Fault identification method and system for buried optical fiber

Abstract

The invention provides a fault identification method and system for a buried optical fiber, and relates to the technical field of computers, the polarization mode dispersion change rate and the time-frequency double-domain disturbance coefficient of an optical signal of a to-be-detected buried optical fiber can be obtained, and the time-frequency double-domain disturbance coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal; under the condition that the polarization mode dispersion change rate is greater than a preset first threshold value and the time-frequency double-domain disturbance coefficient is greater than a preset second threshold value, acquiring environment state information and physical excitation parameters of the buried optical fiber; inputting the environment state information, the polarization mode dispersion change rate, the physical excitation parameter and the amplitude fluctuation coefficient into a pre-trained micro-bending strain identification model to obtain a micro-bending strain state index of the buried optical fiber; and inputting the micro-bending strain state index, the environment state information and the time-frequency double-domain disturbance coefficient into a preset optical fiber fault classification model to obtain a fault identification result of the buried optical fiber. And fault identification can be carried out on the buried optical fiber comprehensively and accurately.

Description

Technical Field

[0001] This application relates to the field of computer technology, and particularly to a method and system for fault identification of buried optical fibers. Background Art

[0002] With the rapid development of information technology, buried optical fibers, as important carriers of information transmission, are widely used in many fields such as communication, power, and transportation. Buried optical fibers provide a solid guarantee for the stable and high-speed transmission of various types of data with their advantages of large transmission capacity, strong anti-interference ability, and good confidentiality. For example, in long-distance communication trunks, buried optical fibers undertake the task of cross-regional transmission of massive data; in smart grids, buried optical fibers are used to transmit the operation status data of power systems in real time to achieve intelligent monitoring and management of the power grid.

[0003] However, due to the complex laying environment of buried optical fibers, faults are likely to occur. Traditional methods for detecting buried optical fiber faults mainly rely on single-parameter monitoring, such as using an optical time domain reflectometer (OTDR) to detect the loss of optical signals. Although this method can detect obvious faults such as fiber breaks, its detection sensitivity for some hidden faults is relatively low. Hidden faults will gradually affect the transmission performance of the optical fiber and may ultimately lead to serious faults. Summary of the Invention

[0004] The technical problem to be solved by this application is to provide a method and system for fault identification of buried optical fibers, which can comprehensively and accurately identify faults in buried optical fibers. The specific solutions are as follows: A method for fault identification of buried optical fibers includes: In response to a fault detection instruction, obtaining the polarization mode dispersion change rate and the time-frequency dual-domain perturbation coefficient of the optical signal of the buried optical fiber to be detected, where the time-frequency dual-domain perturbation coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal; When the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency dual-domain perturbation coefficient is greater than a preset second threshold, obtaining the environmental state information and physical excitation parameters of the buried optical fiber; Inputting the environmental state information, the polarization mode dispersion change rate, the physical excitation parameters, and the amplitude fluctuation coefficient into a pre-trained micro-bending strain identification model to obtain the micro-bending strain state index of the buried optical fiber; Inputting the micro-bending strain state index, the environmental state information, and the time-frequency dual-domain perturbation coefficient into a preset optical fiber fault classification model to obtain the fault identification result of the buried optical fiber.

[0005] For the above method, optionally, the process of determining the time-frequency dual-domain perturbation coefficient according to the amplitude fluctuation coefficient of the optical signal and the optical signal frequency offset includes: Perform wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time-domain sub-band energy distribution matrix; the energy distribution matrix characterizes the energy distribution characteristics under different time-domain sub-bands; Perform window Fourier transform on the frequency offset of the optical signal, extract the singularity index of each frequency point in the time-frequency matrix, and screen the abnormal frequency point set through a preset singularity threshold. The singularity index reflects the local mutation characteristics of the frequency-domain signal; the abnormal frequency point set includes abnormal frequency points with a singularity index less than the singularity threshold; Adjust the weight coefficient of each time-domain sub-band according to the variance characteristics of each time-domain sub-band in the time-domain sub-band energy distribution matrix; determine the frequency-domain abnormal intensity according to the maximum frequency offset in the abnormal frequency point set; determine the statistical correlation characteristics of the time-domain sub-band energy distribution matrix after adjusting the weight coefficient and the abnormal frequency point set, and perform non-linear fusion on the adjusted weight coefficient of each time-domain sub-band, the frequency-domain abnormal intensity, and the statistical correlation characteristics through a dynamic coupling factor to generate an initial perturbation coefficient; Generate a compensation factor for the initial perturbation coefficient according to the change rate of the ambient temperature where the buried optical fiber is located, and adjust the initial perturbation coefficient according to the compensation factor to obtain a time-frequency dual-domain perturbation coefficient.

[0006] For the above method, optionally, the micro-bending strain recognition model includes a temporal convolutional network, a bidirectional long short-term memory network, and a task learning module. The step of inputting the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into a pre-trained micro-bending strain recognition model to obtain the micro-bending strain state index of the buried optical fiber includes: Input the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the temporal convolutional network in the micro-bending strain recognition model to obtain the temporal features output by the temporal convolutional network; Input the temporal features into the bidirectional long short-term memory network in the micro-bending strain recognition model to obtain the context-aware features output by the bidirectional long short-term memory network; Input the context-aware features into the task learning module to obtain the micro-bending strain state index of the buried optical fiber.

[0007] For the above method, optionally, the training process of the micro-bending strain recognition model includes: Obtain an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and the sample label of each first training sample; the first training sample includes a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence; Select a first target training sample currently used for training from the first training sample set; Input the first target training sample into the initial recognition model to obtain the recognition result output by the initial recognition model; Calculate the loss function value according to the recognition result and the sample label of the first target training sample; Update the model parameters of the initial recognition model by using the loss function value; When the initial recognition model after updating the model parameters does not meet the first training completion condition, return to execute the step of selecting a first target training sample currently used for training from the first training sample set; When the initial recognition model after updating the model parameters meets the first training completion condition, determine the initial recognition model that meets the first training completion condition as the micro-bending strain recognition model.

