A method and system for predicting the wear of a submarine cable outer sheath by multi-source data fusion

Abstract

This invention relates to the field of wear prediction technology, specifically to a method and system for predicting wear of submarine cable outer sheaths using multi-source data fusion. The method includes the following steps: performing spatiotemporal joint analysis of ship trajectories to obtain a near-cable ship threat index; calculating periodic bending stress loads on local micro-elements of the submarine cable based on a simplified fluid-structure interaction model to obtain a time-varying flow-induced dynamic curvature sequence; extracting time-varying baseline vulnerability based on the cable's inherent properties and historical maintenance data, and fusing the near-cable ship threat index, the flow-induced dynamic curvature sequence, and the time-varying baseline vulnerability to obtain a comprehensive real-time risk vector; using the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion as the core, fusing evidence theory to obtain long-term environmental risk trend characteristics; and combining the actual mechanical strain distribution, the comprehensive real-time risk vector, and the long-term environmental risk trend characteristics to predict the location and extent of wear on the submarine cable outer sheath through hierarchical decision-making.

Description

Technical Field

[0001] This invention relates to the field of wear prediction technology, specifically to a method and system for predicting wear of submarine cable outer sheaths using multi-source data fusion. Background Technology

[0002] Submarine cables, as the core carriers of marine communication and offshore energy transmission, have outer sheaths that directly protect the internal cable cores from marine environmental erosion, ship disturbances, seabed friction, and other factors. The integrity of the outer sheath directly determines the safe and stable operation of the submarine cable. Currently, the monitoring and prediction of wear on the outer sheath of submarine cables mainly relies on traditional methods such as manual inspection and underwater robot detection, which have many drawbacks: First, the inspection cycle is long and costly, making real-time monitoring impossible and easily overlooking early wear hazards; second, the prediction accuracy is low, mostly regional qualitative predictions, unable to achieve precise positioning at the micro-element level, and difficult to quantify the degree of wear; third, the fusion of multi-source data is poor, with key data affecting wear such as ship trajectory, marine environment, and submarine cable strain being independent and not effectively fused, resulting in incomplete risk identification; fourth, traditional algorithms are not adapted to the characteristics of submarine cable scenarios, resulting in large errors and weak generalization ability, and failing to accurately capture the cumulative and periodic characteristics of wear on the outer sheath of submarine cables.

[0003] Therefore, there is an urgent need for an algorithm that can integrate multi-source data, adapt to submarine cable scenarios, and achieve high-precision prediction, so as to solve the shortcomings of traditional methods, provide a scientific basis for preventive operation and maintenance of submarine cables, and reduce the failure rate and operation and maintenance costs of submarine cables. Summary of the Invention

[0004] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides a method for predicting wear of the outer sheath of submarine cables using multi-source data fusion, comprising the following steps: A spatiotemporal joint analysis of ship trajectories is performed to obtain a near-cable ship threat index. Based on a simplified fluid-structure interaction model, the periodic bending stress load on local micro-elements of the submarine cable is calculated by combining the known spatial coordinates and mechanical parameters of the cable, obtaining a time-varying flow-induced dynamic curvature sequence. Time-varying baseline vulnerability is extracted based on the cable's inherent properties and historical maintenance data. The near-cable ship threat index, the flow-induced dynamic curvature sequence, and the time-varying baseline vulnerability are fused to obtain a comprehensive real-time risk vector. Taking the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion as the core, long-term environmental risk trend characteristics are obtained by fusing evidence theory. Combining the actual mechanical strain distribution, the comprehensive real-time risk vector, and the long-term environmental risk trend characteristics, hierarchical decision-making is used to predict the location and extent of wear on the cable's outer sheath.

[0005] Optionally, the step of performing spatiotemporal joint analysis of ship trajectories to obtain the near-lined ship threat index includes the following steps: The spatial distribution density of ship trajectories along the submarine cable is quantified by kernel density estimation to obtain spatial density characteristics; the DBSCAN clustering algorithm is optimized by introducing a time dimension to distinguish ship behaviors of different risk levels and obtain spatiotemporal clustering risk characteristics; trajectory anomaly characteristics are extracted based on trajectory anomaly detection of short-time energy zero-crossing rate; and the near-cable ship threat index is obtained by fusing the spatial density characteristics, the spatiotemporal clustering risk characteristics, and the trajectory anomaly characteristics.

[0006] Optionally, the step of calculating the periodic bending stress load on local micro-elements of the submarine cable based on the simplified fluid-structure interaction model, combined with the known spatial coordinates and mechanical parameters of the submarine cable, to obtain the time-varying flow-induced dynamic curvature sequence includes the following steps: Based on a simplified fluid-structure interaction model, the hydrodynamic effect of the ocean current field on the submarine cable is transformed into periodic bending stress of local micro-elements. According to the physical relationship between stress and curvature, the dynamic bending deformation of the submarine cable under the action of the current field is quantified, and the flow-induced dynamic curvature sequence changing with time is obtained.

[0007] Optionally, the step of extracting time-varying baseline vulnerability based on the cable's inherent properties and historical maintenance data includes the following steps: Set the variables and parameters of the survival analysis algorithm; quantify and embed the prior knowledge in the historical maintenance data into the survival analysis algorithm, and correct the innate vulnerability coefficient to obtain the basic vulnerability coefficient that takes into account both innate attributes and the cumulative effects of acquired factors; use the basic vulnerability coefficient to calculate the time-varying baseline vulnerability.

[0008] Optionally, the process of fusing the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability to obtain a comprehensive real-time risk vector includes the following steps: The risk weights of the attention mechanism are dynamically calculated; and a comprehensive real-time risk vector is obtained by combining the risk weights, the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability.

[0009] Optionally, the method of obtaining long-term environmental risk trend characteristics by fusing the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion using evidence theory includes the following steps: Spatial features are obtained by extending the heat map characteristics of near-cable risk contact with time series, and temporal features are obtained by encoding the periodic residual erosion index sequence through time series. Based on the DS evidence theory synthesis rules, the spatial features and the temporal features are used as two independent evidence sources for confidence allocation and fusion to obtain long-term environmental risk trend features.

[0010] Optionally, extracting the near-cable risk contact heatmap features includes the following steps: Register multi-source trajectories to obtain a preliminary fused ship trajectory dataset; based on the ship trajectory dataset, obtain high-precision fused trajectory data by optimizing acoustic positioning accuracy; quantify near-cable space risk according to the high-precision fused trajectory data to obtain a spatial risk distribution matrix; construct a near-cable risk contact heat map using the spatial risk distribution matrix to obtain near-cable risk contact heat map features.

[0011] Optionally, the step of predicting the location and extent of wear on the outer sheath of the submarine cable through hierarchical decision-making, combining the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, includes the following steps: A standardized feature matrix is ​​obtained by standardizing the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics; the uncertainty of the standardized feature matrix is ​​quantified, and an error correction coefficient is introduced to obtain a corrected feature matrix; a hierarchical decision fusion model is constructed, and the location and extent of wear on the outer sheath of the submarine cable are predicted by combining the hierarchical decision fusion model and the corrected feature matrix.

[0012] Optionally, obtaining the actual mechanical strain distribution includes the following steps: The flow-induced dynamic curvature sequence is loaded into the finite element simulation model of the submarine cable to generate the theoretical strain distribution sequence of each micro-element; the simulation model is corrected through data assimilation, and the actual mechanical strain distribution is extracted using the corrected simulation model.

[0013] This invention integrates multi-source data such as ship trajectory, marine environment, and submarine cable strain. Through clustering, data assimilation, evidence theory, and a hierarchical decision-making model, it completes multi-source data preprocessing, micro-element-level feature extraction, multi-modal feature fusion, and uncertainty correction. This enables precise location of micro-element-level wear on the submarine cable outer sheath and quantitative prediction of light, moderate, and severe wear. It effectively addresses the pain points of low accuracy and incomplete risk identification in traditional submarine cable wear prediction methods, accurately capturing various wear risks, providing a scientific basis for preventative maintenance of submarine cables, effectively reducing cable failure rates, lowering maintenance costs, ensuring the safe and stable operation of submarine cables, and supporting the normal operation of marine communications and related fields.

