Communication optical fiber multi-parameter sensing data analysis method based on multi-layer fusion modeling

Abstract

This invention relates to the field of safety monitoring technology, specifically disclosing a method for analyzing multi-parameter sensor data in optical fiber communication based on multi-layer fusion modeling. First, the invention maps the collected data to mileage and time dimensions using two-dimensional grid cells to achieve spatiotemporal uniformity. Then, for the standardized preprocessed data, cross-fitting of various data types is achieved by combining the Brillouin coupling equation, mean square error, and relative error constraints. Finally, overdetermined equations based on the Brillouin coupling equation are used for temperature-strain decoupling. Electromagnetic interference judgment and sensitivity coefficients for adaptive operating conditions are superimposed to ensure the effectiveness of decoupling. This achieves efficient and accurate multi-parameter data decoupling, thereby ensuring the effectiveness of multi-layer fusion modeling and risk assessment and early warning based on decoupled parameters and coupling characteristics. This solves the problems of spatiotemporal misalignment of data, low fault location accuracy in single-parameter analysis, and cross-sensitivity in multi-parameter analysis.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, and in particular to a method for analyzing multi-parameter sensor data of optical fiber communication based on multi-layer fusion modeling. Background Technology

[0002] Optical fiber cables are crucial infrastructure in power communication networks, carrying information for dispatch control, relay protection, production management, and status awareness. With the continuous increase in important transmission channels, inter-regional backbone lines, and sections in complex terrain, the operating environment of optical fiber cables is characterized by long distances, strong disturbances, and multiple operating conditions. Single-parameter monitoring is no longer sufficient to meet the needs of refined operation and maintenance. Relying on distributed optical fiber sensing technology, multiple status information such as temperature, strain, and vibration can be continuously acquired along the same optical cable channel, providing the conditions for the transformation of optical fiber cables from "post-event repair" to "real-time monitoring, intelligent early warning, and proactive maintenance."

[0003] Existing technologies primarily rely on Brillouin Optical Time Domain Reflectometry (BOTDR) and Brillouin optical time domain analysis based on Brillouin scattering for distributed temperature and strain measurements, Rayleigh scattering-based phase-sensitive optical time domain reflectometry for vibration detection, distributed temperature measurement based on Raman or Brillouin scattering, and distributed strain measurement based on distributed strain sensing. Domestic and international research has explored various aspects of transmission line icing, galloping, lightning strikes, partial discharge, and external disturbance monitoring, proposing multiple solutions such as combined Rayleigh and Brillouin measurements, Brillouin and Raman scattering, and multi-sensor multiplexing.

[0004] However, from the perspective of multi-parameter fusion analysis of optical fiber communication cables, existing research still has significant shortcomings: data fusion usually remains at the level of simple splicing or rule superposition, and does not adequately consider issues such as temporal misalignment, spatial mismatch, noise propagation, and feature redundancy. In addition, parameters such as temperature, strain, and vibration have cross-sensitivities and coupling effects, making it difficult for single-parameter analysis to accurately locate the source of risks.

[0005] Therefore, there is an urgent need to provide a method for analyzing multi-parameter sensing data in optical fiber communication to solve the above problems. Summary of the Invention

[0006] To address the problems existing in the prior art, this application provides a method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling, which achieves efficient and accurate temperature-strain decoupling.

[0007] In a first aspect, embodiments of this application provide a method for analyzing multi-parameter sensing data of optical fibers based on multi-layer fusion modeling, including: Two-dimensional grid cells are used to map the communication fiber multi-parameter sensing data obtained based on Brillouin scattering and Rayleigh scattering to reduce the mismatch between multi-channel data and obtain intermediate multi-parameter sensing data with unified spatiotemporal scale. Standardized preprocessing and time-series cross-fitting are performed on intermediate multi-parameter sensing data to improve the consistency and effectiveness of intermediate multi-parameter sensing data and obtain standard multi-parameter sensing data. Temperature-strain decoupling is performed on standard multi-parameter sensing data to generate decoupled non-cross-sensitive parameters. Multi-layer fusion modeling is then performed in conjunction with synchronously generated coupling features to conduct risk assessment and early warning.

[0008] Secondly, embodiments of this application provide a communication fiber optic multi-parameter sensing data analysis system based on multi-layer fusion modeling, including: The mapping processing module is used to map the communication fiber multi-parameter sensing data obtained based on Brillouin scattering and Rayleigh scattering using two-dimensional grid cells to obtain intermediate multi-parameter sensing data with unified spatiotemporal scale. The standardization processing module is used to perform standardization preprocessing and time-series cross-fitting on intermediate multi-parameter sensing data to obtain standard multi-parameter sensing data. The decoupling module is used to decouple temperature and strain from standard multi-parameter sensing data, generate decoupled non-cross-sensitive parameters, and perform multi-layer fusion modeling in combination with synchronously generated coupling features for risk assessment and early warning.

[0009] Thirdly, embodiments of this application provide a computer-readable storage medium for storing a computer program used in the method described.

[0010] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This application first maps the collected multi-parameter sensing data of communication optical cables using two-dimensional mesh cells to ensure the spatiotemporal uniformity of the data from the source. Then, through standardized preprocessing and time-series cross-fitting, based on the Brillouin coupling equation, mean square error, and relative error joint constraints, cross-validation of different types of data is achieved to obtain qualified standard multi-parameter sensing data. Furthermore, the overdetermined equation established based on the Brillouin coupling equation is used to achieve temperature-strain decoupling, ensuring the usability of decoupling. Finally, multi-layer fusion modeling and risk assessment and early warning can be performed based on the decoupled parameters and coupling characteristics, solving the problems existing in the prior art.

[0011] 2. In this application, the division of two-dimensional grid cells is based on the rule of first spatial axis and then time axis. When dividing the spatial axis cells, the bidirectional Hausdorff distance is used to measure the similarity of the data to optimize the data division. When dividing the time axis cells, the low-frequency sampling period is used as the benchmark to determine the two-dimensional grid cell to which the data belongs. This solves the problem of data misalignment caused by the inconsistency of sampling frequency and spatial point position of multi-source data. The gridded data form further ensures the effectiveness of subsequent temperature-strain decoupling.

[0012] 3. This application establishes corresponding time series fitting models based on the coefficients of variation of various types of data for the standardized preprocessed data, so as to balance fitting accuracy and computing power. Then, using different source data as samples, the results of pairwise cross-fitting of the time series fitting models are minimized through Brillouin coupling equation to obtain the fitted dataset. Finally, the degree of difference with the original data is compared to determine the similar data that need to be merged, thereby further ensuring the reliability of the data in each two-dimensional grid cell.

[0013] 4. In this application, an overdetermined equation system is constructed by superimposing the converted Rayleigh dynamic strain value with the Brillouin coupling equation, which solves the problem of underdetermined traditional single equations. Then, the gridded and time-series cross-fitted data is input into the overdetermined equation system, and the non-cross-sensitive parameters are solved by the least squares method, so as to achieve effective separation of cross-sensitive parameters. Based on the pre-electromagnetic interference judgment and the post-decoupling residual, the effectiveness of temperature-strain decoupling can be further improved, thereby ensuring the accuracy of fault tracing.

