Cable optical fiber temperature measurement monitoring system based on big data AI fusion

Abstract

The invention discloses a cable optical fiber temperature measurement monitoring system based on big data AI fusion, particularly relates to the technical field of cable optical fiber temperature measurement monitoring, and aims to realize thermal-electric dynamic coupling monitoring through high-density distributed optical fiber temperature measurement and big data AI fusion modeling and introduction of current load synchronous acquisition, and realize high-density distributed optical fiber temperature measurement and big data AI fusion by means of time synchronization and space relocation processing. The method comprises the following steps: identifying a weak response signal of a thermal inertia hysteresis characteristic and an initial temperature rise stage, constructing a cross-scale time sequence recurrent neural network through multi-dimensional enhancement and vector fusion, outputting a continuous scoring sequence reflecting abnormal credibility, and constructing a risk factor group reflecting a fault evolution path in combination with a scoring fluctuation amplitude, a continuous period and a response delay characteristic, and carrying out positioning offset correction through correlation mapping of risk factors and along-cable temperature distribution and by utilizing a topological relation between optical fiber path coordinates and a cable space, and realizing space reverse projection of an abnormal peak value, thereby recovering an accurate position of a fault heat source and a latent hidden danger section.

Description

Technical Field

[0001] This invention relates to the field of cable fiber optic temperature monitoring technology, and more specifically, to a cable fiber optic temperature monitoring system based on big data and AI fusion. Background Technology

[0002] In modern power systems, cables serve as the core transmission carrier, and their operational status directly impacts grid security and energy efficiency. With increasing dynamic fluctuations in transmission loads, traditional methods relying on periodic inspections and single-point temperature measurements are no longer sufficient to meet the demand for early warning of cable anomalies. In recent years, temperature monitoring methods based on the fusion of fiber optic sensing and big data AI have gradually become mainstream. By constructing a high-density temperature sensing network and combining it with intelligent algorithms for feature extraction and trend recognition, the continuity and accuracy of anomaly detection have been greatly improved.

[0003] However, in practical applications, this type of method still faces key technical challenges when dealing with the high-dynamic load environment of cables during operation. When the cable body experiences drastic load fluctuations, its temperature response typically exhibits a significant thermal hysteresis effect, meaning that the actual overload risk has not yet manifested through temperature rise. The input obtained by the AI ​​model remains in an initial stage of feature ambiguity and signal weakening. This "time mismatch between thermal inertia and temperature rise response" directly prolongs the weak feature window perceived by the model, amplifying the ambiguity in anomaly identification, and consequently causing the model to be unable to accurately distinguish the boundary between early-stage potential problems and normal fluctuations.

[0004] More complexly, the uncertainty in data labels introduced during this weak signal phase can also negatively impact the learning ability of AI models, making them unable to effectively capture the dynamic evolution of temperature rise and the patterns of hysteresis responses during training. This causal coupling ultimately forms a training bias loop: the insignificance of features caused by thermal inertia makes it difficult for the model to learn hysteresis patterns, while the weakened model recognition ability further exacerbates misjudgments of thermal inertia scenarios. This mechanism has become a key bottleneck restricting the sensitivity and early warning reliability of existing AI cable temperature measurement methods. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a cable fiber optic temperature monitoring system based on big data AI fusion to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The cable fiber optic temperature monitoring system based on big data and AI integration includes a temperature and load joint acquisition module, a temperature and load coupling trajectory construction module, a multi-dimensional feature extraction module, a temperature and load coupling anomaly identification module, a risk classification module, and a fault reconstruction and location module. The temperature and load joint acquisition module is used to lay distributed temperature measuring optical cables on the surface of the cable body with high density in a sinusoidal path, to build a continuous thermal sensing channel, and simultaneously acquire high-resolution temperature data at multiple points along the cable and the cable load current signal within the corresponding time window, generating a thermal-electric dynamic coupling original data body with multi-dimensional time-space correspondence. The temperature load coupling trajectory construction module is used to perform time synchronization and spatial relocation processing on the original data volume of thermal-electric dynamic coupling, and construct the cable temperature evolution trajectory driven by load disturbance. The multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuation from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously detect perturbation signals in the initial fluctuation stage of temperature rise, extract weak signal feature parameters in the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters. The temperature load coupling anomaly identification module is used to fuse thermal inertia delay parameters with enhanced weak response features at the vector level to form a coupled feature group for early risk identification. This group is then input into the time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model and outputs a weak response anomaly confidence score sequence that reflects the model's perception sensitivity. The risk grading module is used to identify the weak response failure range of intelligent sensing capability based on the weak response anomaly confidence score sequence output by the model, and to construct a risk factor group characterizing the failure evolution path by combining the score fluctuation amplitude, duration period and response delay characteristics. The fault restoration and location module is used to correlate and map the risk factor group with the temperature distribution along the cable, analyze the location offset using the optical fiber path coordinates and the cable spatial topology relationship, and project the abnormal peak value of the score back to the actual spatial path to restore the precise location of the fault heat source and the section with potential hidden dangers.

[0007] In a preferred embodiment, in the temperature load joint acquisition module, distributed temperature-measuring optical cables are densely laid on the surface of the cable body using a sinusoidal wave path to construct a continuous thermal sensing channel, as detailed below: Based on the cable laying direction and structural characteristics, a coverage strategy for the thermal sensing path is planned. Distributed optical fiber temperature sensors are arranged in three dimensions along the cable surface in the form of equal amplitude sine waves to obtain a continuous thermal sensing channel. Simultaneously collect high-resolution temperature data from multiple points along the cable and cable load current signals within the corresponding time window to generate a thermal-electric dynamic coupling raw data body with multi-dimensional time-space correspondence; The temperature and cable load data are processed by sampling window normalization and time series resampling, respectively, and a preliminary noise reduction filtering algorithm is applied to the temperature signal. The calibrated temperature differential profile data along the cable and the cable load data in the same window are precisely bound at the sampling time and the optical fiber spatial index coordinates to construct a two-dimensional temperature-electricity comparison table structure. Cable structure parameters are introduced for preprocessing enhancement, and finally a thermal-electric dynamic coupling original data body with a complete physical coordinate system, dynamic timestamp chain and multi-dimensional operation indicators is generated.

