A communication fault diagnosis method and device

Abstract

This invention belongs to the field of communications and provides a method and apparatus for diagnosing communication faults, comprising: acquiring a backscattered time-domain signal from a communication optical cable, obtaining multiple intrinsic mode functions through variational mode decomposition, selecting the signal with the largest kurtosis as the analysis signal, reconstructing the phase space of the analysis signal, and determining a set of relevant integral radii; calculating the relevant integral for a preset embedding dimension and candidate delay time values ​​to obtain difference statistics and correlation statistics, constructing a global index function and determining the optimal delay time, extracting the optimal delay time, corresponding curvature, and standard deviation of the difference statistics to form a real-time state feature vector, calculating the Mahalanobis distance based on the normal state benchmark mean vector and covariance matrix, comparing the distance with a preset threshold, and completing the diagnosis of communication optical cable faults.

Description

Technical Field

[0001] This application belongs to the field of communications, and in particular relates to a method and apparatus for diagnosing communication faults. Background Technology

[0002] During installation and long-term operation, optical cables are inevitably affected by factors such as geological subsidence, external construction damage, temperature and humidity changes, and aging, resulting in various faults such as breakage, excessive bending, and deterioration of connection points. In particular, early, minor faults, if not detected and located in time, may gradually evolve into serious faults, leading to communication interruptions and causing huge economic losses.

[0003] Currently, optical time-domain reflectometry (OTDR) is a technical means for monitoring and diagnosing optical cable conditions by injecting optical pulses into the cable and analyzing the returned backscattered signals. Traditional OTDR technology relies on interpreting changes in signal amplitude, which has limited ability to identify early or minor faults with indistinct signal characteristics. Furthermore, the backscattered signals themselves have low signal-to-noise ratios and are easily affected by environmental noise, making accurate fault diagnosis a technical challenge. Phase space reconstruction technology based on nonlinear dynamics theory can identify the inherent dynamic characteristics of signals. The accuracy of phase space reconstruction is highly dependent on the selection of the embedding dimension *m* and the delay time *τ*. The CC algorithm can simultaneously determine the optimal embedding dimension and delay time and has been widely used in time series analysis. However, traditional CC algorithms typically use a fixed radius sequence when calculating the correlation integral, making them insufficiently sensitive to signal scale changes and noise levels. The criteria for determining the optimal delay time are relatively simple, usually taking the first zero or minimum value of a certain statistic. Such criteria are easily affected by signal fluctuations in strong noise backgrounds, reducing the accuracy and reliability of fault diagnosis. Therefore, improving the method for determining the delay time and combining it with feature extraction and state assessment models is an urgent problem to be solved in order to improve the performance of optical cable fault diagnosis. Summary of the Invention

[0004] To address the problem that existing technologies are not sensitive enough to changes in signal scale and noise levels, and are easily affected by signal fluctuations in strong noise backgrounds.

[0005] In a first aspect, the present invention provides a communication fault diagnosis method, comprising:

[0006] The backscattered time-domain signal of the communication optical cable is obtained, and the time-domain signal is decomposed into multiple intrinsic mode functions using variational mode decomposition. The intrinsic mode function with the largest kurtosis value is selected as the analysis signal.

[0007] The analyzed signal is reconstructed in phase space. Based on the distance distribution between data points in the reconstructed phase space, a set of radii for calculating the correlation integral is determined. For multiple preset embedding dimensions and candidate values ​​of delay time, the correlation integral in phase space is calculated using the set of radii. Based on the correlation integral, a difference statistic representing the difference in correlation integral under different embedding dimensions and a correlation statistic representing the correlation between points in phase space are calculated for each candidate value of delay time. Combining the dispersion of the difference statistic under multiple embedding dimensions with the value of the correlation statistic, a global index function that varies with delay time t is constructed, and the t at which the global index function reaches its minimum value is determined as the optimal delay time.

[0008] The optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions are extracted, and after standardization, they are used to jointly constitute the real-time state feature vector of the optical cable. A benchmark feature distribution model of the communication optical cable under normal conditions is pre-established. The model includes a benchmark mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the benchmark mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the communication optical cable can be diagnosed.

[0009] Optionally, the step of acquiring the backscattered time-domain signal of the communication optical cable, decomposing the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and selecting the intrinsic mode function with the largest kurtosis value as the analysis signal includes:

[0010] The variational mode decomposition is set to 5 modes and the penalty factor is 2000. The center frequency is initialized, and the acquired backscattered time-domain signal is decomposed to obtain five intrinsic mode function components.

[0011] Calculate the kurtosis values ​​of the five intrinsic mode function components, where the kurtosis value is the expected value of the fourth power of the difference between the signal data point and the signal mean, divided by the fourth power of the signal standard deviation.

[0012] By comparing the five calculated kurtosis values, the intrinsic mode function component with the largest kurtosis value is determined as the analysis signal.

[0013] Optionally, the step of reconstructing the phase space of the analyzed signal and determining the set of radii for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space includes:

[0014] A preliminary phase space reconstruction is performed with an embedding dimension of 2 and a delay time of 1 to obtain a phase space data point set;

[0015] For each data point in the phase space, calculate the Euclidean distance between the data point and all other data points, find the fifth nearest neighbor data point and record this fifth nearest neighbor distance, and after traversing all data points, form the k nearest neighbor distance distribution.

[0016] Calculate the mean and standard deviation of the k-nearest neighbor distance distribution;

[0017] The value obtained by subtracting 1.5 times the standard deviation from the mean, and the maximum value between it and 0, is used as the lower limit of the radius interval. The value obtained by adding 1.5 times the standard deviation to the mean is used as the upper limit of the radius interval. The interval is then divided into 20 equal parts to obtain twenty radius values, which constitute the radius set used to calculate the relevant integral.