[0014] 5. The time-series cross-fitting and temperature-strain decoupling processes in this application both involve the Brillouin coupling equation. Addressing the limitations of the sensitivity coefficients in the Brillouin coupling equation, this application determines corresponding segmented intervals based on the temperature and strain values ​​at each time point, and then determines whether the corresponding data belongs to the critical interval of each segmented interval. To address the instantaneous jitter problem that can easily occur during decoupling in critical intervals, a suitable sensitivity coefficient is obtained through linear interpolation; otherwise, the sensitivity coefficient of the segmented interval is directly searched. The accuracy of the sensitivity coefficients ensures the effectiveness of the time-series cross-fitting, thereby improving the reliability and multi-condition applicability of the temperature-strain decoupling. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0016] Figure 1 A flowchart of the communication optical fiber multi-parameter sensing data analysis method based on multi-layer fusion modeling provided by the present invention; Figure 2 This is a schematic diagram of the structure of the communication optical fiber multi-parameter sensing data analysis system based on multi-layer fusion modeling provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0018] Firstly, such as Figure 1 As shown in the embodiments of this application, a method for analyzing multi-parameter sensing data of communication optical fibers based on multi-layer fusion modeling is provided, including the following steps: The first step is to acquire multi-parameter sensing data of the communication optical fiber through the sensing layer, and then process the multi-parameter sensing data of the communication optical fiber using a unified coordinate mapping mechanism to obtain intermediate multi-parameter sensing data with unified spatiotemporal scale.

[0019] In this step, the sensing layer employs a dual-mechanism synergistic scheme of Brillouin scattering and Rayleigh scattering. Brillouin scattering primarily handles distributed measurement tasks for temperature and strain. Rayleigh scattering mainly handles vibration, disturbance, and transient anomaly sensing tasks, suitable for identifying events such as external forces such as construction, impacts, dragging, and vibration shocks. Brillouin scattering and Rayleigh scattering differ in sampling rate and spatial resolution; the former has a sampling rate in the Hz range and a spatial resolution in the meter range, while the latter has a sampling rate in the kHz to MHz range and a spatial resolution in the sub-meter range.

[0020] In implementation, data based on Brillouin scattering sampling is denoted as low-frequency multi-parameter sensing data, including Brillouin temperature values ​​and Brillouin static axial strain values ​​calculated based on optical cable mileage, acquisition time, and Brillouin frequency shift. Waveform data based on Rayleigh scattering sampling is denoted as high-frequency multi-parameter sensing data, including Rayleigh dynamic vibration values ​​and Rayleigh instantaneous strain abrupt change values. In some other embodiments, it may also include Raman temperature values ​​acquired by an optical time domain reflectometer (OTDR) and fixed-point strain or temperature values ​​acquired by multiplexed external sensors. That is, communication fiber optic multi-parameter sensing data is a multi-source data set synchronously acquired through BOTDR and DAS equipment using communication optical fiber as the sensing carrier, and may also include data acquired by FBG.

[0021] The second step involves performing standardized preprocessing and time-series cross-fitting on the intermediate multi-parameter sensing data through a preprocessing layer to obtain standardized multi-parameter sensing data. The standardized preprocessing is used to improve the consistency and effectiveness of the intermediate multi-parameter sensing data.

[0022] Understandably, the perception layer only performs data partitioning and aggregation, and high-frequency dimensionality reduction and spatiotemporal alignment. Therefore, a preprocessing layer is needed to further standardize the data, including five steps: time synchronization, spatial registration, noise suppression, anomaly repair, and quality evaluation. The data after quality evaluation is then subjected to time-series cross-fitting to obtain the standard multi-parameter sensing data that meets the quality standards.

[0023] The third step is to decouple the standard multi-parameter sensing data by temperature and strain, generate three types of non-cross-sensitive parameters after decoupling, and perform multi-layer fusion modeling by combining the synchronously generated coupling features, so as to conduct risk assessment and early warning.

[0024] In implementation, this application employs two parallel verification methods: one is a physical decoupling method based on overdetermined equations; the other is a learning-based decoupling method based on neural network regression, which learns the nonlinear mapping between temperature, strain, and multi-source observations through labeled samples. The former has strong interpretability, while the latter has better fitting ability under complex working conditions, and the two can be used for mutual verification.

[0025] In a specific embodiment of the present invention, a unified coordinate mapping mechanism is used to process multi-parameter sensing data, including two aspects: firstly, data collected by different sensing devices are uniformly mapped to a two-dimensional "distance-time" grid cell; secondly, data with different sampling frequencies are synchronously windowed and then associated with data at the same time scale. This avoids the dimensional inconsistencies and noise amplification caused by directly splicing the original data, resulting in intermediate multi-parameter sensing data with a unified spatiotemporal scale. The specific process is as follows: Step 11: Using the physical laying distance of the optical cable as the spatial coordinate axis, determine continuous spatial axis cells based on the pre-set spatial grid step size; the physical mileage of data belonging to the same spatial axis cell belongs to the same mileage interval.

[0026] Step 12: Determine continuous time axis cells based on the set global time axis and fusion time window length.

[0027] For example, the physical laying distance of the optical cable is set to 0 to L (km), and the spatial grid step size is... Generally, it can be set to 1m. If it is equally spaced, then the spatial mileage interval of the kth spatial axis cell is... Set the fusion time window length. It is 2 seconds, and the starting time is If the intervals are equal, then the time interval of the nth time axis cell is... .

[0028] Step 13: Construct a two-dimensional grid cell for multi-parameter sensor data based on spatial axis cells and time axis cells. Each grid cell uniquely corresponds to a fixed mileage interval and time interval of the optical cable; Step 14: Determine the two-dimensional grid cell to which the heterogeneous sensing device belongs based on the spatial location of the heterogeneous sensing device that collects multi-parameter sensing data from the communication fiber and the time of data collection.

[0029] This step ensures that all parameters used in the temperature-strain coupling solution originate from the same spatial point, eliminating spatial misalignment issues caused by differences in device ranging. If ranging errors from heterogeneous sensing devices are further considered, these deviations need to be corrected based on actual data from the completed optical cable installation.

[0030] Step 15: Using the spatial mileage interval and time interval of the two-dimensional grid cell as the filtering conditions, traverse each two-dimensional grid cell to filter and obtain high-frequency multi-parameter sensing data. The remaining data are low-frequency multi-parameter sensing data. Step 16: Extract time-domain and frequency-domain features from the selected high-frequency multi-parameter sensing data, obtain multi-dimensional time-domain and frequency-domain features respectively, and then stitch them together. The stitched data and the low-frequency multi-parameter sensing data are recorded as intermediate multi-parameter sensing data.

[0031] For example, time-domain features include window average energy, peak amplitude, envelope upper and lower extrema, and zero-crossing energy. By performing a Fast Fourier Transform on the selected high-frequency multi-parameter sensor data, a single-sided spectrum is obtained, thereby extracting frequency-domain features, including the dominant frequency, spectral peak amplitude, and effective bandwidth cumulative energy. The time-domain and frequency-domain features are concatenated to form a high-frequency multi-parameter sensor data feature vector. For instance, with a Rayleigh scattering sampling rate of 10kHz and 20,000 original sampling points within a single 2s window, eight floating-point feature values ​​are obtained after processing, achieving a dimensionality reduction from tens of thousands to eight dimensions. This feature vector is then bound to low-frequency multi-parameter sensor data.