[0008] In a preferred embodiment, in the temperature load coupling trajectory construction module, time synchronization and spatial relocation processing are performed on the original thermal-electric dynamic coupling data volume to construct the cable temperature evolution trajectory driven by load disturbance, as follows: Perform time base unification and clock drift correction on the raw data volume of thermo-electric dynamic coupling; The original thermal-electric dynamic coupling data volume after unifying the time base is sampled and resampled, and then the isochronous step sequence data volume is output. Spatial positioning mapping and fiber-to-cable coordinate registration are performed on the isochronous sequence data volume to generate a spatially mapped time-space matrix. Noise suppression and outlier removal are performed on the spatially mapped time-space matrix, and a cleaned and missing time-space matrix is ​​output. Load disturbance event detection is performed in parallel on the cleaned time-space matrix and the isochronous load data sequence, and a set of disturbance events with event labels is output. Using the set of disturbance events as a time reference, the corresponding temperature subsequences are extracted from the cleaned time-space matrix, and each subsequence is synchronously aligned with the event starting point as the alignment anchor point to construct a set of temperature evolution trajectories driven by load disturbance.

[0009] In a preferred embodiment, the multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuations from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously perform perturbation signal detection on the initial fluctuation stage of temperature rise, extract weak signal feature parameters of the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters, as follows: Baseline correction and trend separation are performed on the cable temperature evolution trajectory to generate a detrended and normalized clean trajectory. The cleaning trajectory is segmented and cut off by the anchor point of the disturbance event, resulting in a set of trajectory segments aligned by the event. The time-domain cross-correlation of temperature and load data sequences is calculated in parallel for the trajectory segment set, and the hysteresis candidate value sequence of each segment is initially estimated and the corresponding confidence level is recorded. Perform dynamic time alignment on the hysteresis candidate value sequence and output a refined thermal inertia hysteresis parameter curve; Multi-scale time-frequency denoising and signal decomposition are performed on the early segments of the trajectory segment set, and the signal is template matched and enhanced using historical early fault templates to generate weak signal segments amplified by signal-to-noise gating. Perturbation indices are extracted from the enhanced weak signal segments, and spatial consistency measures of adjacent points along the cable are calculated. : ,in For trajectory fragment points The set of adjacent points, The number of adjacent points, For trajectory fragment points and trajectory fragments The Pearson correlation coefficient is used to form the weak signal feature vector for each segment; Feature extension and derivation are performed on the feature vector of the weak signal, and the hysteresis-response ratio is calculated based on the thermal inertia hysteresis parameter curve corresponding to the segment. : ,in Thermal inertia hysteresis parameter The time required to rise to the peak value is used to output the enhanced coupling feature set for discrimination. Significance tests are applied to the enhanced coupling feature set to generate a weak response significance score for each event; The weak response significance score and the refined thermal inertia hysteresis parameters are jointly judged by aligning them with time to generate the final set of thermal inertia hysteresis parameters and the enhanced weak response feature set.

[0010] In a preferred embodiment, the temperature load coupled anomaly identification module is used to perform vector-level fusion of thermal inertia delay parameters and enhanced weak response features to form a coupled feature set for early risk identification. This set is then input into a time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model. The output is a weak response anomaly confidence score sequence reflecting the model's perception sensitivity, as follows: The thermal inertia hysteresis parameters and the enhanced weak response features are aligned line by line according to the time index and spatial coordinates and robustly scaled to generate a normalized feature matrix of the same dimension and free of bias. The normalized feature matrix is ​​concatenated into a time series vector by time step, and the sequence is segmented by a sliding window on the time axis to output a multi-scale feature sequence set covering multiple time scales. Local compression and representation enhancement are performed on the multi-scale feature sequence set respectively. Lightweight temporal convolutional layers or variational autoencoders are used to reduce the dimensionality and suppress noise for each window, generating scale feature tensors with low-dimensional semantic representation. The scale feature tensor is encoded with temporal location information in chronological order, and a hysteresis mask is constructed to represent thermal inertia information. The output is a temporal input tensor with temporal location information and hysteresis channel. The temporal input tensor is fed into a stacked recursive structure, and multi-layer LSTM / GRU units are used in parallel combined with dilated recursive connections and cross-layer residual direct connections to expand the receptive field and generate hidden state temporal sequences. The scale attention and temporal attention fusion is applied to the hidden state time sequence. The attention weights for the hidden states at different scales at each time step are calculated and summed according to the weights to generate a context representation sequence that integrates cross-scale information. The context representation sequence is fed into two discriminant heads in parallel: one discriminant head is used to output the anomaly confidence score for each time step, and the other uncertainty estimation head is used to output the time series uncertainty estimate; The abnormal confidence scores and uncertainty sequences are smoothed and time-windowed over time, and the final weak response abnormal confidence score sequence is generated by weighted summation of the abnormal confidence scores and uncertainties.

