Fault positioning method, device and system for optical fiber line

Abstract

The invention discloses a fault positioning method, device and system for an optical fiber line, and relates to the technical field of optical fiber communication, and the method comprises the steps: controlling an optical time domain reflectometer of each station to transmit an optical pulse signal to the optical fiber line, and receiving an optical pulse reflection signal; performing time-frequency transformation on the light pulse reflection signal to obtain a reflection signal two-dimensional time-frequency diagram set; performing feature extraction on the reflected signal two-dimensional time-frequency diagrams in the reflected signal two-dimensional time-frequency diagram set to obtain a reflected signal time-frequency feature vector set; extracting a reflected signal time-frequency feature vector corresponding to the station to be evaluated from the reflected signal time-frequency feature vector set as a query feature vector; performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors to obtain a matching result representation feature matrix; and optimizing the matching result representation feature matrix, and judging whether the to-be-evaluated site is a fault site or not. According to the invention, the accuracy of optical fiber line fault positioning can be improved.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber communication technology, and in particular to a method, device and system for locating faults in an optical fiber line. Background Art

[0002] With the rapid development of power grid construction and the continuous improvement of the automation level of power system equipment, the power optical fiber communication network has achieved unprecedented development. Optical fiber communication has many advantages, such as low transmission loss and long transmission distance, so it has been widely promoted and applied.

[0003] Fiber optic switching devices are key devices in fiber optic communication networks. They are responsible for managing and controlling the transmission path of optical signals. Such devices can achieve automatic or manual switching between fiber optic lines, thereby enhancing the reliability and flexibility of the network. Fiber optic switching devices play a vital role in fiber optic communication systems, especially in data centers, telecommunication networks, and radio and television.

[0004] However, when a fiber optic line encounters a fault, such as being cut off due to a construction accident, traditional fiber optic switching devices are inadequate in locating the fault. They often require time-consuming segment-by-segment detection, which prolongs the fault response time and thus affects the efficiency of communication. In addition, fiber optic switching devices usually lack automated fault diagnosis capabilities and rely on manual fault identification and location. This method is not only inefficient, but may also lead to inaccurate diagnostic results due to environmental factors or human error, which further affects the recovery speed and overall reliability of the fiber optic communication network.

[0005] Patent text CN118199719 A discloses a fault location method for optical fiber links, electronic equipment, and computer-readable storage media, wherein the method includes: using an optical time domain reflectometer to perform multi-level optical fiber detection and echo reference denoising on the optical fiber link to be tested, and obtaining a denoised signal echo data group; performing multi-dimensional space reconstruction and attention time-frequency analysis on the denoised signal echo data group to obtain an echo time-frequency feature group; performing characteristic fault location on the echo time-frequency feature group to obtain an optical fiber fault distance group; obtaining a link data map of the optical fiber link to be tested, performing line identification and layout analysis on the link data map to obtain an optical fiber layout parameter group; calculating a standard fault distance based on the optical fiber layout parameter group and the optical fiber fault distance group, and using the standard fault distance to perform fault location annotation on the link data map. This method locates faults by distance, and also relies on the link data map of the optical fiber to be able to locate, and the method is complex and has low reliability. Summary of the invention

[0006] The present invention provides a method, device and system for locating optical fiber line faults, which can improve the accuracy of locating optical fiber line faults.

[0007] A first aspect of the present invention provides a method for locating a fault in an optical fiber line, which is applied to an optical fiber communication system, wherein the optical fiber communication system includes a plurality of sites, each of which is provided with an optical time domain reflectometer; the method includes:

[0008] Controlling the optical time domain reflectometer of each station to transmit optical pulse signals to the optical fiber line and receive optical pulse reflection signals;

[0009] Performing time-frequency transformation on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal;

[0010] Extracting features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals;

[0011] Extracting the reflection signal time-frequency feature vector corresponding to the site to be evaluated from the reflection signal time-frequency feature vector set as a query feature vector;

[0012] Performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set to obtain a matching result representation feature matrix;

[0013] The matching result representation feature matrix is ​​optimized, and whether the site to be evaluated is a faulty site is determined based on the optimized matching result representation feature matrix.

[0014] In combination with the first aspect, performing time-frequency transformation on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal includes:

[0015] Discrete wavelet transform is performed on each optical pulse reflection signal to obtain a two-dimensional time-frequency diagram of the reflection signal, and a plurality of the two-dimensional time-frequency diagrams of the reflection signal constitute the set of the two-dimensional time-frequency diagrams of the reflection signal.

[0016] In combination with the first aspect, further, fine-grained decoupling and feature matching are performed on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, including:

[0017] Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to obtain a time-frequency feature vector subsequence;

[0018] Performing enhanced encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector;

[0019] Calculating the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors to obtain a time-frequency response covariance matrix;

[0020] Calculating the matrix trace of the time-frequency response covariance matrix to obtain a time-frequency response trace measurement factor;

[0021] Activating the time-frequency response trace measurement factor to obtain the time-frequency response trace attention weight;

[0022] The weighted sum of the time-frequency response covariance matrix is ​​calculated based on the attention weight of the time-frequency response trace to obtain a matching result representation feature matrix.