[0008] For the above method, optionally, after obtaining the fault recognition result of the buried optical fiber, it further includes: Generate a fault warning message according to the fault recognition result and output the fault warning message.

[0009] A fault recognition system for a buried optical fiber, including: A first acquisition unit, configured to, in response to a fault detection instruction, acquire the polarization mode dispersion change rate and the time-frequency double-domain perturbation coefficient of the optical signal of the buried optical fiber to be detected, where the time-frequency double-domain perturbation coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal; A second acquisition unit, configured to, when the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency double-domain perturbation coefficient is greater than a preset second threshold, acquire the environmental state information and physical excitation parameters of the buried optical fiber; A first execution unit, configured to input the environmental state information, the polarization mode dispersion change rate, the physical excitation parameters, and the amplitude fluctuation coefficient into a pre-trained micro-bending strain recognition model to obtain the micro-bending strain state index of the buried optical fiber; A second execution unit, configured to input the micro-bending strain state index, the environmental state information, and the time-frequency double-domain perturbation coefficient into a preset optical fiber fault classification model to obtain the fault recognition result of the buried optical fiber.

[0010] For the above system, optionally, the first acquisition unit includes: A decomposition subunit, configured to perform wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time-domain sub-band energy distribution matrix; the energy distribution matrix characterizes the energy distribution characteristics under different time-domain sub-bands; A transform sub - unit, configured to perform window Fourier transform on the frequency offset of the optical signal, extract the singularity index of each frequency point in the time - frequency matrix, and screen out the set of abnormal frequency points through a preset singularity threshold. The singularity index reflects the local mutation characteristics of the frequency - domain signal; the set of abnormal frequency points contains abnormal frequency points with singularity indices less than the singularity threshold; A first execution sub - unit, configured to adjust the weight coefficients of each time - domain sub - band according to the variance characteristics of each time - domain sub - band in the time - domain sub - band energy distribution matrix; determine the frequency - domain abnormal intensity according to the maximum frequency offset in the set of abnormal frequency points; determine the statistical correlation characteristics between the adjusted time - domain sub - band energy distribution matrix and the set of abnormal frequency points, and perform non - linear fusion on the adjusted weight coefficients of each time - domain sub - band, the frequency - domain abnormal intensity, and the statistical correlation characteristics through a dynamic coupling factor to generate an initial perturbation coefficient; A second execution sub - unit, configured to generate a compensation factor for the initial perturbation coefficient according to the change rate of the ambient temperature where the buried optical fiber is located, and adjust the initial perturbation coefficient according to the compensation factor to obtain a time - frequency dual - domain perturbation coefficient.

[0011] For the above - mentioned system, optionally, the micro - bend strain recognition model includes a temporal convolutional network, a bidirectional long - short - term memory network, and a task learning module. The first execution unit includes: A third execution sub - unit, configured to input the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the temporal convolutional network in the micro - bend strain recognition model to obtain the temporal features output by the temporal convolutional network; A fourth execution sub - unit, configured to input the temporal features into the bidirectional long - short - term memory network in the micro - bend strain recognition model to obtain the context - aware features output by the bidirectional long - short - term memory network; A fifth execution sub - unit, configured to input the context - aware features into the task learning module to obtain the micro - bend strain state index of the buried optical fiber.

[0012] For the above - mentioned system, optionally, the first execution unit includes: An acquisition sub - unit, configured to acquire an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and the sample label of each first training sample; the first training sample includes a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence; A selection sub - unit, configured to select a current first target training sample from the first training sample set; The sixth execution subunit is configured to input the first target training sample into the initial recognition model to obtain a recognition result output by the initial recognition model; The calculation subunit is configured to calculate a loss function value according to the recognition result and the sample label of the first target training sample; The update subunit is configured to update the model parameters of the initial recognition model by using the loss function value; The seventh execution subunit is configured to, when the initial recognition model after updating the model parameters does not meet the first training completion condition, return to execute the step of selecting the current first target training sample for training in the first training sample set; The determination subunit is configured to, when the initial recognition model after updating the model parameters meets the first training completion condition, determine the initial recognition model that meets the first training completion condition as the micro-bending strain recognition model.

[0013] For the above system, optionally, it further includes: The third execution unit is configured to generate a fault warning message according to the fault recognition result and output the fault warning message.

[0014] Based on the above-mentioned fault recognition method and system for buried optical fibers provided by the present application, the polarization mode dispersion change rate and the time-frequency double-domain perturbation coefficient of the optical signal of the buried optical fiber to be detected can be obtained, and the time-frequency double-domain perturbation coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal; when the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency double-domain perturbation coefficient is greater than a preset second threshold, obtain the environmental state information and physical excitation parameters of the buried optical fiber; input the environmental state information, the polarization mode dispersion change rate, the physical excitation parameters, and the amplitude fluctuation coefficient into a pre-trained micro-bending strain recognition model to obtain the micro-bending strain state index of the buried optical fiber; input the micro-bending strain state index, the environmental state information, and the time-frequency double-domain perturbation coefficient into a preset optical fiber fault classification model to obtain the fault recognition result of the buried optical fiber. It can comprehensively and accurately identify the faults of buried optical fibers. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0016] Figure 1 It is a method flow chart of a fault recognition method for a buried optical fiber provided by the present application; Figure 2 A flowchart of a process for determining a time-frequency dual-domain perturbation coefficient provided by this application; Figure 3 A flowchart of a process for obtaining a microbending strain state index of a buried optical fiber provided by this application; Figure 4 A schematic structural diagram of a fault identification system for a buried optical fiber provided by this application. Specific embodiments

[0017] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0018] In this application, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, the element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0019] An embodiment of the present invention provides a fault identification method for a buried optical fiber, which is applied to an electronic device. The method flowchart of the method is as Figure 1 shown, and specifically includes: S101: In response to a fault detection instruction, obtain the polarization mode dispersion change rate and the time-frequency dual-domain perturbation coefficient of the optical signal of the to-be-detected buried optical fiber, where the time-frequency dual-domain perturbation coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal.

