Optical fiber long-distance self-checking method and system

Abstract

The invention relates to the technical field of optical fiber sensing, in particular to an optical fiber long-distance self-checking method and system, and the method comprises the steps of vibration phase collection, data fusion, adaptive adjustment and anomaly detection. The method comprises the following steps: firstly, collecting a vibration phase and an environment image along an optical fiber, and generating multi-modal convolution fusion data through quantum interference detection; then, self-adaptive adjustment is carried out according to environmental conditions, and a multi-modal recognition result adapting to environmental changes is generated. Furthermore, the recognition result is input into a double-path depth enhancement anomaly detection module, anomaly trend data is generated, and an early warning signal is generated. According to the invention, quantum interference and a multi-mode fusion technology are combined, so that the method has extremely high detection precision, quick response capability and noise adaptability, and can be widely applied to long-distance optical fiber state monitoring and safety early warning systems.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber sensing technology, and in particular to an optical fiber long-distance self-test method and system. Background Art

[0002] The long-distance self-test technology of optical fiber is of great significance in optical fiber monitoring and maintenance, especially in application scenarios where the environment is complex and the optical fiber is susceptible to external interference. It can realize real-time detection and early warning of vibration, stress and other conditions along the optical fiber, greatly improving the safety and reliability of the optical fiber. At present, the mainstream technology of long-distance optical fiber sensing (Chinese invention patent, publication number: CN118465692A, name: TDOA vibration source vertical distance detection method and device for long-distance sensing optical fiber) mostly adopts TDOA (arrival time difference) method, which collects the time difference of vibration signals through multiple optical fiber sensing units, and then calculates the vibration source distance in the vertical direction of the optical fiber in combination with the trigonometric function model. However, this method has certain limitations, including the need for cumbersome calibration of the nearest sensing unit, high computational complexity, and sensitivity to noise and changes in environmental media, resulting in limited accuracy and difficulty in meeting the needs of fast and accurate detection in complex environments. Summary of the invention

[0003] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for long-distance self-inspection of optical fiber. The present invention combines quantum interference detection with a multi-modal data fusion method to first collect vibration, stress and other data along the optical fiber in multiple frequency bands and multiple modes; then, the collected multi-modal data is fused, denoised and adaptively adjusted through a convolutional neural network to form a structured recognition result; finally, based on a dual-path deep enhanced anomaly detection process, abnormal events are accurately identified and early warning signals are generated. The present invention effectively solves the sensitivity of traditional optical fiber self-inspection systems to noise and environmental interference, and significantly improves detection accuracy and response speed.

[0004] An optical fiber long-distance self-test method comprises the following steps:

[0005] The vibration phase changes along the optical fiber are collected to generate resonance phase vibration data, echo characteristic data is collected based on the stress state of the internal structure of the optical fiber, and the vibration signal is layered and collected in multiple frequency bands to generate weighted vibration data;

[0006] Input the resonance phase vibration data, echo characteristic data and weighted vibration data into a multi-layer dynamic feedback network to generate adaptive feedback data, and identify the changes in the internal stress of the optical fiber through nonlinear signal processing to form layered feedback vibration data for multi-band feedback;

[0007] Collect environmental images along the optical fiber to generate environmental image data, and fuse the environmental image data with adaptive feedback data to generate multimodal convolution fusion data; adaptively adjust the multimodal convolution fusion data according to environmental conditions to generate multimodal recognition result data;

[0008] The multimodal recognition result data is input into the anomaly detection process to generate anomaly judgment data, and the anomaly trend data is generated through deep enhanced anomaly detection. The stability of the system is evaluated based on the multimodal joint entropy and early warning signal data is generated.

[0009] The resonance phase vibration data, echo feature data, multimodal recognition result data and warning signal data generated during the detection process are summarized, and key features are extracted to generate a self-inspection report file.

[0010] Preferably, the process of collecting the vibration phase change along the optical fiber to generate the resonant phase vibration data includes: expressing the vibration signal as a phase change amount in the quantum interference detection process, and calculating the resonant response amount through piezoelectric response and phase adjustment, wherein the phase change amount Δφ is expressed by the following formula:

[0011] Δφ=k·A·sin(ω·t)

[0012] Among them, Δφ is the phase change of the vibration signal; k is the piezoelectric constant of the piezoelectric response; A is the amplitude of the vibration signal; ω is the angular frequency of the vibration signal; and t is time.

[0013] The phase change Δφ is used to enhance the detection sensitivity of tiny vibration signals to generate resonant phase vibration data containing phase characteristics.

[0014] Preferably, the resonant phase vibration data generation process includes: inputting the collected vibration phase change data into the piezoelectric response module, adjusting the piezoelectric resonance according to the vibration response characteristics at different frequencies, thereby gradually optimizing the phase characteristics of the vibration signal, and generating resonant phase vibration data including the optical fiber stress state and internal phase change.

[0015] Preferably, the resonant phase vibration data, echo feature data and weighted vibration data are input into a multi-layer dynamic feedback network, a graph neural network is used to establish a transmission path between signal nodes, and the feedback data is dynamically adjusted based on the timing model of the recursive neural network to generate adaptive feedback data, thereby characterizing the layered feedback vibration data in different frequency bands.

[0016] Preferably, the nonlinear signal processing step includes: performing segmented cross-correlation analysis on echo feature data, calculating the similarity and difference of different segments of the echo signal, extracting key data points of the optical fiber stress characteristics, and fusing the key data points of the optical fiber stress characteristics into the adaptive feedback data.

[0017] Preferably, environmental images along the optical fiber are collected to generate environmental image data, and the environmental image data are denoised and edge feature enhanced through a convolutional neural network to generate image feature data, and the image feature data is fused with adaptive feedback data in a multimodal convolution interaction layer to generate multimodal convolution fusion data.

[0018] Preferably, adaptively adjusting the multimodal convolution fusion data according to environmental conditions includes: adjusting the image feature data, quantum phase vibration data and echo feature data in the multimodal convolution fusion data in real time according to the parameters of environmental temperature, humidity and noise, so as to generate multimodal recognition result data that adapts to environmental changes.

[0019] Preferably, the multimodal recognition result data is input into the anomaly detection process, anomaly judgment data is generated through the dual-path deep enhanced anomaly detection process, and preliminary anomaly features are identified through the fast detection path, and the changing trend of the abnormal signal is further identified in the deep detection path to generate abnormal trend data containing abnormal types, which is used to determine the type and frequency of abnormal events.

[0020] Preferably, the stability of the multimodal joint entropy evaluation system includes: calculating the quantum phase vibration data, the echo feature data and the image feature data into corresponding entropy values, respectively expressed as S q , S e and S i , the mutual information between multimodal data is determined by the joint entropy calculation formula, where the joint entropy S joint The calculation formula is:

[0021]

[0022] Among them, S joint represents the joint entropy of multimodal data; p j Represents the joint probability distribution value of quantum phase vibration data, echo feature data and image feature data.

[0023] When the joint entropy S joint When the preset stability threshold is exceeded, early warning signal data is generated to indicate abnormal conditions of the internal stress state of the optical fiber or changes in environmental characteristics.

[0024] A system for implementing the optical fiber long-distance self-test method, comprising:

[0025] Quantum interference detection module, used to collect vibration phase changes along the optical fiber and generate resonant phase vibration data;

[0026] A stress state acquisition module, used to acquire echo characteristic data based on the stress state of the internal structure of the optical fiber;

[0027] Multi-frequency layered acquisition module, used to perform layered acquisition of low-frequency, medium-frequency and high-frequency bands of vibration signals to generate weighted vibration data;

[0028] A multi-layer dynamic feedback network module is used to receive resonance phase vibration data, echo characteristic data and weighted vibration data, generate adaptive feedback data, and identify changes in internal stress of the optical fiber through nonlinear signal processing to form layered feedback vibration data for multi-band feedback;

[0029] An image acquisition module, used for acquiring environmental images along the optical fiber to generate environmental image data;

[0030] A multimodal convolution fusion module is used to fuse the environmental image data with the adaptive feedback data to generate multimodal convolution fusion data, and to adaptively adjust the multimodal convolution fusion data according to the environmental conditions to generate multimodal recognition result data;

[0031] Anomaly detection module, used to receive multimodal recognition result data, generate anomaly determination data, generate anomaly trend data through deep enhanced anomaly detection, and generate early warning signal data based on the stability of the multimodal joint entropy evaluation system;

[0032] The data aggregation and report generation module is used to aggregate the resonance phase vibration data, echo feature data, multimodal recognition result data and warning signal data generated during the detection process, extract key features and generate a self-inspection report file.