[0023] In combination with the first aspect, further, a subsequence semantic enhancement encoder based on the Bi-LSTM model performs enhancement encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector.

[0024] In combination with the first aspect, further, calculating the matrix trace of the time-frequency response covariance matrix to obtain the time-frequency response trace measurement factor includes:

[0025] The elements on the diagonal of the time-frequency response covariance matrix are summed to obtain the time-frequency response trace quantity factor.

[0026] In combination with the first aspect, further, the time-frequency response trace measurement factor is activated based on the Sigmoid activation function to obtain the time-frequency response trace attention weight.

[0027] In combination with the first aspect, further, optimizing the matching result representation feature matrix includes:

[0028] Inputting the matching result representation feature matrix into a pre-trained classifier to obtain an initial probability that the site to be evaluated is a faulty site;

[0029] The matching result representation feature matrix is ​​activated by a probabilistic activation function to obtain a probabilistic feature matrix;

[0030] Subtract the probabilized feature matrix from the all-one feature matrix to obtain a subtraction result matrix;

[0031] Constructing a power function feature matrix according to the point subtraction result matrix and the initial probability;

[0032] Performing a dot multiplication of the power function feature matrix and the probabilistic feature matrix to obtain an intermediate feature matrix;

[0033] Calculating biased eigenvalues ​​according to the intermediate feature matrix and the initial probability;

[0034] The bias eigenvalue is multiplied by the hyperparameter and then added to the intermediate feature matrix to obtain an optimized matching result representation feature matrix.

[0035] In combination with the first aspect, further, according to the point subtraction result matrix and the initial probability, a power function feature matrix is ​​constructed, including:

[0036] Each element in the point subtraction result matrix is ​​used as a base, and the initial probability is used as an exponent to construct a power function, and the value of each power function is calculated, and the power function feature matrix is ​​constructed according to the calculation results.

[0037] In combination with the first aspect, further, calculating the biased eigenvalue according to the intermediate feature matrix and the initial probability includes:

[0038] The absolute value of each element of the intermediate feature matrix is ​​used as the base, and the reciprocal of the initial probability is used as the exponent to construct a power function, calculate the value of each power function, and sum the calculation results to obtain the biased eigenvalue.

[0039] In combination with the first aspect, further, the classifier determines whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0040] A second aspect of the present invention provides a fault location device for an optical fiber line applied to the above method, the device comprising:

[0041] A signal acquisition module is used to control the optical time domain reflectometer of each station to transmit an optical pulse signal to the optical fiber line and receive an optical pulse reflection signal;

[0042] A time-frequency conversion module, used for performing time-frequency conversion on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal;

[0043] A first feature extraction module, configured to extract features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals;

[0044] A second feature extraction module is used to extract the reflection signal time-frequency feature vector corresponding to the site to be evaluated from the reflection signal time-frequency feature vector set as a query feature vector;

[0045] A decoupling module, used for performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, to obtain a matching result representation feature matrix;

[0046] The optimization module is used to optimize the matching result representation feature matrix, and determine whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0047] Furthermore, the time-frequency conversion module performs time-frequency conversion on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal, including:

[0048] Discrete wavelet transform is performed on each optical pulse reflection signal to obtain a two-dimensional time-frequency diagram of the reflection signal, and a plurality of the two-dimensional time-frequency diagrams of the reflection signal constitute the set of the two-dimensional time-frequency diagrams of the reflection signal.

[0049] Furthermore, the decoupling module performs fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, including:

[0050] Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to obtain a time-frequency feature vector subsequence;

[0051] Performing enhanced encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector;

[0052] Calculating the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors to obtain a time-frequency response covariance matrix;

[0053] Calculating the matrix trace of the time-frequency response covariance matrix to obtain a time-frequency response trace measurement factor;

[0054] Activating the time-frequency response trace measurement factor to obtain the time-frequency response trace attention weight;

[0055] The weighted sum of the time-frequency response covariance matrix is ​​calculated based on the attention weight of the time-frequency response trace to obtain a matching result representation feature matrix.

[0056] Furthermore, the decoupling module performs enhancement encoding on the time-frequency feature vector subsequence based on the subsequence semantic enhancement encoder of the Bi-LSTM model to obtain a local correlation feature vector.

[0057] Furthermore, the decoupling module calculates the matrix trace of the time-frequency response covariance matrix to obtain the time-frequency response trace measurement factor, including:

[0058] The elements on the diagonal of the time-frequency response covariance matrix are summed to obtain the time-frequency response trace quantity factor.

[0059] Furthermore, the decoupling module activates the time-frequency response trace measurement factor based on the Sigmoid activation function to obtain the time-frequency response trace attention weight.