[0020] In this embodiment, the fault detection instruction may be an instruction generated when the acceleration standard deviation detected by a three-dimensional microelectromechanical system sensor exceeds the acceleration threshold in a continuous plurality of sampling periods (sampling rate 500 Hz), the interval temperature gradient between adjacent regions monitored by a distributed optical fiber temperature measurement system > the temperature gradient threshold, or the data collected by an optical time domain reflectometer indicates that the optical power drops by more than the preset power threshold within a preset time.

[0021] In this embodiment, the polarization mode dispersion change rate may refer to the change rate of the propagation time delay difference between two orthogonal polarization modes of an optical signal in an optical fiber due to the birefringence effect. The polarization mode dispersion change rate can be calculated by a polarization diversity receiver monitoring the Stokes parameters in real time.

[0022] Optionally, the time-frequency dual-domain perturbation coefficient is a comprehensive perturbation index that combines the time-domain amplitude fluctuation and the frequency-domain frequency offset of the fused optical signal, characterizing the intensity and frequency response characteristics of external interference.

[0023] In this embodiment, the amplitude fluctuation coefficient can be the degree of change of the optical signal amplitude over time; the optical signal can be converted into an electrical signal using a photodetector, and after sampling by a high-speed analog-to-digital converter, the statistical value can be calculated; data can also be obtained by combining an optical power meter or an optical time-domain reflectometer to obtain the amplitude fluctuation coefficient. In some embodiments, the amplitude fluctuation coefficient can be obtained by calculating the standard deviation or variance of the amplitude time series and then normalizing it by dividing by the average amplitude.

[0024] In this embodiment, the frequency offset can be the deviation between the actual frequency and the nominal frequency of the optical signal; the frequency offset can be calculated by the phase detection method or the spectrum analysis method, such as using a phase-locked loop to track the phase change; or analyzing the spectrum to detect the center frequency offset. It can also be directly measured using a spectrum analyzer, or the phase change can be analyzed after demodulating the signal by a coherent receiver, and the frequency offset can be calculated by combining digital signal processing techniques.

[0025] S102: When the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency dual-domain perturbation coefficient is greater than a preset second threshold, obtain the environmental state information and physical excitation parameters of the buried optical fiber.

[0026] In this embodiment, the first threshold is set according to the type of the buried optical fiber, and the preset configuration table can be queried according to the type of the buried optical fiber to obtain the first threshold; the second threshold is set according to the burial depth of the buried optical fiber, and the setting method of the second threshold is as follows:

[0027] wherein, is the second threshold, and d is the burial depth of the buried optical fiber.

[0028] Optionally, the environmental state information can include at least one of soil temperature, chemical corrosion parameters, electromagnetic environment parameters, biological activity parameters, and groundwater level, etc.; the physical excitation parameters can include at least one of mechanical dynamics parameters (such as shock energy spectral density, stress wave propagation speed, and plastic deformation index, etc.), three-dimensional vibration acceleration, and sound pressure level, etc.

[0029] In this embodiment, the polarization mode dispersion change rate reflects the change rate of the propagation time delay difference of the optical signal in the optical fiber due to the birefringence effect, and is sensitive to the change of the internal structure of the optical fiber; the time-frequency dual-domain perturbation coefficient combines the time-domain amplitude fluctuation and the frequency-domain frequency offset characteristics of the optical signal, and characterizes the intensity and frequency response characteristics of the external interference. The combination of the two can more accurately judge whether the optical fiber is in an abnormal state, reduce the possibility of misjudgment and missed judgment, and effectively improve the accuracy of fault identification.

[0030] In this embodiment, the first threshold is set according to the type of the buried optical fiber. Due to differences in materials, structures, etc. of different types of optical fibers, their sensitivities to faults are different. The corresponding threshold is obtained by querying the preset configuration table, making the detection more targeted; the second threshold is set according to the burial depth of the buried optical fiber. Considering that the degree of influence of the optical fiber by the environment is different for different burial depths, for example, the optical fiber with a shallower burial depth is more likely to be affected by ground activities. This way of setting the threshold according to the actual installation parameters further improves the accuracy of fault judgment.

[0031] S103: Input the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into a pre-trained microbend strain recognition model to obtain the microbend strain state index of the buried optical fiber.

[0032] In this embodiment, the microbend strain recognition model includes a temporal convolutional network, a bidirectional long short-term memory network, and a task learning module. The temporal convolutional network can be used to extract temporal features, and the bidirectional long short-term memory network can be used to extract context-aware features. The task learning module can be a multi-task learning module, which can output the microbend strain state index, pre-classification result, and the confidence of the microbend strain state index, etc. based on the context-aware features. That is, the output results of the microbend strain recognition model can include the microbend strain state index, pre-classification result, and the confidence of the microbend strain state index, etc.

[0033] Optionally, the microbend strain state index can comprehensively reflect the degree of small bending deformation of the buried optical fiber caused by external mechanical stress or environmental changes and its potential risks. The microbend strain state index can take values between [0, 1], and the larger the value, the higher the risk.

[0034] In this embodiment, the microbend strain recognition model is a time series parameter, and the time series parameter includes environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient.

[0035] In this embodiment, the micro-bending strain recognition model consists of a temporal convolutional network, a bidirectional long short-term memory network, and a multi-task learning module. The temporal convolutional network effectively extracts the temporal features of the input parameters through a specific convolutional structure, and can capture the variation laws of environmental states, optical signal parameters, etc. over time; the bidirectional long short-term memory network uses its bidirectional structure to fully consider the front and back information of the time series, extracts context-aware features, and enhances the understanding ability of complex time series data; the multi-task learning module outputs the micro-bending strain state index, pre-classification results, and confidence based on the context-aware features, and evaluates the micro-bending strain from multiple perspectives. This way of multi-module collaborative work can evaluate the degree of micro-bending deformation and its potential risks of buried optical fibers caused by external mechanical stress or environmental changes more comprehensively and accurately than traditional single-structure models, providing a reliable basis for subsequent fault classification.

[0036] S104: Input the micro-bending strain state index, the environmental state information, and the time-frequency dual-domain perturbation coefficient into a preset optical fiber fault classification model to obtain the fault recognition result of the buried optical fiber.