[0033] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0034] The present invention realizes highly sensitive detection of phase changes of optical fiber vibration signals through quantum interference detection and piezoelectric resonance adjustment, can accurately capture tiny vibrations and stress state changes, and improve the accuracy of vibration source detection;

[0035] The present invention adopts multi-modal data convolution fusion and adaptive adjustment methods to achieve unified processing of multiple data types (such as vibration data, echo characteristics and environmental image data), effectively reduce the interference of environmental noise on the signal, and ensure the robustness of data fusion;

[0036] The present invention realizes efficient anomaly recognition through dual-path deep enhanced anomaly detection. Combined with the multimodal joint entropy evaluation system, it can identify potential abnormal trends in advance and generate early warning signals, significantly improving the sensitivity and early warning capability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the process of the present invention;

[0038] Figure 2It is a schematic diagram of the dual-path deep enhanced anomaly detection process in the present invention;

[0039] Figure 3 It is an internal relationship diagram of dual-path deep enhanced anomaly detection in the present invention;

[0040] Figure 4 It is a feedback relationship diagram of quantum phase and multi-modal data fusion in the present invention;

[0041] Figure 5 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0042] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0043] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0044] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0045] like Figure 1 As shown, a method for long-distance self-test of optical fiber comprises the following steps:

[0046] The vibration phase changes along the optical fiber are collected to generate resonance phase vibration data, echo characteristic data is collected based on the stress state of the internal structure of the optical fiber, and the vibration signal is layered and collected in multiple frequency bands to generate weighted vibration data;

[0047] In the present invention, the process of collecting vibration phase changes along the optical fiber to generate resonant phase vibration data, collecting echo characteristic data based on the stress state of the internal structure of the optical fiber, and performing multi-band layered collection of vibration signals to generate weighted vibration data is one of the key steps in long-distance self-test of optical fiber. This step combines quantum interference technology, piezoelectric response and multi-band modulation technology to achieve detection of optical fiber vibration, stress changes and multi-band characteristics, and finally forms comprehensive self-test data for subsequent data analysis and feedback control.

[0048] Specifically, the process of collecting vibration phase changes along the optical fiber to generate resonant phase vibration data utilizes the high-sensitivity detection capability of quantum interference technology to capture tiny vibration phase changes. In principle, the phase of the light wave transmitted by the optical fiber is accurately monitored by a quantum interference detector. When the optical fiber is subjected to tiny external vibrations at its position along the line, the phase of the optical fiber will shift accordingly. The offset is represented as vibration phase change data, and these phase changes are adjusted by the piezoelectric response module. The piezoelectric response module adaptively amplifies the phase change signal generated by quantum interference based on the response characteristics of the vibration signal at different frequencies to generate resonant phase vibration data. The resonant phase vibration data has high accuracy and can reflect the subtle vibration characteristics of the position along the optical fiber.

[0049] On this basis, echo characteristic data is collected based on the stress state of the internal structure of the optical fiber, and the change information of the internal structure of the optical fiber is obtained by ultrasonic signal feedback. The stress state of the optical fiber will produce certain ultrasonic reflections (i.e., echo signals) inside the optical fiber, and these signals are captured and processed by a nonlinear ultrasonic echo feedback module. For the echo characteristic data in the present invention, nonlinear ultrasonic technology will convert different stress state characteristics into different ultrasonic echo modes, and the differences in these modes reflect the local stress conditions or structural defects inside the optical fiber. Therefore, the echo characteristic data is used as a reliable stress detection method in the present invention, which is not only used to analyze the local health state of the optical fiber, but also provides an important basis for subsequent feedback and abnormality detection.

[0050] The process of performing multi-band layered acquisition of vibration signals to generate weighted vibration data is mainly to separate the characteristic information of vibration signals in different frequency bands. Multi-band acquisition uses multi-frequency modulation technology. By layered acquisition of low-frequency bands, mid-frequency bands and high-frequency bands of vibration signals, the vibration signals are classified according to frequency characteristics. This layered acquisition method can effectively distinguish the contribution of different frequency characteristics in the vibration signal. For example, low-frequency band signals may correspond to the slow deformation of the optical fiber as a whole or a longer section, while high-frequency band signals may be related to instantaneous impacts in local stress concentration areas. Therefore, through the combination of layered acquisition and multi-frequency modulation technology, the present invention can not only capture the vibration characteristics of the optical fiber as a whole, but also detect local high-frequency vibration information.

[0051] The layered acquisition module processes the vibration signals of the low-frequency, medium-frequency and high-frequency bands respectively, and assigns specific weights to the vibration signals of each frequency band through the frequency weighting algorithm to generate weighted vibration data. After modulation and weight distribution, the weighted vibration data can reflect the contribution of signals of different frequency bands to the overall vibration characteristics of the optical fiber. This data not only plays an important role in the subsequent dynamic feedback, but can also be used in the adaptive adjustment process to ensure that the feedback signal can more accurately reflect the actual state of the vibration source.

[0052] Preferably, the process of collecting the vibration phase change along the optical fiber to generate the resonant phase vibration data includes: expressing the vibration signal as a phase change amount in the quantum interference detection process, and calculating the resonant response amount through piezoelectric response and phase adjustment, wherein the phase change amount Δφ is expressed by the following formula:

[0053] Δφ=k·A·sin(ω·t)

[0054] Among them, Δφ is the phase change of the vibration signal; k is the piezoelectric constant of the piezoelectric response; A is the amplitude of the vibration signal; ω is the angular frequency of the vibration signal; and t is time.

[0055] The phase change Δφ is used to enhance the detection sensitivity of tiny vibration signals to generate resonant phase vibration data containing phase characteristics.

[0056] The step of collecting vibration phase changes along the optical fiber to generate resonant phase vibration data is the key to achieving accurate detection of tiny vibrations. This step is based on quantum interference technology, combined with the adjustment of the piezoelectric response module, to achieve high-sensitivity detection and regulation of optical fiber vibration signals. In optical fiber detection applications, since the amplitude changes of vibration signals are extremely subtle, quantum interference detectors convert these tiny vibrations into phase changes through optical phase shifts to form quantum phase vibration data. This process uses precise phase monitoring to capture and quantify the subtle changes of each vibration source along the optical fiber.

[0057] In actual operation, the quantum interference detector tracks the phase of the light wave in the optical fiber. When external vibration acts on the optical fiber, the phase of the light wave will shift accordingly, resulting in a phase change Δφ. The piezoelectric response module adjusts this phase change so that the phase response matches the actual vibration signal. Specifically, the piezoelectric response module will respond and adjust according to the different frequency characteristics of the vibration signal, and finally form resonant phase vibration data to reflect the vibration characteristics and changes.

[0058] The mathematical formula of phase change shows that the phase change Δφ is determined by the characteristic constant k of the piezoelectric response module, the amplitude A of the vibration signal, and the angular frequency ω. By performing piezoelectric adjustment on vibration signals of different frequencies and intensities, this process not only enhances the detection sensitivity of tiny signals, but also makes the generated resonant phase vibration data more accurately reflect the actual condition of the optical fiber vibration.

[0059] In specific applications, assuming that there are a series of vibration sources of different frequencies along the optical fiber, such as equipment vibration and environmental noise along the optical fiber transmission path, the piezoelectric response module will respond and adjust these signals one by one to ensure that the vibration signal of each frequency band can produce an accurate phase change. The role of this frequency adjustment is that the piezoelectric module can dynamically adapt to the characteristics of different vibration sources to avoid signal distortion or interference, so that the resonant phase vibration data finally generated can be used for subsequent hierarchical analysis and feedback control. Finally, through the combination of the above-mentioned quantum interference detection and the piezoelectric response module, the present invention realizes the high-precision capture of tiny vibration changes along the optical fiber, provides basic data for long-distance self-inspection of optical fiber, and ensures the accuracy and real-time performance of the detection results.

[0060] Preferably, the resonant phase vibration data generation process includes: inputting the collected vibration phase change data into the piezoelectric response module, adjusting the piezoelectric resonance according to the vibration response characteristics at different frequencies, thereby gradually optimizing the phase characteristics of the vibration signal, and generating resonant phase vibration data including the optical fiber stress state and internal phase change.