[0018] Optionally, the step of combining the dispersion of the difference statistic across multiple embedding dimensions with the value of the correlation statistic to construct a global index function that varies with the delay time t includes:

[0019] The embedding dimension is set to a range of 2 to 5, and the search range for the delay time is set to 1 to 50.

[0020] For each delay time value and each embedded dimension value, calculate the value of the relevant integral under the radius set, and calculate the difference between the maximum and minimum values ​​of the integral value under the radius set to obtain the difference statistic;

[0021] For each delay time value, the standard deviation of the difference statistic is calculated at embedding dimensions 2 to 5, and used as a dispersion index.

[0022] For each delay time value, the mean of the correlation integral obtained over all radii and embedding dimensions at embedding dimensions 2 to 5 will be used as the correlation statistic.

[0023] For each delay time value, the dispersion index is added to the correlation statistic to obtain the global index function value corresponding to the delay time.

[0024] Optionally, determining the optimal delay time as the t that minimizes the global metric function includes:

[0025] Iterate through the delay times from 1 to 50 with a step size of 1.

[0026] For each delay time value, calculate the value of the global index function corresponding to that time value;

[0027] After the traversal is complete, compare all the calculated global index function values, find the delay time that makes the global index function reach its global minimum value, and determine the delay time value as the optimal delay time.

[0028] Optionally, the extraction of the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions, and the standardization processing thereof, together constitute the real-time state feature vector of the optical cable, including:

[0029] The determined optimal delay time is used as the first component of the feature vector;

[0030] The curve of correlation statistic changing with delay time is taken as the correlation statistic curve. The first and second derivatives of the curve at the optimal delay time point are calculated using the numerical second-order central difference method. The curvature is then calculated as the second component of the eigenvector according to the curvature formula.

[0031] The standard deviation of the difference statistic at the optimal delay time with embedding dimensions of 2 to 5 is obtained as the third component of the feature vector;

[0032] The three components are standardized and then combined to form a real-time state feature vector.

[0033] Optionally, the pre-established baseline feature distribution model of the communication optical cable under normal conditions, the model including a baseline mean vector and a covariance matrix, calculates the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and compares the distance with a preset fault diagnosis threshold to achieve fault diagnosis of the communication optical cable, including:

[0034] 100 sets of backscattered time-domain signals of communication optical cables under normal conditions were collected, and 100 normal-state feature vectors were calculated for each set of signals.

[0035] The mean value of the 100 normal state feature vectors is calculated to obtain the benchmark mean vector, and the covariance matrix is ​​calculated to form the benchmark feature distribution model.

[0036] For the real-time state feature vector to be diagnosed, the square of the Mahalanobis distance between the vector and the baseline distribution is calculated according to the Mahalanobis distance formula.

[0037] The fault diagnosis threshold is the critical value of the chi-square distribution with 3 degrees of freedom at a confidence level of 99%. If the squared Mahalanobis distance calculated is greater than the threshold, the communication optical cable is determined to be faulty.

[0038] In another aspect, the present invention also provides a communication fault diagnosis device, comprising the following modules:

[0039] The selection module is used to acquire the backscattered time-domain signal of the communication optical cable, decompose the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and select the intrinsic mode function with the largest kurtosis value as the analysis signal.

[0040] A construction module is used to reconstruct the phase space of the analyzed signal. Based on the distance distribution between data points in the reconstructed phase space, a set of radii for calculating the correlation integral is determined. For multiple preset embedding dimensions and candidate values ​​of delay time, the correlation integral in the phase space is calculated using the set of radii. Based on the correlation integral, a difference statistic representing the difference in correlation integrals under different embedding dimensions and a correlation statistic representing the correlation between points in the phase space are calculated for each candidate value of delay time. Combining the dispersion of the difference statistic under multiple embedding dimensions and the value of the correlation statistic, a global index function that varies with the delay time t is constructed, and the t at which the global index function reaches its minimum value is determined as the optimal delay time.

[0041] The comparison module is used to extract the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions. After standardization, these parameters are used to jointly construct the real-time state feature vector of the optical cable. A baseline feature distribution model of the communication optical cable under normal conditions is pre-established. The model includes a baseline mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the communication optical cable can be diagnosed.

[0042] Preferably, the step of acquiring the backscattered time-domain signal of the communication optical cable, decomposing the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and selecting the intrinsic mode function with the largest kurtosis value as the analysis signal includes:

[0043] The variational mode decomposition is set to 5 modes and the penalty factor is 2000. The center frequency is initialized, and the acquired backscattered time-domain signal is decomposed to obtain five intrinsic mode function components.

[0044] Calculate the kurtosis values ​​of the five intrinsic mode function components, where the kurtosis value is the expected value of the fourth power of the difference between the signal data point and the signal mean, divided by the fourth power of the signal standard deviation.

[0045] By comparing the five calculated kurtosis values, the intrinsic mode function component with the largest kurtosis value is determined as the analysis signal.

[0046] Preferably, the step of reconstructing the phase space of the analyzed signal, and determining the set of radii for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space, includes:

[0047] A preliminary phase space reconstruction is performed with an embedding dimension of 2 and a delay time of 1 to obtain a phase space data point set;

[0048] For each data point in the phase space, calculate the Euclidean distance between the data point and all other data points, find the fifth nearest neighbor data point and record this fifth nearest neighbor distance, and after traversing all data points, form the k nearest neighbor distance distribution.

[0049] Calculate the mean and standard deviation of the k-nearest neighbor distance distribution;

[0050] The value obtained by subtracting 1.5 times the standard deviation from the mean, and the maximum value between it and 0, is used as the lower limit of the radius interval. The value obtained by adding 1.5 times the standard deviation to the mean is used as the upper limit of the radius interval. The interval is then divided into 20 equal parts to obtain twenty radius values, which constitute the radius set used to calculate the relevant integral.