[0032] The above steps address the heterogeneity of multi-source data arising from the low-frequency, sparse spatiotemporal sampling characteristics of Brillouin sensors and the high-frequency, dense spatiotemporal sampling characteristics of Rayleigh sensors. They abandon the fusion method of directly stitching together the original waveforms based on Rayleigh scattering, instead constructing a two-dimensional grid cell. Random impulse noise is smoothed using statistical features, preventing high-frequency Rayleigh noise from directly coupling to the Brillouin temperature-strain parameter. Through this processing, precise correlation of multi-parameter sensor data based on their common origin, location, and time period is achieved, fundamentally solving the technical problems of dimensional mismatch, spatiotemporal displacement, and noise amplification through cross-parameter coupling. Simultaneously, it provides standardized, high-precision common-source data input for subsequent solving of the temperature-strain coupled overdetermined equations.

[0033] In the above scheme, determining the corresponding two-dimensional mesh element includes: Step 21: Extract the spatial mileage intervals of each spatial axis cell, and form a spatial mileage set by the center of each spatial mileage interval.

[0034] Step 22: Extract the mileage from the multi-parameter sensor data of the communication optical fiber to form a discrete set of mileage points.

[0035] Step 23: Obtain the bidirectional Hausdorff distance between the mileage point set and the spatial mileage set to measure the spatial overlap between the communication fiber multi-parameter sensing data and the mileage of this grid. If the obtained bidirectional Hausdorff distance is less than the preset spatial distance threshold, the corresponding data will be assigned to the corresponding spatial axis cell and similar data will be merged. Otherwise, no additional processing will be performed. In implementation, assume a set of mileage points. and spatial mileage set There exists a one-way Huffman distance between the sets, i.e. ; ; In the formula, and They are respectively and any point, This represents the Euclidean distance between two points. and This indicates selecting the maximum and minimum values ​​from the set of distances. Two-way Hausdorff distance. It is the larger of the two one-way Hausdorff distances, expressed as: .

[0036] It is important to understand that if data in the multi-parameter sensing data of the communication fiber optic cable is not found to fall within any spatial axis cell mileage interval, the single-point Euclidean distance between its mileage and the center of each spatial mileage interval is calculated. Data whose single-point Euclidean distance is not greater than the preset Euclidean distance threshold is assigned to the corresponding spatial axis cell; otherwise, such data is returned to manual verification.

[0037] For example, the preset Euclidean distance is 0.3m, and the center of the spatial mileage interval is 1000.5m. The single-point Euclidean distances of the FBG sensor corresponding to the actual mileages of 1000.3 and 1000.6 are both less than the preset Euclidean distance, which are considered to be temperature measurements from the same source and are entered into the merge list. Meanwhile, the discrete data of mileage 1001.2 is assigned to the next adjacent spatial axis cell.

[0038] Based on the above-mentioned spatial axis cell categorization, further consideration can be given to data redundancy, namely: Step 24: Extract the total length of the optical cable and the vibration energy of each spatial axis grid cell to determine whether to perform redundant data assessment, i.e., similar data that need to be merged.

[0039] Generally, if the total length of the optical cable does not exceed 50m or the vibration energy exceeds the limit, it means that redundant data will not be judged and the original data will be fully retained.

[0040] Step 25: If a determination is made and redundant data exists in the same spatial axis grid cell, the original data is replaced by the signal-to-noise ratio weighted fusion result of the corresponding data. If no determination is made or it is determined that there is no redundant data, no additional processing is performed.

[0041] Step 26: After synchronizing the clocks of multiple devices, determine the time axis cells of the multi-parameter sensor data of the communication fiber optic cable based on the sampling time, and combine them with the spatial axis cells to form corresponding two-dimensional grid cells. During implementation, BeiDou or GPS is used for unified time synchronization to control the clock errors of each sensor device. It is important to understand that the process of dividing the spatial axis cells first and then determining the time axis cells in the above scheme cannot be reversed. If the data is divided according to the time axis cells first, it will lead to scattered spatial data within the same time slice, or multiple operating conditions mixed within the same time period, affecting the subsequent temperature-strain decoupling accuracy and fault location, etc.

[0042] It should be understood that the above solution is not applicable to underground optical cables and equipment bay optical cables because the laying mileage is short and there is no duplicate data from the same source.

[0043] In the above steps, spatial axis cells are divided based on bidirectional Hausdorff distance for multi-parameter sensing data from communication optical fibers. Simultaneously, it is determined whether redundant data needs to be identified. If so, signal-to-noise ratio (SNR) weighting is applied to redundant data from the same source; otherwise, the data is retained. The deployment of multiple sets of multiplexed sensing devices along the same path is common in existing projects. This often leads to multiple sets of identical samples appearing in the same spatial axis grid cell. After weighted merging, the number of feature list fields is significantly reduced, and the subsequent computational workload decreases accordingly. Furthermore, through weighting, the weight of low SNR anomalous samples is reduced, and high SNR effective data dominates the results. This avoids the noise superposition caused by simple arithmetic averaging of multiple sets of identical data with random noise, especially suppressing abnormal jumps caused by fiber optic connector loss and short-term pulse interference. Redundant data entering subsequent model training for identification and classification can easily cause the model to remember repeated noise and generalize poorly. Redundancy removal can reduce the probability of model overfitting and improve the accuracy of early warning for unknown on-site conditions.

[0044] In a specific embodiment of the present invention, in terms of time synchronization, relying on two-dimensional grid cells, a unified acquisition clock, interpolation alignment, and sliding time window slicing are used to achieve the same time reference for multi-source data.

[0045] For spatial registration, data from each channel are unified to the line mileage coordinates, and resampling is performed at a unified spatial resolution when necessary. For low-frequency data such as BOTDR, linear interpolation or spline interpolation can be used; for high-frequency vibration data, intra-window statistics can be extracted before dimensionality reduction and alignment.

[0046] Noise suppression employs a combined strategy of wavelet threshold denoising, Savitzky-Golay smoothing, and adaptive median filtering. Wavelet threshold denoising is suitable for removing high-frequency random noise, Savitzky-Golay smoothing helps maintain trend patterns and peak positions, and median filtering can suppress occasional spike noise. For strongly non-stationary disturbance data, empirical mode decomposition or variational mode decomposition can be introduced to separate the background trend term from the anomalous shock term.

[0047] Regarding anomaly repair and missing data interpolation, different approaches are taken for situations such as jump points, breakpoints, sudden changes in fusion loss, and packet loss during data acquisition: isolated anomalies are corrected using the 3σ criterion, Hampel filtering, or local outlier detection; short-term missing data are interpolated using spline interpolation, Kalman filtering, or time-series model interpolation; and long-segment anomalies are marked and included in subsequent models to avoid mistakenly deleting fault features as noise.

[0048] After preprocessing, establish indicators for quality evaluation, including at least signal-to-noise ratio, data integrity rate, outlier ratio, synchronization error, and spatial registration error, to provide data credibility labels for subsequent models.

[0049] In a specific embodiment of the present invention, multi-layer fusion modeling, relying solely on the original time-series values, is insufficient to achieve stable identification results. Therefore, a feature library is first constructed from multiple dimensions, followed by feature selection. For fusion, a "three-layer fusion" modeling approach is adopted: the first layer is feature-level fusion, which involves concatenating or weighting temperature, strain, vibration, and their derived features within a unified time window and spatial location to form a joint feature vector; the second layer is model-level fusion, which involves training models suitable for different tasks separately, and then performing a comprehensive judgment through stacking or weighted outputs; the third layer is decision-level fusion, which combines physical thresholds, statistical rules, and model probability outputs to form a final risk conclusion. This design can take into account the advantages of different models and avoid burdening all tasks on a single model.