[0011] In a preferred embodiment, the fault reconstruction and location module is used to correlate and map the risk factor group with the temperature distribution along the cable, analyze the location offset using the optical fiber path coordinates and the cable spatial topology, and back-project the abnormal peak value of the score to the actual spatial path to restore the precise location of the fault heat source and the section with potential hidden dangers, as follows: Identify weak response anomaly confidence score peaks from the temperature distribution along the cable and risk factor groups, label each peak with fiber distance index, timestamp, peak intensity and corresponding risk factor group, and generate a structured list of anomaly peaks; Taking the list of abnormal peaks as input, the fiber distance index is mapped to the cable axial coordinate and the corresponding radial / phase offset estimate is calculated, and the preliminary axial position estimate list is output. Using the preliminary list of axial position estimates as input, spatiotemporal aggregation is performed on multiple peaks occurring in adjacent time windows and adjacent measurement points: isolated anomalies are removed using RANSAC clustering, and robust positions are obtained by weighting according to the confidence scores of weak response anomalies, and aggregated position clusters are output. Using the cluster of aggregation locations and the corresponding temperature profile along the cable as input, the temperature profile is physically driven to invert / deconvolve. The regularized inversion method is used to obtain the true axial position and intensity of the heat source, and the thermal inversion location estimate and inversion residual index are output. Using thermal inversion location estimation, aggregated location clusters, and OTDR / OBR reflection / loss labels as inputs, evidence-level fusion is performed: Bayesian update is used to fuse different pieces of evidence into the posterior location distribution of each event, and the fused posterior location set is output. Using the posterior distribution of location and cable spatial topology as input, the positioning offset is analyzed and corrected: the axial posterior is projected onto the real three-dimensional line coordinates, and the geographic location point / segment is output. Using geographic location points / segments as input, the posterior distribution of neighboring events is aggregated across time windows to identify persistent hotspots: weighted overlay is used to merge the posteriors, identify and output segments with potential hazards.

[0012] The technical effects and advantages of this invention are as follows: 1. This invention constructs a continuous and accurate thermal sensing network along the entire length of the cable by fusing high-density distributed fiber optic temperature measurement with big data AI modeling. It also incorporates synchronous current load acquisition to achieve dynamic thermo-electric coupling monitoring. Through time synchronization and spatial relocation processing, it not only fully depicts the temperature evolution trajectory driven by load disturbances but also identifies thermal inertia hysteresis characteristics and weak response signals in the early stages of temperature rise. Furthermore, through multi-dimensional enhancement and vector-level fusion, early risk characteristics are fully amplified and preserved during the AI ​​modeling stage. Based on this, the constructed cross-scale temporal recurrent neural network maintains high sensing sensitivity even when features are still in the weak signal stage, outputting a continuous scoring sequence reflecting the credibility of anomalies. Further, by combining the scoring fluctuation amplitude, duration period, and response delay characteristics, a risk factor group reflecting the fault evolution path is constructed. Through the correlation mapping between risk factors and temperature distribution along the cable, and by using fiber optic path coordinates and cable spatial topology for positioning offset correction, spatial back-projection of abnormal peaks is achieved, thereby restoring the precise location of the fault heat source and the section with potential hazards.

[0013] 2. This invention can effectively capture and locate potential anomalies in the early stages before the temperature rise signal caused by thermal inertia becomes significant, significantly shortening the early warning response time, improving the robustness and accuracy of fault identification, and avoiding missed detections and misjudgments caused by feature ambiguity. At the same time, the spatial reverse projection and robust positioning strategy ensures the high-precision traceability of the fault source location, providing maintenance personnel with immediate and actionable decision-making basis, thus having significant comprehensive advantages in ensuring power grid safety, reducing maintenance costs, and extending cable service life. Attached Figure Description

[0014] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of the system according to an embodiment of the present invention. Detailed Implementation

[0015] 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.