[0051] Preferably, the step of constructing a global index function that varies with the delay time t by combining the dispersion of the difference statistic across multiple embedding dimensions with the value of the correlation statistic includes:

[0052] The embedding dimension is set to a range of 2 to 5, and the search range for the delay time is set to 1 to 50.

[0053] For each delay time value and each embedded dimension value, calculate the value of the relevant integral under the radius set, and calculate the difference between the maximum and minimum values ​​of the integral value under the radius set to obtain the difference statistic;

[0054] For each delay time value, the standard deviation of the difference statistic is calculated at embedding dimensions 2 to 5, and used as a dispersion index.

[0055] For each delay time value, the mean of the correlation integral obtained over all radii and embedding dimensions at embedding dimensions 2 to 5 will be used as the correlation statistic.

[0056] For each delay time value, the dispersion index is added to the correlation statistic to obtain the global index function value corresponding to the delay time.

[0057] Preferably, determining the optimal delay time as the t that minimizes the global index function includes:

[0058] Iterate through the delay times from 1 to 50 with a step size of 1.

[0059] For each delay time value, calculate the value of the global index function corresponding to that time value;

[0060] After the traversal is complete, compare all the calculated global index function values, find the delay time that makes the global index function reach its global minimum value, and determine the delay time value as the optimal delay time.

[0061] Preferably, the step of extracting the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions, and then standardizing them to jointly constitute the real-time state feature vector of the optical cable, includes:

[0062] The determined optimal delay time is used as the first component of the feature vector;

[0063] The curve of correlation statistic changing with delay time is taken as the correlation statistic curve. The first and second derivatives of the curve at the optimal delay time point are calculated using the numerical second-order central difference method. The curvature is then calculated as the second component of the eigenvector according to the curvature formula.

[0064] The standard deviation of the difference statistic at the optimal delay time with embedding dimensions of 2 to 5 is obtained as the third component of the feature vector;

[0065] The three components are standardized and then combined to form a real-time state feature vector.

[0066] Preferably, the pre-established baseline feature distribution model of the communication optical cable under normal conditions includes a baseline mean vector and a covariance matrix. The model is used to diagnose the fault state of the communication optical cable by calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing this distance with a preset fault diagnosis threshold.

[0067] 100 sets of backscattered time-domain signals of communication optical cables under normal conditions were collected, and 100 normal-state feature vectors were calculated for each set of signals.

[0068] The mean value of the 100 normal state feature vectors is calculated to obtain the benchmark mean vector, and the covariance matrix is ​​calculated to form the benchmark feature distribution model.

[0069] For the real-time state feature vector to be diagnosed, the square of the Mahalanobis distance between the vector and the baseline distribution is calculated according to the Mahalanobis distance formula.

[0070] The fault diagnosis threshold is the critical value of the chi-square distribution with 3 degrees of freedom at a confidence level of 99%. If the squared Mahalanobis distance calculated is greater than the threshold, the communication optical cable is determined to be faulty.

[0071] This invention processes the original signal through variational mode decomposition combined with kurtosis criterion, effectively filtering out noise interference and accurately extracting the core signal components representing the optical cable's state. In determining the optimal delay time, the process of selecting the radius of the correlation integral is improved, and a global index function fusing the dispersion of difference statistics and correlation statistics is constructed, enhancing the accuracy and reliability of optimal delay time calculation. By constructing a multidimensional feature vector containing the optimal delay time, the curvature of correlation statistics, and the standard deviation of difference statistics, and combining it with Mahalanobis distance for state discrimination, the operating state of the optical cable can be represented more comprehensively. This not only considers the intrinsic correlation between various features but also improves the sensitivity to early or weak fault identification, thereby achieving stable diagnosis of communication optical cable faults and reducing false alarm and missed alarm rates. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the communication optical cable fault diagnosis system provided by the present invention;

[0073] Figure 2 A flowchart of a communication fault diagnosis method provided by the present invention;

[0074] Figure 3 This is a schematic diagram illustrating the construction of the three-dimensional real-time state feature vector provided by the present invention. Detailed Implementation

[0075] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0076] It should be understood that the terms “comprising” and “having”, and any variations thereof, in the embodiments of this specification are intended to cover but not exclude inclusion. For example, a product or device that includes a series of components is not necessarily limited to those components that are explicitly listed, but may include other components that are not explicitly listed or that are inherent to such product or device.

[0077] Communication optical cable fault diagnosis system architecture diagram, such as Figure 1 As shown. The architecture, from bottom to top, is divided into a data acquisition layer, a processing layer, a feature layer, and a diagnostic layer. Specifically, in embodiment one, this invention proposes a communication fault diagnosis method, such as... Figure 2 As shown, it includes the following steps:

[0078] S1, Obtain the backscattered time-domain signal of the communication optical cable, decompose the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and select the intrinsic mode function with the largest kurtosis value as the analysis signal;

[0079] An optical time-domain reflectometer (OTDR) is used to emit light pulses along the communication optical cable and receive the returned Rayleigh backscattered light signals. These light signals are then converted into a one-dimensional time-domain signal sequence, denoted as the original signal X. The variational mode decomposition algorithm is set with the number of modes K and a quadratic penalty factor. and convergence tolerance After considering hyperparameters, the system calls tools such as the vmdpy utility library in Python's scientific computing library to perform variational mode decomposition on the original signal X, obtaining K intrinsic mode function (IMF) components. For each IMF component, the kurtosis value is calculated using functions such as the kurtosis function in the scipy.stats library, which is based on the fourth central moment of the signal divided by the square of the variance. The kurtosis values ​​of all IMF components are compared, and the IMF component with the largest kurtosis value is selected as the analysis signal for subsequent analysis.