[0050] For example, the dimensions for constructing the feature library include four categories. The first category is statistical features, including mean, variance, range, skewness, kurtosis, rate of change, slope, gradient, and cumulative increment, which are suitable for characterizing the slow evolution of temperature and strain. The second category is time-frequency features, namely the aforementioned time-domain and frequency-domain features. The third category is spatial features, including spatial gradients along the path, differences between adjacent points, local spatial continuity, and the length of anomalous sections, used to locate anomalous sections and identify spatial diffusion trends. The fourth category is coupling features, namely cross-parameter features such as temperature-strain correlation coefficients, vibration energy-strain mutation coupling terms, and time lag correlations, used to reflect the linkage relationship of multiple physical quantities.

[0051] In the feature selection stage, a coarse screening is first performed using methods such as correlation analysis, variance screening, and mutual information. A secondary screening is then conducted using recursive feature elimination or XGBoost feature importance assessment to avoid excessively high feature dimensionality, which could lead to increased model complexity and decreased generalization ability. For deep learning models, raw time-series segments or spectrograms can be introduced in addition to manually generated features to improve recognition capabilities under complex conditions. To enhance the model's engineering transferability, a unified feature dictionary and feature naming rules are established. For example, features are coded according to "data source-window length-statistical method-spatial range" for subsequent model training, performance comparison, and standard interface definition. The feature library is not only used for offline training but can also directly support online inference modules, enabling feature reuse and simplified deployment.

[0052] For example, in anomaly detection tasks, for scenarios with small sample sizes and high requirements for feature interpretability, models such as support vector machines, random forests, and XGBoost can be preferred; for complex temporal and multimodal input scenarios, one-dimensional convolutional neural networks, long short-term memory networks, temporal convolutional networks, or Transformers can be used.

[0053] It is important to understand that the environments of different sections of optical cable vary. For example, the environment of overhead optical cable is characterized by strong winds and frequent transient vibrations, while the environment of duct optical cable is characterized by static temperature changes and occasional excavation disturbances. The frequency of disturbances, noise levels, and parameter change rates are completely different in each section.

[0054] In the above scenario, if the spatial axis cells are divided using a fixed spatial grid step size and the time axis cells are divided using a fixed fusion time window length, it will lead to data redundancy and wasted computing power in small grids in static and stable sections, while the spatial positioning accuracy will be insufficient in fault-prone sections with large grids. In addition, extracting time-domain and frequency-domain features with a fixed fusion time window length is a low-pass smoothing in the time domain. This means that millisecond-level instantaneous impacts, micro-deformation abrupt changes, and local instantaneous strain jumps, such as fiber optic cable collisions, will be averaged out by the entire window, and high-frequency transient fault features will be treated as random fluctuations, i.e., fault information will be lost, and the low-latency output requirements cannot be met.

[0055] Preferably, the spatial axis cells and time axis cells are cells with equal or adaptive intervals. The determination process includes: Step 31: Extract the Brillouin temperature values ​​from each two-dimensional grid cell after dividing according to the preset step size, calculate the temperature gradient value, and simultaneously extract the window average energy of the two-dimensional grid cell.

[0056] Step 32: Use the calculated temperature gradient value and the extracted window average energy as dual-condition decision data to determine the cells with equal or adaptive intervals.

[0057] Step 33: If the data for both conditions do not exceed the corresponding judgment values, then the corresponding two-dimensional grid cells are defined using equally spaced cells.

[0058] In some implementations, the spatial axis cells and time axis cells, determined by a fixed spatial grid step size and a fusion time window length, are equally spaced cells, which is suitable for routine monitoring of optical cables in stable environments, such as inter-provincial trunk pipeline optical cables and indoor equipment rooms.

[0059] The fixed spatial grid step size is set by professionals. Generally, this data falls between the minimum spatial resolution of the distributed fiber optic hardware and the effective ranging step size of the BOTDR; otherwise, data misalignment across grids will occur. For example, in the analysis of multi-parameter sensing data from overhead optical cables, a spatial grid step size of 0.1m is often used to balance the inherent sampling requirements of Brillouin and Reliami positioning. Similarly, the fusion time window length is used to unify high-frequency time-granularity data to low-frequency time-granularity. For example, during regular BOTDR sampling, it is set to 2s, which is twice the sampling period, to avoid single-point sampling jumps.

[0060] Step 34: If the data with dual conditions exceeds the corresponding judgment value, the corresponding two-dimensional grid cell is labeled with an adaptive interval cell. That is, the corresponding spatial axis cell and time axis cell need to be adjusted. It can be directly fed back to manual processing, or the data of one or more sliding windows can be superimposed for re-judgment until the conditions are met.

[0061] In other implementations, for optical cables in non-steady-state environments, a fixed spatial grid step size can lead to excessively small spatial axis cells in stable sections and excessively large spatial axis cells in sections prone to disturbances. In such cases, the corresponding cells are determined by relying on an adaptive spatial grid step size and a fusion time window.

[0062] For example, if a 1m spatial grid step size and a 2s fusion time window are set for a undisturbed pipeline optical cable, it will lead to massive repeated sampling and a surge in invalid computing power consumption caused by redundant features. In mountainous areas with frequent construction, small grids will mix multiple independent disturbances into the same two-dimensional grid cell, resulting in cross-location mismatch of temperature strain coupling equations, decoupling distortion, and inaccurate subsequent fault identification due to the average energy of the whole window approaching normal.

[0063] During implementation, the corresponding segmentation length can be determined based on the obtained temperature gradient value and the window average energy mapping, or the fine grid can be directly switched. That is, qualified spatial axis cells correspond to a coarse grid, such as 4m, and unqualified spatial axis cells correspond to a fine grid, such as 1m. Time axis cells correspond to a coarse grid, such as 2s, and a fine grid, such as 0.5s.

[0064] For example, in situations involving continuous high-frequency vibrations in fiber optic cable excavation sections in suburban farmland, it is necessary to simultaneously reduce the length of both the spatial and time axis cell lengths. For fiber optic cables in ducts with stable temperature changes and occasional single-point pulses from bird strikes, the length of the time axis cell length should be reduced to capture instantaneous change characteristics. Similarly, for fiber optic cables buried in mountains undergoing slow stretching, with continuous static strain drift and low vibration energy, it is necessary to reduce the spatial axis cell length to improve the accuracy of static strain positioning.

[0065] This step allows us to reduce data redundancy by using a coarse grid in stable segments and improve spatial positioning accuracy by using a fine grid in disturbed segments, thus eliminating the spatial granularity mismatch caused by a globally uniform step size.

[0066] In the above scheme, the waveform characteristics of fiber optic connector loss jumps, random ambient temperature drift, and actual damage faults are highly similar, and the reliability of direct filtering is limited. Therefore, cross-coupling verification is needed to determine the true cause of data deviation, thereby distinguishing between environmental noise and faults. In addition, the time-domain and frequency-domain features extracted from the data collected based on Rayleigh scattering cannot reflect millisecond-level fault characteristics, and a cross-validation method is also needed to capture transient anomalies.

[0067] Preferably, the time-series cross-fitting includes: Step 41: Construct the corresponding heterogeneous time series set for the standardized preprocessed data in each two-dimensional grid cell.

