Mobile monitoring commissioning method and system

Abstract

The application provides a mobile monitoring and debugging method and system, relates to the technical field of optical fiber monitoring, and comprises the following steps: positioning a mobile carrier to a target monitoring point, dividing a monitoring section and setting a virtual reference point, constructing a fault feature curve based on a parameter difference atlas to locate a target fault point, constructing a digital twin model and using Monte Carlo simulation to calculate a debugging parameter sensitivity matrix, and adopting a hybrid optimization algorithm to obtain an optimal debugging parameter combination. The application can accurately locate an optical fiber fault point, improve the debugging efficiency, reduce manual intervention, reduce the maintenance cost, and prolong the service life of the optical fiber.

Description

Technical Field

[0001] This invention relates to the field of fiber optic monitoring technology, and in particular to a mobile monitoring and debugging method and system. Background Technology

[0002] Fiber optic communication networks play a crucial role in modern communication systems, with their high bandwidth, low loss, and resistance to electromagnetic interference making them the primary carrier of information transmission. As fiber optic networks continue to expand and application scenarios become increasingly complex, the monitoring, diagnosis, and troubleshooting of fiber optic faults are becoming increasingly important. Traditional fiber optic monitoring primarily relies on a combination of fixed monitoring equipment and manual inspection, using devices such as optical time-domain reflectometers to test and analyze fiber optic lines.

[0003] Currently, fiber optic monitoring and debugging technologies mainly include optical time-domain reflectometry, optical frequency-domain reflectometry, and distributed fiber optic sensing technology based on Brillouin scattering. These technologies measure the scattered and reflected signals in the fiber optic cable, analyze the time, frequency, and intensity characteristics of the optical signal, and thus locate and diagnose fault points in the fiber optic cable. Meanwhile, with the development of data processing technology, some intelligent algorithms and data analysis methods have also been applied to fiber optic fault diagnosis, improving the accuracy of fault identification. Summary of the Invention

[0004] The embodiments of the present invention provide a mobile monitoring and debugging method and system, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a mobile monitoring and debugging method, comprising:

[0006] Receive monitoring instructions and control the mobile carrier to move to the target monitoring point of the optical fiber to be monitored;

[0007] Based on the length and laying path of the optical fiber to be monitored, the optical fiber to be monitored is divided into multiple monitoring segments and virtual reference points are set; the optical signal parameters of each monitoring segment are collected and compared with the reference parameters of the corresponding virtual reference points to generate a parameter difference map;

[0008] Based on the parameter difference map, a fault feature curve is constructed. Anomalies are located by curve fitting, and the fault probability value of each anomaly is calculated to determine the target fault point.

[0009] Based on the physical and optical parameters of the target fault point, a digital twin model of the target fault point is constructed. Multiple sets of debugging parameter combinations are executed in parallel on the digital twin model. The debugging parameter sensitivity matrix of the target fault point is calculated using a Monte Carlo simulation algorithm.

[0010] Based on the sensitivity matrix of the debugging parameters, an objective function is constructed, and a hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until the preset performance threshold of the target fault point is reached, thus obtaining the optimal combination of debugging parameters.

[0011] Generate a monitoring and debugging report that includes fault diagnosis results and parameter debugging process.

[0012] Based on the length and laying path of the optical fiber to be monitored, the optical fiber is divided into multiple monitoring segments and virtual reference points are set; the optical signal parameters of each monitoring segment are collected and compared with the reference parameters of the corresponding virtual reference points to generate a parameter difference map, including:

[0013] Obtain the laying path and end-to-end transmission distance of the fiber to be monitored, calculate the critical transmission length based on the attenuation coefficient and dispersion characteristics of the fiber to be monitored, and divide the fiber to be monitored into multiple monitoring segments starting from the beginning of the fiber, with the critical transmission length as the reference.

[0014] Probe light signals are sequentially injected into each monitoring section along the laying path. Backscattered light generated during the transmission of the probe light signals is received. The energy attenuation curve of the backscattered light is calculated. The position with the minimum energy attenuation in each monitoring section is selected as a virtual reference point. The optical power value, wavelength value and signal-to-noise ratio value of the probe light signal at the virtual reference point are extracted as the reference parameters of each monitoring section.

[0015] Optical signal parameters in each monitoring segment are collected, and the deviation values ​​from the corresponding reference parameters are calculated to generate optical power deviation curves, wavelength deviation curves, and signal-to-noise ratio deviation curves. These curves are then superimposed in the time domain and fused in the frequency domain to obtain the parameter difference map of the monitoring segment, and finally, the parameter difference map of the entire monitoring segment is generated.

[0016] Based on the parameter difference map, a fault characteristic curve is constructed. Anomalies are located by curve fitting, and the fault probability value of each anomaly is calculated to determine the target fault point, including:

[0017] A continuous wavelet transform is performed on the parameter difference spectrum to generate a wavelet power spectrum, which is then integrated along the time axis to obtain a time-frequency energy distribution map. The boundary trajectory of the energy accumulation region is extracted, and the projection of the boundary trajectory in the time domain is used as the fault feature curve. Piecewise cubic Hermite interpolation is used to fit the fault feature curve, and the curvature of each point on the curve is calculated. Points where curvature changes abruptly are marked as anomalies.

[0018] Random probability sampling is performed on each anomaly point to obtain an initial fault probability distribution. The initial fault probability distribution is smoothed based on kernel density estimation, and a maximum likelihood estimation function is constructed. The likelihood estimation function value of each anomaly point is calculated as the fault probability value, and the anomaly point with the highest fault probability value is selected as the target fault point.

[0019] Based on the physical and optical parameters of the target fault point, a digital twin model of the target fault point is constructed. Multiple sets of debugging parameter combinations are executed in parallel on the digital twin model. Using a Monte Carlo simulation algorithm, the sensitivity matrix of the debugging parameters of the target fault point is calculated, including:

[0020] The physical and optical parameters of the target fault point are obtained, and basis function groups are set respectively. By dynamically adjusting the weight coefficients of each basis function, a mapping relationship between the physical and optical parameters is established. Based on the mapping relationship, the spatiotemporal evolution characteristics of the physical and optical parameters are mapped to the digital twin space to generate a digital twin model of the target fault point.

[0021] Within the parameter space of the digital twin model, multiple sets of debugging parameter combinations are generated based on each historical fault case, and computing resources are allocated according to the computational complexity of each debugging parameter combination to generate a parallel execution task queue.

[0022] Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters. The debugging parameters of each random sampling point are perturbed by a preset positive and negative step size to generate multiple sets of simulation samples. The change in performance index corresponding to each simulation sample is recorded. The parameter sensitivity is calculated based on the change in performance index to obtain the debugging parameter sensitivity matrix.

[0023] Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters. The debugging parameters of each random sampling point are perturbed by a preset positive and negative step size to generate multiple sets of simulation samples. The changes in performance indicators corresponding to each simulation sample are recorded, including:

[0024] Multiple combinations of debugging parameters for the parallel execution task are obtained and reconstructed into parameter vectors. The mean function of the parameter vectors is calculated as prior knowledge, and a kernel function is generated based on the Euclidean distance between the parameter vectors. The prior knowledge and the kernel function are input into a Gaussian process regression model to obtain the probability distribution of the parameter vectors.

[0025] Within the multiple combinations of debugging parameters, a preset number of random sampling points are generated using the Latin hypercube sampling method. The random sampling points are then input into the probability distribution to obtain the initial performance index of each random sampling point. Based on the initial performance index of the random sampling points, a covariance matrix is ​​calculated. A parameter perturbation step size distribution is generated according to the covariance matrix. The joint probability density of the parameter perturbation step size distribution and the current parameter state is calculated. The perturbation acceptance probability is determined based on the joint probability density.