[0080] As an optional implementation, the step of acquiring the backscattered time-domain signal of the communication optical cable, decomposing the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and selecting the intrinsic mode function with the largest kurtosis value as the analysis signal includes:

[0081] The variational mode decomposition is set to 5 modes and the penalty factor is 2000. The center frequency is initialized, and the acquired backscattered time-domain signal is decomposed to obtain five intrinsic mode function components.

[0082] Calculate the kurtosis values ​​of the five intrinsic mode function components, where the kurtosis value is the expected value of the fourth power of the difference between the signal data point and the signal mean, divided by the fourth power of the signal standard deviation.

[0083] By comparing the five calculated kurtosis values, the intrinsic mode function component with the largest kurtosis value is determined as the analysis signal.

[0084] Discrete backscattered time-domain signals x(n) of length N, for example, N=4096, are acquired using an optical time-domain reflectometer (OTDR). To suppress noise and extract vibration or disturbance information, variational mode decomposition (VMD) is employed. Key VMD parameters are set as follows: mode decomposition number K=5, penalty factor... =2000, convergence tolerance is 1e-7. The algorithm initializes the center frequencies of each mode, for example, uniformly distributed within the normalized frequency range of [0, 0.5], and iteratively updates them using the alternating direction multiplier method, decomposing the original signal x(n) into five intrinsic mode functions (IMF) components. k=1,2,3,4,5. To quantitatively assess the amount of shock or abrupt change information contained in each IMF component, kurtosis values ​​are calculated. ,in and These are components The mean and standard deviation are given, and E represents the expected value. For example, the calculated five kurtosis values ​​are [3.1, 2.8, 8.5, 4.2, 2.9]. Since the third component has the largest kurtosis value of 8.5, it indicates that this component contains non-Gaussian or impulsive characteristics, therefore it is selected... The optimal analysis signal s(n) is used for subsequent analysis.

[0085] S2, perform phase space reconstruction on the analyzed signal, and determine the radius set for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space; for multiple preset embedding dimensions and candidate values ​​of delay time, calculate the correlation integral in the phase space using the radius set, and based on the correlation integral, calculate the difference statistic representing the difference in correlation integral under different embedding dimensions and the correlation statistic representing the correlation between points in the phase space for each candidate value of delay time; combine the dispersion of the difference statistic under multiple embedding dimensions with the value of the correlation statistic to construct a global index function that varies with the delay time t, and determine the t when the global index function reaches its minimum value as the optimal delay time;

[0086] The embedding dimension *m* is defined within a range, typically from 2 to 5, along with a candidate range for the delay time *t*, for example, from 1 to 5% of the signal length. For each candidate delay time *t* and each embedding dimension *m*, the Takens embedding theorem is used to reconstruct the phase space of the analyzed signal, generating a set of points in the reconstructed phase space. Randomly selected pairs of data points are then used in the reconstructed phase space to calculate the Euclidean distance between them, and the standard deviation of these distance values ​​is determined. Set the radius set to a series of equally spaced values, for example, take the radius , where i ranges from 1 to 4. For each radius Calculate the relevant integral C( ): That is, the distance in space is less than The proportion of point pairs to the total number of point pairs. For each combination of t and m, calculate the difference statistic. The value of the difference statistic is the total radius of the current combination. The corresponding integral C( The difference between the maximum and minimum values ​​of ). For each candidate delay time t, two core statistics are calculated. The first is the dispersion of the difference statistic across multiple embedding dimensions m, specifically calculated as follows: , , , The standard deviation of this set of data is denoted as . The second is the correlation statistic. The value of the correlation statistic is , , , The arithmetic mean of this set of data. Construct a global indicator function. for and The sum. Iterate through all candidate delay times t and calculate a series of... The value is found using the argmin function. The time t that reaches its minimum value is the optimal delay time.

[0087] As an alternative implementation, the step of reconstructing the phase space of the analyzed signal and determining the set of radii for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space includes:

[0088] A preliminary phase space reconstruction is performed with an embedding dimension of 2 and a delay time of 1 to obtain a phase space data point set;

[0089] For each data point in the phase space, calculate the Euclidean distance between the data point and all other data points, find the fifth nearest neighbor data point and record this fifth nearest neighbor distance, and after traversing all data points, form the k nearest neighbor distance distribution.

[0090] Calculate the mean and standard deviation of the k-nearest neighbor distance distribution;

[0091] The value obtained by subtracting 1.5 times the standard deviation from the mean, and the maximum value between it and 0, is used as the lower limit of the radius interval. The value obtained by adding 1.5 times the standard deviation to the mean is used as the upper limit of the radius interval. The interval is then divided into 20 equal parts to obtain twenty radius values, which constitute the radius set used to calculate the relevant integral.

[0092] After selecting the intrinsic mode function with the largest kurtosis as the analysis signal s(n), a data-driven method is used to determine the radius set to avoid the subjectivity of radius selection in the CC algorithm. A low-dimensional preliminary phase space reconstruction is performed, fixing the embedding dimension m=2 and the delay time t=1, mapping the length N=4096 of the one-dimensional time series s(n) to N-1 two-dimensional phase space points Y(i)=[s(i),s(i+1)]. For each point Y(i) in the phase space, the Euclidean distance d(i,j) between it and all other points Y(j)j≠i is calculated. For each point Y(i), the N-2 distance values ​​corresponding to that point are arranged in ascending order, and the fifth smallest distance value, i.e., the fifth nearest neighbor distance, is recorded. After traversing all N-1 data points, a k-nearest neighbor distance distribution (k=5) containing N-1 distance values ​​is obtained. Calculate the statistical properties of this distribution, such as obtaining the mean. =0.08 and standard deviation =0.02. Based on 3 The idea behind the criterion is to define a reasonable radius search interval, with a lower bound. =0.05, upper limit =0.11. Divide this interval [0.05, 0.11] into 20 equal parts linearly to generate a set R consisting of 20 radius values, for example, R = {0.050, 0.053, 0.056, ..., 0.110}. This set will be used for subsequent integral calculations under all embedding dimensions and delay times.