[0016] Example: The present invention provides, as follows Figure 1 The cable fiber optic temperature monitoring system shown is based on big data and AI fusion and includes a temperature and load joint acquisition module, a temperature and load coupling trajectory construction module, a multi-dimensional feature extraction module, a temperature and load coupling anomaly identification module, a risk classification module, and a fault reconstruction and location module. The temperature and load joint acquisition module is used to lay distributed temperature measuring optical cables on the surface of the cable body with high density in a sinusoidal path, to build a continuous thermal sensing channel, and simultaneously acquire high-resolution temperature data at multiple points along the cable and the cable load current signal within the corresponding time window, generating a thermal-electric dynamic coupling original data body with multi-dimensional time-space correspondence. The temperature load coupling trajectory construction module is used to perform time synchronization and spatial relocation processing on the original data volume of thermal-electric dynamic coupling, and construct the cable temperature evolution trajectory driven by load disturbance. The multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuation from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously detect perturbation signals in the initial fluctuation stage of temperature rise, extract weak signal feature parameters in the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters. The temperature load coupling anomaly identification module is used to fuse thermal inertia delay parameters with enhanced weak response features at the vector level to form a coupled feature group for early risk identification. This group is then input into the time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model and outputs a weak response anomaly confidence score sequence that reflects the model's perception sensitivity. The risk grading module is used to identify the weak response failure range of intelligent sensing capability based on the weak response anomaly confidence score sequence output by the model, and to construct a risk factor group characterizing the failure evolution path by combining the score fluctuation amplitude, duration period and response delay characteristics. The fault restoration and location module is used to correlate and map the risk factor group with the temperature distribution along the cable, analyze the location offset using the optical fiber path coordinates and the cable spatial topology relationship, and back-project the abnormal peak value of the score to the actual spatial path to restore the accurate location of the fault heat source and the section with potential hidden dangers. In the temperature load joint acquisition module, distributed temperature-measuring optical cables are laid in high-density bonding on the surface of the cable body using a sinusoidal wave path to construct a continuous thermal sensing channel, as detailed below: Based on the cable laying direction and structural characteristics, a coverage strategy for the thermal sensing path is planned. Under the premise of ensuring continuous sensing, distributed optical fiber temperature sensors are arranged in three dimensions along the cable surface in the form of equal amplitude sine waves. High-density coverage of the longitudinal and radial thermal diffusion areas of the cable is achieved through laying methods such as spiral winding / waveform implantation / layered staggering, resulting in a continuous thermal sensing channel. Simultaneously acquire high-resolution temperature data from multiple points along the cable and cable load current signals within corresponding time windows to generate a raw data volume of thermal-electric dynamic coupling with multi-dimensional time-space correspondence, as detailed below: A fiber optic distributed temperature sensing system (such as OTDR / OBR based on Raman / Brillouin scattering) is used to collect temperature data at each point of the fiber in the bonding path with high temporal frequency and high spatial resolution, forming a differential temperature profile dataset along the cable with meter-level resolution and second-level refresh capability. A synchronization mechanism is established to connect temperature data acquisition with the cable operation status monitoring system. Through industrial communication protocols (such as IEC 61850, Modbus, etc.), cable load data information that is completely aligned with the temperature sampling time window is acquired in real time, including cable load current data, conductor resistance change data and power supply waveform characteristics, to ensure that thermal-electric data have a unified timestamp reference and sampling period at the acquisition end. The temperature and cable load data are sampled and normalized and resampled over time to eliminate data misalignment caused by sampling frequency differences. Preliminary denoising filtering algorithms (such as Kalman filtering and wavelet denoising) are applied to the temperature signal to ensure that the thermoelectric signal has structural correspondence with equal step size, equal time window and same spatial location. The calibrated temperature differential profile data along the cable and the cable load data in the same window are precisely bound at the sampling time and the optical fiber spatial index coordinates to construct a two-dimensional temperature-electricity comparison table structure. Cable structural parameters (such as cross-sectional area, insulation structure, thermal conductivity) are introduced for preprocessing enhancement, and finally a thermal-electric dynamic coupling original data body with a complete physical coordinate system, dynamic timestamp chain and multi-dimensional operation indicators is generated. In the temperature load coupling trajectory construction module, time synchronization and spatial relocation processing are performed on the original thermal-electric dynamic coupling data volume to construct the cable temperature evolution trajectory driven by load disturbance, as follows: Perform time base unification and clock drift correction on the original thermal-electric dynamic coupling data volume to obtain a unified timestamp reference for each sampling point and generate the original thermal-electric dynamic coupling data volume after time base unification. The original thermal-electric dynamic coupling data volume after unifying the time base is sampled and resampled to eliminate the sampling frequency differences of different devices and output the isochronous sequence data volume. Spatial positioning mapping and fiber-to-cable coordinate registration are performed on the isochronous sequence data volume. The fiber optic mileage and the axial / radial relationship of the cable are corrected using the laying design information, OTDR / OBR ranging anchor points and known geometric markers, and a spatially mapped time-space matrix is ​​generated. Noise suppression and outlier removal are performed on the spatially mapped time-space matrix. A window-based robust filtering, local anomaly detection, and signal reconstruction method are used to correct abrupt changes and missing values, and a cleaned and missing time-space matrix is ​​output. Load disturbance event detection is performed in parallel on the cleaned time-space matrix and the isochronous load data sequence. The disturbance window is labeled based on the amplitude, duration and spectral characteristics of the load mutation and a set of disturbance events with event labels is output. The process involves parallel load disturbance event detection on the cleaned time-space matrix and isochronous load data sequence. Disturbance windows are labeled based on load mutation amplitude, duration, and spectral characteristics, and a set of labeled disturbance events is output, as detailed below: The cleaned time-space matrix and the isochronous load data sequence are normalized, and the instantaneous load difference, rolling mean and rolling standard deviation of each sampling point are calculated to generate a normalized load change sequence with relative amplitude representation and steady-state noise baseline. The normalized load change sequence is input into the multi-scale time-domain mutation detector. Based on short window difference, sliding Z-score and robust statistics (such as MAD), candidate perturbation windows that exceed the adaptive amplitude threshold and meet the minimum duration condition are identified. The output is a preliminary set of candidate perturbation windows labeled by time index. The preliminary candidate perturbation window set and the normalized load change sequence are fed into the spectrum analysis process in parallel. For each candidate window, the short-time Fourier transform (STFT) spectrum is calculated, and the dominant frequency energy, spectral energy ratio, spectral entropy and harmonic / interharmonic components are extracted to generate a candidate perturbation feature set with spectral signature. The candidate perturbation feature set is fused with time and frequency features, and the comprehensive perturbation intensity score of each candidate window is calculated. The comprehensive perturbation intensity score is obtained by weighted summation of amplitude component, duration component, spectral energy ratio and signal-to-noise ratio, and the candidate perturbation list with comprehensive perturbation intensity score is output. The candidate perturbation list with comprehensive perturbation intensity score is aggregated and pruned according to temporal proximity and overlap. The time intersection-union ratio (IoU) is used to merge overlapping or nearest neighbor windows, and the minimum duration threshold and minimum energy threshold are used to prune short-term low-energy noise segments. The merged and deduplicated perturbation event set is output. The merged set of disturbance events is input into the event type identification process. Based on time-domain morphological features (step, ramp, pulse, oscillation), spectral signature (single low frequency, broadband interference, harmonic group), and phase / phase consistency rules, rule determination is performed. A lightweight supervised classifier is called to perform secondary classification of complex morphologies and output a set of disturbance events with type labels. Cross-reference the set of perturbation events with type labels with the temperature data sequence to obtain the lag time and peak increment of the temperature response after the event is triggered. The lag time and peak increment are weighted and summed to generate a thermal coupling verification score, and a candidate list of perturbation events with thermal coupling verification labels is output. The final list of candidate disturbance events is converted into a standardized event record format. The record fields include: event start and end time, main amplitude, duration, main frequency band and harmonic features, event type label, thermal coupling verification score, sampling point index and spatial coordinates, and exported as a disturbance event set file for subsequent localization, alarm and model training. Using the set of disturbance events as a time reference, the corresponding temperature subsequences are extracted from the cleaned time-space matrix, and each subsequence is synchronously aligned with the event starting point as the alignment anchor point to construct a set of temperature evolution trajectories driven by load disturbance. Normalization, scale standardization, and multi-view characterization are performed on the temperature evolution trajectory set to extract trajectory-level temporal features (such as rise start time, peak delay, peak amplitude, recovery rate, and energy accumulation), and the characterization results are written into the trajectory feature library for subsequent time series modeling and anomaly identification. The multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuations from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously detect perturbation signals during the initial fluctuation stage of temperature rise, extract weak signal feature parameters during the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters, as detailed below: Baseline correction and trend separation are performed on the cable temperature evolution trajectory to remove long-term environmental drift and periodic background, generating a detrended and normalized clean trajectory. The cleaning trajectory is segmented and cut off by the anchor point of the disturbance event, resulting in a set of trajectory segments aligned by the event. The time-domain cross-correlation of temperature and load data sequences is calculated in parallel for the trajectory segment set, and the hysteresis candidate value sequence of each segment is initially estimated and the corresponding confidence level is recorded. The time-domain cross-correlation between temperature and load data sequences is calculated in parallel on the trajectory segment set. The hysteresis candidate value sequence of each segment is initially estimated and the corresponding confidence level is recorded, as follows: For each lag time Calculate the normalized cross-correlation function : ,in The temperature sequence in the trajectory segment. This is a sequence of concurrent load data within a trajectory segment. The length of the trajectory segment. The mean of the temperature series, The mean of the load data series. For discrete-time indexing, ; The lag that maximizes the normalized cross-correlation is selected as the candidate lag for cross-correlation. : And record the maximum cross-correlation value. : ; The sequence of lag candidate values ​​is obtained: ,in Let K be the k-th sub-window of the trajectory segment, where K is the number of sub-windows. The ratio of the amplitude of the cross-correlation peak to the second peak is used as the confidence level of the hysteresis candidate value sequence of the trajectory segment. : ,in This is to prevent division by zero by a very small