[0014] Secondly, to efficiently execute the multi-source data fusion method for predicting wear of submarine cable outer sheaths provided by this invention, this invention also provides a multi-source data fusion system for predicting wear of submarine cable outer sheaths, comprising: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory stores program instructions used for the multi-source data fusion method for predicting wear of submarine cable outer sheaths. The multi-source data fusion system for predicting wear of submarine cable outer sheaths provided by this invention has a compact structure and stable performance, and can stably execute the multi-source data fusion method for predicting wear of submarine cable outer sheaths provided by this invention, further improving the overall applicability and practical application capability of this invention. Attached Figure Description

[0015] Figure 1 A flowchart of a method for predicting wear of submarine cable outer sheath by multi-source data fusion provided in an embodiment of the present invention; Figure 2 This is a framework diagram of a multi-source data fusion system for predicting wear of submarine cable outer sheaths, provided in an embodiment of the present invention. Detailed Implementation

[0016] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0017] Throughout this specification, references to an embodiment, example, or illustration mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, phrases appearing in various places throughout the specification, such as "in one embodiment," "in an embodiment," "an example," or "an illustration," do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in any suitable combination and / or sub-combination in one or more embodiments or examples. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0018] Please see Figure 1 To address the aforementioned problems, this invention provides a method for predicting wear of submarine cable outer sheaths using multi-source data fusion, such as... Figure 1 As shown, in one embodiment, the method includes the following steps: S1. Perform spatiotemporal joint analysis of ship trajectories to obtain the near-cable ship threat index.

[0019] In this embodiment, the step of performing spatiotemporal joint analysis of ship trajectories to obtain the near-lined ship threat index includes the following steps: S11. The spatial distribution density of ship trajectories along the submarine cable is quantified by kernel density estimation to obtain spatial density characteristics.

[0020] First, valid near-cable trajectory data is screened, and invalid interference information is removed to provide high-quality data support for subsequent density calculation, spatiotemporal clustering, and anomaly detection. Specifically, the original ship AIS trajectory data is thoroughly cleaned, focusing on three types of problematic data: First, null data, which is supplemented with missing values ​​of key parameters such as trajectory point coordinates and speed using linear interpolation to ensure trajectory continuity; second, abnormal data, which is scientifically judged based on the navigation rules of the submarine cable laying area—speed is set to 0 knots (ship is anchored, no movement trajectory, cannot reflect near-cable threat) or >25 knots (exceeding the navigation speed limit of the sea area, likely a false alarm or a non-operational / navigational vessel, posing a very low threat to the submarine cable), and a sudden change in heading >180° (abnormal jump in trajectory point, not in line with normal ship navigation patterns, indicating data distortion), and data points judged as abnormal are directly removed; third, redundant data, which is deduplicated for repeated trajectory points with time intervals <10 seconds to avoid redundant calculations in subsequent calculations.

[0021] After data cleaning, based on the geospatial coordinate data of the submarine cable (including route direction, protected area, burial depth, etc.), the Euclidean distance calculation method is used to accurately calculate the spatial distance between each ship trajectory point and the centerline of the submarine cable route. The trajectories of near-cable vessels entering the submarine cable protected area and the surrounding 1 nautical mile range are screened out. This range is set according to the actual risk pattern of wear on the outer sheath of the submarine cable. When ships operate within this range, their anchoring, navigation disturbances, and other behaviors may cause mechanical wear on the outer sheath of the submarine cable. Beyond this range, the threat is negligible.

[0022] Finally, the selected near-cable vessel trajectories were standardized in format, and the timestamps and coordinate system (WGS84 coordinate system, consistent with the geographic coordinates of the submarine cable) were unified. This set a clear effective range and laid the data foundation for accurate calculations in subsequent stages.

[0023] Based on this, the spatial distribution density of ship trajectories along the submarine cable is quantified by kernel density estimation, which accurately identifies high-risk areas with frequent ship activity and provides a spatial density basis for subsequent spatiotemporal clustering.

[0024] Specifically, the effective near-cable vessel trajectory points obtained after processing the near-cable trajectory data are uniformly projected into a two-dimensional spatial coordinate system around the submarine cable to construct a two-dimensional spatial grid adapted to the linear routing characteristics of the submarine cable. The grid division standard is comprehensively set by combining the length of the submarine cable, the accuracy of the vessel trajectory, and the computational efficiency. A grid size of 100 meters × 100 meters is adopted, which can accurately reflect the spatial distribution differences of the vessel trajectory, and avoid the problems of excessive computation caused by too fine a grid and density distortion caused by too coarse a grid.

[0025] Subsequently, to address the issue of mismatch between the fixed circular kernel function window in traditional kernel density estimation and the linear routing characteristics of submarine cables, which easily leads to inaccurate density calculations, a scenario-based improvement was implemented: the fixed circular kernel function window was optimized and adjusted to an elliptical window dynamically adapted to the cable route. The major axis of the ellipse aligns with the centerline of the cable route, and its length is set to the width of the cable protection zone (typically 200 meters), ensuring coverage of the core area of ​​vessel activity around the cable. The minor axis of the ellipse is perpendicular to the cable route, and its length is set to 100 meters, adapting to the lateral activity range of vessels in the near-cable area. This effectively avoids density calculation errors caused by an excessively large (including areas unrelated to the near-cable) or excessively small (omitting high-risk areas near the cable) circular window coverage. A Gaussian kernel function adapted to spatial density estimation was selected, and the kernel bandwidth was calibrated using cross-validation (calibrated to 50 meters based on the amount and distribution characteristics of near-cable vessel trajectory data), ensuring the accuracy and rationality of the density calculation.

[0026] Finally, the density of ship trajectory points in each two-dimensional spatial grid is calculated using the kernel density estimation algorithm, generating a spatial density distribution matrix of ship trajectories along the submarine cable. Each element in the matrix corresponds to the density value of a grid. The higher the density value, the more frequent the ship activity in that grid and the higher the potential risk of near-cable threat. This provides core data support for the spatial neighborhood determination of subsequent spatiotemporal clustering.

[0027] S12. Introduce a time dimension to optimize the DBSCAN clustering algorithm and obtain spatiotemporal clustering risk features by distinguishing ship behaviors of different risk levels.

[0028] Traditional spatial DBSCAN clustering can only cluster based on the spatial distance of trajectory points, failing to distinguish between the behavioral differences of a ship temporarily passing through a near-mooring area and continuously lingering in that area. Therefore, this invention introduces temporal dimension features into the traditional spatial DBSCAN clustering algorithm to achieve spatiotemporal joint clustering of ship trajectories, accurately distinguishing ship behaviors at different risk levels.

[0029] First, the core input for clustering is the spatial density distribution matrix obtained by calculating the spatial density of the near-cable track, as well as the processed ship track timestamp data, to ensure that the clustering process takes into account both spatial distribution and temporal continuity.

[0030] Subsequently, two time-neighborhood-specific parameters were introduced and calibrated based on the activity patterns of vessels near the submarine cable: First, the time interval between trajectory points is ≤1 minute, which is adapted to the regular update frequency of AIS data (usually 1 minute / time) to ensure that the continuous dynamics of vessel navigation can be captured and to avoid trajectory breaks and clustering distortion caused by excessive time intervals; Second, the continuous dwell time is ≥30 minutes. Based on historical data of submarine cable operation and maintenance, vessels that stay in the near-cable area for more than 30 minutes are likely to be operating vessels, anchored vessels, etc. Their activities (such as anchoring and towing, and operational disturbances) are likely to cause wear and tear on the outer sheath of the submarine cable, which is judged as high-risk behavior. Vessels that stay for less than 30 minutes are mostly passing through temporarily and have lower risks.

[0031] Based on the above parameters, the clustering conditions are reconstructed to meet the density standard within the spatial neighborhood and ensure the continuity of trajectory points within the temporal neighborhood. The specific clustering process is as follows: First, based on the spatial density distribution matrix, the spatial neighborhood radius (combined with the grid size, calibrated to 150 meters) and density threshold (grid density value ≥ 5 trajectory points / hour) are set to select spatial core points; then, the temporal continuity of the spatial core points is verified, and core points with a continuous trajectory point time interval ≤ 1 minute and a cumulative continuous dwell time ≥ 30 minutes are retained to form high-risk clusters; trajectory points with spatial density meeting the standard but insufficient temporal continuity are formed into low-risk clusters; and discrete trajectory points with spatial density not meeting the standard are determined to be risk-free trajectory points.

[0032] Ultimately, the clustering results were clearly divided into three categories: persistent loitering clusters (high risk) (mostly operating vessels and anchored vessels), temporary passing clusters (low risk) (mostly navigating vessels), and discrete trajectory points (no risk) (mostly data misreporting or long-distance navigating vessels). The spatial range, number of trajectories, and risk level of individual vessels for each cluster were initially determined.

[0033] S13. Based on the trajectory anomaly detection of short-time energy zero-crossing rate, extract trajectory anomaly features.

[0034] First, for ships within clusters (no risk-free discrete trajectory points need to be detected, reducing computational load), the near-moor trajectory data of a single ship is transformed into a two-dimensional time series of speed and heading. The sampling frequency of the time series is consistent with the AIS data update frequency (1 minute / sampling point) to ensure that the dynamic changes of the ship can be fully captured.