[0093] As an alternative implementation, the step of constructing a global index function that varies with the delay time t by combining the dispersion of the difference statistic across multiple embedding dimensions with the value of the correlation statistic includes:

[0094] The embedding dimension is set to a range of 2 to 5, and the search range for the delay time is set to 1 to 50.

[0095] For each delay time value and each embedded dimension value, calculate the value of the relevant integral under the radius set, and calculate the difference between the maximum and minimum values ​​of the integral value under the radius set to obtain the difference statistic;

[0096] For each delay time value, the standard deviation of the difference statistic is calculated at embedding dimensions 2 to 5, and used as a dispersion index.

[0097] For each delay time value, the mean of the correlation integral obtained over all radii and embedding dimensions at embedding dimensions 2 to 5 will be used as the correlation statistic.

[0098] For each delay time value, the dispersion index is added to the correlation statistic to obtain the global index function value corresponding to the delay time.

[0099] Construct a global index function that can comprehensively represent the characteristics of phase space. This process is implemented through a double loop: the outer loop iterates over candidate values ​​t for the delay time, ranging from 1 to 50, with a step size of 1; the inner loop iterates over the embedding dimension m, ranging from 2 to 5. For each (t, m) combination, a phase space point set Y(i) = [s(i), s(i+t), ..., s(i+(m-1)t)] is generated based on s(n). Using the previously determined set R of 20 radius values, 20 corresponding correlation integrals C( (m,t), where m=2,3,4,5. Based on the 20 relevant integral values, calculate the difference statistic. After the inner loop finishes, for the current delay time t, four difference statistics are obtained { , , , }. Calculate the standard deviation of the four values ​​to obtain the dispersion index. Simultaneously, the arithmetic mean of the 80 relevant integral values ​​corresponding to all four embedding dimensions and 20 radii at the current time t is calculated to obtain the correlation statistic. Add the two metrics above to construct the global metric function value corresponding to the current delay time t: . This indicates the consistency of local density differences in phase space across different dimensions, while This indicates the overall correlation strength of points in phase space.

[0100] As an alternative implementation, determining the optimal delay time as the t that minimizes the global metric function includes:

[0101] Iterate through the delay times from 1 to 50 with a step size of 1.

[0102] For each delay time value, calculate the value of the global index function corresponding to that time value;

[0103] After the traversal is complete, compare all the calculated global index function values, find the delay time that makes the global index function reach its global minimum value, and determine the delay time value as the optimal delay time.

[0104] The calculation yielded 50 global index function values ​​with delay times t ranging from 1 to 50. Then, a discrete index sequence S={ , ,..., The minimum point of this sequence corresponds to the phase space having optimal expansion characteristics and minimal redundancy information at that delay time. Determining the optimal delay time... The operation is a search process: iterate through the index sequence S and compare all 50... Find the global minimum value within the given sequence. For example, if the calculated sequence is [0.85, 0.72, ..., 0.15, 0.18, ...], and the global minimum value is found within the entire sequence... =0.15 is the only minimum value, so the output 12 is taken as the optimal delay time, that is =12. The aforementioned The value will be used as a key parameter for subsequent feature extraction and fault diagnosis.

[0105] S3. Extract the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions. After standardization, these are used to construct the real-time state feature vector of the optical cable. A baseline feature distribution model of the optical cable under normal conditions is pre-established. The model includes a baseline mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the optical cable can be diagnosed.

[0106] The first feature is the optimal delay time obtained in the previous step. The second characteristic is the correlation statistic. The curve is The curvature at a given point is calculated using numerical differentiation methods, such as the `gradient` function from the NumPy library. Calculate the first and second derivatives of the data points and substitute them into the curvature formula. The curvature value is obtained. The third feature is the standard deviation of the difference statistic at the optimal delay time, i.e. The values ​​are then standardized to form a three-dimensional real-time state feature vector, such as... Figure 3 As shown. A large number of backscattered signals from normally operating optical cables are pre-collected. For each signal, the above steps are repeated to extract feature vectors, resulting in a feature vector sample set for normal operation. The mean vector and covariance matrix of this sample set are calculated using functions such as `mean` and `cov` from the NumPy library, forming a baseline feature distribution model. For the real-time acquired feature vector `x`, the formula is used to calculate the relationship between the feature vector and the baseline mean vector. The square of the Mahalanobis distance between The calculation process is as follows: calculate x and Given the difference vector, the inverse of the benchmark covariance matrix can be calculated using the `numpy.linalg.inv` function. The squared Mahalanobis distance is calculated. Compared with the preset fault diagnosis threshold, if If the value exceeds a threshold, the communication optical cable is considered faulty; otherwise, it is considered normal. This threshold can be set based on the Mahalanobis distance distribution of the normal sample set, using, for example, the 3σ criterion.

[0107] As an optional implementation, the extraction of the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions, and the standardization processing of these parameters, together constitute the real-time state feature vector of the optical cable, including:

[0108] The determined optimal delay time is used as the first component of the feature vector;

[0109] The curve of correlation statistic changing with delay time is taken as the correlation statistic curve. The first and second derivatives of the curve at the optimal delay time point are calculated using the numerical second-order central difference method. The curvature is then calculated as the second component of the eigenvector according to the curvature formula.

[0110] The standard deviation of the difference statistic at the optimal delay time with embedding dimensions of 2 to 5 is obtained as the third component of the feature vector;

[0111] The three components are combined to form a real-time state feature vector.