[0061] In the optical fiber long-distance self-test of the present invention, the generation process of the resonant phase vibration data mainly depends on the adjustment and optimization of the piezoelectric response module. The process first inputs the collected vibration phase change data into the piezoelectric response module, and then adjusts the signal for piezoelectric resonance according to the vibration response characteristics at different frequencies, thereby gradually optimizing the phase characteristics of the vibration signal, and finally generating the resonant phase vibration data containing the optical fiber stress state and internal phase change. The design purpose of this process is to optimize the phase change of the vibration signal within different frequency ranges, ensure that the phase characteristics of different frequency bands can be effectively retained, and make the subsequent stress detection more accurate.

[0062] Specifically, the vibration phase change data input into the piezoelectric response module is actually the light phase change caused by the tiny vibrations along the optical fiber. The piezoelectric response module adjusts the resonance state of the signal according to the piezoelectric response amplitude and sensitivity at different vibration frequencies by adjusting the physical properties of the piezoelectric material. In order to achieve optimal processing of the signal, the present invention adopts a frequency division adjustment method, that is, the vibration phase change data is subdivided according to the frequency characteristics, so that the signal of each frequency band obtains the best response at the corresponding piezoelectric resonance frequency. This resonance adjustment not only enhances the response amplitude of the signal, but also effectively filters out the noise of non-target frequencies, thereby retaining key phase information.

[0063] In specific implementation, piezoelectric resonance regulation can be described by the following formula:

[0064] Δφ′=k·A·sin(ω·t)·f(ω)

[0065] Among them, Δφ′ is the phase change after piezoelectric resonance adjustment; k is the response constant of the piezoelectric material; A is the amplitude of the original vibration signal; ω is the angular frequency of the vibration signal; t is time; f(ω) is the frequency-dependent resonance response function, which is used to adjust the resonance characteristics of the signal at different frequencies.

[0066] Through the above formula, the piezoelectric response module can selectively amplify signals of different frequencies. Assume that there are two vibration sources in the detection area of ​​the optical fiber, namely high-frequency mechanical vibration and low-frequency environmental noise. The piezoelectric response module can amplify the high-frequency vibration source more strongly according to the resonant response characteristics at high and low frequencies, while attenuating the low-frequency noise, so that the resonant phase vibration data finally generated can better reflect the characteristics of the high-frequency vibration source. In this way, the resonant phase vibration data processed by the piezoelectric response module not only contains the phase change information under the stress state of the optical fiber, but also reflects the phase characteristics of vibration signals in different frequency bands, thereby realizing the optimized processing of vibration signals.

[0067] From the application effect, the resonant phase vibration data generated by piezoelectric resonance adjustment can significantly improve the accuracy and signal quality of vibration signal detection along the optical fiber. This data provides highly reliable data support for subsequent dynamic feedback and stress analysis.

[0068] Input the resonance phase vibration data, echo characteristic data and weighted vibration data into a multi-layer dynamic feedback network to generate adaptive feedback data, and identify the changes in the internal stress of the optical fiber through nonlinear signal processing to form layered feedback vibration data for multi-band feedback;

[0069] In the optical fiber long-distance self-test method of the present invention, the resonance phase vibration data, echo characteristic data and weighted vibration data are input into a multi-layer dynamic feedback network to generate adaptive feedback data, and the changes in the internal stress of the optical fiber are identified through nonlinear signal processing. This process aims to fuse and process the multi-modal vibration and stress data through a dynamic feedback network, thereby constructing an adaptive system that can identify and feedback stress changes in different frequency bands, and realize accurate monitoring and response to the internal stress state of the optical fiber.

[0070] Specifically, the resonant phase vibration data, echo characteristic data and weighted vibration data represent the comprehensive characteristics of the optical fiber in vibration phase, stress reflection characteristics and multi-band vibration respectively. The multi-layer dynamic feedback network inputs these data into its multiple network layers at the same time to realize the synchronous processing and correlation analysis of multimodal data. The dynamic feedback network adopts a combination of graph neural network and recursive neural network, in which the role of graph neural network is to establish the signal transmission path between different data nodes and realize the correlation modeling of multimodal data; recursive neural network is used for time series modeling, and the input multimodal data is iteratively processed in the time dimension, so that the feedback network can capture the dynamic change characteristics of the data and generate adaptive feedback data with high sensitivity to different vibration frequency bands.

[0071] In practical applications, each layer of the multi-layer dynamic feedback network uses the output of the current layer as the input of the next layer. Each layer of network nodes will not only gradually integrate the vibration and stress data, but also gradually form a feedback structure through adaptive adjustment of the signal to reflect the stress changes inside the optical fiber. Through this layer-by-layer iterative processing, the adaptive feedback data generated by the multi-layer dynamic feedback network can be dynamically adjusted according to the real-time input vibration and stress information, thereby ensuring that the feedback results can accurately reflect the stress state inside the optical fiber. In order to improve the accuracy of the adaptive feedback data, the present invention also uses nonlinear signal processing technology to perform segmented analysis on the echo feature data in the feedback network, and uses cross-correlation analysis to extract the stress change characteristics in the signal, thereby generating layered feedback vibration data that can adapt to the feedback requirements of different frequency bands.

[0072] Specifically, the processing of feedback data by nonlinear signal processing technology can be described as the following formula:

[0073]

[0074] Among them, F feedback (t) represents the feedback intensity of the layered feedback vibration data at time t; f j represents the nonlinear adjustment factor for the j-th layer feedback, which is used to adjust the response of different feedback layers to multi-band signals; C j(t) represents the timing signal response of the j-th layer feedback node, which is used to reflect the characteristics of the current feedback data at this layer.

[0075] Through this formula, the feedback network can adaptively adjust the response in the layered feedback structure according to the characteristics of vibration signals in different frequency bands to achieve flexible reflection of stress changes inside the optical fiber. For example, when a sudden high-frequency vibration occurs at a certain position along the optical fiber, the feedback network can respond quickly in the high-frequency feedback layer and adjust the feedback strength to generate feedback data reflecting this sudden high-frequency signal, thereby presenting the stress concentration area inside the optical fiber in the layered feedback vibration data.

[0076] After being processed by a multi-layer dynamic feedback network, the generated adaptive feedback data can not only accurately reflect the stress changes along the optical fiber, but also has high sensitivity to multi-band signals. The layered feedback vibration data generated thereby ensures the real-time and accuracy of the optical fiber long-distance self-inspection by responding to signals in different frequency bands, and provides a reliable feedback mechanism for subsequent stress monitoring and abnormal warning.

[0077] Preferably, the resonant phase vibration data, echo feature data and weighted vibration data are input into a multi-layer dynamic feedback network, a graph neural network is used to establish a transmission path between signal nodes, and the feedback data is dynamically adjusted based on the timing model of the recursive neural network to generate adaptive feedback data, thereby characterizing the layered feedback vibration data in different frequency bands.

[0078] In the present invention, the process of inputting the resonant phase vibration data, echo feature data and weighted vibration data into the multi-layer dynamic feedback network is to realize the dynamic modeling and timing adjustment of the data by combining the graph neural network and the recursive neural network. Specifically, the graph neural network is used to establish the transmission path between each signal node in the feedback network, so as to effectively integrate and analyze the vibration and stress data along the optical fiber. At the same time, the recursive neural network dynamically processes these signals in time series, so that the feedback system can adaptively reflect the stress and vibration changes inside the optical fiber in different time and frequency dimensions.

[0079] First, the role of the graph neural network is to treat multimodal data (i.e., resonant phase vibration data, echo feature data, and weighted vibration data) as connected signals between different nodes. In the graph structure, each node represents the vibration or stress data at a specific location, and the edges between nodes represent the signal propagation relationship between these locations. Specifically, the graph neural network spatially propagates the vibration and stress signals throughout the optical fiber structure through the transmission path design between nodes, thereby forming a multi-layer feedback structure. By adjusting the weights between nodes, the graph neural network realizes the weighted propagation of different signals on different paths. For example, for low-frequency vibration signals, the graph neural network can enhance long-distance propagation through larger weights, while appropriately weakening the weights for high-frequency signals, thereby improving the feedback network's ability to spatially distinguish signals in different frequency bands.

[0080] Next, the recursive neural network processes the feedback data in the time dimension, allowing the network to identify and track the dynamic changes of the signal. Based on the timing model, the recursive neural network uses the data output at each moment as the input of the next time node, and combines the adaptive feedback process to achieve dynamic adjustment of multimodal data. For example, in long-distance self-inspection of optical fibers, the recursive neural network can capture the changing trend of stress signals in a continuous time series. Through the recursive calculation of the time step, the system can identify the changes in the vibration signal at a certain location at different times, which is particularly critical for capturing sudden stress events or continuous vibration characteristics.