[0068] For example, for For example, heterogeneous time series sets include Brillouin temperature time series. Brillouin static strain time series Rayleigh feature conversion of dynamic strain time series ,Right now: ; ; ; In the formula, , These represent the 1st and pth Brillouin temperatures, respectively. , These represent the 1st and pth Brillouin static axial strain values, respectively. , These represent the 1st and pth Rayleigh dynamic strain values, respectively. p is the number of time-series sampling points within a single time window.

[0069] That is, each heterogeneous time series set is a data sequence obtained by sorting the data according to the same physical quantity and the same data source. Different physical quantities, or the same physical quantity but different data sources, constitute different heterogeneous time series sets.

[0070] It should be noted that the Rayleigh dynamic strain value is calculated based on the temporal features extracted in the above steps. The temporal features are already dimensionality-reduced features, thus ensuring that the three types of temporal features are completely aligned on the time axis and meet the input conditions for subsequent cross-validation.

[0071] Step 42: Calculate the coefficient of variation for each heterogeneous time series set. If the calculated coefficient of variation is less than the coefficient of variation threshold, a second-order polynomial is selected to construct a time series fitting model for the heterogeneous time series set to characterize the variation law of physical quantities in the set over time. Otherwise, an autoregressive model is selected to construct a time series fitting model for the heterogeneous time series set.

[0072] In implementation, in response to the above , , Establish a time series fitting model , , , respectively representing temperature time series Fitting model, static strain Fitting model, dynamic strain Fitting Model. The model is automatically selected based on the stability of the data within a single window. For example, in the modeling of static temperature and strain time series of optical fibers, a second-order polynomial can be used to describe the relationship between temperature or strain and time. That is, a polynomial is selected for slow temperature change, and a regression model such as Rayleigh dynamic strain value is selected for large disturbance fluctuations.

[0073] Specifically, the coefficient of variation (COP) of each heterogeneous time series set is first calculated. This COP is represented by the ratio of the set's standard deviation to its mean, where the standard deviation and mean reflect the data's fluctuation range and average level, respectively. Then, when the COP is less than a threshold, a second-order polynomial is selected; otherwise, an autoregressive model is chosen. This selection mechanism satisfies the embedded real-time computing power requirements of the communication fiber optic cable edge, avoiding the stuttering and latency caused by complex high-order models.

[0074] For example, retrieve the discrete-time index of the current window. Corresponding Brillouin temperature value Then, the coefficients of the polynomial are fitted using least squares to obtain the second-order polynomial, i.e., the fitted model. .

[0075] Step 43: For each heterogeneous time series set, a time series fitting model is used to fit the data of each remaining heterogeneous time series set, thereby obtaining the initial fitting dataset for each remaining heterogeneous time series set.

[0076] During implementation, through Fitting each , To obtain the fitted dataset , ;pass Fitting each , To obtain the fitted dataset , ;pass Fitting each , To obtain the fitted dataset , This refers to the initial fitted dataset after standardized preprocessing. For example, through... Fitting each To obtain the fitted dataset Soon Timing input ,at this time The coefficient is known, so we can obtain Thus forming .

[0077] Specifically, for each heterogeneous time series set, a time series fitting model is used to fit the data of each remaining heterogeneous time series set, resulting in the initial fitting datasets for the remaining heterogeneous time series sets, including: For any heterogeneous time series set i, a time series fitting model is constructed for each heterogeneous time series set other than i, with the same function structure but different coefficients. In the example above, i is 1, 2, and 3.

[0078] The data from each heterogeneous time series set are sequentially input into the time series fitting models determined by heterogeneous time series set i, and the fitting results of each data in each heterogeneous time series set are obtained respectively.

[0079] Calculate the mean square error between the fitting result and the actual value of the data in each heterogeneous time series set, and take the weighted sum of the mean square errors of each heterogeneous time series set as the comprehensive error of heterogeneous time series set i; among them, the mean square error of heterogeneous time series set i has the largest weight when weighted summing.

[0080] The time series fitting models of each heterogeneous time series set determined by the heterogeneous time series set i are retained if the comprehensive error is less than the comprehensive error threshold.

[0081] By fitting the data of the corresponding heterogeneous time series set with a time series fitting model other than heterogeneous time series set i, multiple initial fitting datasets other than heterogeneous time series set i are obtained.

[0082] When fitting a model to a specific heterogeneous time series set, the mean squared errors are weighted and summed. The physical quantity corresponding to that heterogeneous time series set is used as the dominant loss term, while the other physical quantities are used as auxiliary labels in training. Therefore, the predicted values ​​of non-dominant physical quantities in the initial fitted dataset obtained by the model output are not equivalent to the original data. Although the mathematical form of the models fitted to different heterogeneous time series sets is the same, the dominant loss weights are different, and the coefficients obtained are also different. Given the same input time point, the complete set of predicted physical quantities output differs, thus generating multiple different initial fitted datasets, providing multiple candidate datasets for subsequent Brillouin coupling equation verification and optimal dataset selection.

[0083] Understandably, it's difficult to distinguish between real-world operating condition changes and random noise / glitch jumps when looking at a single time series. For example, looking only at a static strain time series, a spike might appear, making it impossible to differentiate between real external force impact and sensor noise. By training the time series fitting model with temperature, static strain, and dynamic strain as the dominant parameters, when a noise anomaly appears in one time series, the models trained on other physical quantities can provide reference prediction results based on the multi-parameter coupling relationship, avoiding being biased by a single abnormal data point. In other words, noise points and abrupt change anomalies that significantly deviate from the multi-parameter laws in the original data are removed using this method, and a few missing data points can be obtained through cross-prediction of other models to obtain predicted values ​​that conform to physical laws.

[0084] Step 44: For the same heterogeneous time series set, calculate the degree of difference between the multiple initial fitted datasets and the heterogeneous time series set, determine the initial fitted dataset with the smallest degree of difference as the optimal fitted dataset of the heterogeneous time series set, calculate the relative error between the optimal fitted dataset and the heterogeneous time series set, merge the data with relative errors less than the error threshold, and rewrite them to the corresponding two-dimensional grid cells.

[0085] Specifically, for the same heterogeneous time series set, the degree of difference between multiple initial fitted datasets and the heterogeneous time series set is calculated, including: For the same heterogeneous time series set, if the initial fitted dataset satisfies the Brillouin coupling equation, the initial fitted dataset is retained; otherwise, the initial fitted dataset is deleted. The Brillouin coupling equation is used to describe the Brillouin frequency shift being coupled by temperature and strain.

[0086] It should be noted that in temperature and strain sensing applications, a linearized coupling model is often used to describe the relationship between the Brillouin frequency shift and temperature and strain, i.e., the Brillouin coupling equation, which is expressed as: ; in, The Brillouin frequency shift is the change in frequency. The change in strain The change in temperature , , respectively, are the sensitivity coefficients of the Brillouin frequency shift to temperature and strain. This equation simplifies the complex physical relationships through linearization, facilitating parameter demodulation and measurement in practical sensing systems.

[0087] In implementation, let the data in the initial fitted dataset be... , , Then each data should satisfy If the conditions are not met, fine-tuning is performed, and the initial fitting dataset that satisfies the Brillouin coupling equation is used as the fitting dataset. Considering practical engineering applications, the corresponding measured values ​​are used directly, while the experiment focuses more on variations.