[0060] Furthermore, the optimization module optimizes the matching result representation feature matrix, including:

[0061] Inputting the matching result representation feature matrix into a pre-trained classifier to obtain an initial probability that the site to be evaluated is a faulty site;

[0062] The matching result representation feature matrix is ​​activated by a probabilistic activation function to obtain a probabilistic feature matrix;

[0063] Subtract the probabilized feature matrix from the all-one feature matrix to obtain a subtraction result matrix;

[0064] Constructing a power function feature matrix according to the point subtraction result matrix and the initial probability;

[0065] Performing a dot multiplication of the power function feature matrix and the probabilistic feature matrix to obtain an intermediate feature matrix;

[0066] Calculating biased eigenvalues ​​according to the intermediate feature matrix and the initial probability;

[0067] The bias eigenvalue is multiplied by the hyperparameter and then added to the intermediate feature matrix to obtain an optimized matching result representation feature matrix.

[0068] Furthermore, the optimization module constructs a power function feature matrix according to the point subtraction result matrix and the initial probability, including:

[0069] Each element in the point subtraction result matrix is ​​used as a base, and the initial probability is used as an exponent to construct a power function, and the value of each power function is calculated, and the power function feature matrix is ​​constructed according to the calculation results.

[0070] Furthermore, the optimization module calculates the bias eigenvalue according to the intermediate feature matrix and the initial probability, including:

[0071] The absolute value of each element of the intermediate feature matrix is ​​used as the base, and the reciprocal of the initial probability is used as the exponent to construct a power function, calculate the value of each power function, and sum the calculation results to obtain the biased eigenvalue.

[0072] Furthermore, the classifier determines whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0073] A third aspect of the present invention provides a fault location system for an optical fiber line, comprising an optical fiber communication system, a processor and a storage device, wherein the optical fiber communication system comprises a plurality of sites, each of which is provided with an optical time domain reflectometer; the storage device stores a plurality of instructions, and the processor is used to read the instructions and execute the above method.

[0074] The optical fiber line fault location method, device and system provided by the present invention have at least the following beneficial effects:

[0075] (1) Perform time-frequency conversion and feature extraction on the optical pulse reflection signals collected by the optical time domain reflectometer of each site, so as to intelligently determine whether the site is a faulty site based on the time-frequency features of the optical pulse reflection signals of each site and the query matching results between other sites. In this way, the optical pulse reflection signals can be quickly analyzed, which can significantly reduce the time and response time required for fault location, thereby reducing the possibility of manual intervention and human error, and improving the efficiency and intelligence of fault location;

[0076] (2) Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to capture and extract the local detail information of the optical pulse reflected signal time-frequency, providing rich input for subsequent subtle difference matching;

[0077] (3) For the feature matrix represented by the matching result, based on the convergence limit of the probability distribution of the power series in the Cauchy-Hadamard form, the feature set of the feature matrix represented by the matching result is represented by a power series based on the convergence probability, and a set convergence bounded bias represented by the convergence radius based on each eigenvalue of the feature matrix represented by the matching result is added. Under the unified probability convergence limit standard, the matching result representation feature matrix can be prevented from deviating from the distribution pattern of the matching result representation feature matrix due to the gradient propagation of the model parameters based on the convergence probability during the regression process based on the classifier, resulting in invalid overflow of the classification result, thereby improving the accuracy of the fault location result obtained by the classifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 The present invention is a flowchart of an embodiment of a method for locating a fault in an optical fiber line provided by the present invention.

[0079] Figure 2 The present invention provides a flowchart of an embodiment of feature matching in the optical fiber line fault location method provided by the present invention.

[0080] Figure 3 The present invention provides a flowchart of an embodiment of optimizing a matching result representation feature matrix in a method for locating a fault in an optical fiber line provided by the present invention.

[0081] Figure 4 A schematic structural diagram of an embodiment of a fault locating device for an optical fiber line provided by the present invention.

[0082] Figure 5 A schematic structural diagram of an embodiment of a fault location system for an optical fiber line provided by the present invention. DETAILED DESCRIPTION

[0083] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0084] refer to Figure 1 In some embodiments, a method for locating a fault in an optical fiber line is provided, which is applied to an optical fiber communication system, wherein the optical fiber communication system includes a plurality of sites, each of which is provided with an optical time domain reflectometer; the method includes:

[0085] S1, controlling the optical time domain reflectometer of each site to transmit an optical pulse signal to the optical fiber line and receive an optical pulse reflection signal;

[0086] S2, performing time-frequency transformation on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal;

[0087] S3, extracting features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals;

[0088] S4, extracting the reflected signal time-frequency feature vector corresponding to the site to be evaluated from the reflected signal time-frequency feature vector set as a query feature vector;

[0089] S5, performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, to obtain a matching result representation feature matrix;

[0090] S6. Optimize the matching result representation feature matrix, and determine whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0091] Specifically, in step S1, the optical time domain reflectometer transmits a series of optical pulse signals to the optical fiber line. When the optical pulse signals are transmitted in the optical fiber, they will be scattered and reflected due to the properties of the optical fiber itself, connectors, joints, bends, etc. The returned optical pulse reflection signals are received and analyzed, and the loss, joints, breakpoints, etc. in the optical fiber are judged based on information such as signal strength and time delay. In general, by analyzing and processing the collected optical pulse reflection signals, the key information features in the reflection signals can be obtained, and then the state of the optical fiber and the fault location can be accurately judged, thereby achieving more accurate fault location.