[0081] In order to integrate the characteristics of graph neural networks and recurrent neural networks into the hierarchical feedback network, the feedback data generation of the present invention can be described by the following formula:

[0082]

[0083] Among them, F adaptive (t) is the adaptive feedback data generated at time t; W ij represents the transfer weight between node i and node j in the graph neural network, which is used to adjust the transmission strength of signals between different nodes; h j (t-1) is the output of node j at time t-1; f RNN is the recursive function of the recursive neural network, which is used to generate the time series data of the current node; F i (t-1) is the input data of node i at time t-1.

[0084] Through the above formula, the present invention updates all node data at time t, so that the feedback network can adaptively adjust the output according to the change at the moment, thereby realizing the generation of multi-band layered feedback vibration data.

[0085] For example, at a certain position on the optical fiber, if the system detects low-frequency background vibration, the graph neural network can assign a larger weight to the signal to achieve effective signal propagation over long distances; and when high-frequency local vibration is detected, the recursive neural network can perform high-frequency response processing on the signal, and through adaptive adjustment, the feedback network can track the changes in high-frequency signals more sensitively, thereby generating enhanced feedback for high-frequency signals. Ultimately, through the synergy of graph neural networks and recursive neural networks, the generated adaptive feedback data can be hierarchically adjusted according to the dynamic changes of multi-band signals, thereby more comprehensively and accurately characterizing the stress and vibration characteristics of different frequency bands along the optical fiber, ensuring real-time feedback and efficient monitoring during the long-distance self-inspection of the optical fiber.

[0086] Preferably, the nonlinear signal processing step includes: performing segmented cross-correlation analysis on echo feature data, calculating the similarity and difference of different segments of the echo signal, extracting key data points of the optical fiber stress characteristics, and fusing the key data points of the optical fiber stress characteristics into the adaptive feedback data.

[0087] In the self-test process of the present invention, the nonlinear signal processing step performs segmented cross-correlation analysis on the echo feature data to accurately identify the characteristic points of stress changes inside the optical fiber. The core of this process is to divide the echo signal into different sections, calculate the similarity and difference of the signals in each section through cross-correlation analysis, and thus extract key data points with important stress characteristics. Fusion of these key data points into adaptive feedback data helps to enhance the feedback system's ability to respond to changes in optical fiber stress, and provides high-resolution data support for subsequent stress monitoring and abnormal warning.

[0088] Specifically, the echo feature data is the ultrasonic echo signal generated when the stress in the optical fiber structure changes. The signal usually contains vibration components in multiple frequency bands and small fluctuations caused by changes in the stress state. In order to accurately capture these changes, the nonlinear signal processing step first divides the echo signal into several sections according to the time period and performs cross-correlation analysis on each section. Cross-correlation analysis is a statistical method used to quantify the similarity between two signal sequences, which can effectively identify the similarities and differences between signals in different sections.

[0089] In the present invention, the segmented cross-correlation calculation formula of the echo characteristic data can be expressed as:

[0090]

[0091] Among them, R xy (τ) is the cross-correlation value between paragraphs, which is used to evaluate the similarity between different paragraphs; x(n) and y(n) represent the echo signals in two different paragraphs respectively; τ is the time delay, and the similarity between signals at different times can be calculated by adjusting the delay value.

[0092] In practical applications, suppose that the echo signal of a certain section contains obvious high-frequency fluctuations, while other sections show slow changes at low frequencies. Through cross-correlation analysis, the similarity between this section and other sections will be low, indicating that the signal here may contain a mutation point of stress characteristics. On the contrary, if the echo signal similarity of the two sections is high, it means that the signal here is relatively stable and there may be no significant stress changes.

[0093] After cross-correlation analysis, the solution extracts those sections with low similarity and high difference as key data points of optical fiber stress characteristics. By incorporating these key data points into the adaptive feedback data, these points with prominent stress characteristics can be magnified in the feedback system, thereby achieving more accurate stress change detection. For example, if the echo signal of a certain section of the optical fiber is detected to be significantly different from that of other locations, this difference data can be directly embedded in the adaptive feedback data, allowing the feedback system to quickly identify and locate the area of ​​sudden stress change in the optical fiber.

[0094] Through this nonlinear processing method of segmented cross-correlation analysis, the present invention can capture the subtle stress change characteristics in the optical fiber at high resolution and enhance the sensitivity and response speed of the feedback data. The adaptive feedback data finally generated is more comprehensive and accurate as a whole, which can support the real-time monitoring of the stress state of the optical fiber and the early warning of abnormal changes in the long-distance self-inspection of the optical fiber, and improve the stability and detection effect of the system.

[0095] Collect environmental images along the optical fiber to generate environmental image data, and fuse the environmental image data with adaptive feedback data to generate multimodal convolution fusion data; adaptively adjust the multimodal convolution fusion data according to environmental conditions to generate multimodal recognition result data;

[0096] In the present invention, the relationship between the external environment change and the internal stress of the optical fiber is comprehensively and dynamically reflected through the multimodal fusion of the environmental image and the optical fiber feedback data. This fusion process uses a convolutional neural network (CNN) to process the environmental image data, and fuses the image and feedback data through a multimodal convolution layer, and finally generates multimodal recognition result data that is adaptively adjusted based on environmental conditions.

[0097] Specifically, the process of collecting environmental images along the optical fiber to generate environmental image data depends on image sensors arranged along the optical fiber. These image sensors periodically collect real-time images of the environment along the line and convert key visual information in the environment into digital image data. In the convolutional neural network, the environmental image data will first be preprocessed, mainly including steps such as image denoising and edge feature extraction, to retain important visual features in the environment and eliminate possible noise interference. For example, rainy and foggy weather in the environmental image, displacement of objects around the optical fiber, etc., will have a potential impact on the optical fiber stress. The image processing module will pay special attention to the edge feature information of these influencing factors, thereby generating clear and processed environmental image data.

[0098] Next, the processed environmental image data and adaptive feedback data are input into the multimodal convolutional fusion layer. The function of the multimodal convolutional fusion layer is to perform deep correlation analysis on image features and feedback data through multi-layer convolution operations, thereby forming multimodal fusion data. In this process, the multimodal convolutional fusion layer will extract features from the spatial features of the environmental image and the temporal features of the feedback data respectively, and then perform pixel-by-pixel feature mapping and data registration in the convolutional fusion layer. This convolutional fusion can not only establish a one-to-one spatial association between the environmental image and the feedback data, but also identify the response characteristics of the feedback signal under specific environmental conditions.

[0099] In implementation, the calculation of fused data can be described by the following expression:

[0100]

[0101] Among them, F multi (x, y) represents the fusion result of multimodal convolution fusion data at the pixel position (x, y); I(x+i, y+j) represents the pixel value of the environmental image data in the convolution kernel window; D(x+i, y+j) represents the value of the adaptive feedback data at the corresponding position; W i,j and V i,j is the weight parameter of the convolution kernel, which is used to adjust the fusion weight of the environment image and the feedback data respectively.

[0102] Through this fusion convolution process, the system can generate multimodal convolution fusion data that contains environmental impact and feedback signal characteristics. These fusion data have the ability to simultaneously reflect changes in environmental conditions and dynamic characteristics of feedback signals, providing highly integrated information for the next step of adaptive recognition.

[0103] Finally, the multimodal convolution fusion data is adaptively adjusted according to the environmental conditions to generate multimodal recognition result data. This adaptive adjustment process mainly considers the possible impact of environmental changes along the optical fiber on the feedback data. The system will adjust the weight parameters of the fusion data in real time based on preset environmental conditions, such as temperature, humidity, and light. Specifically, different weight coefficients are assigned to different environmental factors through the adaptive adjustment layer to adapt to dynamic environmental changes. For example, in a high temperature environment, the system will appropriately increase the weight of the temperature on the recognition result, and in high humidity conditions, it will increase the humidity weight. This dynamic adjustment ensures that the final recognition result can truly reflect the actual stress state of the optical fiber and reduce the interference of environmental noise.

[0104] In summary, by processing the multimodal convolution of environmental image data and adaptive feedback data, and adaptively adjusting according to real-time environmental conditions, the multimodal recognition result data finally generated has the ability to accurately identify the stress changes of optical fibers under complex environmental conditions, providing comprehensive and accurate environmental and stress correlation feedback for long-distance optical fiber self-inspection. This recognition result data provides a comprehensive reference for subsequent stress analysis and abnormal warning, and improves the adaptability and recognition accuracy of the optical fiber self-inspection system in different environments.