[0035] Subsequently, the core parameters of the short-time energy zero-crossing rate algorithm were reconstructed for specific scenarios to adapt to the time scale of ship navigation dynamics: First, the short-time frame length was set to 5 minutes. Combined with the dynamic characteristics of ship navigation, the 5-minute time window can capture instantaneous anomalies (such as sudden turns) while avoiding false detections due to excessively short frame lengths (such as minor speed fluctuations) and missed anomaly detections due to excessively long frame lengths. Second, the energy threshold was set based on the variance of the average speed of ships around the cable. The specific calculation process was as follows: first, statistical analysis of the average speed variance of ships in the near-cable area over the past 3 months was performed. With ship speed data, the average speed and variance are calculated, and the energy threshold is set as the average speed ± 2 times the variance. Speed ​​changes exceeding this threshold are judged as abnormal energy changes (such as a sudden drop in speed from 20 knots to 5 knots, or a sudden increase in speed from 5 knots to 25 knots). Third, the zero-crossing rate is redefined as the number of course abrupt changes, and the course abrupt change threshold is set at 30°. When the number of course abrupt changes is ≥ 2 within a short frame (5 minutes), it is judged as an abnormal zero-crossing rate (such as frequent ship turning, which is likely due to operation or anchoring towing behavior).

[0036] Finally, based on the reconstructed parameters, the speed-heading time series is detected: the energy value and zero-crossing rate of each short frame are calculated. When the energy value exceeds the threshold or the zero-crossing rate is abnormal, it is determined that there is a trajectory anomaly. Combined with the degree of anomaly (such as the magnitude of the speed drop and the number of abrupt changes in heading), the trajectory anomaly index of a single ship is calculated using a normalization method (0~1, the higher the value, the more severe the anomaly and the greater the risk).

[0037] S14. By integrating the spatial density characteristics, the spatiotemporal clustering risk characteristics, and the trajectory anomaly characteristics, a near-cable vessel threat index is obtained.

[0038] By transforming the spatial distribution, behavioral types, and instantaneous anomalies of ships into quantifiable threat indices, we can not only quantify the risks of individual ships but also obtain the regional risk distribution along the submarine cable, providing core input features for subsequent comprehensive risk fusion.

[0039] First, the weighting coefficients were determined using historical accident data from submarine cable maintenance (selecting accident cases caused by ship factors among submarine cable outer sheath wear accidents in the past 5 years, and statistically analyzing the impact weights of spatial distance, cluster type, and trajectory anomalies on the accidents): First, spatial distance weighting: the closer the ship is to the centerline of the submarine cable route, the greater the threat of wear and tear on the outer sheath of the cable. The weight is set in a linear decreasing manner (0~500 meters from the cable, weight 1.0; 500~1000 meters, weight 0.6; 1000~1500 meters, weight 0.3; 1500~1852 meters (1 nautical mile), weight 0.1) to ensure that the weight matches the actual threat level. Second, cluster risk weights are determined by combining the risk differences in the clustering results. The weight of the cluster that continuously loiters is set at 0.7 (most of these vessels are operating vessels and pose the highest threat), the weight of the cluster that temporarily passes through is set at 0.3 (most of these vessels are navigable vessels and pose a lower threat), and the weight of the risk-free discrete trajectory points is set at 0. Third, the weight of the anomaly index is dynamically set. The higher the trajectory anomaly index, the greater the weight (anomaly index 0~0.3, weight 0.2; 0.3~0.7, weight 0.5; 0.7~1.0, weight 0.8), highlighting the high-risk characteristics of instantaneous anomaly behavior.

[0040] After the weights are calibrated, a multi-weight linear fusion method is used to calculate the threat index of a single vessel: Threat index of a single vessel = spatial distance weight × normalized spatial density value + clustering risk weight × clustering risk coefficient (1.0 for continuous loitering and 0.5 for temporary passing) + anomaly index weight × trajectory anomaly index. After fusion, the index is normalized to the range of 0 to 1. The higher the value, the greater the near-line threat of a single vessel.

[0041] Subsequently, the overall density of the clusters was quantified, and the regional threat index along the submarine cable was calculated: the submarine cable route was divided into segments of 1 kilometer each, and the spatial density values ​​of all clusters and the mean threat index of ships within each cluster were statistically analyzed. A weighted summation method (spatial density weight 0.4, ship threat index mean weight 0.6) was used to calculate the threat index of each segment, thus obtaining the regional threat index distribution along the submarine cable. This not only achieves accurate risk quantification for individual ships but also identifies high-risk areas along the submarine cable, providing core quantitative indicators for the ship threat dimension for subsequent multi-source feature fusion and wear prediction.

[0042] S2. Based on the simplified fluid-structure interaction model, the periodic bending stress load on the local micro-element of the submarine cable is calculated by combining the known spatial coordinates and mechanical parameters of the submarine cable, and the flow-induced dynamic curvature sequence varies with time.

[0043] In this embodiment, the step of calculating the periodic bending stress load on a local micro-element of the submarine cable based on a simplified fluid-structure interaction model, combined with the known spatial coordinates and mechanical parameters of the submarine cable, to obtain a time-varying fluid-induced dynamic curvature sequence includes the following steps: S21. Based on the simplified fluid-structure interaction model, the hydrodynamic effect of the ocean current field on the submarine cable is transformed into the periodic bending stress of local micro-elements.

[0044] The submarine cable is transformed into a local micro-element that can be accurately calculated, and key geometric features affecting the stress analysis are extracted to provide clear geometric boundary conditions for subsequent hydrodynamic and bending stress calculations.

[0045] Specifically, based on the geospatial coordinates of the submarine cable and the topographic data of the seabed, the route of the submarine cable is divided into uniform micro-elements. In the example, a division standard of 1 meter / segment is determined. This standard is calibrated in combination with the actual characteristics of wear on the outer sheath of the submarine cable and the engineering calculation requirements.

[0046] After the micro-element is divided, the core geometric features of each micro-element are extracted segment by segment. Each feature corresponds to a key input for subsequent stress calculations: First, the micro-element orientation angle, represented by the angle with the centerline of the submarine cable route (range 0°~180°), directly affects the decomposition direction of ocean current force and determines the effective component of lateral current force; Second, the topographic slope, calculated from seabed topographic data (accuracy 0.1°). The greater the slope, the more obvious the stress unevenness of the submarine cable micro-element, and the easier it is to generate stress concentration; Third, the micro-element burial depth, accurate to 0.1 meters. The burial depth is directly related to the value of the subsequent flow field attenuation coefficient, affecting the accuracy of actual hydrodynamic calculations; Fourth, the marking of inflection points / slope change points, with clear judgment criteria established—a route turning angle > 30° is judged as an inflection point, and a sudden change in topographic slope > 15° is judged as a slope change point. These locations are the core areas of sudden changes in submarine cable stress, and subsequent stress calculations require separate reinforcement of boundary conditions.

[0047] All extracted geometric features will be organized into a matrix of infinitesimal geometric parameters, which will serve as the basic geometric boundary input for subsequent fluid-structure interaction modeling and bending stress calculation.

[0048] Furthermore, the coordinate transformation method is used to decompose the three-dimensional flow field vector into two components: one parallel to the seabed and the other perpendicular to the seabed. Then, the lateral component perpendicular to the direction of the micro-element is further selected and projected onto the local two-dimensional plane of the micro-element (parallel to the seabed and perpendicular to the direction of the micro-element), retaining only the lateral flow force that has the core influence on the bending of the submarine cable.

[0049] Furthermore, a flow field attenuation coefficient k is introduced, the value of which is directly related to the burial depth and substrate type. Following the rule that the deeper the burial and the denser the substrate, the larger the coefficient and the more significant the flow field attenuation. Specific values ​​are calibrated based on measured data from different scenarios: for example, when the burial depth is >3 meters, k=0.2 (flow field attenuation 80%); when the burial depth is 1-3 meters, k=0.5 (flow field attenuation 50%); and when the burial depth is <1 meter, k=0.8 (flow field attenuation 20%). Simultaneously, the substrate type correction coefficient is set as follows: silt substrate > sand and gravel substrate > rock substrate (e.g., for silt substrate, the coefficient increases by 0.1 based on the corresponding burial depth), because silt substrate is denser and has a stronger hindering and attenuating effect on the flow field.

[0050] By introducing this coefficient, the actual hydrodynamic force on the submarine cable micro-element is corrected (corrected hydrodynamic force = original calculated value × attenuation coefficient), effectively avoiding the deviation in force calculation caused by ignoring the burial depth.

[0051] Ultimately, a simplified fluid-structure interaction model adapted to the underwater laying scenario of submarine cables was constructed through two simplifications, which balances computational efficiency and engineering accuracy, and also conforms to the core influencing factors of wear on the outer sheath of submarine cables.