[0112] Once the optimal delay time is determined ,For example =12, and three highly sensitive physical quantities can be extracted from the calculation process to form a three-dimensional feature vector x=[ , , The first component represents the real-time status of the optical cable. Choose the optimal delay time value. =12. The second component It is a correlation statistic curve At t= The curvature at a given point can sensitively represent the local inflection point characteristics of a curve. Because... It is a discrete sequence, and the derivative of the sequence is estimated numerically. The second-order central difference method is used, and the first-order derivative is... Second derivative For example, if =0.08, =0.05, =0.06, then ≈-0.01, ≈0.04. According to the curvature formula Calculations yielded ≈0.04. The third component It is a global indicator function. One of the components, namely, at the optimal delay time At m=2 to 5, the standard deviation of the difference statistic, also known as the dispersion index. This value has already been determined. Calculated in time, for example =0.11. The mean of the optimal delay time component is obtained based on normal sample statistics. =10, standard deviation =2.5, mean of curvature components =0.03, standard deviation =0.012, the mean of the standard deviation of the difference statistic =0.10, standard deviation =0.03; Each component was calculated separately using the Z-score normalization formula: =(12-10) / 2.5=0.8, =(0.04-0.03) / 0.012≈0.83, =(0.11-0.10) / 0.03≈0.33; finally, the standardized real-time state feature vector is obtained. =[0.8,0.83,0.33].

[0113] As an optional implementation, the pre-established baseline feature distribution model of the communication optical cable under normal conditions includes a baseline mean vector and a covariance matrix. The model is used to diagnose the fault state of the communication optical cable by calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing this distance with a preset fault diagnosis threshold. This includes:

[0114] 100 sets of backscattered time-domain signals of communication optical cables under normal conditions were collected, and 100 normal-state feature vectors were calculated for each set of signals.

[0115] The mean value of the 100 normal state feature vectors is calculated to obtain the benchmark mean vector, and the covariance matrix is ​​calculated to form the benchmark feature distribution model.

[0116] For the real-time state feature vector to be diagnosed, the square of the Mahalanobis distance between the vector and the baseline distribution is calculated according to the Mahalanobis distance formula.

[0117] The fault diagnosis threshold is the critical value of the chi-square distribution with 3 degrees of freedom at a confidence level of 99%. If the squared Mahalanobis distance calculated is greater than the threshold, the communication optical cable is determined to be faulty.

[0118] During the offline modeling phase, M=100 sets of OTDR backscattered signals from communication optical cables under clearly defined normal and fault-free conditions were collected. For each of the 100 sets of signals, the aforementioned VMD decomposition, optimal parameter determination, feature extraction, and standardization processes were fully executed to obtain a 100x3 normal state feature vector matrix. Based on this matrix, the column mean was calculated to obtain a 3x1 baseline mean vector. and a 3x3 baseline covariance matrix. The two statistics mentioned above ( , Together, they defined the distribution model of the normal state of the optical cable in the feature space. During the online diagnostic phase, for a newly acquired real-time OTDR signal to be diagnosed, the corresponding real-time state feature vector is also calculated. Calculate the squared Mahalanobis distance between this vector and the normal state distribution model: ,in It is the inverse of the covariance matrix. The calculated... The result is compared to a preset fault diagnosis threshold T, for example, a fault diagnosis threshold of 11.34. This threshold T is set according to statistical principles, taking the critical value of a chi-square distribution with 3 degrees of freedom, a feature vector dimension, and a confidence level of 99%, i.e., T=11.34. If... A value >11.34 indicates that the current state vector deviates significantly from its normal distribution, thus indicating a fault in the optical cable; conversely, if... If the value is ≤11.34, the optical cable is considered to be in normal condition.

[0119] The acquired OTDR backscattered time-domain signal of length N=4096 was decomposed using variational mode decomposition (VMD), with the mode number K=5 and a penalty factor set. =2000, resulting in five intrinsic mode functions (IMFs). The kurtosis of each IMF is calculated, for example, [3.1, 2.8, 8.5, 4.2, 2.9], and IMF3 with the largest kurtosis of 8.5 is selected as the analysis signal s(n). To determine the optimal delay time for the CC algorithm, a set of radii for calculating the correlation integral needs to be determined first. Therefore, a preliminary phase space reconstruction is performed on the signal s(n), setting the embedding dimension m=2 and the delay time t=1, generating N-1 phase space points Y(i)=[s(i), s(i+1)]. For each point Y(i), the Euclidean distance from that point to all other points is calculated, and the distance to the fifth nearest neighbor is found. Calculate the mean of a distribution formed by the fifth nearest neighbor distances of all points. For example, 0.08 and standard deviation For example, 0.02. Set the lower limit of the radius interval. =0.05, upper limit =0.11. Dividing the interval [0.05, 0.11] linearly into 20 equal parts, we obtain a set of radii R = {...} , ,..., Within the search range of delay time t [1, 50] with a step size of 1 and the range of embedding dimension m [2, 5], a global index function is constructed. For each t value, calculate the dispersion index corresponding to that value. and correlation statistics For each (t,m) combination, calculate the correlation integral C(t,m) over 20 radii. The difference statistic is obtained by considering m and t. ; That is { , , , The standard deviation of}; This is the mean of all 80 relevant integral values. The global index function is defined as follows: By iterating through t from 1 to 50, the following can be calculated: Find the curve and the delay time corresponding to the global minimum point, for example, when t=12. The minimum value determines the optimal delay time. =12.