[0105] Preferably, environmental images along the optical fiber are collected to generate environmental image data, and the environmental image data are denoised and edge feature enhanced through a convolutional neural network to generate image feature data, and the image feature data is fused with adaptive feedback data in a multimodal convolution interaction layer to generate multimodal convolution fusion data.

[0106] In the present invention, the visual information of the optical fiber's surrounding environment and the optical fiber's internal stress feedback signal are integrated into a multi-modal convolution structure to improve the accuracy and adaptability of the self-inspection system under complex environmental conditions.

[0107] First, the environmental image data generated by collecting environmental images along the optical fiber is completed by image sensors arranged along the optical fiber. These sensors regularly collect real-time images of the environment along the line, including external factors such as the terrain, climate change, and weather conditions outside the optical fiber, and convert the environmental information of the optical fiber into a continuous image data stream. The image data contains various visual information that may affect the stress of the optical fiber. For example, when the optical fiber passes through a bridge or a construction area, the vehicle or mechanical vibration in the environment will affect the stress state of the optical fiber. Therefore, digitizing this information into image data helps to capture changes in the external environment of the optical fiber during the self-inspection process.

[0108] Next, the environmental image data is denoised and edge feature enhanced through a convolutional neural network (CNN) to generate high-quality image feature data for subsequent analysis. In this process, the convolutional neural network first denoises the image data through multiple convolutional layers and pooling layers to eliminate noise interference in the image and retain key environmental information. This denoising process is particularly important for external factors such as weather conditions or dynamically changing objects in the environment, which can enhance the stability of the fiber environment image. The denoised image data is then further processed through the edge feature enhancement layer. In this step, the convolutional neural network enhances important edge information in the image, such as the outline of structures around the fiber path and areas of terrain mutation, so that the system can identify the potential impact of these environmental features on fiber stress.

[0109] The edge feature enhancement processing of the convolutional neural network on the image can be expressed by the following convolution expression:

[0110]

[0111] Among them, E(x,y) is the image feature data after edge enhancement processing; K i,j Represents the weight of the convolution kernel, which is used to extract image edge features; I(x+i,y+j) is the pixel value of the image after denoising.

[0112] Through this processing, the system obtains denoised and enhanced image feature data, which helps to improve the ability to identify key environmental factors in subsequent analysis.

[0113] In the multimodal convolution interaction layer, the image feature data is fused with the adaptive feedback data to generate multimodal convolution fusion data. The core of this process is to achieve dynamic interaction between the fiber's surrounding environment and its internal stress feedback through multimodal fusion. The adaptive feedback data is the internal feedback data obtained by dynamically adjusting the stress and vibration signals along the optical fiber, while the image feature data provides visual information of the external environment. Through the multimodal convolution interaction layer, the system inputs these different modal data into the same convolution structure to achieve cross-modal feature matching and fusion.

[0114] In the specific implementation, the multimodal convolution interaction layer will perform feature mapping on the input data of each layer, and use the convolution kernel to perform pixel-by-pixel weighted fusion of the environmental image feature data and the adaptive feedback data. The fusion process can be expressed as the following formula:

[0115]

[0116] Among them, F fusion(x, y) represents the fusion result of multimodal convolution fusion data at the pixel position (x, y); E(x+i, y+j) is the pixel value of the image feature data in the convolution window; D(x+i, y+j) is the value of the adaptive feedback data at the corresponding position; W i,j and V i,j are the fusion weight parameters of image feature data and adaptive feedback data, which control the convolution weighting of the two types of data respectively.

[0117] In this fusion convolution process, the system can integrate the corresponding information of image features and feedback data into multimodal convolution fusion data, realizing real-time dynamic interaction between environment and stress feedback. For example, when encountering sudden changes in terrain or severe climate along the optical fiber, the fusion data can simultaneously reflect the drastic changes in the external environment of the optical fiber and the stress response of the internal stress, thereby providing comprehensive detection results.

[0118] The multimodal convolution fusion data finally generated not only includes the changing characteristics of the external environment of the optical fiber, but also reflects the dynamic response of the internal stress feedback, so that the optical fiber long-distance self-inspection system can achieve higher detection accuracy and adaptability in complex environments. This data fusion method improves the robustness of the self-inspection system, ensures that the system can timely identify potential stress influencing factors, provide comprehensive and accurate feedback information, and facilitate timely response to abnormal detection and early warning.

[0119] Preferably, adaptively adjusting the multimodal convolution fusion data according to environmental conditions includes: adjusting the image feature data, quantum phase vibration data and echo feature data in the multimodal convolution fusion data in real time according to the parameters of environmental temperature, humidity and noise, so as to generate multimodal recognition result data that adapts to environmental changes.

[0120] In the present invention, according to dynamic parameters such as ambient temperature, humidity and noise, the image feature data, quantum phase vibration data and echo feature data in the multimodal convolution fusion data are adjusted in real time, thereby generating multimodal recognition result data that can adapt to different environmental changes. The purpose of this adaptive adjustment process is to enable the detection system to more accurately reflect the stress state and vibration characteristics inside the optical fiber under complex external conditions, and ensure the stability and accuracy of the detection.

[0121] Specifically, the multimodal convolution fusion data includes the characteristic data of the environmental image along the optical fiber, the quantum phase data of the optical fiber vibration, and the stress state data obtained by echo feature analysis. The performance of these data in a complex environment is affected by factors such as temperature, humidity and noise. For example, a high temperature environment will cause the expansion of the optical fiber material, humidity changes will change the propagation speed of sound waves in the optical fiber, and environmental noise will affect the vibration detection accuracy of the optical fiber. In order to achieve environmental adaptability, the present invention uses an adaptive weight adjustment module to assign different weights to each modal data in the multimodal convolution fusion data to adapt to the current environmental conditions.

[0122] In actual implementation, the adaptive adjustment module sets the weight adjustment function based on the real-time environmental parameters and adjusts each modal data in the multimodal fusion data through weighted calculation. For example, when the ambient temperature is high, the adaptive adjustment module will reduce the weight of the image feature data and increase the weight of the quantum phase vibration data to compensate for the impact of temperature on image detection accuracy. The adjustment calculation formula can be expressed as:

[0123] F adaptive =α(T)·F image +β·(H)·F quantum +γ(N)·F ech o

[0124] Among them, F adaptive is the multimodal recognition result data after adaptive adjustment; α(T) is the weight adjustment coefficient related to the ambient temperature T, which adjusts the image feature data F image β·(H) is the weight adjustment coefficient related to the ambient humidity H, which adjusts the quantum phase vibration data F quantum γ(N) is the weight adjustment coefficient related to the environmental noise N, which adjusts the echo feature data F ech o The weight of .

[0125] Through this weighted fusion, the system can dynamically adjust the contribution ratio of various types of data in the fusion process according to environmental parameters. For example, assuming that in an environment with high humidity, the noise level of quantum phase vibration data is high, the weight coefficient β·(H) of the humidity parameter will be reduced accordingly to reduce the interference of humidity changes on the recognition results. In an environment with strong noise, the weight of the echo feature data will be appropriately reduced, making the recognition results of the optical fiber self-inspection system more stable and accurate.

[0126] For example, in a hot and humid environment in summer, the combined effects of temperature and humidity along the optical fiber may cause image features to become blurred and the signal-to-noise ratio of the adaptive feedback data to decrease. Under such conditions, the system will dynamically adjust the weight parameters so that the weight of the quantum phase vibration data increases, while the weight of the image feature data decreases. In this way, even if the environmental image along the optical fiber is not clear enough, the recognition result can still reflect the actual stress state inside the optical fiber, effectively avoiding the accumulation of errors caused by changes in temperature and humidity.

[0127] The resulting adaptive multimodal recognition result data has higher stability and accuracy under different environmental conditions. This adaptive adjustment based on real-time environmental parameters ensures the adaptability of the fiber self-test system to various complex environments, enhances the detection accuracy during long-distance fiber self-test, enables the system to effectively provide real-time feedback and abnormal warning, and ensures the safe operation of the fiber.

[0128] The multimodal recognition result data is input into the anomaly detection process to generate anomaly judgment data, and the anomaly trend data is generated through deep enhanced anomaly detection. The stability of the system is evaluated based on the multimodal joint entropy and early warning signal data is generated.

[0129] like Figure 2 As shown, in the present invention, through a multi-level detection and evaluation process, a comprehensive analysis of data such as internal stress of the optical fiber and external environmental impact is achieved, thereby providing high-precision anomaly detection and early warning. This process enables the system to quickly and accurately identify potential anomalies through deep enhanced anomaly detection and joint entropy evaluation, and then issue an early warning before the system risks occur, effectively ensuring the safety and stability of the optical fiber.