[0052] Based on this, the simplified fluid-structure interaction model is combined with actual flow field data to calculate the transverse hydrodynamics element by element and time point by time, thus obtaining a dynamic hydrodynamic sequence. The specific implementation process consists of three steps: First, spatial matching of flow field data with micro-elements: Ocean flow field monitoring data is usually a time series of discrete monitoring points (e.g., one data point every 10 minutes). Spatial interpolation is required to match the discrete monitoring data to each submarine cable micro-element. Kriging interpolation is used, based on the coordinates of each flow field monitoring point, combined with the coordinates of the micro-element center for interpolation calculation. The flow velocity and flow direction data are accurately assigned to each micro-element to ensure that each micro-element has corresponding flow field parameters, and the interpolation error is controlled within 5 cm / s.

[0053] Secondly, the decomposition and calculation of flow forces: Based on the extracted direction angle of the micro-element, the flow field velocity and direction vector after matching are decomposed into a transverse component (core force component) perpendicular to the direction of the micro-element and an axial component (which has been ignored in the model simplification) parallel to the direction of the micro-element. Only the transverse velocity component is retained and substituted into the simplified fluid-structure interaction model. The Morison equation commonly used in engineering is used to calculate the transverse hydrodynamic force on the micro-element.

[0054] Finally, a hydrodynamic time series is generated: the lateral hydrodynamic force of each micro-element is calculated point by point according to the timestamp (consistent with the flow field monitoring data, 10 minutes / data point), and the hydrodynamic sequence of each micro-element over time is obtained. Each micro-element corresponds to a one-dimensional time series array, and the array elements are the magnitude of the lateral hydrodynamic force at each time point. This sequence fully reflects the dynamic force changes of each micro-element of the submarine cable under the action of the flow field.

[0055] S22. Based on the physical relationship between stress and curvature, the dynamic bending deformation of the submarine cable under the action of the flow field is quantified to obtain the flow-induced dynamic curvature sequence that changes with time.

[0056] The obtained micro-element hydrodynamic time series is transformed into a periodic bending stress series actually borne by the outer sheath. The key is to correct the stress transmission deviation caused by the composite structure of the submarine cable, ensuring that the stress calculation closely matches the actual stress condition of the outer sheath. The specific implementation process consists of two steps: First, preliminary calculation of bending stress: Combining the mechanical parameters of the submarine cable (elastic modulus E, moment of inertia I, bending strength of the outer sheath, etc., all taken from measured parameters provided by the submarine cable manufacturer, such as the elastic modulus of the polyethylene outer sheath E=0.8GPa), the classical bending stress calculation formula of mechanics of materials is used. The calculation is performed, where M is the bending moment on the micro-element (obtained by multiplying the lateral hydrodynamic force by the lever arm, which is half the length of the micro-element, i.e., 0.5 meters; since the micro-element is uniformly segmented, the force can be approximated as concentrated at the center of the micro-element), and W is the section modulus of the submarine cable (calculated by the section moment of inertia I and the maximum distance between the sections, W=I / y, where y is the distance from the center of the section to the surface of the outer sheath). Using this formula, the lateral hydrodynamic force at each time point is converted into the corresponding bending stress, resulting in a preliminary bending stress sequence for each micro-element.

[0057] Secondly, stress transfer correction for composite structures: Submarine cables have a core + outer sheath composite structure. The core is mostly steel, and its elastic modulus is much higher than that of the outer sheath (the elastic modulus of steel is about 200 GPa, which is more than 250 times that of the polyethylene outer sheath). Under actual stress, the core will bear most of the bending stress. If the total bending stress is directly regarded as the stress borne by the outer sheath, it will lead to excessive calculation errors and seriously affect the accuracy of subsequent wear prediction. Therefore, a stress transfer coefficient is introduced. Calibration based on measured stress data of submarine cable composite structures That is, only 30% of the total bending stress is extracted as the effective stress of the outer sheath, and the remaining 70% is borne by the cable core. This coefficient is verified by comparing the measured stress values ​​of the outer sheath and the cable core of the submarine cable, and the error is controlled within 8%.

[0058] After correction, the effective bending stress of the outer sheath = the total bending stress calculated in the preliminary calculation × the stress transfer coefficient. Based on this, the periodic bending stress sequence of each micro-element over time is obtained. Each value in the sequence corresponds to the actual bending stress (unit MPa) borne by the outer sheath, and exhibits a periodic change consistent with the flow field period (12-hour tidal period).

[0059] Furthermore, the bending stress sequence is transformed into a quantifiable, flow-induced dynamic curvature sequence that can be used for subsequent fusion. The core of this process is establishing a physical relationship between stress and curvature, while simultaneously achieving dimensional unification with other features through normalization. The specific implementation process consists of three steps: First, the physical relationship between bending stress and curvature is established: based on Hooke's law and the bending deformation formula in mechanics of materials, the bending stress is... With curvature The relationship between them is given by the following formula: , The corrected effective bending stress of the outer sheath is given by L, where L is the length of the micro-element (1 meter), E is the elastic modulus of the outer sheath, and I is the moment of inertia of the submarine cable section. This formula converts the bending stress value at each time point into a corresponding curvature value (unit: rad / m). A larger curvature value indicates more significant bending deformation of the submarine cable micro-element, more severe mechanical fatigue of the outer sheath, and a higher risk of wear.

[0060] Secondly, normalization of curvature: Due to differences in the outer sheath materials and cross-sectional specifications of different submarine cables, their allowable curvature (i.e., the maximum curvature to which the outer sheath does not undergo plastic deformation) varies. To achieve dimensional consistency with other subsequent characteristics (such as the ship threat index and strain values, all within the 0-1 range), normalization is required using the allowable curvature of the submarine cable's outer sheath as a threshold. Allowable curvature The allowable bending radius is determined based on the performance parameters of the outer sheath material. For example, the allowable bending radius for a polyethylene outer sheath is typically 0.02 rad / m, and the normalized calculation formula is as follows: ,like (i.e., bending deformation exceeds the allowable range), then the normalized value is taken as 1.0, and it is marked as high-risk bending deformation; if If the value is 0 (no bending deformation), then the value is taken as 0. The larger the normalized value, the closer the bending deformation is to the allowable limit, and the higher the risk of wear.

[0061] Finally, the generation and integration of the dynamic curvature sequence: The normalized curvature values ​​of each micro-element are sorted by timestamp to obtain the flow-induced dynamic curvature sequence of each local micro-element of the submarine cable over time. This sequence fully reflects the dynamic bending deformation law of each micro-element of the submarine cable under the action of the flow field, characterizing both the instantaneous degree of bending deformation and the periodic variation characteristics. It is the core physical feature for subsequent comprehensive real-time risk fusion and actual mechanical strain calculation, providing accurate quantitative indicators of flow-induced bending deformation for subsequent steps.

[0062] S3. Extract time-varying baseline vulnerability based on the cable's inherent properties and historical maintenance data, and integrate the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability to obtain a comprehensive real-time risk vector.

[0063] In one embodiment, the extraction of time-varying baseline vulnerability based on the cable's inherent properties and historical maintenance data includes the following steps: S311. Set the variables and parameters for the survival analysis algorithm.

[0064] First, the dependent variable was redefined as the duration during which the outer sheath of the submarine cable did not experience severe wear. The criteria for determining severe wear were based on submarine cable operation and maintenance specifications and wear level classification, namely, the wear depth of the outer sheath ≥ 0.5 mm or the wear level reaching moderate or above. This standard was calibrated using submarine cable maintenance historical data from the past 5 years to ensure consistency with subsequent wear predictions.

[0065] The independent variables focus on the inherent properties of the submarine cable itself, and the three types of parameters that have the most significant impact on wear susceptibility are selected. All of them are taken from the submarine cable production and laying archives to ensure the authenticity of the data: First, the laying period (accurate to the month, ranging from 0 to 20 years, covering the conventional service life of submarine cables), second, the characteristics of the outer sheath material (the core is wear resistance and elastic modulus, such as the wear resistance of polyethylene outer sheath ≥15MPa and elastic modulus 0.8~1.0GPa), and third, the laying method (divided into three categories: buried laying, exposed laying, and pipeline laying. The stress and environmental contact of the outer sheath of different methods are significantly different. Buried laying is affected by soil friction, while exposed laying is more directly affected by flow field and ship disturbance.

[0066] Secondly, for model selection and parameter calibration, the Kaplan-Meier survival analysis model was selected to adapt to truncated data (some submarine cables have not experienced severe wear and are still in service, belonging to right-truncated data). This model does not require assumptions about data distribution and is more in line with the randomness characteristics of submarine cable wear data. The model parameters were fitted and calibrated using historical maintenance data of submarine cables. Complete maintenance records of 10 submarine cables of the same type in the past 5 years were selected, and the intrinsic attributes of each micro-element of each submarine cable and the corresponding duration of no severe wear were extracted. These were substituted into the model for fitting training, and the weight coefficients of each intrinsic attribute were calculated (laying years weight 0.5, material properties weight 0.3, laying method weight 0.2, the weights are based on the gray-scale correlation analysis method, combined with the degree of correlation between each attribute and wear accidents).

