula is as follows:
[0081] S xy (F i )=m x (F j )×m y (F j )+m x (F j )×m y (θ)+m x (θ)×m y (F j ), xy∈i
[0082]
[0083] Among them, S xy (F i ) represents the sum of the first joint support, m xy (F i K represents the intermediate support. xy The first conflict coefficient;
[0084] The secondary fusion calculation module is used to fuse the intermediate support with the remaining evidence, calculate the second joint support sum according to the joint support sum calculation formula, and calculate the second conflict coefficient to reflect the degree of contradiction among the three pieces of evidence; and calculate the final confidence level based on the second joint support sum and the second conflict coefficient.
[0085] S xyz (F i )=m xy (F j )×m z (F j )+m xy (F j )×m z (θ)+m xy (θ)×m z (F j ), xyz∈i
[0086]
[0087] Among them, S xyz (F i ) represents the sum of the second joint support, m xyz (F i K represents the final confidence level. xyz This is the second conflict coefficient.
[0088] The AI analysis engine specifically includes a fault location module and a fault classification module.
[0089] Fault location module: such as Figure 4 As shown, this is used to trigger the synchronous acquisition of optical signals by the two OTDR devices after an abnormal signal is detected at the edge node, obtaining backscattering curves S1(x) and S2(x); wavelet denoising (db4 wavelet, 3-level decomposition) is performed on the curves to identify loss abrupt change points x1 and x2, and loss gradient features are extracted; the loss gradient-physical distance mapping relationship is established using an SVM regression model, and the calculation formula is as follows:
[0090]
[0091] Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber (1.467). Positioning accuracy is ≤50 meters (50% improvement over traditional single-ended OTDRs).
[0092] Fault classification module: Used to collect a fault sample library and build a fault classification model based on a BP neural network structure, such as... Figure 5 As shown; it is also used to input abnormal signals detected by edge nodes into a trained fault classification model and output the fault type.
[0093] In this embodiment, a fault sample library containing 8 types of faults (≥1000 samples per type) is established; the input parameters corresponding to the abnormal signals include 12 features such as strain peak value, vibration dominant frequency, temperature gradient, and humidity threshold; the fault sample library is shown in the table below:
[0094]
[0095] The BP neural network structure consists of an input layer of 12 neurons, two hidden layers (each with 30 / 15 neurons), and an output layer of 8 neurons (corresponding to the fault type). It is trained using the Adam optimizer (learning rate 0.001) until the classification accuracy is ≥98%.
[0096] A health monitoring method for fiber optic networks used in water diversion projects includes the following steps:
[0097] Sensing steps: Simultaneously acquire fiber optic environmental parameters through edge computing nodes and a multi-source sensor array; the multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors, and tensile sensors;
[0098] Transmission network construction steps: Connect adjacent edge computing nodes to form a star network via LoRa for short-distance networking; use an industrial-grade PON fiber optic network to enable edge nodes to access the optical line terminal (OLT) via a splitter to build a point-to-multipoint backbone transmission network.
[0099] Health monitoring steps: The data acquired by the transmission network construction module is preprocessed to obtain vibration signals, strain data, and temperature and humidity data; the preprocessed data is input into a preset DS evidence theory fusion model to obtain a warning trigger signal for whether a fault warning is triggered. In the DS evidence theory fusion model, vibration signals, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculation and confidence calculation are performed according to the BPA function of a single evidence body. When the final confidence reaches a set threshold, a warning trigger signal is generated; it is also used to locate and classify faults through a trained fault classification model after an abnormal signal is detected at the edge node, and output the fault location and fault type.