constant (generally taken as...). ); Perform dynamic time alignment (such as DTW) on the hysteresis candidate value sequence to eliminate spurious correlations, correct nonlinear displacements, and output refined thermal inertia hysteresis parameter curves; Multi-scale time-frequency denoising and signal decomposition (such as wavelet denoising, empirical mode decomposition or singular spectrum analysis) are performed on the early segments (temperature rise start window) of the trajectory segment set, and the signal is template matched and enhanced using historical early fault templates to generate weak signal segments amplified by signal-noise gating. For the enhanced weak signal segments, perturbation indices (e.g., initial rise slope, short-time energy accumulation, local spectral entropy, harmonic components, kurtosis / skewness) are extracted, and spatial consistency measures of adjacent points along the cable are calculated. : ,in For trajectory fragment points The set of adjacent points, The number of adjacent points, For trajectory fragment points and trajectory fragments The Pearson correlation coefficient is used to form the weak signal feature vector for each segment; Feature extension and derivation (including time-delay embedding, derivative features, normalized energy ratio, and accumulation factor) are performed on the feature vector of the weak signal, and the hysteresis-response ratio is calculated based on the thermal inertia hysteresis parameter curve corresponding to the segment. : ,in Thermal inertia hysteresis parameter The time required to rise to the peak value is used to output an enhanced coupling feature set for discrimination, which includes time delay embedding, derivative features, normalized energy ratio, energy accumulation value, and hysteresis-response ratio; Significance tests (such as Mahalanobis distance, quantile thresholds, or confidence calibration based on historical distributions) are applied to the enhanced coupling feature set to generate a weak response significance score for each event. The weak response significance score and the refined thermal inertia hysteresis parameters are jointly judged according to time alignment to generate the final thermal inertia hysteresis parameter set and the enhanced weak response feature set (including confidence, time window, spatial index and source template matching score), and written into the event library for subsequent time series modeling and self-learning updates. The temperature load coupled anomaly identification module is used to perform vector-level fusion of thermal inertia delay parameters and enhanced weak response features to form a coupled feature set for early risk identification. This set is then input into the time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model. The output is a weak response anomaly confidence score sequence that reflects the model's perception sensitivity, as detailed below: The thermal inertia hysteresis parameters and the enhanced weak response features are aligned line by line according to time index and spatial coordinate and robustly scaled (e.g., extremum removal based on quantiles and Z-score / robust scaling) to generate a normalized feature matrix of the same dimension and free from bias. The normalized feature matrix is ​​concatenated into a time series vector by time step, and the sequence is segmented by a sliding window on the time axis (overlapping windows of short / medium / long scales) to output a multi-scale feature sequence set covering multiple time scales. Local compression and representation enhancement are performed on the multi-scale feature sequence set respectively. Lightweight temporal convolutional layers or variational autoencoders are used to reduce the dimensionality and suppress noise for each window, generating scale feature tensors with low-dimensional semantic representation. Encode the scale feature tensor with temporal location information in temporal order (e.g., relative timestamp / absolute location encoding or time difference mask), and construct a hysteresis mask so that thermal inertia information exists as an explicit hysteresis channel, outputting a temporal input tensor with temporal location information and hysteresis channel; The temporal input tensor is fed into a stacked recursive structure, and multi-layer LSTM / GRU units are used in parallel to combine dilated recursive connections and cross-layer residual direct connections to expand the receptive field, generating hidden state temporal sequences that are sensitive to both short-term fast responses and long-term lag evolution. At this stage, a temporal convolutional front-end is used in parallel to improve the ability to capture short-term patterns. The scale attention and temporal attention fusion is applied to the hidden state time sequence. The attention weights for the hidden states at different scales at each time step are calculated and summed according to the weights to generate a context representation sequence that integrates cross-scale information. The context representation sequence is fed in parallel into two discriminant heads: one discriminant head outputs the anomaly confidence score at each time step, and the other uncertainty estimation head outputs the temporal uncertainty estimate. The discriminant head uses sigmoid / softmax to output continuous anomaly confidence scores, and employs a time-penalized loss (early detection loss) during training to improve sensitivity to early weak responses. The uncertainty estimation head uses heteroscedastic regression or Monte Carlo dropout / deep ensemble methods to estimate uncertainty, and outputs the anomaly confidence score and uncertainty sequence. The abnormal confidence scores and uncertainty sequences are smoothed and time-windowed (e.g., exponential smoothing + max pooling rule) over time to reduce instantaneous jitter, and the final weak response abnormal confidence score sequence is generated by weighted summation of the abnormal confidence scores and uncertainties. The weak response anomaly confidence score sequence is confidence-calibrated with historical labeled events (e.g., Platt scaling or Winslow reliability calibration), and the calibrated scores are written back as a calibrated confidence score sequence with probabilistic semantics. At the same time, the time series confidence interval of each score is output for downstream decision-making. The calibration confidence score sequence, the corresponding hidden state context, and the event localization error are used to form training samples and written into the training queue. A training strategy of semi-supervised prior reconstruction + supervised fine-tuning is adopted to update the model online / periodically (including adversarial enhancement to simulate weak signal perturbations), thereby improving the weak response recognition capability in a closed loop and outputting iterative improvement logs and model performance reports for operation and maintenance. The risk grading module is used to identify the weak response failure range of the intelligent sensing capability based on the weak response anomaly confidence score sequence output by the model, and to construct a group of risk factors characterizing the failure evolution path by combining the score fluctuation amplitude, duration period and response delay characteristics, as follows: The weak response anomaly confidence score sequence output by the model is time-aligned and probabilistically calibrated to obtain a calibrated confidence score sequence with a unified time reference and probabilistic semantics. The calibration confidence score sequence is smoothed and denoised, and baseline is estimated to generate a smooth confidence curve with a de-jittered and comparable baseline. The smooth confidence curve is divided into several candidate intervals according to an adaptive segmentation strategy. The start and end times and duration of each candidate interval are labeled, and the set of candidate weak response failure intervals is output. The candidate weak response failure interval set is time-event aligned with the temperature evolution trajectory, disturbance event record and load change sequence within the corresponding time window to form an interval cascade object with event context; Extract the score fluctuation amplitude (the score fluctuation amplitude is the standard deviation of the weak response anomaly confidence score), duration period (the duration period is the duration of the candidate weak response failure interval), and response delay (the response delay is the lag time of the temperature response after the event is triggered) of the interval cascade object, and construct a risk factor group that characterizes the failure evolution path by weighted summation of the score fluctuation amplitude, duration period, and response delay. The fault reconstruction and location module is used to correlate and map risk factor groups with temperature distribution along the cable. It analyzes the location offset using fiber optic path coordinates and cable spatial topology, projects abnormal peak values ​​back onto the actual spatial path, and restores the precise location of the fault heat source and potential hazard sections, as detailed below: Identify weak response anomaly confidence score peaks from the temperature distribution along the cable and risk factor groups, label each peak with fiber distance index, timestamp, peak intensity and corresponding risk factor group, and generate a structured list of anomaly peaks; Using the list of abnormal peaks as input, based on the optical cable laying documents (sine wave amplitude, period, phase) and the distance between the TDR / OBR and the anchor point, the fiber distance index is mapped to the cable axial coordinate and the corresponding radial / phase offset estimate is calculated, and a preliminary axial position estimate list is output. Using the preliminary list of axial position estimates as input, spatiotemporal aggregation is performed on multiple peaks that occur in adjacent time windows and adjacent measurement points: RANSAC clustering is used to remove isolated outliers, and robust positions are obtained by weighting the weak response outlier confidence scores (the weighted mean is used as the candidate robust representative position for each position cluster), and the aggregated position clusters are output (each cluster contains a representative position and samples within the cluster). Using the cluster of aggregation locations and the corresponding temperature profile along the cable as input, the temperature profile is physically driven to invert / deconvolve based on the known thermal diffusion characteristics of the cable. The normalized inversion method is used to obtain the true axial position and intensity of the heat source, and the output is the thermal inversion location estimate and inversion residual index (residual, goodness of fit). Using thermal inversion location estimation, clustered location clusters, and OTDR / OBR reflection / loss labels as inputs, evidence-level fusion is performed: Bayesian update is used to fuse different evidence (thermal inversion confidence, cluster confidence, OTDR SNR) into the posterior location distribution (mean, covariance / confidence interval) of each event, and outputs the fused posterior location set; Using the posterior distribution of location and the spatial topology of cables (coordinates of pipe gallery / ditch, burial depth / overhead offset, bending section information, GIS layer) as input, the positioning offset is analyzed and corrected: the axial posterior is projected onto the real three-dimensional line coordinates (considering the radial offset and arc length mapping error caused by sinusoidal laying), and the geographic positioning point / segment is output with the uncertainty ellipse and the source of positioning error explained. Using geographic location points / segments as input, the posterior distribution of neighboring events is aggregated across time windows to identify persistent hotspots: weighted overlay is used to merge the posteriors, identify and output segments with potential hazards; Using the potential hazard section as input, a location index record is generated according to the operational standard: including point / segment coordinates, uncertainty range, summary of main evidence (heat map, OTDR features, risk factor composition), priority and recommended actions (partial discharge detection, on-site infrared, excavation / inspection), and the record is written into the GIS / maintenance work order system for work order dispatch and visualization. This invention constructs a continuous and accurate thermal sensing network along the entire length of the cable by fusing high-density distributed fiber optic temperature measurement with big data AI modeling. It also incorporates synchronous current load acquisition to achieve dynamic thermo-electric coupling monitoring. Through time synchronization and spatial relocation processing, it not only fully depicts the temperature evolution trajectory driven by load disturbances but also identifies thermal inertia hysteresis characteristics and weak response signals in the early stages of temperature rise. Furthermore, through multi-dimensional enhancement and vector-level fusion, early risk characteristics are fully amplified and preserved during the AI ​​modeling stage. Based on this, the constructed cross-scale temporal recurrent neural network maintains high sensing sensitivity even when features are still in the weak signal stage, outputting a continuous scoring sequence reflecting the credibility of anomalies. Further, by combining the scoring fluctuation amplitude, duration period, and response delay characteristics, a risk factor group reflecting the fault evolution path is constructed. Through the correlation mapping between risk factors and temperature distribution along the cable, and by using fiber optic path coordinates and cable spatial topology for positioning offset correction, spatial back-projection of abnormal peaks is achieved, thereby restoring the precise location of the fault heat source and the section with potential hazards.