[0037] In this embodiment, the micro-bending strain state index, the environmental state information, and the time-frequency dual-domain perturbation coefficient are synchronously input into a preset optical fiber fault classification model, and the fault recognition result of the buried optical fiber is generated through multi-modal feature fusion and non-linear decision boundary calculation; among them, the micro-bending strain state index characterizes the strain energy accumulation state of the microscopic deformation of the optical fiber, the environmental state information includes soil temperature and humidity, groundwater level, and ground stress distribution parameters, and the time-frequency dual-domain perturbation coefficient is jointly calculated based on the time-domain derivative of the optical signal amplitude fluctuation and the frequency-domain power spectrum entropy; the optical fiber fault classification model is pre-trained through a generative adversarial network, and its decision space is divided into four orthogonal sub-spaces of mechanical damage, chemical corrosion, geological deformation, and environmental interference, and the output result is a fault type label with confidence evaluation, and can also include a positioning topology map.

[0038] In this embodiment, the optical fiber fault classification model synchronously inputs the micro-bending strain state index, the environmental state information, and the time-frequency dual-domain perturbation coefficient through multi-modal feature fusion, fully utilizes the information carried by different parameters, and comprehensively characterizes the state of the optical fiber. Through non-linear decision boundary calculation, complex feature relationships can be processed, and the fault recognition result can be accurately generated. This method can identify the fault type more accurately and improve the accuracy of fault classification compared with traditional simple classification methods.

[0039] Optionally, three parallel input channels can be established to perform normalization encoding on the micro-bending strain state index, the environmental state information, and the time-frequency dual-domain perturbation coefficient respectively.

[0040] In some embodiments, the output result of the micro-bending strain recognition model, the environmental state information, and the time-frequency dual-domain perturbation coefficient may be input into a preset optical fiber fault classification model to obtain the fault recognition result of the buried optical fiber.

[0041] Applying the method provided in the embodiments of the present application can comprehensively and accurately identify faults in buried optical fibers.

[0042] In an embodiment provided by the present application, based on the above solution, optionally, the process of determining the time-frequency dual-domain perturbation coefficient according to the amplitude fluctuation coefficient and the optical signal frequency offset of the optical signal is as Figure 2 shown and includes: S201: Perform wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time-domain sub-band energy distribution matrix; the energy distribution matrix characterizes the energy distribution characteristics under different time-domain sub-bands.

[0043] In this embodiment, perform K-layer wavelet packet decomposition on the optical signal amplitude fluctuation coefficient, use the multi-Bessel 4 wavelet basis function, and adaptively select the optimal basis function from the basis function library through the minimum energy entropy criterion. After decomposition, k time-domain sub-bands are generated, and each sub-band corresponds to a different time resolution level (for example, the first layer corresponds to the instantaneous fluctuation of 0-10 ms, and the kth layer corresponds to the long-term trend of 1-10 s). Integrate the energy of each sub-band signal to construct a time-domain sub-band energy distribution matrix , where e i represents the normalized energy value of the i-th layer sub-band, k is an integer greater than or equal to 2, for example, it can be 8; i is an integer not greater than k. Through multi-scale time-frequency analysis, separate the transient and steady components of the signal, overcoming the limitation of traditional wavelet decomposition that only decomposes low frequencies. Select the basis function that makes the sub-band energy distribution most concentrated, suppress noise interference, and enhance the characteristics of effective signals.

[0044] S202: Perform window Fourier transform on the frequency offset of the optical signal, extract the singularity index of each frequency point in the time-frequency matrix, and screen the abnormal frequency point set through a preset singularity threshold. The singularity index reflects the local mutation characteristics of the frequency-domain signal; the abnormal frequency point set contains abnormal frequency points with a singularity index less than the singularity threshold.

[0045] In this embodiment, perform Hamming window short-time Fourier transform on the optical signal frequency offset, with a window length of 256 points and an overlap rate of 75% to generate a time-frequency matrix . Calculate the singularity index of each frequency point, specifically as follows:

[0046] Among them, is the singularity index, and this singularity index can be the Hlder index; s is the scale parameter, represents time, represents frequency.

[0047] Optionally, filter out abnormal frequency points that satisfy < 0.5 to form a set of abnormal frequency points .

[0048] In this embodiment, the singularity threshold can be a preset critical value of the Hölder index (such as 0.5) for distinguishing normal frequency points from abnormal frequency points.

[0049] In this embodiment, the set of abnormal frequency points contains all frequency points that satisfy < 0.5, which characterizes frequency domain burst interference events.

[0050] S203: Adjust the weight coefficients of each time domain sub-band according to the variance characteristics of each time domain sub-band in the time domain sub-band energy distribution matrix; determine the frequency domain anomaly intensity according to the maximum frequency offset in the set of abnormal frequency points; determine the statistical correlation characteristics between the time domain sub-band energy distribution matrix after adjusting the weight coefficients and the set of abnormal frequency points, and perform non-linear fusion on the adjusted weight coefficients, the frequency domain anomaly intensity, and the statistical correlation characteristics of each time domain sub-band through a dynamic coupling factor to generate an initial perturbation coefficient.

[0051] In this embodiment, the variance of the energy sequence of each time domain sub-band can be calculated , and the weight coefficient to suppress the noise interference of high-fluctuation sub-bands.

[0052] Optionally, the maximum frequency offset in the set of abnormal frequency points can be extracted to quantify the frequency domain anomaly intensity, as follows:

[0053] where represents the maximum frequency offset, represents the frequency offset with respect to the frequency in absolute value. represents the frequency offset at the frequency , that is, the difference between the actual frequency and the reference frequency. Taking the absolute value is to uniformly measure the magnitude of the frequency offset without considering the offset direction. is the set of abnormal frequency points.

[0054] Optionally, calculate the Pearson correlation coefficient ρ between the time domain sub-band energy distribution matrix and the abnormal frequency distribution, and filter out the significantly correlated sub-bands with |ρ| > 0.7.