[0026] Random sampling points with a perturbation acceptance probability greater than a preset acceptance probability threshold are subjected to parameter perturbation in both the positive and negative directions to generate multiple sets of simulation samples. The change in the performance index of the simulation samples is calculated by combining the initial performance index of each random sampling point.

[0027] Based on the sensitivity matrix of the debugging parameters, an objective function is constructed. A hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until a preset performance threshold for the target fault point is reached, thus obtaining the optimal combination of debugging parameters, including:

[0028] Based on the sensitivity of each debugging parameter, the corresponding performance indicators are weighted and combined to obtain the objective function;

[0029] Singular value decomposition is performed on the debugging parameter sensitivity matrix to calculate the parameter weight vector; based on the parameter weight vector, the parameter gradient vector of the objective function is calculated, the parameter gradient vector is projected onto the main feature space of the debugging parameter sensitivity matrix to obtain the main search direction, and the learning rate is dynamically adjusted to iteratively optimize each debugging parameter along the main search direction. When the improvement rate of the objective function is less than a preset threshold after multiple consecutive iterations, an orthogonal search direction group is constructed based on the debugging parameter combination of the current iteration.

[0030] Calculate the spectral radius of the sensitivity matrix of the debugging parameters and use it as the initial temperature. At each temperature level, generate the disturbance direction according to the orthogonal search direction group and generate the disturbance amplitude according to the current temperature. Combine the cooling coefficient of each debugging parameter for iterative optimization and calculate the objective function value of each debugging parameter combination. When the objective function value reaches the preset performance threshold of the target fault point, output the debugging parameter combination of the current generation as the optimal debugging parameter combination.

[0031] A second aspect of the present invention provides a portable monitoring and debugging system, comprising:

[0032] The first unit is used to receive monitoring instructions and control the mobile carrier to move to the target monitoring point of the optical fiber to be monitored.

[0033] The second unit is used to divide the fiber to be monitored into multiple monitoring segments and set virtual reference points based on the length and laying path of the fiber to be monitored; to collect the optical signal parameters of each monitoring segment and compare them with the reference parameters of the corresponding virtual reference points to generate a parameter difference map.

[0034] The third unit is used to construct a fault feature curve based on the parameter difference map, locate abnormal points through curve fitting, calculate the fault probability value of each abnormal point, and determine the target fault point.

[0035] The fourth unit is used to construct a digital twin model of the target fault point based on the physical and optical parameters of the target fault point, execute multiple sets of debugging parameter combinations in parallel on the digital twin model, and calculate the debugging parameter sensitivity matrix of the target fault point through Monte Carlo simulation algorithm.

[0036] The fifth unit is used to construct an objective function based on the sensitivity matrix of the debugging parameters, and to optimize the combination of debugging parameters using a hybrid optimization algorithm that combines gradient descent and simulated annealing until the preset performance threshold of the target fault point is reached, thereby obtaining the optimal combination of debugging parameters.

[0037] The sixth unit is used to generate a monitoring and debugging report that includes fault diagnosis results and parameter debugging process.

[0038] A third aspect of the present invention,

[0039] An electronic device is provided, comprising:

[0040] processor;

[0041] Memory used to store processor-executable instructions;

[0042] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0043] Fourth aspect of the embodiments of the present invention,

[0044] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0045] The beneficial effects of this application are as follows:

[0046] The mobile monitoring and debugging method provided by this invention can flexibly locate the target monitoring point through a mobile carrier, and combined with virtual reference point technology and parameter difference spectrum analysis, it can achieve accurate location and diagnosis of fiber optic faults, thereby improving the accuracy and efficiency of fault detection.

[0047] This invention uses digital twin technology to construct a virtual model of the target fault point. By executing multiple sets of debugging parameter combinations and Monte Carlo simulation algorithms in parallel, the debugging parameter sensitivity matrix is ​​calculated, avoiding the blind debugging process in traditional methods and reducing debugging costs and time.

[0048] This invention constructs an optimization objective function based on the sensitivity matrix of debugging parameters and adopts a hybrid optimization algorithm that combines gradient descent and simulated annealing. This algorithm can quickly converge to the optimal combination of debugging parameters in complex and ever-changing fiber optic environments, thereby improving the reliability and stability of the system. At the same time, the generated monitoring and debugging report facilitates technical personnel in tracking problems and accumulating knowledge. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the portable monitoring and debugging method according to an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the architecture of a debugging parameter optimization method based on a hybrid optimization algorithm. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the mobile monitoring and debugging method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0054] Receive monitoring instructions and control the mobile carrier to move to the target monitoring point of the optical fiber to be monitored;

[0055] Based on the length and laying path of the optical fiber to be monitored, the optical fiber to be monitored is divided into multiple monitoring segments and virtual reference points are set; the optical signal parameters of each monitoring segment are collected and compared with the reference parameters of the corresponding virtual reference points to generate a parameter difference map;

[0056] Based on the parameter difference map, a fault feature curve is constructed. Anomalies are located by curve fitting, and the fault probability value of each anomaly is calculated to determine the target fault point.

[0057] Based on the physical and optical parameters of the target fault point, a digital twin model of the target fault point is constructed. Multiple sets of debugging parameter combinations are executed in parallel on the digital twin model. The debugging parameter sensitivity matrix of the target fault point is calculated using a Monte Carlo simulation algorithm.

[0058] Based on the sensitivity matrix of the debugging parameters, an objective function is constructed, and a hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until the preset performance threshold of the target fault point is reached, thus obtaining the optimal combination of debugging parameters.

[0059] Generate a monitoring and debugging report that includes fault diagnosis results and parameter debugging process.

[0060] In one optional implementation, based on the length and laying path of the optical fiber to be monitored, the optical fiber is divided into multiple monitoring segments and virtual reference points are set; optical signal parameters of each monitoring segment are collected and compared with the reference parameters of the corresponding virtual reference points to generate a parameter difference map, including:

[0061] Obtain the laying path and end-to-end transmission distance of the fiber to be monitored, calculate the critical transmission length based on the attenuation coefficient and dispersion characteristics of the fiber to be monitored, and divide the fiber to be monitored into multiple monitoring segments starting from the beginning of the fiber, with the critical transmission length as the reference.

[0062] Probe light signals are sequentially injected into each monitoring section along the laying path. Backscattered light generated during the transmission of the probe light signals is received. The energy attenuation curve of the backscattered light is calculated. The position with the minimum energy attenuation in each monitoring section is selected as a virtual reference point. The optical power value, wavelength value and signal-to-noise ratio value of the probe light signal at the virtual reference point are extracted as the reference parameters of each monitoring section.

[0063] Optical signal parameters in each monitoring segment are collected, and the deviation values ​​from the corresponding reference parameters are calculated to generate optical power deviation curves, wavelength deviation curves, and signal-to-noise ratio deviation curves. These curves are then superimposed in the time domain and fused in the frequency domain to obtain the parameter difference map of the monitoring segment, and finally, the parameter difference map of the entire monitoring segment is generated.

[0064] In practical applications, the mobile carrier can adopt a four-wheel drive vehicle structure, equipped with a GPS positioning module, gyroscope and lidar obstacle avoidance system. It navigates to the target location through a preset path planning algorithm. After the mobile carrier arrives at the target monitoring point, it fixes its position through an automatic locking mechanism to ensure stability during the monitoring process.