[0130] Specifically, the step of inputting the multimodal recognition result data into the anomaly detection process and generating the anomaly judgment data first performs basic anomaly detection on the recognition result data. The multimodal recognition result data contains data of multiple modes such as environmental image features, quantum phase vibration information, and echo features. These data have reflected the internal stress changes and environmental conditions of the optical fiber through previous fusion and adaptive adjustment. Based on these data, the anomaly detection module uses specific thresholds and discriminant models to identify potential abnormal signals. In this process, the anomaly judgment data is used to identify stress states or vibration characteristics that may exceed the normal range. For example, when the intensity of the vibration signal of a specific frequency is detected to exceed the standard range, or the stress data under specific environmental conditions suddenly changes, the system will identify it as an anomaly and generate anomaly judgment data as input for further in-depth detection.

[0131] On this basis, deep enhanced anomaly detection further analyzes the anomaly judgment data through a deep learning model to generate anomaly trend data. Deep enhanced anomaly detection uses a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to identify the time series trends and spatial features of abnormal signals. Specifically, CNN is responsible for extracting characteristic patterns in abnormal signals, such as a sharp rise in vibration amplitude or a sudden change in image features, while RNN is used to process the time evolution trend of these features and determine the change speed and trend direction of abnormal signals. By modeling the trend of abnormal signals, deep enhanced detection can distinguish between sudden anomalies and gradually changing anomalies, and the generated abnormal trend data reflects the evolution of abnormal signals over time. For example, for a gradually increasing stress data trend, deep detection can identify its upward trend at an early stage, so that appropriate early warning measures can be taken before the anomaly escalates into a serious failure.

[0132] The process of evaluating the stability of the system and generating early warning signal data based on multimodal joint entropy evaluates the current stability of the system by calculating the joint entropy of multimodal data. The multimodal joint entropy reflects the mutual information between quantum phase vibration data, echo feature data and image feature data, thereby judging the overall consistency and stability of the system.

[0133] By calculating the joint entropy, the system can identify the degree of coupling between multimodal data, thereby generating a warning signal when the data is inconsistent or has low correlation. Specifically, when the joint entropy value is low, it means that the correlation between the modal data is high and the system stability is good; when the joint entropy value exceeds the preset stability threshold, the system generates warning signal data, indicating that there may be unstable fiber stress or sudden changes in the external environment. For example, under normal operating conditions, the joint entropy of quantum phase vibration data, echo feature data, and image feature data may remain within a stable range; but when the ambient temperature rises sharply and causes abnormal vibration data, the entropy value will increase accordingly, thereby triggering a system warning.

[0134] Through the above process, the warning signal data finally generated can not only timely remind the system of abnormal status, but also judge the severity of the abnormality based on the abnormal trend and the stability of multimodal data. This multi-level abnormality detection and evaluation method enables the optical fiber long-distance self-inspection system to monitor and identify potential risks in real time in complex environments, and achieve high-precision and high-sensitivity fault warning, ensuring that the system responds to sudden abnormalities at an early stage and takes preventive measures.

[0135] Preferably, Figure 3As shown, the multimodal recognition result data is input into the anomaly detection process, anomaly judgment data is generated through the deep enhanced anomaly detection process of the dual paths, and preliminary anomaly features are identified through the fast detection path. The changing trend of the abnormal signal is further identified in the deep detection path, and anomaly trend data containing anomaly types is generated to determine the type and frequency of abnormal events.

[0136] In the optical fiber long-distance self-inspection system, the multimodal recognition result data is input into the anomaly detection process, and the anomaly judgment data is generated through the dual-path deep enhanced anomaly detection process to further conduct in-depth analysis of the abnormal state of the optical fiber. This process uses the combination of fast detection path and deep detection path to achieve efficient preliminary anomaly identification and detailed trend analysis, thereby providing data support for subsequent monitoring and ensuring that the optical fiber can timely identify the type and frequency of abnormal events under long-distance monitoring.

[0137] Specifically, the dual-path deep enhanced anomaly detection process is divided into two main detection paths: a fast detection path and a deep detection path. First, the core of the fast detection path is to quickly process the input multimodal recognition result data based on the convolutional neural network (CNN) so as to identify preliminary abnormal features in the shortest time. This path focuses on the initial features and sudden changes of abnormal signals. Through simple convolution kernel calculations and fast threshold judgments, potential abnormal events can be efficiently identified without increasing the computational complexity. For example, during the detection process, if an instantaneous high-frequency signal appears in the echo feature data of a certain area, the fast detection path will directly identify the sudden feature through the set abnormal threshold and input it as abnormal judgment data into the next step of deep analysis.

[0138] The calculation process of the fast detection path can be expressed by the following formula:

[0139]

[0140] Among them, A fast (x, y) represents the preliminary abnormal feature output of the fast detection path for the input data at position (x, y); K i,y is the convolution kernel weight, which is used to extract fast detection features; D(x+i,y+j) is the value of the multimodal recognition result data in the convolution window.

[0141] The design of this fast detection path enables the system to capture the initial characteristics of anomalies in the shortest possible time, ensuring a rapid response when an emergency occurs along the optical fiber.

[0142] After the initial abnormal feature recognition is completed, the system will further input the quickly detected abnormal judgment data into the deep detection path, and use the recursive neural network (RNN) in the deep detection path to conduct a detailed analysis of the time series of abnormal signals to identify the changing trend of abnormal signals. The deep detection path mainly models the time series changes of signals and identifies the evolution laws of abnormal signals at multiple time points, thereby generating abnormal trend data containing abnormal types. In this process, the recursive neural network can capture subtle trend changes in abnormal signals, help the system distinguish different types of abnormalities, such as instantaneous abnormalities and gradually growing abnormalities, and count the frequency of each abnormal type.

[0143] The trend analysis process of recurrent neural network can be expressed by the following formula:

[0144] T depth (t) = f RNN (A fast (t-1),A fast (t))

[0145] Among them, T depth (t) is the abnormal trend data generated by the deep detection path at time t; A fast (t-1) and A fast (t) are the preliminary abnormal feature data of the fast detection path at time points t-1 and t respectively; f RNN is the recursive function of the recursive neural network, which is used to calculate the changes in abnormal trends.

[0146] By analyzing the abnormal trend through the deep detection path, the system can effectively identify different types of abnormal events. For example, if a high-frequency vibration signal continues to appear at a certain position along the optical fiber, the system will identify the signal as a "continuous abnormality" through trend analysis and count its frequency to assess the impact. These abnormal trend data include the type, frequency and development trend of the abnormal signal, which facilitates the system to more comprehensively understand the stress state inside the optical fiber.

[0147] The abnormal trend data finally generated not only includes the accurate classification of abnormal events (such as sudden or incremental abnormalities), but also reveals their frequency of occurrence, providing a detailed basis for further abnormal warning and maintenance. This dual-path detection method combines the efficiency of the fast detection path with the accuracy of the deep detection path, making the abnormal identification and analysis of the optical fiber long-distance self-inspection system in complex environments more comprehensive, helping to ensure the safe operation of the system and the real-time detection.

[0148] Preferably, the stability of the multimodal joint entropy evaluation system includes: calculating the quantum phase vibration data, the echo feature data and the image feature data into corresponding entropy values, respectively expressed as S q , Se and S i , the mutual information between multimodal data is determined by the joint entropy calculation formula, where the joint entropy S joint The calculation formula is:

[0149]

[0150] Among them, S joint represents the joint entropy of multimodal data; p j Represents the joint probability distribution value of quantum phase vibration data, echo feature data and image feature data.

[0151] When the joint entropy S joint When the preset stability threshold is exceeded, early warning signal data is generated to indicate abnormal conditions of the internal stress state of the optical fiber or changes in environmental characteristics.

[0152] In the present invention, the overall stability and consistency of the system are evaluated in real time by joint entropy analysis of quantum phase vibration data, echo feature data and image feature data. The core of this process is to quantify the mutual information between different modal data to determine whether they are coordinated and consistent, and then detect abnormal changes in the internal stress state of the optical fiber and environmental characteristics. By calculating the joint entropy of multimodal data, the system can identify potential abnormal patterns in the optical fiber self-test data and generate early warning signals to indicate significant changes in the optical fiber structure state or external environment.

[0153] Specifically, this step first calculates the multimodal data into corresponding entropy values, which are defined as follows: S q : Entropy value of quantum phase vibration data, indicating the randomness and uncertainty of quantum phase change characteristics along the optical fiber. e : Entropy value of echo characteristic data, indicating the complexity of stress distribution inside the optical fiber reflected by the ultrasonic signal. i : The entropy value of the image feature data reflects the change information of the image features of the optical fiber surrounding environment, including terrain, obstacles and other environmental factors.