[0092] Furthermore, in step S2, the optical pulse reflection signal is subjected to time-frequency transformation to obtain a set of two-dimensional time-frequency diagrams of the reflection signal, including:

[0093] Discrete wavelet transform is performed on each optical pulse reflection signal to obtain a two-dimensional time-frequency diagram of the reflection signal, and a plurality of the two-dimensional time-frequency diagrams of the reflection signal constitute the set of the two-dimensional time-frequency diagrams of the reflection signal.

[0094] Specifically, the frequency information of the optical pulse reflection signal is different in different local time scales. Discrete wavelet transform (DWT) is a signal processing technology used to decompose the signal into different frequency components. Unlike the traditional Fourier transform, wavelet transform can provide time and frequency information at the same time, which is suitable for analyzing non-stationary signals. The optical pulse reflection signal may contain fault information such as optical fiber breakage and poor connection. These faults usually produce short-term high-frequency components in the signal. DWT can effectively capture these local features and provide time-frequency information of the fault. Therefore, in this embodiment, the optical pulse reflection signal returned by each station is subjected to discrete wavelet transform to capture and extract the frequency characteristics within each local time point, and obtain a set of two-dimensional time-frequency diagrams of the reflection signal.

[0095] Furthermore, in step S3, feature extraction is performed on the two-dimensional time-frequency graphs of the reflected signal in the set of two-dimensional time-frequency graphs of the reflected signal to obtain a set of time-frequency feature vectors of the reflected signal. Specifically, feature extraction can be performed on the two-dimensional time-frequency graphs of the reflected signal in the set of two-dimensional time-frequency graphs of the reflected signal by a signal time-frequency feature extractor of a dilated convolutional neural network model.

[0096] Specifically, in order to capture and mine the frequency characteristics of the optical pulse reflection signal in the local time dimension from each reflection signal two-dimensional time-frequency diagram, in this embodiment, each reflection signal two-dimensional time-frequency diagram in the set of the reflection signal two-dimensional time-frequency diagram is respectively passed through a signal time-frequency feature extractor based on a hole convolutional neural network model to obtain a reflection signal time-frequency feature vector set. In particular, the receptive field in the two-dimensional time-frequency diagram of the optical pulse detection reflection signal is expanded by increasing the spacing of elements in the convolution kernel using the hole convolution, so as to capture a wider range of contextual information.

[0097] Further, in step S4, the reflection signal time-frequency feature vector corresponding to the site to be evaluated is extracted from the reflection signal time-frequency feature vector set as a query feature vector. Specifically, considering that the optical pulse reflection signal of each site contains unique time-frequency characteristics, it can reflect the specific status of the optical fiber at the site. Therefore, in order to be able to more carefully analyze and consider the time-frequency characteristics of the optical pulse reflection signal of each site, so as to quickly locate potential faults or abnormal conditions in the optical fiber, in this embodiment, the reflection signal time-frequency feature vector corresponding to the site to be evaluated is extracted from the reflection signal time-frequency feature vector set as a query feature vector. In this way, the specific information of the site can be focused on to analyze the optical fiber status of the site, thereby improving the accuracy of fault detection and location.

[0098] Further, refer to Figure 2 In step S5, the query feature vector is finely decoupled and feature matched with other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, including:

[0099] S51, performing fine-grained decoupling on other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set to obtain a time-frequency feature vector subsequence;

[0100] S52, performing enhanced encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector;

[0101] S53, calculating the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors to obtain a time-frequency response covariance matrix;

[0102] S54, calculating the matrix trace of the time-frequency response covariance matrix to obtain a time-frequency response trace measurement factor;

[0103] S55, performing activation processing on the time-frequency response trace measurement factor to obtain the time-frequency response trace attention weight;

[0104] S56. Calculate the weighted sum of the time-frequency response covariance matrix based on the time-frequency response trace attention weight to obtain a matching result representation feature matrix.

[0105] Specifically, in step S51, in order to more accurately and clearly analyze and understand the degree of matching between the time-frequency characteristics of the optical pulse reflection signal of the site to be evaluated and other sites, in this embodiment, the other reflection signal time-frequency feature vectors in the reflection signal time-frequency feature vector set are fine-grainedly decoupled to capture and refine the local detail information of the time-frequency of the optical pulse reflection signal, providing rich input for subsequent subtle difference matching.