[0055] Optionally, through the dynamic coupling factor Perform weighted superposition on the time-domain weight coefficient, the frequency-domain anomaly intensity, and the statistical correlation characteristics to generate an initial perturbation coefficient, and the initial perturbation coefficient is expressed as follows:

[0056] Among them, tanh represents the hyperbolic tangent function, which is used to map the integral result to the interval (−1, 1), playing the role of normalization and nonlinear transformation. Through this nonlinear transformation, the response of the coupling factor to the frequency offset can better meet the actual requirements, avoid the limitations that may be brought by linear transformation, and enhance the adaptability of the model to complex situations. T represents the length of the historical time window, and T can be determined according to the application scenario.

[0057] The dynamic coupling factor can adaptively adjust the fusion ratio of the time-domain and frequency-domain features according to the historical frequency offset trend. Specifically, it can be based on the normalization parameter of the frequency offset integral value to balance the fusion weights of the time-domain and frequency-domain features.

[0058] S204: Generate a compensation factor for the initial perturbation coefficient according to the change rate of the ambient temperature of the buried optical fiber, and adjust the initial perturbation coefficient according to the compensation factor to obtain a time-frequency dual-domain perturbation coefficient.

[0059] In this embodiment, the change rate of temperature ΔT / Δt along the optical fiber can be obtained through a distributed optical fiber temperature measurement system.

[0060] Optionally, a logical function can be used to calculate the temperature-sensitive compensation factor: Among them, is the temperature-sensitive compensation factor, is the reference temperature change rate, k is the slope coefficient, and t is the time.

[0061] In this embodiment, temperature compensation can be performed on the initial perturbation coefficient:

[0062] Among them, γ is the temperature-sensitive coefficient (default 0.1), which is calibrated according to the thermal expansion coefficient of the optical fiber material.

[0063] In this embodiment, the signal parameter drift caused by the natural change of the ambient temperature can be suppressed, and the stability of the perturbation detection can be improved.

[0064] In an embodiment provided by the present application, based on the above - mentioned solution, optionally, the micro - bend strain recognition model includes a temporal convolutional network, a bidirectional long - short - term memory network, and a task learning module. The process of inputting the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the pre - trained micro - bend strain recognition model to obtain the micro - bend strain state index of the buried optical fiber is as Figure 3 shown and includes: S301: Input the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the temporal convolutional network in the micro - bend strain recognition model to obtain the temporal features output by the temporal convolutional network.

[0065] In this embodiment, the temporal convolutional network (TCN) consists of five layers of dilated causal convolutional layers, and the size of the convolutional kernel for each layer is set to 5. Its dilation coefficients increase sequentially by layer, which are [1, 2, 4, 8, 16]. Through this structural design, the network can effectively capture the dependence relationships and local features of the input parameters at different time scales.

[0066] Specifically, the first - layer convolutional layer performs a convolution operation on the input data with a convolutional kernel of size 5 and a dilation coefficient of 1 to initially extract features on a relatively short time scale; as the number of layers increases, the dilation coefficient increases, and subsequent convolutional layers can gradually capture feature information on longer time scales.

[0067] For example, when processing data on the change of soil temperature over time, the TCN can extract features from short - term temperature fluctuations (such as hourly temperature changes) to long - term temperature trends (such as daily temperature changes), and thus output a temporal feature vector containing information on different time scales.

[0068] S302: Input the temporal features into the bidirectional long - short - term memory network in the micro - bend strain recognition model to obtain the context - aware features output by the bidirectional long - short - term memory network.

[0069] In this embodiment, the bidirectional long - short - term memory network (Bi - LSTM) has two hidden layers, and the dimension of each hidden layer is set to 64. The Bi - LSTM can consider both the forward and backward information of the input sequence, thereby better capturing context - dependence relationships. When receiving the temporal features output by the TCN, the forward and backward hidden layers of the Bi - LSTM process the temporal features respectively. The forward hidden layer starts from the beginning of the sequence and gradually learns the information before the current moment; the backward hidden layer starts from the end of the sequence and obtains the information after the current moment.

[0070] For example, when analyzing physical excitation parameters (such as the change of vibration frequency over time), the Bi-LSTM can not only know the historical changes of the vibration frequency, but also consider the possible future change trends. After fusing this information, it outputs a context-aware feature vector, which synthesizes the context information of the input parameters over the entire time series, providing a more comprehensive feature representation for accurately evaluating the micro-bending strain state subsequently.

[0071] S303: Input the context-aware feature into the task learning module to obtain the micro-bending strain state index of the buried optical fiber.

[0072] In this embodiment, the task learning module is a multi-task learning module, which adopts a dual-task branch structure. One branch outputs the micro-bending strain state index through an activation function. The value range of this index is between [0, 1]. The larger the value, the higher the degree of micro-bending deformation and its potential risk of the buried optical fiber caused by external mechanical stress or environmental changes. The other branch outputs pre-classification labels, such as "normal", "micro-bending", "squeezing", etc., through a normalization exponential function to preliminarily judge the fault type of the optical fiber. At the same time, to balance the training processes of the two tasks, the loss functions of the two tasks are adjusted through dynamic weight coefficients.

[0073] For example, in the initial stage of training, since the accuracy requirement for the micro-bending strain state index is relatively high, the weight of the loss function of this task can be appropriately increased; as the training progresses, when the accuracy improvement space of the pre-classification label is relatively large, the weight of the loss function of the pre-classification task is increased. Through this multi-task learning mechanism, the model can more effectively learn the complex relationship between the input features and the micro-bending strain state, so as to accurately output the micro-bending strain state index, providing an important basis for subsequent optical fiber fault classification.

[0074] In an embodiment provided by the present application, based on the above solution, optionally, the training process of the micro-bending strain recognition model includes: Obtain an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and the sample label of each first training sample; the first training sample includes a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence; Select a first target training sample currently used for training from the first training sample set; Input the first target training sample into the initial recognition model to obtain the recognition result output by the initial recognition model; Calculate the loss function value according to the recognition result and the sample label of the first target training sample; Update the model parameters of the initial recognition model by using the loss function value; In the case that the initial recognition model after updating the model parameters does not meet the first training completion condition, return to execute the step of selecting the current first target training sample from the first training sample set; In the case that the initial recognition model after updating the model parameters meets the first training completion condition, determine the initial recognition model that meets the first training completion condition as the micro-bending strain recognition model.