[0120] Determining the optimal delay time Then, a three-dimensional real-time state feature vector for state diagnosis can be constructed. The first component... This is the optimal delay time. ,For example =12. The second component Correlation statistic curve At t= The curvature k at that point. This curvature is calculated numerically: the first and second derivatives are estimated using the second-order central difference formula, for example... and If calculated =-0.01, =0.04, then the curvature ≈0.04, that is =0.04. The third component At the optimal delay time The standard deviation of the difference statistic at embedding dimensions 2 to 5, i.e. ,For example =0.11. The three components, after standardization, constitute the real-time state feature vector. =[0.8,0.83,0.33]. The diagnostic process is divided into offline modeling and online diagnosis. In the offline stage, OTDR signals are collected, for example, 100 sets of signals from optical cables operating normally. For each set of signals, the above steps are repeated to calculate 100 normal state feature vectors, and a baseline mean vector is calculated based on this sample set. and the 3x3 baseline covariance matrix These together constitute the baseline feature distribution model. During online diagnosis, feature vectors are calculated for newly acquired signals. The squared distance between the eigenvector and the baseline distribution is calculated using the Mahalanobis distance formula: The calculated The value is compared with a preset fault diagnosis threshold T. This threshold T is determined based on the critical value of a chi-square distribution with 3 degrees of freedom and eigenvector dimensions at a 99% confidence level, i.e., T = 11.34. If the value is greater than 11.34, then the communication optical cable is determined to be faulty; if If the value is ≤11.34, the optical cable is considered to be in normal condition.

[0121] In a second specific embodiment, the present invention also provides a communication fault diagnosis device, comprising the following modules:

[0122] The selection module is used to acquire the backscattered time-domain signal of the communication optical cable, decompose the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and select the intrinsic mode function with the largest kurtosis value as the analysis signal.

[0123] A construction module is used to reconstruct the phase space of the analyzed signal. Based on the distance distribution between data points in the reconstructed phase space, a set of radii for calculating the correlation integral is determined. For multiple preset embedding dimensions and candidate values ​​of delay time, the correlation integral in the phase space is calculated using the set of radii. Based on the correlation integral, a difference statistic representing the difference in correlation integrals under different embedding dimensions and a correlation statistic representing the correlation between points in the phase space are calculated for each candidate value of delay time. Combining the dispersion of the difference statistic under multiple embedding dimensions and the value of the correlation statistic, a global index function that varies with the delay time t is constructed, and the t at which the global index function reaches its minimum value is determined as the optimal delay time.

[0124] The comparison module is used to extract the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions. After standardization, these parameters are used to jointly construct the real-time state feature vector of the optical cable. A baseline feature distribution model of the communication optical cable under normal conditions is pre-established. The model includes a baseline mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the communication optical cable can be diagnosed.

[0125] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and to aid in understanding a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art, upon reading this specification, may readily identify some of the devices as separate embodiments. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. And the content of each secondary embodiment is valid even if it contains fewer than all the features of a single foregoing disclosed embodiment.

Claims

1. A communication fault diagnosis method, characterized in that, Includes the following steps: The backscattered time-domain signal of the communication optical cable is obtained, and the time-domain signal is decomposed into multiple intrinsic mode functions using variational mode decomposition. The intrinsic mode function with the largest kurtosis value is selected as the analysis signal. The analyzed signal is reconstructed in phase space. Based on the distance distribution between data points in the reconstructed phase space, a set of radii for calculating the correlation integral is determined. For multiple preset embedding dimensions and candidate values ​​of delay time, the correlation integral in phase space is calculated using the set of radii. Based on the correlation integral, a difference statistic representing the difference in correlation integral under different embedding dimensions and a correlation statistic representing the correlation between points in phase space are calculated for each candidate value of delay time. Combining the dispersion of the difference statistic under multiple embedding dimensions with the value of the correlation statistic, a global index function that varies with delay time t is constructed, and the t at which the global index function reaches its minimum value is determined as the optimal delay time. The optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions are extracted, and after standardization, they are used to jointly constitute the real-time state feature vector of the optical cable. A benchmark feature distribution model of the communication optical cable under normal conditions is pre-established. The model includes a benchmark mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the benchmark mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the communication optical cable can be diagnosed.

2. The method according to claim 1, characterized in that, The process of acquiring the backscattered time-domain signal of the communication optical cable, decomposing the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and selecting the intrinsic mode function with the largest kurtosis value as the analysis signal includes: The variational mode decomposition is set to 5 modes and the penalty factor is 2000. The center frequency is initialized, and the acquired backscattered time-domain signal is decomposed to obtain five intrinsic mode function components. Calculate the kurtosis values ​​of the five intrinsic mode function components, where the kurtosis value is the expected value of the fourth power of the difference between the signal data point and the signal mean, divided by the fourth power of the signal standard deviation. By comparing the five calculated kurtosis values, the intrinsic mode function component with the largest kurtosis value is determined as the analysis signal.

3. The method according to claim 2, characterized in that, The step of reconstructing the phase space of the analyzed signal, and determining the set of radii for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space, includes: A preliminary phase space reconstruction is performed with an embedding dimension of 2 and a delay time of 1 to obtain a phase space data point set; For each data point in the phase space, calculate the Euclidean distance between the data point and all other data points, find the fifth nearest neighbor data point and record this fifth nearest neighbor distance, and after traversing all data points, form the k nearest neighbor distance distribution. Calculate the mean and standard deviation of the k-nearest neighbor distance distribution; The value obtained by subtracting 1.5 times the standard deviation from the mean, and the maximum value between it and 0, is used as the lower limit of the radius interval. The value obtained by adding 1.5 times the standard deviation to the mean is used as the upper limit of the radius interval. The interval is then divided into 20 equal parts to obtain twenty radius values, which constitute the radius set used to calculate the relevant integral.