[0106] Furthermore, in step S52, a subsequence semantic enhancement encoder based on a Bi-LSTM model performs enhancement encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector.

[0107] Specifically, the time-frequency feature vector subsequence is enhanced and encoded through the subsequence semantic enhancement encoder of the Bi-LSTM model, so as to extract and mine the expressions between the contexts of the time-frequency features of the reflection signal between each semantic span, and obtain the local correlation feature vector.

[0108] Furthermore, in step S53, the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors is calculated to obtain a time-frequency response covariance matrix to further quantify the linear time-frequency feature correlation relationship between the two and reflect the degree of matching between the two.

[0109] Furthermore, in step S54, the matrix trace of the time-frequency response covariance matrix is ​​calculated to obtain the time-frequency response trace measurement factor, including:

[0110] The elements on the diagonal of the time-frequency response covariance matrix are summed to obtain the time-frequency response trace quantity factor.

[0111] The overall time-frequency feature correlation between the query feature vector and each of the obtained local correlation feature vectors is evaluated by calculating the matrix trace of the time-frequency response covariance matrix.

[0112] Furthermore, in step S55, the time-frequency response trace measurement factor is activated based on the Sigmoid activation function to obtain the time-frequency response trace attention weight.

[0113] Finally, in step S56, the weighted sum of each time-frequency response covariance matrix is ​​calculated using the time-frequency response trace attention weight as the weight to comprehensively adjust the contribution of each feature to obtain the final matching result representation feature matrix.

[0114] Specifically, the process formula for fine-grained decoupling and feature matching of the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is as follows:

[0115] S={S1,S2,...,S n}; (1)

[0116] S i =Decompose(K); (2)

[0117] {E i}=Bi_LSTM({S i}); (3)

[0118] {C i = {Cov(Q, E i )}; (4)

[0119] {τ i}={Tr(C i )}; (5)

[0120] {a i}={σ(τ i )}; (6)

[0121] M=∑ i a i C i ; (7)

[0122] Where K represents the other time-frequency feature vectors of the reflected signal in the set of time-frequency feature vectors of the reflected signal, Decompose(·) is a fine-grained decoupling operation, S represents the set of subsequences of time-frequency feature vectors, and S irepresents the i-th time-frequency feature vector subsequence, Bi_LSTM(·) represents the Bi-LSTM encoder, E i represents the local correlation feature vector, Q represents the query feature vector, Cov(·,·) represents the operation of calculating the covariance matrix, C i represents the time-frequency response covariance matrix, Tr(·) represents the calculation matrix trace, τi represents the time-frequency response trace measurement factor, σ(·) is the sigmoid function, a i Represents the attention weight of the time-frequency response trace, and M represents the feature matrix of the matching result.

[0123] Furthermore, in the case where the set of reflection signal time-frequency feature vectors expresses the image semantic features of each reflection signal two-dimensional time-frequency graph, when the reflection signal time-frequency feature vector corresponding to the site to be evaluated performs adaptive coding query response with fine-grained decoupling features from other reflection signal time-frequency feature vectors, the obtained matching result representation feature matrix will also cause a response distribution regression mode shift due to the difference in adaptive coding correspondence under the fine-grained decoupling of image semantic features, that is, it will cause the matching result representation feature matrix to overflow the positioning result of the input classifier, affecting its accuracy.

[0124] Based on this, in this embodiment, reference Figure 3 In step S6, after obtaining the matching result representation feature matrix, the obtained matching result representation feature matrix is ​​optimized, specifically including:

[0125] S61, inputting the matching result representation feature matrix into a pre-trained classifier to obtain an initial probability that the site to be evaluated is a faulty site;

[0126] S62, activating the matching result representation feature matrix through a probabilistic activation function to obtain a probabilistic feature matrix;

[0127] S63, subtracting the probabilized feature matrix from the all-one feature matrix to obtain a subtraction result matrix;

[0128] S64, constructing a power function feature matrix according to the point subtraction result matrix and the initial probability;

[0129] S65, performing a dot multiplication of the power function feature matrix and the probabilistic feature matrix to obtain an intermediate feature matrix;

[0130] S66, calculating bias eigenvalues ​​according to the intermediate eigenmatrix and the initial probability;

[0131] S67: multiply the bias eigenvalue by the hyperparameter, and then add the result to the intermediate feature matrix to obtain an optimized matching result representation feature matrix.

[0132] Furthermore, in step S64, a power function feature matrix is ​​constructed according to the point subtraction result matrix and the initial probability, including:

[0133] Each element in the point subtraction result matrix is ​​used as a base, and the initial probability is used as an exponent to construct a power function, and the value of each power function is calculated, and the power function feature matrix is ​​constructed according to the calculation results.

[0134] Furthermore, in step S66, the biased eigenvalue is calculated according to the intermediate feature matrix and the initial probability, including:

[0135] The absolute value of each element of the intermediate feature matrix is ​​used as the base, and the reciprocal of the initial probability is used as the exponent to construct a power function, calculate the value of each power function, and sum the calculation results to obtain the biased eigenvalue.