[0075] In this embodiment, an initial recognition model and a first training data set are obtained. The initial recognition model is the starting architecture for constructing the micro-bending strain recognition model, which has model parameters that can be adjusted subsequently and provides a basic framework for model training.

[0076] The first training data set includes multiple first training samples and the sample labels corresponding to each of the first training samples. Among them, the first training samples include a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence. These sequence data reflect the relevant characteristics of the buried optical fiber under different past times and environmental conditions, and are important bases for the model to learn the relationship between micro-bending strain and various parameters. The historical environmental state information sequence covers information such as soil temperature, humidity, chemical corrosion parameters, electromagnetic environment parameters, biological activity parameters, and groundwater level, reflecting the comprehensive situation of the environment where the buried optical fiber is located; the historical polarization mode dispersion change rate sequence records the change rate of the propagation time delay difference between two orthogonal polarization modes caused by the birefringence effect of the optical signal over time, and this parameter is sensitive to the micro-bending strain of the optical fiber; the historical physical excitation parameter sequence includes mechanical dynamics parameters (such as shock energy spectral density, stress wave propagation speed, and plastic deformation index, etc.), three-dimensional vibration acceleration, and sound pressure level, etc., reflecting the physical effects of the outside world on the optical fiber; the historical amplitude fluctuation coefficient sequence characterizes the fluctuation of the optical signal amplitude, and its change is related to the micro-bending strain of the optical fiber and external interference. The sample labels clearly mark the actual state of the micro-bending strain corresponding to each training sample, providing an accurate learning target for model training.

[0077] In this embodiment, a sample is selected from the rich first training sample set by means of random sampling or sampling according to specific rules as the input for the current training, aiming to enable the model to gradually learn the features and laws contained in different samples in the data set, avoid the model from overfitting to specific samples, and enhance the generalization ability of the model.

[0078] Optionally, based on its current model parameters, the initial recognition model extracts and analyzes the features of the input first target training sample, attempts to predict the corresponding micro-bending strain state of the sample, and outputs the recognition result. This recognition result is the model's preliminary judgment on the micro-bending strain of the sample based on its current learning state.

[0079] The loss function is used to measure the degree of difference between the recognition result predicted by the model and the actual label of the sample. It is an important indicator for evaluating the current performance of the model. By reasonably selecting the loss function (such as mean squared error loss function, cross-entropy loss function, etc.), the deviation between the predicted value and the true value of the model can be quantified, providing a clear direction for subsequent model parameter updates, that is, by adjusting the model parameters to minimize the loss function value, thereby improving the prediction accuracy of the model.

[0080] Optimization algorithms such as the stochastic gradient descent algorithm are used to calculate the gradients of the model parameters according to the loss function value, and the model parameters are adjusted according to the gradient direction, so that the model can be closer to the true label of the sample in subsequent predictions. This is one of the core steps in model training. By continuously adjusting the parameters, the model gradually learns the complex mapping relationship between the input features and the micro-bending strain state.

[0081] Optionally, the first training completion condition can be that the preset number of training rounds reaches the upper limit, the loss function value converges below a certain threshold, the accuracy of the validation set reaches a specific standard, etc. If the model does not meet this condition, it means that the model has not fully learned the patterns in the data, and new samples need to be selected from the training dataset for training, repeating the above steps to further optimize the model parameters until the first training completion condition is met.

[0082] In an embodiment provided by the present application, based on the above solution, optionally, after obtaining the fault recognition result of the buried optical fiber, it further includes: Generating a fault warning message according to the fault recognition result and outputting the fault warning message.

[0083] In this embodiment, after generating the fault warning message, the system will output the fault warning message through a preset output channel. The output channels can include but are not limited to the following methods: Local display: On the local device responsible for monitoring the buried optical fiber (such as a monitoring terminal), the fault warning message is displayed in the form of a visual interface, facilitating on-site staff to obtain the fault situation in a timely manner. For example, a prompt box pops up on the display screen of the monitoring terminal, clearly showing key information such as the fault type, cause, location, and severity.

[0084] Remote transmission: Transmit the fault warning information to the remote management center through a network (such as a wired network, wireless network, etc.) so that the management personnel can remotely monitor the operation status of the buried optical fiber. For example, send the information to the cloud server through the Internet of Things technology, and the management personnel can log in to the management platform through terminal devices such as computers and mobile phones to view the detailed fault warning content.

[0085] Alarm notification: Send fault alarm notifications to relevant maintenance personnel by means of text messages, voice broadcasts, etc. For example, the system automatically sends text messages containing fault information to the mobile phones of maintenance personnel, or issues a voice alarm through the on-site voice broadcast device to notify the nearby staff to handle the fault in time.

[0086] Through the above steps of generating and outputting the fault warning information, the fault situation of the buried optical fiber can be conveyed to the relevant personnel in a timely and effective manner, so as to take corresponding measures for processing and ensure the normal operation of the buried optical fiber.

[0087] See Figure 4 , which is a schematic structural diagram of a fault identification method system for a buried optical fiber provided by an embodiment of the present application. The system includes: The first acquisition unit 401 is configured to obtain the polarization mode dispersion change rate and the time-frequency double-domain perturbation coefficient of the optical signal of the buried optical fiber to be detected in response to a fault detection instruction, and the time-frequency double-domain perturbation coefficient is determined according to the amplitude fluctuation coefficient and the frequency offset of the optical signal; The second acquisition unit 402 is configured to obtain the environmental state information and physical excitation parameters of the buried optical fiber when the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency double-domain perturbation coefficient is greater than a preset second threshold; The first execution unit 403 is configured to input the environmental state information, the polarization mode dispersion change rate, the physical excitation parameter, and the amplitude fluctuation coefficient into a pre-trained micro-bending strain identification model to obtain the micro-bending strain state index of the buried optical fiber; The second execution unit 404 is configured to input the micro-bending strain state index, the environmental state information, and the time-frequency double-domain perturbation coefficient into a preset optical fiber fault classification model to obtain the fault identification result of the buried optical fiber.