[0067] Finally, the inherent vulnerability coefficient was calculated. Based on the calibrated model, the inherent vulnerability coefficient was calculated element by element (normalized to the 0~1 range). The calculation logic conformed to the influence law of the material properties: the longer the laying period, the coefficient showed a linear increasing trend (the coefficient increased by 0.05 for every additional year of laying period); the worse the material's wear resistance and the lower the elastic modulus, the higher the coefficient (the coefficient increased by 0.2 when the wear resistance strength was below 15MPa); the laying method was in the order of exposed laying > buried laying > pipeline laying, and the coefficient decreased in that order (the baseline value of the coefficient for exposed laying is 0.6, for buried laying is 0.4, and for pipeline laying is 0.2).

[0068] S312. Quantify and embed the prior knowledge in the historical maintenance data into the survival analysis algorithm, and correct the innate vulnerability coefficient to obtain the basic vulnerability coefficient that takes into account both innate attributes and the cumulative impact of acquired factors.

[0069] The inherent vulnerability coefficient only reflects the inherent properties of the submarine cable itself and does not take into account the cumulative effects of historical wear, maintenance and repair during use. Micro-elements that have experienced wear in the past may have damaged outer sheath structures, and the probability of further wear is significantly higher than that of unworn micro-elements. After repair, there are differences in the connection between the repaired part and the original structure, and the wear resistance is reduced, making them susceptible to secondary wear. Wear caused by different types of faults also has different effects on the subsequent vulnerability of micro-elements.

[0070] Therefore, this invention quantifies and embeds prior knowledge from historical maintenance data, and by formulating targeted correction rules, modifies the inherent vulnerability coefficient to obtain a basic vulnerability coefficient that takes into account both inherent attributes and the cumulative effects of acquired factors, thus achieving comprehensive vulnerability quantification. The specific implementation process is as follows: First, prior knowledge was screened and standardized. Complete historical maintenance data of submarine cables over the past 5 years was sorted out to identify three types of prior information directly related to the vulnerability of micro-elements, ensuring that the data is quantifiable and matchable: 1) past wear location data (including wear micro-element number, wear occurrence time, wear depth, and wear cause); 2) repair record data (including repair micro-element number, repair time, repair method, and repair material); and 3) fault type data (categorized into three types according to fault cause: mechanical fatigue, environmental erosion, and ship disturbance, among which mechanical fatigue faults cause the most significant damage to the micro-element structure).

[0071] Subsequently, the three types of prior information were precisely matched with the submarine cable micro-elements. By associating them with the micro-element numbers, it was ensured that the prior information of each micro-element was traceable and without omission. For micro-elements without historical maintenance records, they were marked as having no past wear, no repairs, and no faults, and no additional corrections were made thereafter.

[0072] Secondly, based on the actual operation and maintenance patterns of submarine cables and historical data statistics, three types of quantifiable correction rules were formulated. The correction range was determined by the historical wear recurrence probability to ensure that the correction logic was consistent with reality: ① Correction of micro-elements at past wear locations: For this type of micro-element, the inherent vulnerability coefficient is increased by 0.3. If there are multiple wear records (≥2 times), an additional increase of 0.1 is made, with the maximum corrected coefficient not exceeding 1.0; ② Correction of repaired micro-elements: For this type of micro-element, the inherent vulnerability coefficient is increased by 0.1. If the repair time exceeds 1 year (the repaired part is aged more), an additional increase of 0.05 is made. If it is a partial repair (only the worn part is repaired, not the entire outer sheath is replaced), an additional increase of 0.05 is made; ③ Correction of micro-elements with faults caused by mechanical fatigue: For this type of micro-element, the inherent vulnerability coefficient is increased by 0.2. If it belongs to both past wear and repaired micro-elements, the correction range is calculated cumulatively, but the corrected coefficient must be controlled to not exceed 1.0.

[0073] Finally, the basic vulnerability coefficient is calculated and verified. The aforementioned correction rules are applied element-wise to correct the inherent vulnerability coefficient. The calculation formula is: Basic vulnerability coefficient = Inherent vulnerability coefficient + Various correction values ​​(calculated cumulatively, upper limit 1.0, lower limit 0). After correction, accuracy verification is performed by comparing the corrected coefficient with historical wear recurrence patterns to ensure the goodness of fit R between the corrected coefficient and actual wear susceptibility. 2 After verification, the basic vulnerability coefficient of each micro-element is obtained if the value is ≥0.85.

[0074] S313. Calculate the time-varying baseline vulnerability using the aforementioned basic vulnerability coefficient.

[0075] A dynamic correction factor is introduced to transform the static basic vulnerability coefficient into a dynamic sequence that changes over time, constructing a time-varying baseline vulnerability to accurately match the cumulative and dynamic characteristics of wear on the outer sheath of submarine cables. The specific implementation process is as follows: First, the introduction and calculation of the time decay coefficient quantifies the cumulative effect of vulnerability over time. The time decay coefficient is calibrated based on the aging pattern of the submarine cable outer sheath and maintenance interval requirements, and it mainly includes two dimensions, both of which are dynamically updated over time: First, there is the laying age attenuation factor, which conforms to the aging law of materials. It is set that for every additional year of laying age, the time attenuation coefficient increases by 0.05. This factor is calculated cumulatively from the time when the submarine cable is laid until the current monitoring time. Second, the maintenance interval attenuation factor is designed to reflect the actual effect of maintenance and repair. It is set that the time attenuation coefficient increases by 0.03 for every 6 months since the last maintenance. If more than 3 years have passed since the last maintenance (without any maintenance), the attenuation coefficient will no longer increase and will remain at the maximum value of 0.18 (3 years × 2 × 0.03) to avoid over-correction. This factor is calculated by the difference between the maintenance time in the submarine cable maintenance record and the current monitoring time. For micro-elements without maintenance records, the calculation is accumulated from the time of completion of laying.

[0076] The final value of the time decay coefficient is the sum of the two dimensional factors, normalized to the range of 0 to 0.2, to ensure that the correction range is reasonable and does not obscure the core impact of the basic vulnerability coefficient.

[0077] Secondly, by associating micro-element numbers, for micro-elements in the high strain value region, an additional vulnerability coefficient of 0.2 is added on the basis of the fusion of the basic vulnerability coefficient and the time decay coefficient. If the micro-element is only in the high strain value region for a short period of time (duration < 1 hour), the coefficient is increased by 0.1. If it belongs to a micro-element with rapid strain growth (strain time change rate > 0.05 / hour), an additional coefficient of 0.05 is added to ensure that the correction fits the actual stress situation of the micro-element and highlights the aggravating effect of external stress on the vulnerability of the body.

[0078] Finally, a time-varying baseline vulnerability sequence is constructed, satisfying: Time-varying baseline vulnerability = (basic vulnerability coefficient + time decay coefficient) × (1 + strain high value area correction coefficient). The calculation results are normalized to the range of 0 to 1. The higher the value, the more vulnerable the micro-element is and the greater the susceptibility to wear. Subsequently, the vulnerability value of each micro-element is calculated point by point according to the timestamp, resulting in a time-varying baseline vulnerability sequence of each local micro-element of the submarine cable as a function of time. Each micro-element corresponds to a one-dimensional time series array, and the array elements contain timestamps and corresponding vulnerability values. This sequence not only retains the core influence of the intrinsic properties and subsequent maintenance, but also incorporates the dynamic influence of time accumulation effect and external stress, realizing the time-varying and refined characterization of vulnerability features.

[0079] Furthermore, the process of fusing the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability to obtain a comprehensive real-time risk vector includes the following steps: S321. Dynamically calculate the risk weights of the attention mechanism.

[0080] First, for the near-cable vessel threat index, the original output includes two types of data: regional level (1 km segments along the submarine cable) and micro-element level. The micro-element level data contains some sparse areas (some micro-elements did not have a threat index calculated due to extremely low vessel activity frequency), which need to be supplemented using spatial interpolation. The Kriging interpolation method, adapted to the linear routing characteristics of the submarine cable, is selected. Based on the vessel threat index of each effective micro-element, interpolation calculation is performed using the spatial coordinates of the micro-element to ensure that the interpolated micro-element level threat index closely matches the risk distribution pattern of the surrounding micro-elements. Finally, a micro-element level near-cable vessel threat index sequence (1 meter / segment micro-element, each micro-element corresponding to a unique threat index) is obtained, perfectly matching the submarine cable micro-element division.

[0081] Subsequently, the features were uniformly calibrated in both time and space dimensions: In the time dimension, the near-cable vessel threat index sequence, current-induced dynamic curvature sequence, and time-varying baseline vulnerability sequence were uniformly calibrated to timestamps of 10 minutes per data point. For data points with non-overlapping timestamps, linear interpolation was used to complete the data and avoid fusion errors caused by time deviations. In the spatial dimension, the submarine cable micro-element division standard of 1 meter per segment mentioned above was strictly followed to ensure that the micro-element numbers and spatial coordinates of the three types of features were completely corresponding, and each micro-element could be matched with the corresponding vessel threat index, current-induced dynamic curvature, and time-varying baseline vulnerability values.