[0088] The root mean square error between each initially fitted dataset and the heterogeneous time series set is calculated to characterize the degree of difference between the two datasets.

[0089] In practice, the formula for calculating the root mean square error is as follows: ; In the formula, Let represent the mean squared error of fitting data using the i-th type of heterogeneous time series data as samples, and q represent the number of all heterogeneous time series data other than the samples. Represents the actual value. This represents the fitted value. i = 1, 2, and 3 respectively represent... , , The corresponding data in the database.

[0090] In implementation, the minimum mean square error criterion is used, i.e., to determine... , as well as The minimum value in the mean square error is taken as the set of fitted values ​​corresponding to the fitted model with the smallest mean square error, and the relative error between the fitted values ​​in the optimal fitted dataset and the true values ​​is calculated.

[0091] The above steps describe a cross-validation method based on a combined error threshold and relative distance. This method uses data from different sources as samples for data fitting, and leverages the Brillouin coupling equation, mean square error, and relative error as joint constraints to achieve cross-validation for different types of data, further ensuring data accuracy. Data that meets the criteria is fused, while data that does not meet the criteria is labeled, reducing unnecessary computational consumption and improving the effectiveness of subsequent processing such as temperature-strain decoupling. In other words, the method described in this application can achieve data denoising, identify and remove outlier data, and complete missing data, preventing noise and outliers from affecting temperature-strain decoupling.

[0092] Preferably, temperature-strain decoupling includes: Step 51: Construct an overdetermined set of equations by combining the extracted time-domain features and the Brillouin coupling equations; ; In the formula, Represents the average energy of vibration. Indicates the amplitude of the spectral peak. This represents the envelope extrema, including maxima and minima; It is a mapping function from time-domain features to dynamic strain.

[0093] In the above overdetermined system of equations, the first equation , For unknown quantities, , , All are known quantities. It is obtained through the second equation and is also a known quantity.

[0094] Step 52: The goal is to make the fitted Brillouin frequency shift change approach the actual Brillouin frequency shift change. The actual Brillouin frequency shift change is the data directly obtained based on Brillouin scattering. The solution is to minimize the sum of squared residuals between the two to solve for three types of non-cross-sensitive parameters, namely the decoupled temperature value, the static axial strain value, and the Rayleigh dynamic strain value. These represent the pure ambient temperature after removing frequency shift interference caused by static or dynamic strain, the static axial strain after removing temperature and instantaneous vibration interference, and the instantaneous dynamic strain caused by excavation, impact, etc. In practice, minimizing the sum of squared residuals is expressed as: ; through fixing , , Batch solving, that is, solving in batches Find the unknowns separately , By taking the partial derivatives and setting them to zero, we obtain the decoupling temperature values ​​at each time step. Static axial strain value and Rayleigh dynamic strain value .

[0095] It is important to understand that a sudden increase in the decoupled temperature value may indicate a high-temperature fire, a slow drift in the static axial strain value indicates fiber optic cable creep and settlement, and an instantaneous jump in the Rayleigh dynamic strain value indicates instantaneous disturbances such as mechanical excavation and theft. Based on these data, the problem of not being able to determine the fault type by single frequency shift can be solved.

[0096] Step 53: Write the solved data into the corresponding two-dimensional mesh cell.

[0097] During implementation, decoupling residuals can be further extracted based on the solved data. The expression is: .

[0098] Then, it is determined whether the decoupling residual is greater than the preset decoupling residual threshold. If so, the corresponding two-dimensional mesh unit is marked as decoupling abnormal; otherwise, it is marked as decoupling normal and manual processing is prompted.

[0099] During implementation, decoupling anomalies indicate that the corresponding two-dimensional mesh elements exhibit fiber plastic deformation or localized damage. The solved data can be further used to solve for coupling characteristics, such as the temperature-static strain Pearson correlation coefficient, the vibration energy-strain abrupt change coupling term (i.e., the ratio of the change in Rayleigh dynamic strain value to the average vibration energy), etc.

[0100] The above decoupling operation will complete the separation of cross-sensitive parameters from the standardized heterogeneous data after preprocessing and cross-validation, and introduce dynamic strain constraints based on Rayleigh scattering data to separate the aliased Brillouin frequency shift into non-interfering physical parameters, thus solving the problem of cross-sensitivity of temperature-strain-vibration and the inability of a single parameter to locate the source of the fault.

[0101] The Brillouin coupling equation in the above scheme is a near-linear model based on the sensitivity coefficient of the Brillouin frequency shift to strain and temperature, which is applicable to the range where temperature and total strain are normal. When temperature and total strain change abnormally, the fiber optic sound velocity and refractive index change nonlinearly, and substituting the fixed sensitivity coefficient into the equation will produce a large decoupling error.

[0102] In other words, it is not applicable in certain scenarios, such as excessively high temperatures or large total strain. Exemplary high-temperature scenarios include situations where overhead optical cables pass through farmland where straw is burned, or high-voltage cables break down and catch fire, causing local temperatures to rise to 150-300°C in a short period, rendering the linear assumption completely ineffective. Another example is scenarios with large plastic deformation, such as large-area settlement of roadbeds in mountainous areas, where the optical fiber undergoes permanent tensile deformation, causing the sensitivity coefficient to drift. If the sensitivity coefficient under normal temperature and micro-strain is forcibly used, the high-temperature or large-deformation mesh will produce systematic decoupling errors, resulting in completely distorted fault diagnosis.

[0103] Preferably, to address the issue that the sensitivity coefficient remains stable in the low-temperature and micro-strain ranges, but exhibits continuous nonlinear shifts in both coefficients after reaching the high-temperature and overstrain ranges, the sensitivity coefficient is obtained as follows: Step 61: Extract the Brillouin static axial strain value and Rayleigh dynamic strain value of the two-dimensional mesh element at the current moment, sum them to obtain the strain value and the strain segment interval to which it belongs, and at the same time determine the temperature segment interval based on the Brillouin temperature value at the same moment. This step decomposes the global single linear characteristic of the Brillouin coupling equation into local linear characteristics between multiple cells, approximating the global nonlinearity with piecewise linearity, adapting to any temperature and strain conditions, and improving the consistency of measurements across the entire line. When the same grid cell encounters instantaneous high temperature or large deformation due to external force, it automatically switches to the corresponding interval coefficients, avoiding data distortion under abrupt changes and improving the accuracy of fault identification.

[0104] Step 62: If the strain value and Brillouin temperature value are not in the critical range of the set strain segment interval and temperature segment interval, query the corresponding sensitivity coefficient from the preset data table according to the strain segment interval and temperature segment interval to which they belong. During implementation, professionals pre-establish a data table containing defined two-dimensional intervals representing independent operating conditions and their corresponding sensitivity coefficients. These intervals define the ranges of temperature and strain. In use, the extracted values ​​determine the corresponding two-dimensional interval, which is then used as an index to retrieve the sensitivity coefficient from the data table.

[0105] Step 63: For cases where the above dual conditions are not met, the corresponding baseline sensitivity coefficient is queried from the preset data table based on the critical interval of the corresponding data, and the corresponding sensitivity coefficient is obtained through linear interpolation.