[0065] The optical fiber to be monitored can be a communication optical cable laid in urban areas, with a total length of 50 kilometers, an attenuation coefficient of 0.25 dB / km, and a dispersion coefficient of 17 ps / (nm·km). Based on the attenuation and dispersion characteristics of the optical fiber, the critical transmission length is calculated to be 10 kilometers. Starting from the beginning of the optical fiber, the 50-kilometer optical fiber is divided into 5 monitoring segments: 0-10 kilometers, 10-20 kilometers, 20-30 kilometers, 30-40 kilometers, and 40-50 kilometers.

[0066] For each monitoring segment, a 1550nm wavelength optical pulse with a pulse width of 10ns and a power of 100mW was used as the probe optical signal. The probe optical signal was injected through the fiber optic input and transmitted along the laying path. During transmission, the optical signal generated backscattered light, including Rayleigh scattering, Brillouin scattering, and Raman scattering. These backscattered lights were received by an optical time-domain reflectometer (OTDR), and the change curve of optical power over time was recorded. The propagation speed of light in the optical fiber (approximately 2 × 10⁻⁶ mW) was then used to analyze the backscattered light. 8 The relationship between distance and time is calculated using m / s, and the time-domain data is converted into spatial-domain data to obtain the energy decay curve of the backscattered light.

[0067] Analyze the energy attenuation curve of each monitoring segment to determine the location with the minimum energy attenuation as a virtual reference point. For example, the virtual reference point for the first monitoring segment (0-10 km) is located at 5.2 km, with an optical power of -28.5 dBm, a wavelength of 1550.02 nm, and a signal-to-noise ratio of 35 dB; the virtual reference point for the second monitoring segment (10-20 km) is located at 15.7 km, with an optical power of -32.3 dBm, a wavelength of 1550.05 nm, and a signal-to-noise ratio of 32 dB; and so on to determine the virtual reference points and corresponding benchmark parameters for other monitoring segments.

[0068] After establishing the virtual reference point baseline parameters, the fiber optic parameter monitoring phase begins. Distributed fiber optic sensing technology is employed, collecting optical signal parameters at 100-meter intervals within each monitoring segment. The fiber optic sensing equipment used includes an optical power meter, a spectrum analyzer, and an optical time-domain reflectometer. The optical power measurement accuracy is ±0.1 dBm, the wavelength measurement accuracy is ±0.01 nm, and the signal-to-noise ratio measurement accuracy is ±0.5 dB.

[0069] Taking the first monitoring segment as an example, starting from kilometer 0, a measurement point is collected every 100 meters, resulting in 100 measurement points acquiring optical power, wavelength, and signal-to-noise ratio (SNR) values. These measured values ​​are compared with the baseline parameters of the virtual reference point for this monitoring segment to calculate the deviation. For example, if the optical power measured at kilometer 4.5 is -29.2 dBm, the optical power deviation is -29.2 - (-28.5) = -0.7 dBm; the wavelength is 1550.08 nm, the wavelength deviation is 1550.08 - 1550.02 = 0.06 nm; and the SNR is 33.5 dB, the SNR deviation is 33.5 - 35 = -1.5 dB. The deviation values ​​of these three parameters for all measurement points within the monitoring segment are calculated in this way, and optical power deviation curves, wavelength deviation curves, and SNR deviation curves are plotted respectively.

[0070] The three deviation curves obtained are superimposed in the time domain and fused in the frequency domain. The time domain superposition is to superimpose the three deviation curves in the time dimension with weighted coefficients of 0.4 for optical power deviation, 0.3 for wavelength deviation, and 0.3 for signal-to-noise ratio deviation. The frequency domain fusion is to convert the superimposed time domain signal to the frequency domain, extract the characteristic frequency components through spectrum analysis, remove high-frequency noise interference, and then convert it back to the time domain to form a smoother and more information-rich difference map.

[0071] This processing method yields a parameter difference map of the first monitoring segment. In this map, the x-axis represents the fiber distance (0-10 km), and the y-axis represents the normalized parameter deviation value. Under normal circumstances, the parameter deviation value should be close to zero. When there are abnormalities such as bending, pressure, or microcracks in the fiber at a certain location, the parameter deviation at the corresponding location will fluctuate significantly. For example, at 7.3 km of the first monitoring segment, the parameter difference map shows a sudden increase in the deviation value to 0.85, indicating that there is bending or pressure on the fiber at that location.

[0072] Repeat the above process for all five monitoring segments to generate parameter difference maps for each segment. Then, stitch these maps together to form a complete parameter difference map covering the entire fiber optic length. During the stitching process, it is necessary to eliminate discontinuities at the boundaries of the monitoring segments and use linear interpolation to smooth the data in the transition area.

[0073] The obtained parameter difference map of the entire monitoring section is presented in the form of a color heatmap, where the color intensity represents the magnitude of the deviation. Areas with larger deviation values ​​are marked in red to alert management personnel. This parameter difference map can visualize the health status of the entire optical fiber, which helps to quickly locate potential fault points and improve the maintenance efficiency and reliability of the optical fiber communication network.

[0074] In one optional implementation, a fault characteristic curve is constructed based on the parameter difference map, anomaly points are located through curve fitting, and the fault probability value of each anomaly point is calculated to determine the target fault point, including:

[0075] A continuous wavelet transform is performed on the parameter difference spectrum to generate a wavelet power spectrum, which is then integrated along the time axis to obtain a time-frequency energy distribution map. The boundary trajectory of the energy accumulation region is extracted, and the projection of the boundary trajectory in the time domain is used as the fault feature curve. Piecewise cubic Hermite interpolation is used to fit the fault feature curve, and the curvature of each point on the curve is calculated. Points where curvature changes abruptly are marked as anomalies.

[0076] Random probability sampling is performed on each anomaly point to obtain an initial fault probability distribution. The initial fault probability distribution is smoothed based on kernel density estimation, and a maximum likelihood estimation function is constructed. The likelihood estimation function value of each anomaly point is calculated as the fault probability value, and the anomaly point with the highest fault probability value is selected as the target fault point.

[0077] In this embodiment, a spectrum of differences in equipment operating parameters is required as the basis for analysis. Continuous wavelet transform processing is performed on the acquired parameter difference spectrum. Specifically, the Mexican cap wavelet is selected as the mother wavelet function, with a scale parameter range of 1 to 64 and a step size of 0.5, resulting in 32 scale levels. The scale interval adopts a logarithmically uniform distribution to balance the extraction capabilities of low-frequency and high-frequency features. Each time series point in the parameter difference spectrum is convolved with the scaled and translated wavelet function to obtain a wavelet coefficient matrix reflecting the energy distribution of the signal at different times and frequencies. For a 12-kilometer fiber optic monitoring segment, a total of 24,000 sampling points need to be processed. Performing wavelet transforms at 32 scales on each sampling point involves a large computational load, which can be accelerated using parallel computing technology. On a standard workstation, the processing time can be controlled within 3 minutes.

[0078] The wavelet power spectrum is calculated based on the wavelet coefficient matrix. The process of generating the wavelet power spectrum involves squared-modulus wavelet transform coefficients to obtain the energy distribution at different time points and frequencies (scales). The wavelet power spectrum can intuitively reflect the distribution of signal energy in the time and frequency domains, which helps in identifying fault characteristics. In practice, the wavelet power spectrum is presented as a pseudo-color image, with the horizontal axis representing distance (0-12000 meters), the vertical axis representing scale (1-64), and color depth representing energy magnitude. A 64-level color scale is typically used. For healthy fiber segments, the wavelet power spectrum exhibits a uniform energy distribution; however, energy accumulation occurs near fault points, manifesting as bright areas on the power spectrum. For example, a micro-bend fault 6453 meters from the fiber's origin will produce a significant energy accumulation area at the corresponding location on the wavelet power spectrum, typically concentrated between scales 8 and 16.