[0136] Specifically, let the probabilistic feature matrix be M, then the optimized matching result representation feature matrix M′ is expressed as:

[0137]

[0138] Among them, M is the probabilistic feature matrix, M I is a fully-one eigenvalue matrix with all eigenvalues ​​equal to one, and f i ∈M, is the eigenvalue of each position in the probabilistic feature matrix M, ⊙ is the point multiplication by position, is subtracted by position point, p is the initial probability, (·) ⊙p is a power function with the initial probability as the exponent, It is added by position point, a represents the hyperparameter, and M′ represents the feature matrix of the optimized matching result.

[0139] For the feature matrix represented by the matching result, based on the convergence limit of the probability distribution of the power series in the Cauchy-Hadamard form, the feature set of the feature matrix represented by the matching result is represented by a power series based on the convergence probability, and a set convergence bounded bias represented by the convergence radius based on each eigenvalue of the feature matrix represented by the matching result is added. Under a unified probability convergence limit standard, the feature matrix represented by the matching result can be prevented from causing invalid overflow of the classification result due to the deviation of the distribution pattern of the feature matrix represented by the matching result caused by the gradient propagation of the model parameters based on the convergence probability during the regression process based on the classifier, thereby improving the accuracy of the positioning result obtained by the fault positioning based on the classifier.

[0140] Pre-train the classifier, including:

[0141] Collect sample data, which is the optical pulse reflection signal received after the optical time domain reflectometer of each site with known conditions transmits the optical pulse signal to the optical fiber line. Each site with known conditions includes a small number of faulty sites and most normal sites. After receiving the optical pulse reflection signal, perform time-frequency transformation on it to obtain a set of two-dimensional time-frequency graphs of the reflection signal; extract features of the two-dimensional time-frequency graphs of the reflection signal in the set of two-dimensional time-frequency graphs of the reflection signal to obtain a set of time-frequency feature vectors of the reflection signal; extract the reflection signal time-frequency feature vector corresponding to the faulty site from the set of reflection signal time-frequency feature vectors as the query feature vector, or the reflection signal time-frequency feature vector corresponding to the normal site as the query feature vector, perform fine-grained decoupling and feature matching on the query feature vector and other reflection signal time-frequency feature vectors in the set of reflection signal time-frequency feature vectors, obtain a matching result representation feature matrix, mark the matching result representation feature matrix according to whether the site corresponding to the query feature vector is faulty or not, and obtain sample data. The reflection signal time-frequency feature vectors corresponding to different sites can be selected as the query feature vector, thereby obtaining multiple different sample data, and the classifier is trained based on the sample data until the classifier converges.

[0142] Furthermore, in step S6, the classifier determines whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix, and specifically outputs the probability that the site to be evaluated is a faulty site.

[0143] refer to Figure 4 In some embodiments, a fault location device for an optical fiber line applied to the above method is provided, comprising:

[0144] The signal acquisition module 201 is used to control the optical time domain reflectometer of each site to transmit an optical pulse signal to the optical fiber line and receive an optical pulse reflection signal;

[0145] The time-frequency conversion module 202 is used to perform time-frequency conversion on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal;

[0146] A first feature extraction module 203 is used to extract features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals;

[0147] A second feature extraction module 204 is used to extract the reflection signal time-frequency feature vector corresponding to the site to be evaluated from the reflection signal time-frequency feature vector set as a query feature vector;

[0148] A decoupling module 205 is used to perform fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set to obtain a matching result representation feature matrix;

[0149] The optimization module 206 is used to optimize the matching result representation feature matrix, and determine whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0150] Furthermore, the time-frequency conversion module 202 performs time-frequency conversion on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal, including:

[0151] Discrete wavelet transform is performed on each optical pulse reflection signal to obtain a two-dimensional time-frequency diagram of the reflection signal, and a plurality of the two-dimensional time-frequency diagrams of the reflection signal constitute the set of the two-dimensional time-frequency diagrams of the reflection signal.

[0152] Furthermore, the decoupling module 205 performs fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, including:

[0153] Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to obtain a time-frequency feature vector subsequence;

[0154] Performing enhanced encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector;

[0155] Calculating the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors to obtain a time-frequency response covariance matrix;

[0156] Calculating the matrix trace of the time-frequency response covariance matrix to obtain a time-frequency response trace measurement factor;

[0157] Activating the time-frequency response trace measurement factor to obtain the time-frequency response trace attention weight;

[0158] The weighted sum of the time-frequency response covariance matrix is ​​calculated based on the attention weight of the time-frequency response trace to obtain a matching result representation feature matrix.

[0159] Furthermore, the decoupling module 205 performs enhancement encoding on the time-frequency feature vector subsequence based on the subsequence semantic enhancement encoder of the Bi-LSTM model to obtain a local correlation feature vector.