4. The method according to claim 1, characterized in that, The method combines the dispersion of the difference statistic across multiple embedding dimensions with the value of the correlation statistic to construct a global index function that varies with the delay time t, including: The embedding dimension is set to a range of 2 to 5, and the search range for the delay time is set to 1 to 50. For each delay time value and each embedded dimension value, calculate the value of the relevant integral under the radius set, and calculate the difference between the maximum and minimum values ​​of the integral value under the radius set to obtain the difference statistic; For each delay time value, the standard deviation of the difference statistic is calculated at embedding dimensions 2 to 5, and used as a dispersion index. For each delay time value, the mean of the correlation integral obtained over all radii and embedding dimensions at embedding dimensions 2 to 5 will be used as the correlation statistic. For each delay time value, the dispersion index is added to the correlation statistic to obtain the global index function value corresponding to the delay time.

5. The method according to claim 1, characterized in that, The step of determining the optimal delay time as the t that minimizes the global index function includes: Iterate through the delay times from 1 to 50 with a step size of 1. For each delay time value, calculate the value of the global index function corresponding to that time value; After the traversal is complete, compare all the calculated global index function values, find the delay time that makes the global index function reach its global minimum value, and determine the delay time value as the optimal delay time.

6. The method according to claim 1, characterized in that, The extraction of the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions, after standardization, together constitute the real-time state feature vector of the optical cable, including: The determined optimal delay time is used as the first component of the feature vector; The curve of correlation statistic changing with delay time is taken as the correlation statistic curve. The first and second derivatives of the curve at the optimal delay time point are calculated using the numerical second-order central difference method. The curvature is then calculated as the second component of the eigenvector according to the curvature formula. The standard deviation of the difference statistic at the optimal delay time with embedding dimensions of 2 to 5 is obtained as the third component of the feature vector; The three components are standardized and then combined to form a real-time state feature vector.

7. The method according to claim 5, characterized in that, The pre-established baseline feature distribution model of the communication optical cable under normal conditions includes a baseline mean vector and a covariance matrix. The model is used to diagnose the fault state of the communication optical cable by calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing this distance with a preset fault diagnosis threshold. This includes: 100 sets of backscattered time-domain signals of communication optical cables under normal conditions were collected, and 100 normal-state feature vectors were calculated for each set of signals. The mean value of the 100 normal state feature vectors is calculated to obtain the benchmark mean vector, and the covariance matrix is ​​calculated to form the benchmark feature distribution model. For the real-time state feature vector to be diagnosed, the square of the Mahalanobis distance between the vector and the baseline distribution is calculated according to the Mahalanobis distance formula. The fault diagnosis threshold is the critical value of the chi-square distribution with 3 degrees of freedom at a confidence level of 99%. If the squared Mahalanobis distance calculated is greater than the threshold, the communication optical cable is determined to be faulty.

8. A communication fault diagnosis device, characterized in that, Includes the following modules: The selection module is used to acquire the backscattered time-domain signal of the communication optical cable, decompose the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and select the intrinsic mode function with the largest kurtosis value as the analysis signal. A construction module is used to reconstruct the phase space of the analyzed signal. Based on the distance distribution between data points in the reconstructed phase space, a set of radii for calculating the correlation integral is determined. For multiple preset embedding dimensions and candidate values ​​of delay time, the correlation integral in the phase space is calculated using the set of radii. Based on the correlation integral, a difference statistic representing the difference in correlation integrals under different embedding dimensions and a correlation statistic representing the correlation between points in the phase space are calculated for each candidate value of delay time. Combining the dispersion of the difference statistic under multiple embedding dimensions and the value of the correlation statistic, a global index function that varies with the delay time t is constructed, and the t at which the global index function reaches its minimum value is determined as the optimal delay time. The comparison module is used to extract the optimal delay time, the curvature of the correlation statistic curve at the optimal delay time, and the standard deviation of the difference statistic at the optimal delay time under the multiple embedding dimensions. After standardization, these parameters are used to jointly construct the real-time state feature vector of the optical cable. A baseline feature distribution model of the communication optical cable under normal conditions is pre-established. The model includes a baseline mean vector and a covariance matrix. By calculating the Mahalanobis distance between the real-time state feature vector and the baseline mean vector, and comparing the distance with a preset fault diagnosis threshold, the fault state of the communication optical cable can be diagnosed.

9. The apparatus according to claim 8, characterized in that, The process of acquiring the backscattered time-domain signal of the communication optical cable, decomposing the time-domain signal into multiple intrinsic mode functions using variational mode decomposition, and selecting the intrinsic mode function with the largest kurtosis value as the analysis signal includes: The variational mode decomposition is set to 5 modes and the penalty factor is 2000. The center frequency is initialized, and the acquired backscattered time-domain signal is decomposed to obtain five intrinsic mode function components. Calculate the kurtosis values ​​of the five intrinsic mode function components, where the kurtosis value is the expected value of the fourth power of the difference between the signal data point and the signal mean, divided by the fourth power of the signal standard deviation. By comparing the five calculated kurtosis values, the intrinsic mode function component with the largest kurtosis value is determined as the analysis signal.

10. The apparatus according to claim 8, characterized in that, The step of reconstructing the phase space of the analyzed signal, and determining the set of radii for calculating the correlation integral based on the distance distribution between data points in the reconstructed phase space, includes: A preliminary phase space reconstruction is performed with an embedding dimension of 2 and a delay time of 1 to obtain a phase space data point set; For each data point in the phase space, calculate the Euclidean distance between the data point and all other data points, find the fifth nearest neighbor data point and record this fifth nearest neighbor distance, and after traversing all data points, form the k nearest neighbor distance distribution. Calculate the mean and standard deviation of the k-nearest neighbor distance distribution; The value obtained by subtracting 1.5 times the standard deviation from the mean, and the maximum value between it and 0, is used as the lower limit of the radius interval. The value obtained by adding 1.5 times the standard deviation to the mean is used as the upper limit of the radius interval. The interval is then divided into 20 equal parts to obtain twenty radius values, which constitute the radius set used to calculate the relevant integral.

Images