[0079] Integrating along the time axis involves accumulating the wavelet power spectrum along the frequency (scale) dimension to obtain a one-dimensional energy distribution curve. The integration process employs the trapezoidal integration method, weighting and accumulating the power values ​​at different scales for each time point. In practice, integration can be selectively performed within specific scale ranges based on the characteristics of different fault types to enhance the identification effect of specific fault types. For example, for micro-bending faults, integration is mainly performed within the scale range of 8-16; for connector loss faults, integration is performed within the scale range of 4-12. The integration result forms a time-frequency energy distribution map, with the horizontal axis representing distance and the vertical axis representing the integrated energy value. The energy distribution of normal fiber segments is relatively flat, with fluctuations typically not exceeding 20% ​​of the background energy; while at fault points, significant energy peaks appear, typically 2-5 times the background energy.

[0080] An adaptive threshold segmentation algorithm is used to extract the boundary trajectory of the energy accumulation region. This algorithm calculates the global energy mean and standard deviation of the time-frequency energy distribution map. The mean plus twice the standard deviation is used as the initial threshold. Regions with energy values ​​exceeding the threshold are marked as energy accumulation regions. Morphological operations (including opening and closing operations) are used to remove noise points and connection breaks, smoothing the boundaries. The structuring element size of the morphological operations is set to 3×3 pixels. The boundary extraction uses a contour tracking algorithm, starting from the left edge of the energy accumulation region and tracking the region boundary in a clockwise direction to obtain a complete set of boundary points. In practical applications, a typical boundary of a micro-bending fault energy accumulation region contains 50-100 boundary points, and the coordinates of these points constitute the boundary trajectory.

[0081] Using the projection of the boundary trajectory in the time domain as the fault characteristic curve is a key step in simplifying the two-dimensional boundary trajectory into a one-dimensional feature representation. During the projection process, for each point on the boundary trajectory, its time (distance) coordinates and corresponding energy values ​​are retained, forming a distance-energy two-dimensional relationship curve. To reduce the influence of noise, a mean-mode filter can be applied to the projection curve, with the window size set to 5 sample points. The fault characteristic curve intuitively reflects the distribution of fault points along the optical fiber. The peak value of the curve corresponds to the fault location, and the peak amplitude reflects the severity of the fault. For example, for a micro-bending fault located at 6453 meters, its characteristic curve shows a significant peak value near this location, with the peak energy being 3.5 times the background energy.

[0082] The process of fitting the fault characteristic curve using piecewise cubic Hermitian interpolation considers both the continuity and smoothness of the curve. Hermitian interpolation not only ensures that the fitted curve passes through all sampling points but also maintains the continuity of the curve's derivative, making it suitable for describing the gradual characteristics of fiber optic faults. The piecewise processing method divides the characteristic curve into multiple sub-intervals, each of which undergoes independent cubic Hermitian interpolation. The length of each sub-interval is typically set to 100-200 meters, and derivative matching technology is used at the interval boundaries to ensure a smooth connection of the overall curve. The sampling interval of the interpolation points is set to 0.1 meters, significantly higher than the sampling density of the original data (0.5 meters), providing a more refined curve description. During the fitting process, for each interpolation point, its function value and first derivative value are calculated to generate a high-precision fitted curve. For a micro-bending fault at 6453 meters, the fitted curve can accurately reproduce the fault characteristics, with the fitting error controlled within 3% of the original data amplitude.

[0083] Calculating the curvature at each point on the curve requires utilizing the first and second derivatives of the fitted curve. For the fitted fault characteristic curve, the first and second derivatives are calculated at 0.1-meter intervals for each point. The derivative calculation employs the central difference method, which offers high numerical accuracy. Curvature calculation, based on the relationship between the first and second derivatives, represents the degree of curvature at that point. To reduce fluctuations in numerical calculations, a Gaussian smoothing filter is applied to the calculated curvature sequence, with a filter window width of 5 sample points. The unit of curvature value is 1 / meter, and the numerical range is typically between 0.001 and 10. A larger curvature indicates a higher degree of curve curvature and more drastic changes in the corresponding fault characteristics. The curvature value of a normal fiber segment is usually below 0.05, while the curvature value at the fault point can range from 0.5 to 5, depending on the fault type and severity.

[0084] An adaptive threshold detection method was used to mark curvature abrupt changes as anomalies. The mean and standard deviation of the curvature sequence were calculated, and the mean plus three times the standard deviation was used as the threshold for curvature abrupt changes. Points with curvature values ​​exceeding the threshold were marked as candidate anomalies. To avoid noise interference, adjacent candidate anomalies were clustered. Only when three or more consecutive sample points were marked as candidate anomalies were they confirmed as true anomalies. For each confirmed anomaly, its location information (distance) and curvature value were recorded as the basis for fault judgment. The rate of change of curvature near the anomaly was also calculated as an indicator of fault severity. For example, a micro-bending fault detected at 6453 meters had a peak curvature of 2.3, which was much higher than the background curvature (0.03), and the rate of change of curvature reached 15 / meter, indicating a significant fiber optic anomaly at that location.

[0085] Random probability sampling is performed on each anomaly point using a stratified Monte Carlo sampling method. Based on the anomaly point set obtained from the aforementioned curvature analysis, each anomaly point is assigned an initial weight, which is proportional to the curvature value of the anomaly point. In practice, the curvature values ​​are normalized to ensure they fall within the range of 0 to 1. The normalization method involves dividing the curvature value of each anomaly point by the maximum curvature value. For example, in a 12-kilometer-long fiber optic monitoring segment, 15 anomalies are detected. The maximum curvature value is 2.87, located 6453 meters from the starting point, so the normalized curvature of this point is 1.0. The curvature of the anomaly point located at 8721 meters is 0.93, which is normalized to 0.324. Based on the normalized curvature value, a sampling probability distribution function is constructed, and random sampling is performed. The number of samplings is related to the number of outliers, and is usually set to 3 to 5 times the number of outliers. In this example, it is 45 times. The sampling results record the frequency of each outlier being selected, forming an initial fault probability distribution. For example, the outlier located at 6453 meters was selected 18 times in 45 samplings, with an initial probability of 0.4; while the outlier located at 8721 meters was selected 8 times, with an initial probability of 0.178.

[0086] The initial fault probability distribution is smoothed based on kernel density estimation. A Gaussian kernel function is used for kernel density estimation, and the kernel width parameter has a decisive impact on the smoothing effect. A kernel width that is too small will lead to overfitting and retain too much noise; a kernel width that is too large will result in over-smoothing and loss of effective information. In practical applications, the kernel width is determined using cross-validation, with a typical value between 0.5% and 2% of the fiber length. For a 12-kilometer fiber, a kernel width of 120 meters yields the best results. During the smoothing process, for each outlier, its spatial distance to all other outliers is calculated, and a Gaussian kernel function is applied to perform a weighted average to obtain the smoothed probability value. Before smoothing, the initial probability of the outlier at 6453 meters is 0.4, and the probability of the adjacent outlier at 6381 meters is 0.067. After smoothing, the probability at 6453 meters is adjusted to 0.385, and the probability at 6381 meters is adjusted to 0.092, making the probability distribution more consistent with the spatial continuity characteristics of the fault. Smoothing can also effectively fill the probability gaps between outliers, forming a continuous probability distribution curve, which is convenient for subsequent maximum likelihood estimation.