[0100] The computational analysis of the DS evidence theory fusion model includes the following steps:
[0101] Basic parameter analysis steps: Analyze basic parameters; the basic parameters include:
[0102] Fault type set: θ = {F1, F1, ..., F8};
[0103] Evidence includes: vibration signal E1, strain data E2, and temperature and humidity data E3;
[0104] BPA function for a single evidence body: m i (F j ), representing the support level of the i-th piece of evidence for the j-th type of fault;
[0105] Uncertainty:
[0106] Preliminary fusion calculation steps: Select any two pieces of evidence from vibration signal E1, strain data E2, and temperature and humidity data E3, and calculate the first joint support sum using the joint support sum calculation formula; calculate the first conflict coefficient between the two pieces of evidence to reflect the degree of contradiction between them; calculate the intermediate support based on the first joint support sum and the first conflict coefficient; the joint support sum calculation formula is as follows:
[0107] S xy (F i )=m x (F j )×m y (F j )+m x (F j )×m y (θ)+m x (θ)×m y (F j ), xy∈i
[0108]
[0109] Among them, S xy (F i ) represents the sum of the first joint support, m xy (F i K represents the intermediate support. xy The first conflict coefficient;
[0110] The second fusion calculation steps are as follows: the intermediate support is fused with the remaining evidence, and the second joint support is calculated according to the formula for calculating the joint support sum; the second conflict coefficient is calculated to reflect the degree of contradiction among the three pieces of evidence; the final confidence level is calculated based on the second joint support sum and the second conflict coefficient.
[0111] S xyz (F i )=m xy (F j )×m z (F j )+m xy (F j )×m z (θ)+m xy (θ)×m z (F j )
[0112]
[0113] Among them, S xyz (F i ) represents the sum of the second joint support, m xyz (F i K represents the final confidence level. xyz This is the second conflict coefficient.
[0114] The health monitoring process specifically includes the following steps:
[0115] Fault localization steps: After detecting an abnormal signal at the edge node, trigger the OTDR devices at both ends to synchronously acquire optical signals and obtain backscattering curves S1(x) and S2(x); perform wavelet denoising on the curves, identify loss abrupt change points x1 and x2, and extract loss gradient features; use an SVM regression model to establish the loss gradient-physical distance mapping relationship, the calculation formula is as follows:
[0116]
[0117] Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber.
[0118] Fault classification steps: Collect a fault sample library and build a fault classification model based on the BP neural network structure; input the abnormal signals detected by the edge nodes into the trained fault classification model and output the fault type.
[0119] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. A health monitoring system for fiber optic networks used in water diversion projects, characterized in that, include: Sensing module: used to synchronously acquire fiber optic environmental parameters through edge computing nodes and a multi-source sensor array; the multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors, and tensile sensors; Transmission network construction module: used to connect adjacent edge computing nodes into a star network via LoRa for short-distance networking; An industrial-grade PON fiber optic network is adopted, enabling edge nodes to access the optical line terminal (OLT) via optical splitters, thus constructing a point-to-multipoint backbone transmission network. Cloud management platform: used to preprocess the data acquired by the transmission network construction module to obtain vibration signals, strain data and temperature and humidity data; The preprocessed data is input into a preset DS evidence theory fusion model to obtain a warning trigger signal for whether a fault warning is triggered. In the DS evidence theory fusion model, vibration signal, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculation and confidence calculation are performed according to the BPA function of each evidence body. When the final confidence reaches a set threshold, a warning trigger signal is generated. It is also used to locate and classify faults through a trained fault classification model after an abnormal signal is detected at the edge node, and output the fault location and fault type. The computational analysis of the DS evidence theory fusion model includes the following modules: Basic Parameter Analysis Module: Used to analyze basic parameters; The basic parameters include: Fault type set: ; Evidence includes: vibration signal E1, strain data E2, and temperature and humidity data E3; BPA function for single evidence: , representing the support of the i-th piece of evidence for the j-th type of fault; Uncertainty: ; The preliminary fusion calculation module is used to randomly select two pieces of evidence from vibration signal E1, strain data E2, and temperature and humidity data E3, and calculate the first joint support sum according to the joint support sum calculation formula; it also calculates the first conflict coefficient of the two pieces of evidence, reflecting the degree of contradiction between the two pieces of evidence; and calculates the intermediate support based on the first joint support sum and the first conflict coefficient. The joint support sum calculation formula is as follows: ， in, The sum of the first joint support, For intermediate support, The first conflict coefficient; The secondary fusion calculation module is used to fuse the intermediate support with the remaining evidence, calculate the second joint support sum according to the joint support sum calculation formula, and calculate the second conflict coefficient to reflect the degree of contradiction among the three pieces of evidence; and calculate the final confidence level based on the second joint support sum and the second conflict coefficient. ， in, The sum of the second joint support, For the final confidence level, The second conflict coefficient; The cloud management platform includes a fault location module: The fault location module is used to trigger the synchronous acquisition of optical signals by the two OTDR devices after an abnormal signal is detected at the edge node, and to obtain the backscattering curves S1(x) and S2(x); to perform wavelet denoising on the curves, identify the loss abrupt change points x1 and x2, and extract the loss gradient features; and to establish the loss gradient-physical distance mapping relationship using an SVM regression model, the calculation formula being: Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber; Fault classification module: used to collect a fault sample library and build a fault classification model based on the BP neural network structure; it is also used to input abnormal signals detected by edge nodes into the trained fault classification model and output the fault type.