[0017] This invention can effectively capture and locate potential anomalies in the early stages, before the temperature rise signal caused by thermal inertia becomes significant. This significantly shortens the early warning response time, improves the robustness and accuracy of fault identification, and avoids missed detections and misjudgments caused by feature ambiguity. At the same time, the spatial reverse projection and robust positioning strategy ensures high-precision traceability of the fault source location, providing maintenance personnel with immediate and actionable decision-making basis. Thus, it has significant comprehensive advantages in ensuring power grid safety, reducing maintenance costs, and extending cable service life.

[0018] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0019] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0020] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0021] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0022] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A cable fiber optic temperature monitoring system based on big data and AI fusion, characterized in that: It includes a temperature and load joint acquisition module, a temperature and load coupling trajectory construction module, a multi-dimensional feature extraction module, a temperature and load coupling anomaly identification module, a risk classification module, and a fault restoration and location module; The temperature and load joint acquisition module is used to lay distributed temperature measuring optical cables on the surface of the cable body with high density in a sinusoidal path, to build a continuous thermal sensing channel, and simultaneously acquire high-resolution temperature data at multiple points along the cable and the cable load current signal within the corresponding time window, generating a thermal-electric dynamic coupling original data body with multi-dimensional time-space correspondence. The temperature load coupling trajectory construction module is used to perform time synchronization and spatial relocation processing on the original data volume of thermal-electric dynamic coupling, and construct the cable temperature evolution trajectory driven by load disturbance. The multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuation from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously detect perturbation signals in the initial fluctuation stage of temperature rise, extract weak signal feature parameters in the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters. The temperature load coupling anomaly identification module is used to fuse thermal inertia delay parameters with enhanced weak response features at the vector level to form a coupled feature group for early risk identification. This group is then input into the time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model and outputs a weak response anomaly confidence score sequence that reflects the model's perception sensitivity. The risk grading module is used to identify the weak response failure range of intelligent sensing capability based on the weak response anomaly confidence score sequence output by the model, and to construct a risk factor group characterizing the failure evolution path by combining the score fluctuation amplitude, duration period and response delay characteristics. The fault restoration and location module is used to correlate and map the risk factor group with the temperature distribution along the cable, analyze the location offset using the optical fiber path coordinates and the cable spatial topology relationship, and project the abnormal peak value of the score back to the actual spatial path to restore the precise location of the fault heat source and the section with potential hidden dangers.