[0154] The calculation of entropy is completed by statistically analyzing the probability distribution of different modal data. The larger the entropy, the higher the uncertainty in the data, that is, the system may be in a less stable state. Therefore, independent entropy values ​​can provide a separate stability assessment for each modal data, but cannot reflect the interdependence between the data.

[0155] In order to further judge the synergy and stability between multimodal data, this paper introduces the joint entropy (S joint ), which calculates the mutual information between different data modalities by analyzing the joint probability distribution of quantum phase vibration, echo characteristics and image characteristics. The calculation formula of joint entropy is shown above.

[0156] The size of the joint entropy value indicates the consistency and dependence between multimodal data. joint When the value is low, it means that the correlation between different modal data is high, the system is consistent with the environment, and is in a relatively stable state. When the joint entropy value increases and exceeds the preset stability threshold, it means that the consistency between multimodal data decreases, and the system may be inconsistent. For example, in some areas, the echo feature data suddenly increases, while the quantum phase vibration data remains stable, then the joint entropy value may increase, indicating that there is a sudden abnormality in the internal stress of the optical fiber in this area.

[0157] Through real-time monitoring of the joint entropy value, when S joint When the set threshold is exceeded, the system automatically generates warning signal data, indicating that the stress state of the optical fiber or environmental factors may have a sudden change. For example, if the internal stress of an optical fiber suddenly increases due to geological activities such as earthquakes, the vibration component in the echo feature data increases significantly, but the quantum phase data and environmental image features do not change accordingly. This multimodal inconsistency phenomenon will lead to S joint Exceeding the threshold triggers an alert.

[0158] In summary, the multimodal joint entropy evaluation system effectively incorporates the collaborative information of the internal and external states of the optical fiber into detection and early warning by calculating the joint entropy of quantum phase vibration data, echo feature data, and image feature data. When the joint entropy value exceeds the set threshold, the system generates early warning signal data to achieve a rapid response to abnormal stress inside the optical fiber and sudden changes in the external environment, thereby improving the sensitivity and safety of the optical fiber long-distance self-inspection system.

[0159] The resonance phase vibration data, echo feature data, multimodal recognition result data and warning signal data generated during the detection process are summarized, and key features are extracted to generate a self-inspection report file.

[0160] In the present invention, Figure 4 As shown in the figure, the last step is to summarize and process various data generated during the detection process, including resonant phase vibration data, echo feature data, multimodal recognition result data, and warning signal data, in order to extract key features and generate a comprehensive self-test report file. This process not only provides a comprehensive analysis of the current state of the optical fiber, but also realizes effective monitoring of the system in long-distance detection, ensuring that managers can judge the health status of the optical fiber based on the report information and make necessary maintenance decisions.

[0161] First, the resonant phase vibration data and echo feature data reflect the internal vibration stress and structural stress characteristics of the optical fiber, respectively. The resonant phase vibration data records the phase changes of the optical fiber vibration signal generated by the quantum phase and piezoelectric response, which are particularly important for identifying the response of the optical fiber under external stress. At the same time, the echo feature data is obtained by ultrasonic echo feedback analysis, describing the stress distribution and elastic characteristics of the internal structure of the optical fiber. The combination of the two provides the system with dynamic monitoring of the internal physical state of the optical fiber. This step summarizes the two types of data to identify potential abnormal points and stress change areas, thereby providing raw data support for the key feature analysis of the report.

[0162] Secondly, the multimodal recognition result data is the product of the fusion of environmental image data and vibration stress feedback data, which mainly records the interaction between the external environment changes along the optical fiber and the stress state of the optical fiber. This data is formed through multimodal convolution fusion and adaptive adjustment, which can reflect the immediate impact of the external environment on the optical fiber state, such as the regulation of vibration characteristics by factors such as temperature, humidity and noise. The summary analysis of the multimodal recognition result data can describe the overall impact of the environment in which the optical fiber is located, so that the self-inspection report contains not only the internal structural state, but also the pressure data of the external environment.

[0163] The warning signal data is generated based on the multimodal joint entropy assessment, which is used to identify stress or environmental mutations that are beyond the normal range. When the joint entropy value indicates that the inconsistency of the multimodal data is higher than the set threshold, the system will trigger a warning signal to indicate potential abnormal events. The summary of the warning signal data allows the reporting system to record specific warning events, types, times, locations, and other information in the self-check report file, providing a basis for managers to respond in real time.

[0164] During the data aggregation process, the system automatically identifies representative key feature points from various types of data through feature extraction algorithms. Feature extraction includes the extraction of important information such as abnormal trends, stress concentration points, and sudden vibration patterns, and formats them into readable items in the report. For example, the system can use cluster analysis algorithms to identify high stress areas at multiple locations as "stress concentration points" and mark them in the report to facilitate managers to assess the potential risks of the area.

[0165] After feature extraction is completed, the system summarizes all key features to generate a self-test report file. This report file organizes data according to dimensions such as location, time, and event type, and includes real-time health status of the fiber status, historical abnormal events, and trend analysis. For example, the report may contain detailed information such as phase change statistics for each monitoring location, environmental image comparison analysis, and abnormal frequency charts to help managers understand the operating performance of optical fibers under complex conditions. At the same time, the report also includes time series trends and event markers of warning data, providing data support for judging the overall stability of optical fiber operation.

[0166] The final self-test report file can provide detailed data basis for the maintenance and management of optical fiber. Through comprehensive analysis of different modal data, it provides systematic, accurate and operational test results for long-distance optical fiber monitoring.

[0167] like Figure 5 As shown, a system for implementing the optical fiber long-distance self-test method comprises:

[0168] The quantum interference detection module is used to collect vibration phase changes along the optical fiber and generate resonant phase vibration data; the quantum interference detection module collects vibration phase changes along the optical fiber through a quantum interference detector, and uses the high sensitivity of quantum interference to convert tiny phase changes in the optical fiber into observable vibration data. The phase change of vibration can reflect the impact of external stress on the optical fiber, and detect subtle changes in vibration frequency and amplitude in the phase interference signal. The generated resonant phase vibration data is used for subsequent feedback processing to ensure that the system can respond to vibration changes on the optical fiber in real time and improve the sensitivity and accuracy of optical fiber self-test.

[0169] The stress state acquisition module is used to collect echo characteristic data based on the stress state of the internal structure of the optical fiber. The module detects the stress state of the internal structure of the optical fiber through the stress state sensor, and mainly relies on the echo characteristic data to reflect the internal elastic changes of the optical fiber. The sensor emits an ultrasonic signal and monitors its reflection in the optical fiber. The change of the reflected wave can indicate the stress condition inside the optical fiber. The characteristic data of the ultrasonic echo provides a reference value of the internal stress for the optical fiber self-inspection, so that the system can determine whether the optical fiber is in a normal state or stress overloaded.

[0170] The multi-frequency layered acquisition module is used to perform layered acquisition of vibration signals in low-frequency, medium-frequency and high-frequency bands to generate weighted vibration data; the multi-frequency layered acquisition module uses a layered signal collector to perform layered acquisition of vibration signals in the optical fiber in low-frequency, medium-frequency and high-frequency bands to generate weighted vibration data. The purpose of layered acquisition is to capture different aspects of the optical fiber vibration characteristics through the acquisition of multiple frequency intervals. The layered acquired vibration data can filter out noise interference in different frequency bands, thereby generating more reference-worthy multi-frequency vibration characteristic data while keeping the signal clear. The weighted vibration data can accurately reflect the response characteristics of the optical fiber in each frequency band and is suitable for subsequent feedback analysis.

[0171] The multi-layer dynamic feedback network module is used to receive the resonant phase vibration data, echo characteristic data and weighted vibration data, generate adaptive feedback data, and identify the changes in the internal stress of the optical fiber through nonlinear signal processing, forming layered feedback vibration data for multi-band feedback; the multi-layer dynamic feedback network module is the core computing unit of the system, which receives the resonant phase vibration data, echo characteristic data and weighted vibration data, and generates adaptive feedback data through a multi-layer dynamic feedback mechanism. This module uses nonlinear signal processing methods to analyze the stress changes inside the optical fiber, and forms layered feedback vibration data in different frequency bands through recursion and dynamic feedback mechanisms. This feedback data is used to identify subtle stress changes inside the optical fiber, realize adaptive response of layered feedback, and enhance the multi-band detection capability of the system.