2. The health monitoring system for fiber optic networks used in water diversion projects according to claim 1, characterized in that: The edge computing node adopts a lightweight structure design, including a processor, a LoRa / PON dual communication module architecture, a solar panel, and a lithium battery.
3. A health monitoring system for fiber optic networks used in water diversion projects according to claim 1, characterized in that: The cloud management platform also includes: Visualization module: Based on the fault type output by the fault classification module and the GIS map, the module loads the corresponding fault points onto the GIS map to display the real-time status of the optical cable. It also allows users to view historical data curves within a set time period through the fault points.
4. A health monitoring method for fiber optic networks used in water diversion projects, characterized in that: include: Sensing steps: Synchronously collect fiber optic environmental parameters through edge computing nodes and multi-source sensor arrays; The multi-source sensor array includes strain sensors, vibration sensors, temperature and humidity sensors, and tensile sensors; Transmission network construction steps: Connect adjacent edge computing nodes to form a star network via LoRa for short-distance networking; An industrial-grade PON fiber optic network is adopted, enabling edge nodes to access the optical line terminal (OLT) via optical splitters, thus constructing a point-to-multipoint backbone transmission network. Health monitoring steps: Preprocess the data acquired by the transmission network construction module to obtain vibration signals, strain data, and temperature and humidity data; The preprocessed data is input into a preset DS evidence theory fusion model to obtain a warning trigger signal for whether a fault warning is triggered. In the DS evidence theory fusion model, vibration signal, strain data, and temperature and humidity data are defined as three evidence bodies, and fusion calculation and confidence calculation are performed according to the BPA function of each evidence body. When the final confidence reaches a set threshold, a warning trigger signal is generated. It is also used to locate and classify faults through a trained fault classification model after an abnormal signal is detected at the edge node, and output the fault location and fault type. The computational analysis of the DS evidence theory fusion model includes the following steps: Basic parameter analysis steps: Analyze basic parameters; the basic parameters include: Fault type set: ; Evidence includes: vibration signal E1, strain data E2, and temperature and humidity data E3; BPA function for single evidence: , representing the support of the i-th piece of evidence for the j-th type of fault; Uncertainty: ; Preliminary fusion calculation steps: Select any two pieces of evidence from vibration signal E1, strain data E2, and temperature and humidity data E3, and calculate the first joint support sum using the joint support sum calculation formula; calculate the first conflict coefficient between the two pieces of evidence to reflect the degree of contradiction between them; calculate the intermediate support based on the first joint support sum and the first conflict coefficient; the joint support sum calculation formula is as follows: ， in, The sum of the first joint support, For intermediate support, The first conflict coefficient; The second fusion calculation steps are as follows: the intermediate support is fused with the remaining evidence, and the second joint support is calculated according to the formula for calculating the joint support sum; the second conflict coefficient is calculated to reflect the degree of contradiction among the three pieces of evidence; the final confidence level is calculated based on the second joint support sum and the second conflict coefficient. ， in, The sum of the second joint support, For the final confidence level, The second conflict coefficient; The health monitoring process specifically includes the following steps: Fault localization steps: After detecting an abnormal signal at the edge node, trigger the OTDR devices at both ends to synchronously acquire optical signals and obtain backscattering curves S1(x) and S2(x); perform wavelet denoising on the curves, identify loss abrupt change points x1 and x2, and extract loss gradient features; use an SVM regression model to establish the loss gradient-physical distance mapping relationship, the calculation formula is as follows: Where c is the speed of light, t1 and t2 are the signal transmission times at both ends, and n is the refractive index of the optical fiber; Fault classification steps: Collect a fault sample library and build a fault classification model based on the BP neural network structure; input the abnormal signals detected by the edge nodes into the trained fault classification model and output the fault type.