2. The cable fiber optic temperature monitoring system based on big data and AI fusion as described in claim 1, characterized in that: In the temperature load joint acquisition module, distributed temperature-measuring optical cables are laid in high-density bonding on the surface of the cable body using a sinusoidal wave path to construct a continuous thermal sensing channel, as detailed below: Based on the cable laying direction and structural characteristics, a coverage strategy for the thermal sensing path is planned. Distributed optical fiber temperature sensors are arranged in three dimensions along the cable surface in the form of equal amplitude sine waves to obtain a continuous thermal sensing channel. Simultaneously collect high-resolution temperature data from multiple points along the cable and cable load current signals within the corresponding time window to generate a thermal-electric dynamic coupling raw data body with multi-dimensional time-space correspondence; The temperature and cable load data are processed by sampling window normalization and time series resampling, respectively, and a preliminary noise reduction filtering algorithm is applied to the temperature signal. The calibrated temperature differential profile data along the cable and the cable load data in the same window are precisely bound at the sampling time and the optical fiber spatial index coordinates to construct a two-dimensional temperature-electricity comparison table structure. Cable structure parameters are introduced for preprocessing enhancement, and finally a thermal-electric dynamic coupling original data body with a complete physical coordinate system, dynamic timestamp chain and multi-dimensional operation indicators is generated.

3. The cable fiber optic temperature monitoring system based on big data and AI fusion according to claim 2, characterized in that: In the temperature load coupling trajectory construction module, time synchronization and spatial relocation processing are performed on the original thermal-electric dynamic coupling data volume to construct the cable temperature evolution trajectory driven by load disturbance, as follows: Perform time base unification and clock drift correction on the raw data volume of thermo-electric dynamic coupling; The original thermal-electric dynamic coupling data volume after unifying the time base is sampled and resampled, and then the isochronous step sequence data volume is output. Spatial positioning mapping and fiber-to-cable coordinate registration are performed on the isochronous sequence data volume to generate a spatially mapped time-space matrix. Noise suppression and outlier removal are performed on the spatially mapped time-space matrix, and a cleaned and missing time-space matrix is ​​output. Load disturbance event detection is performed in parallel on the cleaned time-space matrix and the isochronous load data sequence, and a set of disturbance events with event labels is output. Using the set of disturbance events as a time reference, the corresponding temperature subsequences are extracted from the cleaned time-space matrix, and each subsequence is synchronously aligned with the event starting point as the alignment anchor point to construct a set of temperature evolution trajectories driven by load disturbance.