[0088] In an embodiment provided by the present application, based on the above solution, optionally, the first acquisition unit 401 includes: The decomposition subunit is configured to perform wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time-domain sub-band energy distribution matrix; the energy distribution matrix characterizes the energy distribution characteristics under different time-domain sub-bands; A transform sub - unit, configured to perform a window Fourier transform on the frequency offset of the optical signal, extract the singularity index of each frequency point in the time - frequency matrix, and screen out an abnormal frequency point set through a preset singularity threshold. The singularity index reflects the local mutation characteristics of the frequency - domain signal; the abnormal frequency point set contains abnormal frequency points with singularity indices less than the singularity threshold. A first execution sub - unit, configured to adjust the weight coefficients of each time - domain sub - band according to the variance characteristics of each time - domain sub - band in the time - domain sub - band energy distribution matrix; determine the frequency - domain abnormal intensity according to the maximum frequency offset in the abnormal frequency point set; determine the statistical correlation characteristics between the adjusted time - domain sub - band energy distribution matrix and the abnormal frequency point set, and perform non - linear fusion on the adjusted weight coefficients of each time - domain sub - band, the frequency - domain abnormal intensity, and the statistical correlation characteristics through a dynamic coupling factor to generate an initial perturbation coefficient. A second execution sub - unit, configured to generate a compensation factor for the initial perturbation coefficient according to the change rate of the ambient temperature where the buried optical fiber is located, and adjust the initial perturbation coefficient according to the compensation factor to obtain a time - frequency dual - domain perturbation coefficient.

[0089] In an embodiment provided by the present application, based on the above - mentioned solution, optionally, the micro - bend strain recognition model includes a temporal convolutional network, a bidirectional long - short - term memory network, and a task learning module. The first execution unit 403 includes: A third execution sub - unit, configured to input the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the temporal convolutional network in the micro - bend strain recognition model to obtain the temporal features output by the temporal convolutional network. A fourth execution sub - unit, configured to input the temporal features into the bidirectional long - short - term memory network in the micro - bend strain recognition model to obtain context - aware features output by the bidirectional long - short - term memory network. A fifth execution sub - unit, configured to input the context - aware features into the task learning module to obtain the micro - bend strain state index of the buried optical fiber.

[0090] In an embodiment provided by the present application, based on the above - mentioned solution, optionally, the first execution unit 403 includes: An acquisition sub - unit, configured to acquire an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and the sample label of each first training sample; the first training sample includes a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence. A selection sub - unit, configured to select a current first target training sample from the first training sample set. The sixth execution subunit is configured to input the first target training sample into the initial recognition model to obtain a recognition result output by the initial recognition model; The calculation subunit is configured to calculate a loss function value according to the recognition result and the sample label of the first target training sample; The update subunit is configured to update the model parameters of the initial recognition model by using the loss function value; The seventh execution subunit is configured to, when the initial recognition model after updating the model parameters does not meet the first training completion condition, return to execute the step of selecting the current first target training sample for training in the first training sample set; The determination subunit is configured to, when the initial recognition model after updating the model parameters meets the first training completion condition, determine the initial recognition model that meets the first training completion condition as the micro-bending strain recognition model.

[0091] In an embodiment provided by the present application, based on the above solution, optionally, it further includes: The third execution unit is configured to generate a fault warning message according to the fault recognition result and output the fault warning message.

[0092] It should be noted that the embodiments in this specification are all described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0093] Finally, it should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0094] For the convenience of description, when describing the above system, various units are described separately according to functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0095] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0096] The above has introduced in detail a method for fault identification of buried optical fibers provided by this application. Specific examples are used in this article to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for identifying faults of buried optical fibers, characterized in that: include: In response to the fault detection instruction, a polarization mode dispersion change rate and a time-frequency dual-domain disturbance coefficient of an optical signal of an underground optical fiber to be detected are obtained, wherein the time-frequency dual-domain disturbance coefficient is determined according to an amplitude fluctuation coefficient and a frequency offset of the optical signal; When the polarization mode dispersion change rate is greater than a preset first threshold value, and the time-frequency dual-domain disturbance coefficient is greater than a preset second threshold value, obtaining environmental state information and physical excitation parameters of the buried optical fiber; Inputting the environmental state information, polarization mode dispersion change rate, physical excitation parameter and amplitude fluctuation coefficient into a pre-trained microbend strain identification model to obtain a microbend strain state index of the buried optical fiber; The microbend strain state index, the environmental state information and the time-frequency dual-domain disturbance coefficient are input into a preset optical fiber fault classification model to obtain a fault identification result of the buried optical fiber.

2. The method according to claim 1, characterized in that The process of determining the time-frequency dual-domain disturbance coefficient according to the amplitude fluctuation coefficient of the optical signal and the frequency offset of the optical signal includes: Performing wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time domain sub-band energy distribution matrix; the energy distribution matrix represents energy distribution characteristics in different time domain sub-bands; Performing a window Fourier transform on the frequency offset of the optical signal, extracting a singularity index of each frequency point in the time-frequency matrix, and filtering an abnormal frequency point set by a preset singularity threshold, wherein the singularity index reflects the local mutation characteristics of the frequency domain signal; the abnormal frequency point set includes abnormal frequency points whose singularity index is less than the singularity threshold; The weight coefficient of each time domain subband is adjusted according to the variance characteristics of each time domain subband in the time domain subband energy distribution matrix; the frequency domain anomaly intensity is determined according to the maximum frequency offset in the abnormal frequency point set; the statistical correlation characteristics of the time domain subband energy distribution matrix after adjusting the weight coefficient and the abnormal frequency point set are determined, and the adjusted weight coefficient of each time domain subband, the frequency domain anomaly intensity and the statistical correlation characteristics are nonlinearly fused through a dynamic coupling factor to generate an initial disturbance coefficient; A compensation factor of the initial disturbance coefficient is generated according to the change rate of the ambient temperature of the buried optical fiber, and the initial disturbance coefficient is adjusted according to the compensation factor to obtain a time-frequency dual-domain disturbance coefficient.