[0082] Finally, consistency verification was performed on the three types of features after alignment. The matching degree between the micro-element number and the spatial coordinates and the synchronization of the timestamp were checked. Invalid data with matching deviation > 1 meter and time deviation > 1 minute were removed. Finally, complete synchronization of the three types of features at the micro-element level and time level was achieved.

[0083] Furthermore, traditional attention mechanisms are mostly applied to feature weight learning in deep learning, without considering the actual risk scenarios of submarine cable operation and maintenance. This can easily lead to a mismatch between weight allocation and actual risks (e.g., using fixed weights during periods of high ship disturbance, failing to highlight the core impact of ship threats). Therefore, this invention improves the traditional attention mechanism by adapting it to submarine cable scenarios. The core of this improvement is to construct a weight calculation rule that combines historical risk benchmarks with real-time scenario-based dynamic adjustments. This enables dynamic adaptation of the weights of each feature, ensuring that the weight allocation aligns with the actual wear and tear risk patterns of submarine cables.

[0084] First, the basic weight coefficients were determined, using historical risk data from the past five years of submarine cable operation and maintenance as the core basis. The percentage of wear accidents on the outer sheath of submarine cables caused by different risk factors was statistically analyzed: wear accidents caused by ship disturbance (including anchoring and navigation disturbance) accounted for 40%, wear accidents caused by flow-induced mechanical fatigue (periodic bending under the action of the flow field) accounted for 35%, and wear accidents caused by the inherent vulnerability of the submarine cable (inherent attributes + acquired accumulation) accounted for 25%. Based on this, the basic weights of the three types of characteristics were determined as follows: near-cable ship threat index 0.4, flow-induced dynamic bending degree 0.35, and time-varying baseline vulnerability 0.25. The basic weights were further verified by the gray-scale correlation analysis method to ensure that they matched the degree of correlation with actual wear accidents.

[0085] Subsequently, a real-time risk scenario factor is introduced to achieve dynamic adjustment of the weights. The triggering conditions of the scenario factor are combined with the risk thresholds of the three types of features (all calibrated to 0.7, consistent with the high-risk judgment standard mentioned above). The specific adjustment rules are as follows: ① When the near-line vessel threat index of a certain micro-element at a certain time point is >0.7, it indicates that the micro-element is in a high-threat scenario for vessels. The weight of the vessel threat is increased by 20% on the basis of the basic weight (i.e., 0.4×1.2=0.48). If the vessel threat index is >0.7 for 3 or more consecutive timestamps, it is increased by an additional 5% to avoid weight misadjustment caused by instantaneous fluctuations. ② When the flow-induced dynamic curvature is >0.7, it indicates that the micro-element is in a high-disturbance flow field scenario, and the weight of flow-induced curvature is increased by 20% (i.e., 0.35×1.2=0.42). If the curvature is also on an upward trend (difference between adjacent timestamps >0.05), it is increased by an additional 5%. ③ When the time-varying baseline vulnerability is >0.7, it indicates that the micro-element is extremely vulnerable, and the weight of the vulnerability is increased by 20% (i.e., 0.25×1.2=0.3). If the micro-element also belongs to the high strain value region, it is increased by an additional 5%.

[0086] Finally, based on the basic weights and real-time scenario factors, dynamic weight coefficients are calculated for each micro-element and each time point. During the calculation process, it is ensured that the sum of the weight coefficients of the three types of features is 1 (if multiple scenario factors are triggered, causing the sum of weights to exceed 1, normalization is used for calibration). Finally, the dynamic weight coefficients corresponding to each micro-element and each time point are obtained, realizing dynamic adaptation of different weight allocations for different risk scenarios, and ensuring the accuracy of subsequent fusion results.

[0087] S322. By combining the risk weights, the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability, a comprehensive real-time risk vector is obtained.

[0088] First, based on the micro-element level near-cable vessel threat index sequence, micro-element level current-induced dynamic curvature sequence, micro-element level time-varying baseline vulnerability sequence after feature dimension alignment, and the calculated dynamic weight coefficients (vehicle threat weight, current-induced curvature weight, and body vulnerability weight) of each micro-element at each time point, a linear weighted summation method is used for fusion calculation. The specific calculation formula is: Comprehensive instantaneous risk value = (near-cable vessel threat index × vessel threat dynamic weight) + (current-induced dynamic curvature × current-induced curvature dynamic weight) + (time-varying baseline vulnerability × body vulnerability dynamic weight).

[0089] During the calculation process, each micro-element and each timestamp are calculated independently to ensure the refinement of the fusion result. At the same time, the calculated comprehensive real-time risk value is normalized to the range of 0 to 1 (the normalization threshold is calibrated based on the measured data of submarine cable real-time risk over the past 5 years, with the maximum value being the highest historical real-time risk value and the minimum value being 0). The higher the normalized value, the greater the comprehensive real-time wear risk of the micro-element at that time point. Among them, 0 to 0.3 is low real-time risk, 0.3 to 0.7 is medium real-time risk, and 0.7 to 1.0 is high real-time risk.

[0090] Subsequently, a comprehensive real-time risk vector feature is constructed: following the cable route (from the cable's start to its end), the comprehensive real-time risk time series (risk values ​​at all timestamps) of each micro-element is used as an element of the vector, forming a one-dimensional comprehensive real-time risk vector. This vector simultaneously contains time series features and spatial distribution features: the spatial distribution features are reflected in the arrangement order of the vector elements (corresponding to the spatial location of the cable micro-element) and the numerical value of each element (corresponding to the real-time risk level of the micro-element); the time series features are reflected in the changing pattern of the risk value within each vector element over time (such as the upward / downward trend and fluctuation amplitude of the risk value).

[0091] Finally, the effectiveness of the fused integrated real-time risk vector was verified by comparing the integrated real-time risk value with the historical real-time risk records of submarine cable operation and maintenance to ensure the goodness of fit R. 2If the value is ≥0.9, the final comprehensive real-time risk vector feature will be output after successful verification.

[0092] S4. Taking the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion as the core, the long-term environmental risk trend characteristics are obtained by integrating evidence theory.

[0093] In this embodiment, extracting the near-cable risk contact heatmap features includes the following steps: S411. Register multi-source trajectories to obtain a preliminary fused ship trajectory dataset.

[0094] There is a problem of inconsistency between the spatiotemporal reference of ship AIS data and underwater acoustic positioning data. Therefore, the spatial reference is selected from the satellite positioning reference station around the submarine cable, and the time reference is uniformly adopted as UTC time to keep it synchronized with the timestamp standard of ship AIS trajectory processing, so as to avoid time deviation caused by time zone differences and different time sampling frequencies.

[0095] During the spatiotemporal alignment process, clear alignment thresholds were established: time deviation ≤ 1 minute (adapting to the conventional sampling frequency of 1 minute / sample for AIS data and 5-10 minutes / sample for acoustic positioning data, ensuring that the trajectories of the same vessel at similar time points can be matched), and spatial deviation ≤ 50 meters (considering the accuracy requirements of near-cable risk analysis for submarine cables, this deviation is negligible; if the deviation is too large, it indicates that the two types of data correspond to different vessels, and there is no meaning in fusion). The alignment operation adopts the spatiotemporal interpolation matching method, performing linear interpolation to complete data points with non-overlapping timestamps, marking data points with spatial deviations exceeding the threshold, and then removing invalid data with inconsistent spatiotemporal references (such as data with time deviation > 1 minute, spatial deviation > 50 meters, or isolated trajectory points that cannot be matched with the same vessel). Finally, a preliminary fused vessel trajectory dataset is obtained, containing key information such as vessel ID, unified timestamp, positioning coordinates in a unified coordinate system, and trajectory point confidence.

[0096] S412. Based on the ship trajectory dataset, high-precision fused trajectory data is obtained by optimizing acoustic positioning accuracy.

[0097] Underwater acoustic positioning data is easily affected by the marine environment (changes in water temperature and salinity affect the speed of sound wave propagation, and ocean currents can cause slight offsets in positioning sensors). Direct use of this data can lead to significant positioning accuracy deviations (maximum deviations can exceed 100 meters), failing to meet the accuracy requirements for near-cable risk analysis. Therefore, this invention addresses the underwater cable laying scenario by performing scenario-specific corrections to acoustic positioning accuracy, thereby improving the precision of acoustic positioning data.