[0087] Maximum likelihood estimation is based on a known fiber optic fault characteristic model, combined with a smoothed probability distribution, to calculate the probability that each anomaly is a true fault point. When constructing the likelihood function, multiple fault characteristic parameters are considered, including the anomaly's curvature value, rate of change of curvature, energy concentration, and similarity to historical fault characteristics. Each characteristic parameter is assigned a weight coefficient, which is determined through optimization using the training dataset. In practical applications, the curvature value typically has a weight of 0.35, the rate of change of curvature 0.25, the energy concentration 0.3, and the historical similarity 0.1. For the anomaly at 6453 meters in this example, its curvature value is 2.87, the rate of change of curvature is 15 / meter, the energy concentration is 0.78, and the similarity to historical micro-bending fault characteristics is 0.85; while the corresponding values ​​for the anomaly at 8721 meters are 0.93, 6.2 / meter, 0.42, and 0.56, respectively. The likelihood function value of the outlier at 6453 meters was 0.876, which is significantly higher than the 0.412 at 8721 meters.

[0088] An iterative optimization method is used to calculate the likelihood estimation function value of each anomaly point as the fault probability value. The initial iteration value is based on a smoothed probability distribution. The calculation is performed through a maximum of 10 iterations until the probability value change between two adjacent iterations is less than a preset threshold (usually 0.001) or the maximum number of iterations is reached. In each iteration, the likelihood function value is calculated based on the current probability distribution, and the probability distribution is updated according to the likelihood function value. For multi-fault-point situations in complex environments, the expectation-maximization algorithm is used for probability calculation, which can effectively distinguish the impact of multiple fault points. In practical applications, parallel processing technology is used for iterative calculations. For an optical fiber segment containing 15 anomalies, the calculation time is controlled within 200 milliseconds. After iteration, the final fault probability value of the anomaly point at 6453 meters is 0.923, which is much higher than other anomalies, such as 0.337 at 8721 meters and other lower probability values ​​at other locations.

[0089] A fault determination threshold of 0.7 is set. A fault point is only confirmed when the highest probability value exceeds this threshold. Probability values ​​below the threshold are sometimes false anomalies caused by measurement noise or environmental interference; low-confidence anomalies are automatically filtered out. To improve positioning accuracy, the confidence interval for the fault point is calculated, typically using a 95% confidence level. This corresponds to a spatial range approximately 1.96 times the kernel width. In this example, the anomaly at 6453 meters is identified as the target fault point, with a fault probability of 0.923 and a 95% confidence interval of 6335 meters to 6571 meters, indicating a 92.3% probability of a micro-bending fault within this range. The fault type, severity, and discovery time are recorded to form a complete fault record. Fault severity is assessed based on a combination of probability and curvature values, categorized into three levels: mild (0.7-0.8), moderate (0.8-0.9), and severe (0.9-1.0). The fault in this example belongs to the severe level and requires priority handling.

[0090] Furthermore, for multi-fault scenarios, the system can identify multiple local maxima in the probability distribution. When the distance between two adjacent local maxima exceeds a preset minimum fault interval (typically 5% of the monitored segment length) and both probability values ​​exceed a threshold, multiple target fault points are simultaneously marked. For example, in another test case, two fault points were simultaneously detected in a 10-kilometer fiber optic cable, located at 2357 meters (probability value 0.883) and 7625 meters (probability value 0.812), corresponding to a joint loss fault and a micro-bending fault, respectively. By comparing the similarity between fault characteristics and the historical fault database, a preliminary fault type determination can also be made, providing maintenance personnel with more comprehensive fault information. The fault location results are presented through a visual interface, including fault location markings, probability distribution heatmaps, and descriptions of fault type and severity, facilitating rapid response and handling by maintenance personnel.

[0091] In one optional implementation, a digital twin model of the target fault point is constructed based on its physical and optical parameters. Multiple sets of debugging parameter combinations are executed in parallel on the digital twin model. A Monte Carlo simulation algorithm is used to calculate the debugging parameter sensitivity matrix of the target fault point, including:

[0092] The physical and optical parameters of the target fault point are obtained, and basis function groups are set respectively. By dynamically adjusting the weight coefficients of each basis function, a mapping relationship between the physical and optical parameters is established. Based on the mapping relationship, the spatiotemporal evolution characteristics of the physical and optical parameters are mapped to the digital twin space to generate a digital twin model of the target fault point.

[0093] Within the parameter space of the digital twin model, multiple sets of debugging parameter combinations are generated based on each historical fault case, and computing resources are allocated according to the computational complexity of each debugging parameter combination to generate a parallel execution task queue.

[0094] Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters. The debugging parameters of each random sampling point are perturbed by a preset positive and negative step size to generate multiple sets of simulation samples. The change in performance index corresponding to each simulation sample is recorded. The parameter sensitivity is calculated based on the change in performance index to obtain the debugging parameter sensitivity matrix.

[0095] When acquiring the physical and optical parameters of the target fault point, physical parameters include temperature (e.g., 350K), pressure (e.g., 3MPa), humidity (e.g., 45%), and vibration frequency (e.g., 68Hz); optical parameters include light intensity (e.g., 80mW / cm²), wavelength (e.g., 632.8nm), polarization angle (e.g., 30°), and phase (e.g., 45°). Basis function sets are set for both physical and optical parameters. The basis function sets for physical parameters can be polynomial basis functions, Fourier basis functions, wavelet basis functions, etc.; the basis function sets for optical parameters can be Hermite polynomials, Laguerre polynomials, etc. For example, a third-order polynomial basis function {1, T, T} can be used for the temperature parameter. 2 , T 3 For the light intensity parameters, the Hermite polynomial {H0(I), H1(I), H2(I)} is used.

[0096] When establishing a mapping relationship by dynamically adjusting the weight coefficients of each basis function, an adaptive weight optimization algorithm is used. Each basis function is assigned an initial weight value (e.g., 0.5). The error between the predicted and actual values ​​is calculated based on historical data. The weight coefficients are iteratively adjusted using gradient descent until the error is less than a preset threshold (e.g., 0.01). For example, in the mapping relationship between temperature and light intensity, the basis functions for temperature are {1, T, T...}. 2 , T 3 The corresponding weight coefficients are adjusted to {0.2, 0.4, 0.25, 0.15}, and the weights corresponding to the basis functions of light intensity {H0(I), H1(I), H2(I)} are adjusted to {0.35, 0.45, 0.2}.

[0097] Based on the established mapping relationship, the spatiotemporal evolution characteristics of physical and optical parameters are mapped to the digital twin space. The spatiotemporal evolution characteristics include the trend of parameter changes over time (e.g., temperature rise rate of 0.5 K / min) and spatial distribution characteristics (e.g., temperature field gradient distribution of 5 K / m). In specific implementation, a spatiotemporal embedding algorithm is used to record the values ​​of each parameter in each time step (e.g., 1 ms) to construct a four-dimensional spatiotemporal matrix. This matrix is ​​then transformed into a standardized representation in the digital twin space through the mapping relationship to generate a digital twin model of the target fault point. In the digital twin model, each parameter is represented as a function of time and space. For example, temperature can be represented as T(x,y,z,t), and light intensity can be represented as I(x,y,z,t).

[0098] When generating multiple sets of debugging parameter combinations based on historical fault cases within the parameter space of the digital twin model, cases with a similarity higher than 80% to the target fault point are selected from the fault case library (e.g., 15 historical cases are selected). Based on these cases, the effective value range of the debugging parameters is extracted, such as the current adjustment range (0.5A-2.0A), frequency adjustment range (50Hz-200Hz), and optical gain adjustment range (0.8-1.5). The Latin hypercube sampling method is then used to generate multiple sets of parameter combinations within the parameter space (e.g., 100 sets of parameter combinations).