[0160] Furthermore, the decoupling module 205 calculates the matrix trace of the time-frequency response covariance matrix to obtain the time-frequency response trace measurement factor, including:

[0161] The elements on the diagonal of the time-frequency response covariance matrix are summed to obtain the time-frequency response trace quantity factor.

[0162] Furthermore, the decoupling module 205 activates the time-frequency response trace measurement factor based on the Sigmoid activation function to obtain the time-frequency response trace attention weight.

[0163] Furthermore, the optimization module 206 optimizes the matching result representation feature matrix, including:

[0164] Inputting the matching result representation feature matrix into a pre-trained classifier to obtain an initial probability that the site to be evaluated is a faulty site;

[0165] The matching result representation feature matrix is ​​activated by a probabilistic activation function to obtain a probabilistic feature matrix;

[0166] Subtract the probabilized feature matrix from the all-one feature matrix to obtain a subtraction result matrix;

[0167] Constructing a power function feature matrix according to the point subtraction result matrix and the initial probability;

[0168] Performing a dot multiplication of the power function feature matrix and the probabilistic feature matrix to obtain an intermediate feature matrix;

[0169] Calculating biased eigenvalues ​​according to the intermediate feature matrix and the initial probability;

[0170] The bias eigenvalue is multiplied by the hyperparameter and then added to the intermediate feature matrix to obtain an optimized matching result representation feature matrix.

[0171] Furthermore, the optimization module 206 constructs a power function feature matrix according to the point subtraction result matrix and the initial probability, including:

[0172] Each element in the point subtraction result matrix is ​​used as a base, and the initial probability is used as an exponent to construct a power function, and the value of each power function is calculated, and the power function feature matrix is ​​constructed according to the calculation results.

[0173] Furthermore, the optimization module 206 calculates the bias eigenvalue according to the intermediate feature matrix and the initial probability, including:

[0174] The absolute value of each element of the intermediate feature matrix is ​​used as the base, and the reciprocal of the initial probability is used as the exponent to construct a power function, calculate the value of each power function, and sum the calculation results to obtain the biased eigenvalue.

[0175] Furthermore, the classifier determines whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

[0176] Please refer to the above embodiments for the specific working principle, which will not be described in detail here.

[0177] Specifically, the optical fiber line fault location device provided in the above embodiment can be implemented in various terminal devices, such as a server for intelligent switching of optical fiber wiring. In some embodiments, the optical fiber line fault location device provided in the above embodiment can be integrated into the terminal device as a software module and / or a hardware module. For example, the optical fiber line fault location device can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the optical fiber line fault location device can also be one of the many hardware modules of the terminal device.

[0178] In some embodiments, the fault locating device of the optical fiber line and the terminal device may also be separate devices, and the fault locating device of the optical fiber line may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0179] In some embodiments, the terminal equipment can be an intelligent optical fiber switching device, which is equipped with an optical time domain reflectometer, which can realize remote testing of the spare fiber cores of all optical cables in the station. The remote fiber jumping function of the device can also be used to realize chain testing of downstream communication optical cables, which greatly saves manpower and material resources, and shortens the monitoring cycle. Moreover, through the switching device, the main station-side server can cooperate with the substation-side switching system to realize remote switching of optical cable cores, which greatly reduces the pressure on vehicle personnel configuration.

[0180] refer to Figure 5 In some embodiments, a fault location system for an optical fiber line is also provided, including an optical fiber communication system, a processor 2 and a storage device 3. The optical fiber communication system includes multiple sites 1, and the site 1 is provided with an optical time domain reflectometer 111; the storage device 3 stores multiple instructions, and the processor 2 is used to read the instructions and execute the above method.

[0181] The optical fiber line fault location method provided in the above embodiment has at least the following beneficial effects:

[0182] (1) Perform time-frequency conversion and feature extraction on the optical pulse reflection signals collected by the optical time domain reflectometer of each site, so as to intelligently determine whether the site is a faulty site based on the time-frequency features of the optical pulse reflection signals of each site and the query matching results between other sites. In this way, the optical pulse reflection signals can be quickly analyzed, which can significantly reduce the time and response time required for fault location, thereby reducing the possibility of manual intervention and human error, and improving the efficiency and intelligence of fault location;

[0183] (2) Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to capture and extract the local detail information of the optical pulse reflected signal time-frequency, providing rich input for subsequent subtle difference matching;

[0184] (3) For the feature matrix represented by the matching result, based on the convergence limit of the probability distribution of the power series in the Cauchy-Hadamard form, the feature set of the feature matrix represented by the matching result is represented by a power series based on the convergence probability, and a set convergence bounded bias represented by the convergence radius based on each eigenvalue of the feature matrix represented by the matching result is added. Under the unified probability convergence limit standard, the matching result representation feature matrix can be prevented from deviating from the distribution pattern of the matching result representation feature matrix due to the gradient propagation of the model parameters based on the convergence probability during the regression process based on the classifier, resulting in invalid overflow of the classification result, thereby improving the accuracy of the fault location result obtained by the classifier.