[0106] During implementation, when the temperature or strain value falls exactly at the boundary point, the sensitivity coefficient will jump abruptly, causing instantaneous fluctuations in the decoupling results, which can easily be misjudged as a fault. In this step, taking temperature as an example, let the minimum temperature of the previous two-dimensional mesh element be 80℃ and the maximum temperature be 100℃. The Brillouin frequency shift sensitivity coefficients to temperature are respectively... , The temperature segmented critical region is as well as Current Brillouin temperature value The interpolated coefficients are If the temperature is not in the critical zone of the temperature segment, the reference sensitivity coefficient obtained from the query should be used directly.

[0107] Adding linear interpolation in this step can effectively smooth out coefficient jumps, because the area near the boundary point is itself the region with the most significant nonlinear characteristics. Interpolation provides a fine fit to the transition region, resulting in higher accuracy than segmented hard truncation.

[0108] During implementation, temperature and strain are judged separately. For data in the critical range, linear interpolation is performed to obtain the sensitivity coefficient of the corresponding item; otherwise, the reference sensitivity coefficient of the corresponding segmented interval is directly used.

[0109] The above scheme adaptively matches sensitivity coefficients for each operating condition, and substitutes these sensitivity coefficients into the Brillouin coupling equations of the overdetermined equation set to decouple the temperature and strain at the current moment. Simultaneously, linear interpolation is performed on the boundaries, which not only ensures the effectiveness of decoupling under extreme conditions such as high temperature and large deformation, but also reduces the impact of numerical jitter on timing stability, adapting to the different operating conditions of multiple optical cable segments and achieving adaptive compensation of "one line, multiple operating conditions; one grid, one parameter".

[0110] The above scheme relies on two independent data sources—Brillouin shift and Rayleigh phase—to construct an overdetermined system of equations. Both types of equipment include photoelectric demodulation circuits, signal amplification modules, and sampling units, making them highly sensitive to electromagnetic interference. Strong electromagnetic fields can simultaneously intrude into both systems, causing synchronous distortion of the original sampled data, ultimately leading to the failure of both ends of the equation system.

[0111] For example, continuous power frequency electromagnetic interference originating from 110kV / 220kV / 500kV high-voltage substations and the vicinity of high-voltage transmission lines, combined with Brillouin frequency shift time series and periodic power frequency spikes, causes the frequency shift value to fluctuate regularly around the normal value. The Rayleigh phase waveform is modulated by power frequency noise, resulting in an overall increase in energy and spectral peak characteristics. This leads to a fixed offset superimposed on the data on both sides of the overdetermined equation system, causing an overall shift in the decoupling results without significant residual abrupt changes, making it easily mistaken for normal data.

[0112] For example, when overhead optical cables in mountainous areas are struck by lightning or when high-voltage equipment is switched on or off, instantaneous strong pulse interference occurs. Within a very short time, the two types of data change drastically, and a large number of spike pulses appear in the timing sequence. The effective features are completely submerged, which causes the decoupling residual to be directly judged as abnormal, but it is impossible to distinguish between "electromagnetic interference" and "real optical cable fault", resulting in a large number of false alarms.

[0113] Preferably, to address the above-mentioned problems, temperature-strain decoupling is performed, preceded by electromagnetic interference determination. The determination process includes: Step 71: Obtain the average signal-to-noise ratio of the data in each two-dimensional grid cell, that is, calculate the signal-to-noise ratio of each type of data and then take the average to represent the average level. Step 72: If the average signal-to-noise ratio exceeds the signal-to-noise ratio threshold, then perform temperature-strain decoupling and add a label with no electromagnetic interference or a label with slight electromagnetic interference to the corresponding data. During implementation, if the average signal-to-noise ratio (SNR) exceeds the first-level SNR threshold, it indicates no significant electromagnetic interference, and the data is deemed valid and distortion-free. The corresponding data is then labeled as having no electromagnetic interference before temperature-strain decoupling. If the average SNR exceeds the second-level SNR threshold but does not exceed the first-level SNR threshold, it indicates the presence of mild or steady-state power frequency interference. The corresponding data is labeled as having mild electromagnetic interference before temperature-strain decoupling. Subsequent modeling steps use this type of data with reduced weight to avoid interference affecting model accuracy.

[0114] Step 73: If the average signal-to-noise ratio does not exceed the signal-to-noise ratio threshold, it indicates the presence of severe pulse or compound interference. Temperature-strain decoupling is not performed, and the corresponding data is tagged with severe electromagnetic interference before triggering a data anomaly alarm. Generally, it is not involved in fault determination and model training.

[0115] In this step, instantaneous spikes and short-term recovery may be caused by lightning strikes or equipment switching pulse interference, while continuous chaotic and irregular waveforms indicate strong composite interference. During implementation, it is also possible to retrieve data from adjacent intervals, use spline interpolation to complete the data, and then perform temperature-strain decoupling, such as data based on three adjacent spatial axis cells and two time axis cells.

[0116] In this implementation scheme, the electromagnetic interference level of the data is marked by the average signal-to-noise ratio of standard multi-parameter sensor data. The accuracy of temperature-strain decoupling is limited for data with excessive interference. Therefore, decoupling is not performed on such data to reduce unnecessary computing power consumption and reduce the probability of fault misjudgment caused by electromagnetic interference.

[0117] Therefore, this application first unifies the collected multi-parameter sensing data in time and space using a two-dimensional grid cell composed of distance and time. Then, it performs time-series cross-fitting on the standardized preprocessed data to achieve cross-validation between data. Finally, it decouples temperature and strain based on the overdetermined equations constructed by the Brillouin coupling equations to separate cross-sensitive parameters. The superposition of the effects of multiple working conditions during the analysis improves the decoupling accuracy and efficiency, thereby ensuring the accuracy and reliability of multi-layer fusion modeling, risk assessment, and early warning based on decoupled cross-sensitive parameters and coupling characteristics. This overcomes the shortcomings of traditional fiber optic sensing, such as cross-sensitivity, data asynchrony, and high false alarm rate in complex environments, and is applicable to various optical cable scenarios.

[0118] Secondly, such as Figure 2 As shown, this application provides a communication fiber optic multi-parameter sensor data analysis system based on multi-layer fusion modeling, including: The mapping processing module is used to map the communication fiber multi-parameter sensing data obtained based on Brillouin scattering and Rayleigh scattering using two-dimensional grid cells to obtain intermediate multi-parameter sensing data with unified spatiotemporal scale. The standardization processing module is used to perform standardization preprocessing and time-series cross-fitting on intermediate multi-parameter sensing data to obtain standard multi-parameter sensing data. The decoupling module is used to decouple temperature and strain from standard multi-parameter sensing data, generate decoupled non-cross-sensitive parameters, and perform multi-layer fusion modeling in combination with synchronously generated coupling features for risk assessment and early warning.

[0119] Thirdly, embodiments of this application provide a computer-readable storage medium for storing a computer program used in the method described.

[0120] Fourthly, embodiments of this application provide a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform the method described thereon.

[0121] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for analyzing multi-parameter sensing data in optical fiber communication based on multi-layer fusion modeling, characterized in that, Includes the following steps: Two-dimensional grid cells are used to map and process the multi-parameter sensing data of communication optical fiber obtained based on Brillouin scattering and Rayleigh scattering to obtain intermediate multi-parameter sensing data with unified spatiotemporal scale. Standardized preprocessing and time-series cross-fitting are performed on intermediate multi-parameter sensing data to obtain standard multi-parameter sensing data. Temperature-strain decoupling is performed on standard multi-parameter sensing data to generate decoupled non-cross-sensitive parameters. Multi-layer fusion modeling is then performed in conjunction with synchronously generated coupling features to conduct risk assessment and early warning.

2. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 1, characterized in that, The acquisition process of the intermediate multi-parameter sensing data includes: Based on the spatial location of different sensing devices and the sampling time, the two-dimensional grid cell to which the multi-parameter sensing data of the communication fiber optic cable belongs is determined; wherein, the two-dimensional grid cell includes spatial axis cells and time axis cells, and the step size of the time axis cells is adapted to the sampling period of Brillouin scattering; For each two-dimensional grid cell, the data sampled based on Brillouin scattering and Rayleigh scattering are used as low-frequency multi-parameter sensing data and high-frequency multi-parameter sensing data, respectively. The time-domain and frequency-domain features of the high-frequency multi-parameter sensing data of the two-dimensional grid cell are extracted and spliced ​​together. The spliced ​​data is combined with the low-frequency multi-parameter sensing data of the two-dimensional grid cell to form the corresponding intermediate multi-parameter sensing data.

3. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 2, characterized in that, The process of determining the two-dimensional grid cell to which the multi-parameter sensing data of the communication fiber belongs includes: Extract the spatial mileage intervals of each spatial axis cell, and form a spatial mileage set by the center of each spatial mileage interval; Extract the mileage from the multi-parameter sensor data of the communication optical fiber to form a set of mileage points; Obtain the bidirectional Hausdorff distance between the set of mileage points and the set of spatial mileage points. If the obtained bidirectional Hausdorff distance is less than a preset spatial distance threshold, then the corresponding data is assigned to the corresponding spatial axis cell. The time axis cells of the communication fiber optic multi-parameter sensor data, which are synchronized with the clock based on the sampling time, are combined with the spatial axis cells in pairs to form corresponding two-dimensional grid cells.

4. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 1, characterized in that, The time-series cross-fitting includes: The standardized preprocessed data in the same two-dimensional grid cell are classified to form multiple heterogeneous time series sets; each heterogeneous time series set is a data sequence with the same physical quantity, the same data source, and arranged in chronological order of acquisition time. Calculate the coefficient of variation for each heterogeneous time series set. If the calculated coefficient of variation is less than the coefficient of variation threshold, then the second-order polynomial is used as the time series fitting model for the heterogeneous time series set; otherwise, the autoregressive model is used as the time series fitting model for the heterogeneous time series set. The time series fitting model is used to characterize the variation law of physical quantities in the set over time. For each heterogeneous time series set, a time series fitting model is used to fit the data of each remaining heterogeneous time series set, thus obtaining the initial fitting dataset for each remaining heterogeneous time series set. For the same heterogeneous time series set, if the initial fitted dataset satisfies the Brillouin coupling equation, the initial fitted dataset is retained; otherwise, the initial fitted dataset is deleted. The Brillouin coupling equation is used to describe the Brillouin frequency shift being coupled by temperature and strain simultaneously. Calculate the root mean square error between each retained initial fitted dataset and the heterogeneous time series set to characterize the degree of difference between the two datasets. Determine the initial fitted dataset with the smallest degree of difference as the optimal fitted dataset of the heterogeneous time series set. Calculate the relative error between the optimal fitted dataset and the heterogeneous time series set. Merge the data with relative errors less than the error threshold and rewrite them to the corresponding two-dimensional grid cells.

5. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 4, characterized in that, The time series fitting model for each heterogeneous time series set is used to fit the data of the remaining heterogeneous time series sets, resulting in the initial fitting dataset for the remaining heterogeneous time series sets, including: For any heterogeneous time series set i, a time series fitting model is constructed for each heterogeneous time series set other than i, with the same function structure but different coefficients. The data from each heterogeneous time series set are sequentially input into the time series fitting models determined by heterogeneous time series set i, and the fitting results of each data in each heterogeneous time series set are obtained respectively. Calculate the mean square error between the fitting result and the actual value of the data in each heterogeneous time series set, and take the weighted sum of the mean square errors of each heterogeneous time series set as the comprehensive error of heterogeneous time series set i; among them, the mean square error of heterogeneous time series set i has the largest weight when weighted summing. The time series fitting models of each heterogeneous time series set determined by heterogeneous time series set i are retained if the comprehensive error is less than the comprehensive error threshold; By fitting the data of the corresponding heterogeneous time series set with a time series fitting model other than heterogeneous time series set i, multiple initial fitting datasets other than heterogeneous time series set i are obtained.

6. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 5, characterized in that, The temperature-strain decoupling includes: An overdetermined set of equations is constructed by combining the extracted time-domain features and the Brillouin coupling equations. The standard multi-parameter sensing data of each two-dimensional grid cell is input into the overdetermined equation system. The non-cross-sensitive parameters are solved by minimizing the sum of squared residuals between the fitted Brillouin frequency shift change and the actual Brillouin frequency shift change. The non-cross-sensitive parameters include the decoupled temperature value, static axial strain value and Rayleigh dynamic strain value. The solved non-crossing sensitive parameters are written into the corresponding two-dimensional mesh cells.

7. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 5 or 6, characterized in that, The Brillouin coupling equation represents a near-linear model based on the sensitivity coefficients of the Brillouin frequency shift to strain and temperature. The sensitivity coefficient is obtained through the following process: Extract the Brillouin static axial strain value and Rayleigh dynamic strain value of the two-dimensional mesh element at the current moment, sum them to obtain the strain value and the strain segment interval to which it belongs, and at the same time determine the temperature segment interval based on the Brillouin temperature value at the same moment. If neither the strain value nor the Brillouin temperature value is within the corresponding preset critical range, then the corresponding sensitivity coefficient is queried from the preset data table using the corresponding strain segment range and temperature segment range as indexes. Otherwise, the sensitivity coefficient is obtained by linear interpolation based on the sensitivity coefficient retrieved from the preset data table using the preset critical interval where the corresponding data is located as the index.

8. The method for analyzing multi-parameter sensing data of optical fiber based on multi-layer fusion modeling according to claim 6, characterized in that, Before performing temperature-strain decoupling, electromagnetic interference determination is also included. The determination process includes: Obtain the average signal-to-noise ratio of standard multi-parameter sensing data in each two-dimensional grid cell; If the average signal-to-noise ratio is greater than the signal-to-noise ratio threshold, temperature-strain decoupling is performed, and a no-electromagnetic-interference or mild-electromagnetic-interference label is added to the corresponding data. Otherwise, temperature-strain decoupling is paused, and a data anomaly alarm is triggered after adding a severe-electromagnetic-interference label.

9. A communication fiber optic multi-parameter sensor data analysis system based on multi-layer fusion modeling, characterized in that, include: The mapping processing module is used to map the communication fiber multi-parameter sensing data obtained based on Brillouin scattering and Rayleigh scattering using two-dimensional grid cells to obtain intermediate multi-parameter sensing data with unified spatiotemporal scale. The standardization processing module is used to perform standardization preprocessing and time-series cross-fitting on intermediate multi-parameter sensing data to obtain standard multi-parameter sensing data. The decoupling module is used to decouple temperature and strain from standard multi-parameter sensing data, generate decoupled non-cross-sensitive parameters, and perform multi-layer fusion modeling in combination with synchronously generated coupling features for risk assessment and early warning.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for use in the method described in any one of claims 1-9.

Images