[0172] The image acquisition module is used to collect environmental images along the optical fiber to generate environmental image data. The image acquisition module collects image data along the optical fiber through the environmental image sensor to generate image information reflecting the state of the environment around the optical fiber. This module can capture the temperature, humidity or other factors that may affect the state of the optical fiber in the environment where the optical fiber is located. These image data are combined with vibration and stress data as data input for the multimodal convolution fusion module to improve the system's adaptive ability in a changing environment.

[0173] The multimodal convolution fusion module is used to fuse the environmental image data with the adaptive feedback data to generate multimodal convolution fusion data, and adaptively adjust the multimodal convolution fusion data according to the environmental conditions to generate multimodal recognition result data; the module performs convolution fusion processing on the environmental image data and the adaptive feedback data through the multimodal convolution processor, and uses the multi-layer feature extraction and adaptive adjustment functions of the convolutional neural network to fuse the image data and the feedback data into multimodal convolution fusion data. The system adaptively adjusts the fusion data according to the environmental conditions to generate multimodal recognition result data, thereby ensuring the accuracy of optical fiber status recognition in complex environments.

[0174] The anomaly detection module is used to receive multimodal recognition result data, generate anomaly determination data, generate anomaly trend data through deep enhanced anomaly detection, and generate early warning signal data based on the stability of the multimodal joint entropy evaluation system; the anomaly detection module uses the deep enhanced detection unit to analyze the multimodal recognition result data, and generates anomaly determination data and abnormal trend data through a dual-path deep enhanced anomaly detection process. This module combines the multimodal joint entropy evaluation system to analyze the relationship between data, ensure that early warning signals can be issued in advance when the system stability decreases or the environment changes drastically, and realize intelligent detection of abnormal trends.

[0175] The data aggregation and report generation module is used to aggregate the resonant phase vibration data, echo feature data, multimodal identification result data, and warning signal data generated during the detection process, and extract key features to generate a self-test report file. This module aggregates and processes various types of data (including resonant phase vibration data, echo feature data, multimodal identification result data, and warning signal data) generated during the detection process through a data aggregation processor, and extracts key features to generate a self-test report file. The self-test report file finally generated provides information on state changes along the optical fiber, provides data support for system maintenance and fault warning, and ensures the reliability of long-distance optical fiber monitoring.

[0176] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0177] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for long-distance self-test of optical fiber, characterized in that: The following steps are involved: The vibration phase changes along the optical fiber are collected to generate resonance phase vibration data, echo characteristic data is collected based on the stress state of the internal structure of the optical fiber, and the vibration signal is layered and collected in multiple frequency bands to generate weighted vibration data; Input the resonance phase vibration data, echo characteristic data and weighted vibration data into a multi-layer dynamic feedback network to generate adaptive feedback data, and identify the changes in the internal stress of the optical fiber through nonlinear signal processing to form layered feedback vibration data for multi-band feedback; Collect environmental images along the optical fiber to generate environmental image data, and fuse the environmental image data with adaptive feedback data to generate multimodal convolution fusion data; adaptively adjust the multimodal convolution fusion data according to environmental conditions to generate multimodal recognition result data; The multimodal recognition result data is input into the anomaly detection process to generate anomaly judgment data, and the anomaly trend data is generated through deep enhanced anomaly detection. The stability of the system is evaluated based on the multimodal joint entropy and early warning signal data is generated. The resonance phase vibration data, echo feature data, multimodal recognition result data and warning signal data generated during the detection process are summarized, and key features are extracted to generate a self-inspection report file.

2. The optical fiber long-distance self-test method according to claim 1, characterized in that: The process of collecting the vibration phase change along the optical fiber to generate the resonance phase vibration data includes: expressing the vibration signal as a phase change amount in the quantum interference detection process, calculating the resonance response amount through the piezoelectric response and phase adjustment, wherein the phase change amount Δφ is expressed by the following formula: Δφ=k·A·sin(ω·t) Among them, Δφ is the phase change of the vibration signal; k is the piezoelectric constant of the piezoelectric response; A is the amplitude of the vibration signal; ω is the angular frequency of the vibration signal; and t is time. The phase change Δφ is used to enhance the detection sensitivity of tiny vibration signals to generate resonant phase vibration data containing phase characteristics.

3. The optical fiber long-distance self-test method according to claim 2, characterized in that: The resonant phase vibration data generation process includes: inputting the collected vibration phase change data into the piezoelectric response module, adjusting the piezoelectric resonance according to the vibration response characteristics at different frequencies, thereby gradually optimizing the phase characteristics of the vibration signal, and generating resonant phase vibration data including the optical fiber stress state and internal phase change.

4. The optical fiber long-distance self-test method according to claim 1, characterized in that: The resonant phase vibration data, echo feature data and weighted vibration data are input into a multi-layer dynamic feedback network, and a graph neural network is used to establish the transmission path between signal nodes. The feedback data is dynamically adjusted based on the timing model of the recursive neural network to generate adaptive feedback data to characterize the layered feedback vibration data in different frequency bands.

5. The optical fiber long-distance self-test method according to claim 1, characterized in that: The nonlinear signal processing steps include: performing segmented cross-correlation analysis on echo feature data, calculating the similarity and difference of different segments of the echo signal, extracting key data points of the optical fiber stress characteristics, and integrating the key data points of the optical fiber stress characteristics into the adaptive feedback data.

6. The optical fiber long-distance self-test method according to claim 1, characterized in that: Environmental images along the optical fiber are collected to generate environmental image data, and the environmental image data is denoised and edge feature enhanced through a convolutional neural network to generate image feature data, and the image feature data is fused with the adaptive feedback data in a multimodal convolution interaction layer to generate multimodal convolution fusion data.

7. The optical fiber long-distance self-test method according to claim 6, characterized in that: Adaptive adjustment of multimodal convolution fusion data according to environmental conditions includes: real-time weight adjustment of image feature data, quantum phase vibration data and echo feature data in the multimodal convolution fusion data according to parameters of environmental temperature, humidity and noise, to generate multimodal recognition result data that adapts to environmental changes.

8. The optical fiber long-distance self-test method according to claim 1, characterized in that: The multimodal recognition result data is input into the anomaly detection process, and the anomaly judgment data is generated through the dual-path deep enhanced anomaly detection process. The preliminary anomaly features are identified through the fast detection path, and the changing trend of the abnormal signal is further identified in the deep detection path. The abnormal trend data containing the abnormal type is generated to determine the type and frequency of abnormal events.

9. The optical fiber long-distance self-test method according to claim 1, characterized in that: The stability of the multimodal joint entropy evaluation system includes: calculating the quantum phase vibration data, echo feature data and image feature data into corresponding entropy values, respectively expressed as S q , S e and S i , the mutual information between multimodal data is determined by the joint entropy calculation formula, where the joint entropy S joint The calculation formula is: Among them, S joint represents the joint entropy of multimodal data; p j Represents the joint probability distribution value of quantum phase vibration data, echo feature data and image feature data. When the joint entropy S joint When the preset stability threshold is exceeded, early warning signal data is generated to indicate abnormal conditions of the internal stress state of the optical fiber or changes in environmental characteristics.

10. A system for implementing the optical fiber long-distance self-test method according to any one of claims 1 to 9, characterized in that: include: Quantum interference detection module, used to collect vibration phase changes along the optical fiber and generate resonant phase vibration data; A stress state acquisition module, used to acquire echo characteristic data based on the stress state of the internal structure of the optical fiber; Multi-frequency layered acquisition module, used to perform layered acquisition of low-frequency, medium-frequency and high-frequency bands of vibration signals to generate weighted vibration data; A multi-layer dynamic feedback network module is used to receive resonance phase vibration data, echo characteristic data and weighted vibration data, generate adaptive feedback data, and identify changes in internal stress of the optical fiber through nonlinear signal processing to form layered feedback vibration data for multi-band feedback; An image acquisition module, used for acquiring environmental images along the optical fiber to generate environmental image data; A multimodal convolution fusion module is used to fuse the environmental image data with the adaptive feedback data to generate multimodal convolution fusion data, and to adaptively adjust the multimodal convolution fusion data according to the environmental conditions to generate multimodal recognition result data; Anomaly detection module, used to receive multimodal recognition result data, generate anomaly determination data, generate anomaly trend data through deep enhanced anomaly detection, and generate early warning signal data based on the stability of the multimodal joint entropy evaluation system; The data aggregation and report generation module is used to aggregate the resonance phase vibration data, echo feature data, multimodal recognition result data and warning signal data generated during the detection process, extract key features and generate a self-inspection report file.

Images