[0098] Specifically, an environmental correction model adapted to the submarine cable scenario is first established. A multiple linear regression model is selected, with water temperature, salinity, and ocean current speed as input variables and acoustic positioning coordinate deviation as output variable. The model parameters are calibrated using historical positioning calibration data from submarine cable operation and maintenance. Ship trajectories that simultaneously acquire AIS data and acoustic positioning data around the submarine cable over the past three years are selected, and the coordinate deviations of the two types of data are calculated. The model is then fitted and trained with the corresponding water temperature, salinity, and ocean current data to obtain the correction coefficients, ensuring that the correction model can accurately compensate for positioning deviations under different marine environments.

[0099] Subsequently, using this environmental correction model, coordinate deviation compensation was performed on the spatiotemporally registered acoustic positioning data to obtain preliminarily corrected acoustic positioning data. To further improve accuracy, the preliminarily corrected acoustic positioning data was verified using highly reliable AIS data as a benchmark: the spatial distance between each acoustic positioning trajectory point and its corresponding AIS trajectory point was calculated. If the distance deviation was >50 meters (exceeding the accuracy threshold for near-cable risk analysis and thus deemed invalid data), the acoustic positioning point was directly removed; if the deviation was ≤50 meters, the deviation value was retained and recorded for subsequent model optimization.

[0100] Finally, by integrating the verified acoustic positioning data with the corresponding AIS data, high-precision fused trajectory data is obtained. This data retains the high reliability of AIS data while making up for the defect that AIS is easily turned off (acoustic positioning is a passive monitoring method and cannot be manually turned off), providing high-precision trajectory input for the next step of quantifying the spatial risks near the cable.

[0101] S413. Quantify the near-cable spatial risk based on the high-precision fused trajectory data to obtain the spatial risk distribution matrix.

[0102] High-precision fused trajectory data is transformed into quantifiable spatial risk values, enabling grid-level quantification of near-cable vessel contact risk and providing numerical support for subsequent heat map construction.

[0103] First, the area within 2 nautical miles (approximately 3704 meters) surrounding the submarine cable is divided into a 100m x 100m spatial grid. This range is set based on the influence of ship disturbances on the outer sheath of the submarine cable. Anchoring and navigation disturbances of ships within 2 nautical miles of the submarine cable may be transmitted to the submarine cable through seawater. Beyond this range, the disturbance impact is negligible. The 100m x 100m grid size is consistent with the grid size of the kernel density estimation, which can accurately reflect the distribution differences of spatial risks and avoid computational redundancy caused by an overly fine grid and risk distortion caused by an overly coarse grid.

[0104] After grid division, the ship contact risk value is calculated grid by grid based on high-precision fused trajectory data: The ship contact risk value of a single grid = the sum of the near-cable threat indices of all ships within the grid × the reciprocal of the spatial distance between the grid center and the centerline of the submarine cable route. Here, the sum of the ship threat indices within the grid reflects the overall threat level of ships within the grid, and the reciprocal of the spatial distance reflects the distance correlation between the grid and the submarine cable; the closer to the cable, the larger the reciprocal, and the higher the risk value.

[0105] After the calculation is completed, a grid-level spatial risk distribution matrix along the submarine cable is generated. Each element in the matrix corresponds to the risk value of a grid (normalized to the range of 0 to 1). The rows and columns of the matrix correspond to the longitude and latitude coordinates of the grid, respectively, clearly showing the distribution pattern of spatial risks around the submarine cable.

[0106] S414. Construct a heat map of near-cable risk contact using the spatial risk distribution matrix to obtain the characteristics of the heat map of near-cable risk contact.

[0107] The grid-level spatial risk distribution matrix is ​​transformed into a visualized heatmap, enabling spatial visualization of near-cable contact risks between vessels and the cable. This visually presents high-risk areas and provides intuitive support for subsequent heatmap feature extraction and wear location positioning. Specifically, using the centerline of the submarine cable route as the visual center, professional visualization tools (such as ArcGIS and Matplotlib) are employed to render the spatial risk distribution matrix and construct a near-cable risk contact heatmap.

[0108] The color gradient of the heatmap follows the principle that the higher the risk value, the darker the color. A clear color correspondence rule is set: risk values ​​of 0~0.3 correspond to light blue (low risk), 0.3~0.7 correspond to yellow (medium risk), and 0.7~1.0 correspond to dark red (high risk). The color gradient setting is combined with the calibration of historical wear data of submarine cable operation and maintenance, and 0.7 is set as the judgment standard for high-risk grids.

[0109] During the construction of the heatmap, high-risk grids with a risk value > 0.7 are marked, and a unique identifier (including grid coordinates, risk value, number of ships within the grid) is added to each high-risk grid to facilitate subsequent feature extraction and risk tracing. At the same time, key information such as the centerline of the submarine cable route, the boundary of the submarine cable protection zone, and seabed topographic inflection points are overlaid on the heatmap to ensure that the heatmap accurately matches the actual submarine cable laying scenario, intuitively reflects the spatial relationship between high-risk grids and submarine cables, facilitates the rapid identification of high-risk areas by maintenance personnel, and lays a visual foundation for the next step of heatmap feature extraction.

[0110] Furthermore, the visualized heatmap is digitized to extract feature vectors that characterize the spatial risk distribution, thus digitizing the spatial risk. The specific implementation process consists of two steps: First, the heatmap is digitized, converting the color information into corresponding risk values. Using the centerline of the submarine cable route as a reference, risk values ​​are extracted grid by grid according to the cable route (from the start to the end). The two-dimensional grid risk distribution is transformed into a one-dimensional ordered feature vector. The length of this vector is consistent with the number of grids surrounding the cable, and each element in the vector corresponds to the risk value of one grid, ensuring that the feature vector retains the spatial risk distribution order.

[0111] Secondly, statistical feature extraction: based on the one-dimensional feature vector, three core statistical features are extracted to comprehensively characterize the spatial risk distribution pattern: first, the number of high-risk grids (the total number of grids with a risk value > 0.7), reflecting the overall scale of the high-risk area; second, the location characteristics of high-risk grids (the set of center coordinates of all high-risk grids), accurately marking the spatial location of the high-risk area and providing spatial clues for wear location; and third, the average risk value of high-risk grids, reflecting the overall risk level of the high-risk area.

[0112] Finally, the one-dimensional ordered feature vector is integrated with three statistical features to form a near-cable risk contact heat map feature. This feature contains both the distribution order of spatial risks and key statistical information, realizing multi-dimensional digitization of spatial risks.

[0113] Furthermore, based on the periodic and trend characteristics of tides and sea states, the periodic erosion effect of the marine environment on the outer sheath of submarine cables is quantified through eddy identification, periodic component decomposition, and extreme value theory analysis. The periodic residual erosion index is obtained to characterize the environmental erosion risk of the outer sheath of submarine cables, providing core features of the time dimension for long-term environmental risk trend analysis.

[0114] First, considering the characteristics of the submarine cable's outer sheath material (such as the erosion resistance of the polyethylene outer sheath), the cable's laying depth, and the type of seabed sediment, a dual threshold for determining effective erosion eddies was established through fitting analysis of historical submarine cable operation and maintenance data from the past three years with corresponding sea state data. First, the vortex radius is ≥50 meters. This threshold is set based on matching the vortex influence range with the submarine cable micro-element division standard (1 meter / segment). The vortex scouring range of a radius less than 50 meters is limited, and it can only affect a small number of micro-elements. Moreover, the scouring intensity is weak and insufficient to cause significant erosion to the outer sheath of the submarine cable. Second, the vortex velocity is ≥1.5 knots (approximately 0.77 m / s). This threshold, combined with the scour resistance limit calibration of the outer sheath material, means that vortices with a velocity lower than 1.5 knots carry insufficient kinetic energy of sediment particles and cannot cause mechanical wear on the surface of the submarine cable's outer sheath. They will only produce slight water flow friction, which can be ignored.

[0115] During the screening process, the flow field vortex characteristic parameters (vortex radius, flow velocity, rotation direction, duration, etc.) in the long-term sea state data are first called up, and all identified vortices are initially screened based on the above dual thresholds. Then, the screening condition is supplemented - the vortex duration is ≥30 minutes to avoid the interference of instantaneous small-scale vortices, because the scouring effect of instantaneous vortices (duration <30 minutes) is temporary and cannot form a cumulative erosion effect. Finally, combined with the coordinates of the submarine cable route, vortex data with a distance of >1 km from the centerline of the submarine cable route are removed, because the scouring effect of vortices in this range cannot be transmitted to the submarine cable laying area.

[0116] Through multi-dimensional screening, flow field characteristic data such as eddies and backflows that have a real erosive effect on the outer sheath of submarine cables were finally extracted. At the same time, conventional ocean current data that have no erosive effect (such as uniform parallel currents and weakly disturbed currents) were removed to obtain an effective erosion sea state dataset. This dataset will be directly used as the input for subsequent long-term data periodic component decomposition.