4. The cable fiber optic temperature monitoring system based on big data and AI fusion according to claim 3, characterized in that: The multi-dimensional feature extraction module is used to identify the hysteresis characteristics of temperature response relative to load fluctuations from the cable temperature evolution trajectory, extract thermal inertia hysteresis parameters, and simultaneously detect perturbation signals during the initial fluctuation stage of temperature rise, extract weak signal feature parameters during the weak signal stage, and perform multi-dimensional enhancement processing on the weak signal feature parameters, as detailed below: Baseline correction and trend separation are performed on the cable temperature evolution trajectory to generate a detrended and normalized clean trajectory. The cleaning trajectory is segmented and cut off by the anchor point of the disturbance event, resulting in a set of trajectory segments aligned by the event. The time-domain cross-correlation of temperature and load data sequences is calculated in parallel for the trajectory segment set, and the hysteresis candidate value sequence of each segment is initially estimated and the corresponding confidence level is recorded. Perform dynamic time alignment on the hysteresis candidate value sequence and output a refined thermal inertia hysteresis parameter curve; Multi-scale time-frequency denoising and signal decomposition are performed on the early segments of the trajectory segment set, and the signal is template matched and enhanced using historical early fault templates to generate weak signal segments amplified by signal-to-noise gating. Perturbation indices are extracted from the enhanced weak signal segments, and spatial consistency measures of adjacent points along the cable are calculated. : ,in For trajectory fragment points The set of adjacent points, The number of adjacent points, For trajectory fragment points and trajectory fragments The Pearson correlation coefficient is used to form the weak signal feature vector for each segment; Feature extension and derivation are performed on the feature vector of the weak signal, and the hysteresis-response ratio is calculated based on the thermal inertia hysteresis parameter curve corresponding to the segment. : ,in Thermal inertia hysteresis parameter The time required to rise to the peak value is used to output the enhanced coupling feature set for discrimination. Significance tests are applied to the enhanced coupling feature set to generate a weak response significance score for each event; The weak response significance score and the refined thermal inertia hysteresis parameters are jointly judged by aligning them with time to generate the final set of thermal inertia hysteresis parameters and the enhanced weak response feature set.

5. The cable fiber optic temperature monitoring system based on big data and AI fusion according to claim 4, characterized in that: The temperature load coupled anomaly identification module is used to perform vector-level fusion of thermal inertia delay parameters and enhanced weak response features to form a coupled feature set for early risk identification. This set is then input into the time-series feature modeling system, which uses a multi-layer recurrent neural network to construct a cross-scale anomaly evolution identification model. The output is a weak response anomaly confidence score sequence that reflects the model's perception sensitivity, as detailed below: The thermal inertia hysteresis parameters and the enhanced weak response features are aligned line by line according to the time index and spatial coordinates and robustly scaled to generate a normalized feature matrix of the same dimension and free from bias. The normalized feature matrix is ​​concatenated into a time series vector by time step, and the sequence is segmented by a sliding window on the time axis to output a multi-scale feature sequence set covering multiple time scales. Local compression and representation enhancement are performed on the multi-scale feature sequence set respectively. Lightweight temporal convolutional layers or variational autoencoders are used to reduce the dimensionality and suppress noise for each window, generating scale feature tensors with low-dimensional semantic representation. The scale feature tensor is encoded with temporal location information in chronological order, and a hysteresis mask is constructed to incorporate thermal inertia information, outputting a temporal input tensor with temporal location information and a hysteresis channel. The temporal input tensor is fed into a stacked recursive structure, and multi-layer LSTM / GRU units are used in parallel combined with dilated recursive connections and cross-layer residual direct connections to expand the receptive field and generate hidden state temporal sequences. The scale attention and temporal attention fusion is applied to the hidden state time sequence. The attention weights for the hidden states at different scales at each time step are calculated and summed according to the weights to generate a context representation sequence that integrates cross-scale information. The context representation sequence is fed into two discriminant heads in parallel: one discriminant head is used to output the anomaly confidence score for each time step, and the other uncertainty estimation head is used to output the time series uncertainty estimate; The abnormal confidence scores and uncertainty sequences are smoothed and time-windowed over time, and the final weak response abnormal confidence score sequence is generated by weighted summation of the abnormal confidence scores and uncertainties.

6. The cable fiber optic temperature monitoring system based on big data and AI fusion according to claim 5, characterized in that: The fault reconstruction and location module is used to correlate and map risk factor groups with temperature distribution along the cable. It analyzes the location offset using fiber optic path coordinates and cable spatial topology, projects abnormal peak values ​​back onto the actual spatial path, and restores the precise location of the fault heat source and potential hazard sections, as detailed below: Identify weak response anomaly confidence score peaks from the temperature distribution along the cable and risk factor groups, label each peak with fiber distance index, timestamp, peak intensity and corresponding risk factor group, and generate a structured list of anomaly peaks; Taking the list of abnormal peaks as input, the fiber distance index is mapped to the cable axial coordinate and the corresponding radial / phase offset estimate is calculated, and the preliminary axial position estimate list is output. Using the preliminary list of axial position estimates as input, spatiotemporal aggregation is performed on multiple peaks occurring in adjacent time windows and adjacent measurement points: isolated anomalies are removed using RANSAC clustering, and robust positions are obtained by weighting according to the confidence scores of weak response anomalies, and aggregated position clusters are output. Using the cluster of aggregation locations and the corresponding temperature profile along the cable as input, the temperature profile is physically driven to invert / deconvolve. The regularized inversion method is used to obtain the true axial position and intensity of the heat source, and the thermal inversion location estimate and inversion residual index are output. Using thermal inversion location estimation, aggregated location clusters, and OTDR / OBR reflection / loss labels as inputs, evidence-level fusion is performed: Bayesian update is used to fuse different pieces of evidence into the posterior location distribution of each event, and the fused posterior location set is output. Using the posterior distribution of location and cable spatial topology as input, the positioning offset is analyzed and corrected: the axial posterior is projected onto the real three-dimensional line coordinates, and the geographic location point / segment is output. Using geographic location points / segments as input, the posterior distribution of neighboring events is aggregated across time windows to identify persistent hotspots: weighted overlay is used to merge the posteriors, identify and output segments with potential hazards.

Images