[0185] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for locating a fault in an optical fiber line, characterized in that: Applied to an optical fiber communication system, the optical fiber communication system includes a plurality of sites, each of which is provided with an optical time domain reflectometer; the method includes: Controlling the optical time domain reflectometers at each station to transmit optical pulse signals to the optical fiber line and receive optical pulse reflection signals; Performing time-frequency transformation on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal; Extracting features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals; Extracting the reflection signal time-frequency feature vector corresponding to the site to be evaluated from the reflection signal time-frequency feature vector set as a query feature vector; Performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set to obtain a matching result representation feature matrix; The matching result representation feature matrix is ​​optimized, and based on the optimized matching result representation feature matrix, it is determined whether the site to be evaluated is a faulty site.

2. The method according to claim 1, characterized in that The performing time-frequency transformation on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal further comprises the following steps: Discrete wavelet transform is performed on each optical pulse reflection signal to obtain a two-dimensional time-frequency diagram of the reflection signal, and a plurality of the two-dimensional time-frequency diagrams of the reflection signal constitute the set of the two-dimensional time-frequency diagrams of the reflection signal.

3. The method according to claim 1, characterized in that The fine-grained decoupling and feature matching of the query feature vector with other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set further comprises the following steps: Fine-grained decoupling of other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set is performed to obtain a time-frequency feature vector subsequence; Performing enhanced encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector; Calculating the covariance matrix between the query feature vector and each of the obtained local correlation feature vectors to obtain a time-frequency response covariance matrix; Calculating the matrix trace of the time-frequency response covariance matrix to obtain a time-frequency response trace measurement factor; Activating the time-frequency response trace measurement factor to obtain the time-frequency response trace attention weight; The weighted sum of the time-frequency response covariance matrix is ​​calculated based on the attention weight of the time-frequency response trace to obtain a matching result representation feature matrix.

4. The method according to claim 3, characterized in that The subsequence semantic enhancement encoder based on the Bi-LSTM model performs enhancement encoding on the time-frequency feature vector subsequence to obtain a local correlation feature vector.

5. The method according to claim 3, characterized in that: Calculating the matrix trace of the time-frequency response covariance matrix to obtain the time-frequency response trace measurement factor includes: The elements on the diagonal of the time-frequency response covariance matrix are summed to obtain the time-frequency response trace quantity factor.

6. The method according to claim 3, characterized in that The time-frequency response trace measurement factor is activated based on the Sigmoid activation function to obtain the time-frequency response trace attention weight.

7. The method according to claim 1, characterized in that Optimizing the matching result representation feature matrix includes: Inputting the matching result representation feature matrix into a pre-trained classifier to obtain an initial probability that the site to be evaluated is a faulty site; The matching result representation feature matrix is ​​activated by a probabilistic activation function to obtain a probabilistic feature matrix; Subtract the probabilized feature matrix from the all-one feature matrix to obtain a subtraction result matrix; Constructing a power function feature matrix according to the point subtraction result matrix and the initial probability; Performing a dot multiplication of the power function feature matrix and the probabilistic feature matrix to obtain an intermediate feature matrix; Calculating biased eigenvalues ​​according to the intermediate feature matrix and the initial probability; The bias eigenvalue is multiplied by the hyperparameter and then added to the intermediate feature matrix to obtain an optimized matching result representation feature matrix.

8. The method according to claim 7, characterized in that The classifier determines whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

9. A fault location device for an optical fiber line applied to the method according to any one of claims 1 to 8, characterized in that: include: A signal acquisition module is used to control the optical time domain reflectometer of each station to transmit an optical pulse signal to the optical fiber line and receive an optical pulse reflection signal; A time-frequency conversion module, used for performing time-frequency conversion on the optical pulse reflection signal to obtain a set of two-dimensional time-frequency graphs of the reflection signal; A first feature extraction module, configured to extract features from the two-dimensional time-frequency graphs of the reflected signals in the set of two-dimensional time-frequency graphs of the reflected signals to obtain a set of time-frequency feature vectors of the reflected signals; A second feature extraction module is used to extract the reflection signal time-frequency feature vector corresponding to the site to be evaluated from the reflection signal time-frequency feature vector set as a query feature vector; A decoupling module, used for performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors in the reflected signal time-frequency feature vector set, to obtain a matching result representation feature matrix; The optimization module is used to optimize the matching result representation feature matrix, and determine whether the site to be evaluated is a faulty site based on the optimized matching result representation feature matrix.

10. A fault location system for an optical fiber line, characterized in that: It comprises an optical fiber communication system, a processor and a storage device, wherein the optical fiber communication system comprises a plurality of sites, and each site is provided with an optical time domain reflectometer; the storage device stores a plurality of instructions, and the processor is used to read the instructions and execute the method according to any one of claims 1 to 8.

Images