[0099] When allocating computing resources based on the computational complexity of each combination of debugging parameters, the computational complexity is determined by factors such as parameter dimensionality and simulation accuracy. For example, a set of parameter combinations containing 10 parameters and requiring high-precision simulation has a complexity score of 85; while another set containing 5 parameters and requiring low-precision simulation has a complexity score of 35. Based on the complexity scores, a dynamic resource allocation algorithm is adopted. High-complexity tasks are allocated 8 CPU cores and 16GB of memory, while low-complexity tasks are allocated 2 CPU cores and 4GB of memory. The generated parallel execution task queue is sorted from high to low complexity, and tasks with similar complexity (difference less than 10%) are grouped into batches for simultaneous execution.

[0100] When performing Monte Carlo simulations on parallel tasks, a preset number of sampling points are generated within each combination of debug parameters (e.g., 500 sampling points per parameter group). For each sampling point, a parameter perturbation method is used to perturb the debug parameters with preset positive and negative steps. For example, for a sampling point with a current parameter value of 1.5A, a perturbation of ±0.05A is applied, generating two perturbation points at 1.45A and 1.55A respectively; for a sampling point with a frequency parameter value of 120Hz, a perturbation of ±5Hz is applied, generating two perturbation points at 115Hz and 125Hz. Simulation calculations are performed on all perturbation points, and the changes in key performance indicators (such as power efficiency, stability, response time, etc.) are recorded.

[0101] When calculating parameter sensitivity based on changes in performance indicators, a sensitivity value is calculated for each parameter. The sensitivity value is equal to the average ratio of the change in performance indicator to the parameter disturbance. For example, if the power efficiency changes from 92% to 93.5% when the current parameter changes from 1.5A to 1.55A, then the sensitivity at that point is (93.5% - 92%) / (1.55A - 1.5A) = 30% / A. Similarly, the sensitivity when the current parameter changes from 1.5A to 1.45A is calculated, and the average of the two values ​​is taken as the sensitivity of the current parameter to power efficiency at that sampling point. The sensitivity values ​​of all parameters to each performance indicator are summarized to form a parameter sensitivity matrix. In the matrix, each row represents a parameter, each column represents a performance indicator, and the matrix element value indicates the sensitivity of that parameter to that indicator.

[0102] Through the detailed steps above, sensitivity analysis of debugging parameters for target fault points based on digital twin models was achieved, providing a quantitative basis for fault diagnosis and debugging.

[0103] In one optional implementation, Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters, and the debugging parameters of each random sampling point are perturbed by positive and negative preset step sizes to generate multiple sets of simulation samples. The changes in performance indicators corresponding to each simulation sample are recorded, including:

[0104] Multiple combinations of debugging parameters for the parallel execution task are obtained and reconstructed into parameter vectors. The mean function of the parameter vectors is calculated as prior knowledge, and a kernel function is generated based on the Euclidean distance between the parameter vectors. The prior knowledge and the kernel function are input into a Gaussian process regression model to obtain the probability distribution of the parameter vectors.

[0105] Within the multiple combinations of debugging parameters, a preset number of random sampling points are generated using the Latin hypercube sampling method. The random sampling points are then input into the probability distribution to obtain the initial performance index of each random sampling point. Based on the initial performance index of the random sampling points, a covariance matrix is ​​calculated. A parameter perturbation step size distribution is generated according to the covariance matrix. The joint probability density of the parameter perturbation step size distribution and the current parameter state is calculated. The perturbation acceptance probability is determined based on the joint probability density.

[0106] Random sampling points with a perturbation acceptance probability greater than a preset acceptance probability threshold are subjected to parameter perturbation in both the positive and negative directions to generate multiple sets of simulation samples. The change in the performance index of the simulation samples is calculated by combining the initial performance index of each random sampling point.

[0107] In this specific embodiment, it is necessary to obtain multiple combinations of debugging parameters for the parallel execution task. These parameters may include key parameters affecting the execution efficiency of the parallel task, such as the number of threads, memory allocation size, and scheduling priority. Assume the obtained combinations of debugging parameters are {(4, 256MB, 1), (8, 512MB, 2), (16, 1024MB, 3)}, where each combination contains three parameters: the number of threads, memory allocation, and priority. These parameter combinations are reconstructed into parameter vector form, such as [4, 256, 1], [8, 512, 2], and [16, 1024, 3]. The mean function of these parameter vectors is calculated. For the above example, the mean vector is [9.33, 597.33, 2], which serves as prior knowledge for the Gaussian process regression model.

[0108] For kernel function generation, the kernel matrix is ​​calculated based on the Euclidean distance between parameter vectors. For example, the Euclidean distance between vectors [4, 256, 1] and [8, 512, 2] is approximately 256.06. Based on these distance values, a squared exponential kernel function is used to generate the kernel matrix, which describes the similarity between different parameter combinations. After inputting the prior knowledge and kernel function into the Gaussian process regression model, the probability distribution of the parameter vectors in the performance index space is obtained.

[0109] After determining the probability distribution, sampling is performed within the given parameter space. Assuming the parameter space consists of the number of threads [2,32], memory allocation [128MB,2048MB], and priority [1,5], a preset number (e.g., 50) of random sampling points are generated using the Latin hypercube sampling method. This method ensures that the samples are uniformly distributed in the parameter space. For each sampling point, its initial performance indicators, such as response time and throughput, are calculated using the previously established Gaussian process regression model.

[0110] Based on these initial sampling points and their performance indices, the covariance matrix is ​​calculated. For example, among the 50 sampling points, the parameter vectors of the first two points are [6, 300, 2] and [20, 800, 4], respectively, and their initial performance indices are 85ms and 120ms, respectively. The covariance between these two points can then be calculated. By analyzing the covariance relationship between all sampling points, a complete covariance matrix is ​​constructed. This matrix reflects the correlation of performance changes between different parameter combinations.

[0111] Based on the covariance matrix, a perturbation step size distribution is generated. For example, for the number of threads parameter, the determined perturbation step size is ±2 threads; for memory allocation, the perturbation step size is ±128MB; and for priority, the perturbation step size is ±1. The joint probability density of the current parameter state and the perturbation state is calculated, and the perturbation acceptance probability is determined based on this. Assuming that a certain sampling point [12, 768MB, 3] is perturbed, the acceptance probability of the positive perturbation [14, 896MB, 4] is calculated to be 0.85, and the acceptance probability of the negative perturbation [10, 640MB, 2] is 0.72.

[0112] The perturbation acceptance probability is compared with a preset threshold (e.g., 0.7). For sampling points with an acceptance probability greater than the threshold, parameter perturbation is performed. In the example above, both perturbation directions are accepted, generating corresponding simulation samples. By executing actual simulations of parallel tasks, the performance indicators of parameter combinations after each perturbation are recorded. For example, the response time of the original sampling point [12, 768MB, 3] is 100ms, which becomes 90ms after a positive perturbation and 105ms after a negative perturbation. The changes in performance indicators are -10ms and +5ms, respectively.

[0113] By analyzing the performance changes after these perturbations, the performance gradient in the parameter space is evaluated, and the parameters most sensitive to performance improvement and their optimal value ranges are identified. For example, when the number of threads is increased from 12 to 14, the performance improvement is significant; however, when it is reduced from 10 to 8, the performance decreases significantly, indicating that the optimal number of threads for the current task is around 14.

[0114] In practical applications, parameter configurations are automatically adjusted based on these analysis results. For example, for a data processing task, the optimal configuration is determined to be 16 threads, 896MB of memory allocation, and priority 4 using the above method. Compared with the initial configuration, the processing speed is improved by 23% and the resource utilization is improved by 18%.