3. The method according to claim 1, characterized in that The microbend strain identification model includes a temporal convolutional network, a bidirectional long short-term memory network and a task learning module. The environmental state information, the polarization mode dispersion change rate, the physical excitation parameter and the amplitude fluctuation coefficient are input into a pre-trained microbend strain identification model to obtain the microbend strain state index of the buried optical fiber, including: Inputting the environmental state information, polarization mode dispersion change rate, physical excitation parameters, and amplitude fluctuation coefficient into the time series convolution network in the microbending strain identification model to obtain the time series characteristics output by the time series convolution network; Inputting the time series features into a bidirectional long short-term memory network in the microbending strain recognition model to obtain context-aware features output by the bidirectional long short-term memory network; The context-aware features are input into the task learning module to obtain a microbend strain state index of the buried optical fiber.

4. The method according to claim 1, characterized in that The training process of the microbending strain identification model includes: Acquire an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and a sample label of each of the first training samples; the first training samples include a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence; Selecting a first target training sample currently used for training from the first training sample set; Inputting the first target training sample into the initial recognition model to obtain a recognition result output by the initial recognition model; A loss function value is calculated based on the recognition result and the sample label of the first target training sample; Using the loss function value to update the model parameters of the initial recognition model; If the initial recognition model after updating the model parameters does not meet the first training completion condition, returning to the step of selecting a first target training sample currently used for training from the first training sample set; When the initial recognition model after updating the model parameters satisfies the first training completion condition, the initial recognition model satisfying the first training completion condition is determined as the microbending strain recognition model.

5. The method according to claim 1, characterized in that After obtaining the fault identification result of the buried optical fiber, the method further includes: Fault warning information is generated according to the fault identification result, and the fault warning information is output.

6. A buried optical fiber fault identification system, characterized in that: include: A first acquisition unit, configured to acquire, in response to a fault detection instruction, a polarization mode dispersion change rate and a time-frequency dual-domain disturbance coefficient of an optical signal of an underground optical fiber to be detected, wherein the time-frequency dual-domain disturbance coefficient is determined according to an amplitude fluctuation coefficient and a frequency offset of the optical signal; A second acquisition unit is used to acquire the environmental state information and physical excitation parameters of the buried optical fiber when the polarization mode dispersion change rate is greater than a preset first threshold and the time-frequency dual-domain disturbance coefficient is greater than a preset second threshold; A first execution unit is used to input the environmental state information, polarization mode dispersion change rate, physical excitation parameters and amplitude fluctuation coefficient into a pre-trained microbend strain identification model to obtain a microbend strain state index of the buried optical fiber; The second execution unit is used to input the microbend strain state index, the environmental state information and the time-frequency dual-domain disturbance coefficient into a preset optical fiber fault classification model to obtain a fault identification result of the buried optical fiber.

7. The system according to claim 6, characterized in that The first acquisition unit includes: A decomposition subunit, configured to perform wavelet packet decomposition on the amplitude fluctuation coefficient of the optical signal to generate a time domain sub-band energy distribution matrix; the energy distribution matrix represents energy distribution characteristics in different time domain sub-bands; A transform subunit, configured to perform a window Fourier transform on the frequency offset of the optical signal, extract a singularity index of each frequency point in the time-frequency matrix, and filter an abnormal frequency point set by a preset singularity threshold, wherein the singularity index reflects the local mutation characteristics of the frequency domain signal; the abnormal frequency point set includes abnormal frequency points whose singularity index is less than the singularity threshold; The first execution subunit is used to adjust the weight coefficient of each time domain subband according to the variance characteristics of each time domain subband in the time domain subband energy distribution matrix; determine the frequency domain anomaly intensity according to the maximum frequency offset in the abnormal frequency point set; determine the statistical correlation characteristics of the time domain subband energy distribution matrix after adjusting the weight coefficient and the abnormal frequency point set, and perform nonlinear fusion of the adjusted weight coefficient of each time domain subband, the frequency domain anomaly intensity and the statistical correlation characteristics through a dynamic coupling factor to generate an initial disturbance coefficient; The second execution subunit is used to generate a compensation factor of the initial disturbance coefficient according to the change rate of the ambient temperature of the buried optical fiber, and adjust the initial disturbance coefficient according to the compensation factor to obtain the time-frequency dual-domain disturbance coefficient.

8. The system according to claim 6, characterized in that The microbending strain recognition model includes a temporal convolutional network, a bidirectional long short-term memory network and a task learning module. The first execution unit includes: A third execution subunit is used to input the environmental state information, polarization mode dispersion change rate, physical excitation parameters and amplitude fluctuation coefficient into the time series convolution network in the microbending strain identification model to obtain the time series characteristics output by the time series convolution network; A fourth execution subunit is used to input the time series features into the bidirectional long short-term memory network in the microbending strain recognition model to obtain context-aware features output by the bidirectional long short-term memory network; The fifth execution subunit is used to input the context-aware features into the task learning module to obtain the microbend strain state index of the buried optical fiber.

9. The system according to claim 6, characterized in that The first execution unit includes: An acquisition subunit is used to acquire an initial recognition model and a first training data set; the first training data set includes a plurality of first training samples and a sample label of each of the first training samples; the first training samples include a historical environmental state information sequence, a historical polarization mode dispersion change rate sequence, a historical physical excitation parameter sequence, and a historical amplitude fluctuation coefficient sequence; A selection subunit, configured to select a first target training sample currently used for training from the first training sample set; a sixth execution subunit, configured to input the first target training sample into the initial recognition model to obtain a recognition result output by the initial recognition model; A calculation subunit, configured to calculate a loss function value according to the recognition result and the sample label of the first target training sample; An updating subunit, used to update the model parameters of the initial recognition model using the loss function value; a seventh execution subunit, configured to return to the step of selecting a first target training sample currently used for training from the first training sample set if the initial recognition model after updating the model parameters does not meet the first training completion condition; The determination subunit is used to determine the initial recognition model that meets the first training completion condition as the microbending strain recognition model when the initial recognition model after the model parameters are updated meets the first training completion condition.

10. The system according to claim 6, characterized in that Also includes: The third execution unit is used to generate fault warning information according to the fault identification result and output the fault warning information.

Images