[0115] In addition, record the resource consumption under different parameter combinations, such as CPU utilization and memory usage, and combine them with performance indicators to form a multi-dimensional evaluation system to find the best balance between performance and resource consumption. For example, although 32 threads provide the highest throughput, considering the overhead of thread context switching and the overall system load, 16 threads are a better choice.

[0116] This method, which combines Monte Carlo simulation with parameter perturbation analysis, enables efficient exploration of large-scale parameter spaces and significantly improves system performance and resource utilization efficiency.

[0117] Figure 2This is a schematic diagram of the architecture of a debugging parameter optimization method based on a hybrid optimization algorithm. In one optional implementation, an objective function is constructed based on the debugging parameter sensitivity matrix. A hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until a preset performance threshold of the target fault point is reached, thus obtaining the optimal combination of debugging parameters, including:

[0118] Based on the sensitivity of each debugging parameter, the corresponding performance indicators are weighted and combined to obtain the objective function;

[0119] Singular value decomposition is performed on the debugging parameter sensitivity matrix to calculate the parameter weight vector; based on the parameter weight vector, the parameter gradient vector of the objective function is calculated, the parameter gradient vector is projected onto the main feature space of the debugging parameter sensitivity matrix to obtain the main search direction, and the learning rate is dynamically adjusted to iteratively optimize each debugging parameter along the main search direction. When the improvement rate of the objective function is less than a preset threshold after multiple consecutive iterations, an orthogonal search direction group is constructed based on the debugging parameter combination of the current iteration.

[0120] Calculate the spectral radius of the sensitivity matrix of the debugging parameters and use it as the initial temperature. At each temperature level, generate the disturbance direction according to the orthogonal search direction group and generate the disturbance amplitude according to the current temperature. Combine the cooling coefficient of each debugging parameter for iterative optimization and calculate the objective function value of each debugging parameter combination. When the objective function value reaches the preset performance threshold of the target fault point, output the debugging parameter combination of the current generation as the optimal debugging parameter combination.

[0121] For the target fault point, relevant operational data is collected to construct a sensitivity matrix for debugging parameters. Assuming an automated production line has 5 key debugging parameters (P1-P5) and 3 performance indicators (I1-I3), the sensitivity matrix S for debugging parameters is obtained through multiple sampling tests. Each element S in this matrix... ij Indicates parameter P i For performance index I j The degree of influence. For example, in the collected data, P1 has a sensitivity of 0.82 to I1, P2 has a sensitivity of 0.35 to I1, and so on to construct a complete 5×3 sensitivity matrix.

[0122] Based on the sensitivity of each debugging parameter, an objective function F is constructed. For the three performance indicators, weight coefficients W1=0.5, W2=0.3, and W3=0.2 are set respectively to reflect the importance of each indicator. The objective function F is calculated by weighted summation of each performance indicator and its corresponding weight. In the specific implementation, the performance indicators are calculated by substituting the current parameter values ​​into the pre-established performance prediction model.

[0123] Singular value decomposition is performed on the sensitivity matrix of the debugging parameters to obtain eigenvalues ​​and eigenvectors. In this example, the calculated singular values ​​are [2.41, 1.56, 0.63], and the corresponding eigenvectors constitute the parameter weight vectors. Through these weight vectors, the parameter gradient vector G of the objective function is calculated. In the actual calculation, the components of the parameter gradient vector G are [0.42, -0.31, 0.56, 0.28, -0.17], which represent the direction and intensity of the influence of each parameter on the objective function.

[0124] Project the parameter gradient vector G onto the principal feature space of the sensitivity matrix to obtain the main search direction D. Based on the singular value decomposition results, select the two eigenvectors with a contribution rate exceeding 85% to form the principal feature space. After projection, the main search direction D is obtained as [0.38, -0.28, 0.52, 0.25, -0.15].

[0125] During gradient descent, a dynamic learning rate adjustment mechanism is employed. The initial learning rate is set to 0.1, and it is adjusted after each iteration based on the improvement of the objective function. If the improvement rate of the objective function is less than 0.01 for three consecutive iterations, the current search direction is considered to be close to a local optimum, and the search strategy needs to be changed. At this point, based on the current parameter combination P=[75.3, 142.6, 28.9, 56.7, 93.2], an orthogonal search direction group O is constructed. This orthogonal search direction group contains 5 vectors, each corresponding to the direction of change of a parameter, and the vectors are mutually orthogonal.

[0126] During the simulated annealing stage, the spectral radius of the sensitivity matrix was calculated as the initial temperature T0 = 2.41. A cooling coefficient α = 0.95 was set, and different cooling coefficients αi were assigned to each parameter to reflect their different characteristics. The cooling coefficients for P1, P2, P3, P4, and P5 were 0.94, 0.96, 0.93, 0.95, and 0.97, respectively. Multiple perturbation searches were performed at each temperature level.

[0127] In each perturbation search, a direction is randomly selected from the orthogonal search direction group O as the perturbation direction. The perturbation amplitude A is proportional to the current temperature T and is calculated as A=βT, where β is the perturbation factor, set to 0.5. A new parameter combination P'=P+A×O is generated, and the objective function value F' corresponding to P' is calculated. According to the Metropolis criterion, if F' is better than F, the new parameter combination is accepted; otherwise, the new combination is accepted with probability exp((F-F') / T).

[0128] At a temperature level of T=1.85, a perturbation generates a new parameter combination P'=[77.1, 141.8, 29.5, 57.3, 92.6], yielding an objective function value F'=0.862, which is an improvement over the current value F=0.889. Therefore, this combination is accepted. At a temperature level of T=1.24, a perturbation generates a combination P''=[78.3, 140.5, 30.1, 57.0, 93.5], yielding an F''=0.905. Although worse than the current value F'=0.862, this combination is still accepted with a probability of 0.32 to escape the local optimum.

[0129] The optimization process continues, with the temperature gradually decreasing according to T=αT until a preset termination condition is reached. In this example, the target performance threshold is set to 0.95. When the objective function value reaches or exceeds 0.95, the optimization process terminates. After 247 iterations, the optimal combination of debugging parameters P*=[81.5, 138.2, 32.7, 55.4, 91.8] is finally obtained, corresponding to an objective function value F*=0.967, which meets the preset performance requirements. This set of parameters was applied to the actual system debugging, successfully restoring the equipment performance to a normal level, verifying the effectiveness of the hybrid optimization algorithm.

[0130] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mobile monitoring and debugging method, characterized in that, include: Receive monitoring instructions and control the mobile carrier to move to the target monitoring point of the optical fiber to be monitored; Based on the length and laying path of the optical fiber to be monitored, the optical fiber to be monitored is divided into multiple monitoring segments and virtual reference points are set; the optical signal parameters of each monitoring segment are collected and compared with the reference parameters of the corresponding virtual reference points to generate a parameter difference map; Based on the parameter difference map, a fault feature curve is constructed. Anomalies are located by curve fitting, and the fault probability value of each anomaly is calculated to determine the target fault point. Based on the physical and optical parameters of the target fault point, a digital twin model of the target fault point is constructed. Multiple sets of debugging parameter combinations are executed in parallel on the digital twin model. Using a Monte Carlo simulation algorithm, the sensitivity matrix of the debugging parameters of the target fault point is calculated, including: The physical and optical parameters of the target fault point are obtained, and basis function groups are set respectively. An adaptive weight optimization algorithm is used to dynamically adjust the weight coefficients of each basis function to establish a mapping relationship between the physical and optical parameters. Based on the mapping relationship, the spatiotemporal evolution characteristics of the physical and optical parameters are mapped to the digital twin space to generate a digital twin model of the target fault point. Within the parameter space of the digital twin model, multiple sets of debugging parameter combinations are generated based on each historical fault case, and computing resources are allocated according to the computational complexity of each debugging parameter combination to generate a parallel execution task queue. Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters. The debugging parameters of each random sampling point are perturbed by a preset positive and negative step size to generate multiple sets of simulation samples. The change in performance index corresponding to each simulation sample is recorded. The parameter sensitivity is calculated based on the change in performance index. The parameter sensitivity is equal to the average value of the ratio of the change in performance index to the parameter perturbation amount, and the debugging parameter sensitivity matrix is ​​obtained. Based on the sensitivity matrix of the debugging parameters, an objective function is constructed, and a hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until the preset performance threshold of the target fault point is reached, thus obtaining the optimal combination of debugging parameters. Generate a monitoring and debugging report that includes fault diagnosis results and parameter debugging process.

2. The method according to claim 1, characterized in that, Based on the length and laying path of the optical fiber to be monitored, the optical fiber to be monitored is divided into multiple monitoring segments and virtual reference points are set. The optical signal parameters of each monitoring segment are collected and compared with the baseline parameters of the corresponding virtual reference point to generate a parameter difference map, including: Obtain the laying path and end-to-end transmission distance of the fiber to be monitored, calculate the critical transmission length based on the attenuation coefficient and dispersion characteristics of the fiber to be monitored, and divide the fiber to be monitored into multiple monitoring segments starting from the beginning of the fiber, with the critical transmission length as the reference. Probe light signals are sequentially injected into each monitoring section along the laying path. Backscattered light generated during the transmission of the probe light signals is received. The energy attenuation curve of the backscattered light is calculated. The position with the minimum energy attenuation in each monitoring section is selected as a virtual reference point. The optical power value, wavelength value and signal-to-noise ratio value of the probe light signal at the virtual reference point are extracted as the reference parameters of each monitoring section. Optical signal parameters in each monitoring segment are collected, and the deviation values ​​from the corresponding reference parameters are calculated to generate optical power deviation curves, wavelength deviation curves, and signal-to-noise ratio deviation curves. These curves are then superimposed in the time domain and fused in the frequency domain to obtain the parameter difference map of the monitoring segment, and finally, the parameter difference map of the entire monitoring segment is generated.

3. The method according to claim 1, characterized in that, Based on the parameter difference map, a fault characteristic curve is constructed. Anomalies are located by curve fitting, and the fault probability value of each anomaly is calculated to determine the target fault point, including: A continuous wavelet transform is performed on the parameter difference spectrum to generate a wavelet power spectrum, which is then integrated along the time axis to obtain a time-frequency energy distribution map. The boundary trajectory of the energy accumulation region is extracted, and the projection of the boundary trajectory in the time domain is used as the fault feature curve. Piecewise cubic Hermite interpolation is used to fit the fault feature curve, and the curvature of each point on the curve is calculated. Points where curvature changes abruptly are marked as anomalies. Random probability sampling is performed on each anomaly point to obtain an initial fault probability distribution. The initial fault probability distribution is smoothed based on kernel density estimation, and a maximum likelihood estimation function is constructed. The likelihood estimation function value of each anomaly point is calculated as the fault probability value, and the anomaly point with the highest fault probability value is selected as the target fault point.

4. The method according to claim 1, characterized in that, Monte Carlo simulation is performed on the parallel execution task. A preset number of random sampling points are generated within each combination of debugging parameters. The debugging parameters of each random sampling point are perturbed by a preset positive and negative step size to generate multiple sets of simulation samples. The changes in performance indicators corresponding to each simulation sample are recorded, including: Multiple combinations of debugging parameters for the parallel execution task are obtained and reconstructed into parameter vectors. The mean function of the parameter vectors is calculated as prior knowledge, and a kernel function is generated based on the Euclidean distance between the parameter vectors. The prior knowledge and the kernel function are input into a Gaussian process regression model to obtain the probability distribution of the parameter vectors. Within the multiple combinations of debugging parameters, a preset number of random sampling points are generated using the Latin hypercube sampling method. The random sampling points are then input into the probability distribution to obtain the initial performance index of each random sampling point. Based on the initial performance index of the random sampling points, a covariance matrix is ​​calculated. A parameter perturbation step size distribution is generated according to the covariance matrix. The joint probability density of the parameter perturbation step size distribution and the current parameter state is calculated. The perturbation acceptance probability is determined based on the joint probability density. Random sampling points with a perturbation acceptance probability greater than a preset acceptance probability threshold are subjected to parameter perturbation in both the positive and negative directions to generate multiple sets of simulation samples. The change in the performance index of the simulation samples is calculated by combining the initial performance index of each random sampling point.

5. The method according to claim 1, characterized in that, Based on the sensitivity matrix of the debugging parameters, an objective function is constructed. A hybrid optimization algorithm combining gradient descent and simulated annealing is used to optimize the combination of debugging parameters until a preset performance threshold for the target fault point is reached, thus obtaining the optimal combination of debugging parameters, including: Based on the sensitivity of each debugging parameter, the corresponding performance indicators are weighted and combined to obtain the objective function; Singular value decomposition is performed on the debugging parameter sensitivity matrix to calculate the parameter weight vector; based on the parameter weight vector, the parameter gradient vector of the objective function is calculated, the parameter gradient vector is projected onto the main feature space of the debugging parameter sensitivity matrix to obtain the main search direction, and the learning rate is dynamically adjusted to iteratively optimize each debugging parameter along the main search direction. When the improvement rate of the objective function is less than a preset threshold after multiple consecutive iterations, an orthogonal search direction group is constructed based on the debugging parameter combination of the current iteration. Calculate the spectral radius of the sensitivity matrix of the debugging parameters and use it as the initial temperature. At each temperature level, generate the disturbance direction according to the orthogonal search direction group and generate the disturbance amplitude according to the current temperature. Combine the cooling coefficient of each debugging parameter for iterative optimization and calculate the objective function value of each debugging parameter combination. When the objective function value reaches the preset performance threshold of the target fault point, output the debugging parameter combination of the current generation as the optimal debugging parameter combination.

6. A portable monitoring and debugging system for implementing the method as described in any one of claims 1-5, characterized in that, include: The first unit is used to receive monitoring instructions and control the mobile carrier to move to the target monitoring point of the optical fiber to be monitored. The second unit is used to divide the fiber to be monitored into multiple monitoring segments and set virtual reference points based on the length and laying path of the fiber to be monitored; to collect the optical signal parameters of each monitoring segment and compare them with the reference parameters of the corresponding virtual reference points to generate a parameter difference map. The third unit is used to construct a fault feature curve based on the parameter difference map, locate abnormal points through curve fitting, calculate the fault probability value of each abnormal point, and determine the target fault point. The fourth unit is used to construct a digital twin model of the target fault point based on the physical and optical parameters of the target fault point, execute multiple sets of debugging parameter combinations in parallel on the digital twin model, and calculate the debugging parameter sensitivity matrix of the target fault point through Monte Carlo simulation algorithm. The fifth unit is used to construct an objective function based on the sensitivity matrix of the debugging parameters, and to optimize the combination of debugging parameters using a hybrid optimization algorithm that combines gradient descent and simulated annealing until the preset performance threshold of the target fault point is reached, thereby obtaining the optimal combination of debugging parameters. The sixth unit is used to generate a monitoring and debugging report that includes fault diagnosis results and parameter debugging process.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images