[0117] Long-term tidal and sea state data exhibit significant dual characteristics of periodic fluctuations and long-term trends. Directly using them for erosion risk quantification can lead to interference between periodic fluctuations and long-term trends, making it impossible to accurately distinguish between the periodic erosion effects of tides and sea states and the cumulative erosion effects of long-term environmental changes (such as seawater pollution and ocean current shifts). Therefore, this invention employs the STL time series decomposition method, adapted to non-stationary and nonlinear time series data, to scientifically decompose the selected tidal and effective erosion sea state data, accurately extracting periodic and trend components. This provides targeted data support for subsequent extreme value risk quantification and long-term trend prediction. The specific implementation process consists of three steps: First, data preprocessing and alignment are performed to align effective erosion sea state data (eddies, backflow characteristics) with long-term tidal data (tidal level, tidal velocity, tidal direction) according to timestamps. At the same time, time periods with a data missing rate >5% are removed, and linear interpolation is used to complete a small number of missing data to ensure the continuity of time series data.

[0118] Secondly, component decomposition was implemented. Using the STL decomposition method, the aligned time-series data was split into three main components: First, the periodic component, which mainly includes the daily cycle of tides (approximately 12 hours, corresponding to semi-diurnal tide characteristics), the monthly cycle (approximately 29.5 days, corresponding to tidal fluctuations caused by lunar phase changes), and the seasonal cycle of sea states (approximately 3 months, corresponding to fluctuations in eddy activity and ocean current intensity caused by seasonal changes). This component directly corresponds to the periodic erosion effect of the marine environment on the outer sheath of the submarine cable. Second, the trend component, which mainly includes the erosion trend brought about by long-term environmental changes, such as the year-by-year increase in seawater salinity, the year-by-year increase in wave height, and the year-by-year increase in eddy activity. This component corresponds to the long-term cumulative erosion effect of the marine environment on the outer sheath of the submarine cable. Third, the residual component (random disturbance), which mainly includes short-term irregular disturbances caused by sudden extreme weather (such as typhoons and rainstorms). This component has no obvious pattern and the probability of occurrence is extremely low, so it has little impact on the long-term erosion trend and is therefore removed.

[0119] Finally, the periodic characteristic parameters are calculated: For each periodic component obtained by decomposition, the Fourier transform method is used to calculate its period length and amplitude: the period length accurately matches the actual periodic characteristics of tides and sea states, and the amplitude characterizes the fluctuation intensity of the periodic component (such as the amplitude of tidal velocity and the amplitude of vortex velocity). The larger the amplitude, the stronger the periodic erosion. Finally, the periodic component, the trend component and the corresponding periodic characteristic parameters are obtained.

[0120] Furthermore, the extreme sea states (such as strong tides, large eddies, and strong backflows) included in the periodic components, although occurring with low frequency, have extremely high erosion intensity per event, and are one of the core environmental factors leading to severe wear of the submarine cable outer sheath. Traditional mean and variance analyses cannot accurately quantify the erosion risk of such extreme events. Therefore, this invention uses the POT (PeakOverThreshold) model from Extreme Value Theory (EVT) to specifically analyze the probability and intensity of extreme sea states for the decomposed periodic components, quantifying their extreme erosion risk to the submarine cable outer sheath, and obtaining the periodic extreme erosion coefficient, providing a quantitative indicator of the extreme risk dimension for the subsequent construction of the residual erosion index. The specific implementation process is as follows: First, based on historical wear data of the erosion resistance limit of the submarine cable outer sheath, the extreme thresholds of each periodic component were calibrated: the extreme threshold for tidal velocity was set to ≥3 knots (approximately 1.54 m / s), the extreme threshold for vortex velocity was set to ≥2.5 knots (approximately 1.28 m / s), and the extreme threshold for wave height was set to ≥2.5 meters. These thresholds were calibrated through statistical analysis of wear accidents of the submarine cable outer sheath over the past 5 years.

[0121] Secondly, sample data exceeding the above extreme thresholds are extracted from the periodic components to form an extreme sea state sample set. At the same time, key parameters such as the occurrence time, intensity (e.g., tidal speed, eddy current speed), and duration of each extreme sample are recorded, with a focus on the temporal distribution pattern of the samples (e.g., whether they are concentrated in a certain tidal cycle or a certain season).

[0122] Subsequently, the extreme sea state sample set was substituted into the POT model to fit a generalized Pareto distribution (GPD), and the model's shape and scale parameters were estimated: the shape parameters characterize the distribution tail features of the extreme samples, and the scale parameters characterize the wave intensity of the extreme samples. The model parameters were solved using the maximum likelihood estimation method to ensure the model fitting accuracy (goodness of fit R). 2 ≥0.9).

[0123] Finally, based on the model parameters, the probability of extreme sea states (such as the number of occurrences of extreme eddies and strong tides per unit time) and the intensity quantification value are calculated. A weighted summation method is used to fuse the two, yielding the periodic extreme erosion coefficient (normalized to the 0-1 range), which satisfies: Periodic extreme erosion coefficient = Probability of extreme sea state occurrence × Intensity quantification value (the intensity quantification value is the ratio of the extreme sample intensity to the threshold; for example, a eddy current velocity of 2.5 knots corresponds to an intensity quantification value of 1.0, and 3.0 knots corresponds to 1.2, after normalization). A higher coefficient indicates a higher probability and greater intensity of extreme sea states in the periodic component, resulting in a greater risk of extreme erosion to the outer sheath of the submarine cable, accurately characterizing the extreme erosion risk features of the periodic component.

[0124] The decomposed trend component reflects the long-term cumulative erosion direction of the submarine cable's outer sheath by the marine environment. Analyzing only historical trends is insufficient for preventative maintenance of submarine cables; accurate prediction of erosion trends over a future period is necessary to provide forward-looking data support for long-term environmental risk trend analysis. Therefore, this invention, based on the trend component, employs a linear trend extrapolation strategy, combined with verification using historical submarine cable maintenance data, to predict the environmental erosion trend for the next six months, quantifying it to obtain the trend erosion coefficient, characterizing the direction and intensity of long-term environmental erosion. The specific implementation process is as follows: First, the trend components obtained from the decomposition (such as salinity change trend, wave height change trend, and eddy activity change trend) are subjected to linear fitting verification, and the goodness of fit R is calculated. 2 If R 2 A value ≥0.85 indicates that the trend component exhibits significant linear variation characteristics, making linear trend extrapolation suitable; if R... 2 If the value is less than 0.85, the local weighted regression (LOWESS) method is used to smooth the trend component, remove random disturbances, strengthen the linear trend characteristics, and then extrapolate.

[0125] Secondly, using time as the independent variable (unit: month) and the quantitative values ​​of trend components (such as average salinity, average wave height, and average eddy activity) as the dependent variable, a linear trend extrapolation model y=kx+b is constructed, where k is the trend slope (k>0 indicates an upward erosion trend, k<0 indicates a downward erosion trend, and k=0 indicates a stable trend), and b is the intercept. The model parameters are estimated using the least squares method to ensure the model's prediction accuracy (prediction error ≤10%).

[0126] Subsequently, the time points of the next 6 months are substituted into the constructed linear trend extrapolation model to predict the quantitative values ​​of each trend component for the next 6 months. At the same time, the confidence interval of the predicted values ​​(confidence level 95%) is calculated to evaluate the reliability of the prediction results and avoid extreme prediction bias.

[0127] Finally, based on the predicted trend of the trend components for the next 6 months, and combined with the weights of each trend component on the erosion of the cable's outer sheath (salinity weight 0.3, wave height weight 0.4, eddy activity weight 0.3, calibrated from historical erosion data), the trend erosion coefficient (normalized to the 0~1 range) is calculated, satisfying the following: Trend erosion coefficient = Σ (trend slope of each trend component × corresponding weight). If the coefficient > 0.5, it indicates that the erosion trend is on the rise in the next 6 months, and the larger the coefficient, the more obvious the upward trend; if the coefficient ≤ 0.5, it indicates that the erosion trend is stable or declining.

[0128] The periodic extreme erosion coefficient only characterizes the erosion risk of periodic extreme sea conditions, and the trend erosion coefficient only characterizes the long-term trend of environmental erosion. Neither can fully reflect the comprehensive cumulative erosion effect of the marine environment on the outer sheath of submarine cables when used alone. Therefore, this invention weights and fuses the two types of coefficients, while also correcting for chemical corrosion intensity based on seawater salinity and pH, to construct a time-varying periodic residual erosion index sequence. This quantifies the comprehensive erosion risk of the marine environment, providing core temporal characteristics for evidence-theory fusion, and accurately matching the cumulative characteristics of wear on the outer sheath of submarine cables. The specific implementation process is as follows: First, based on historical wear data of submarine cable outer sheaths over the past five years, the fusion weights of the periodic extreme erosion coefficient and the trend erosion coefficient were determined using grayscale correlation analysis: periodic extreme erosion coefficient weight 0.6, trend erosion coefficient weight 0.4. The weighted fusion calculation formula is: preliminary fusion index = periodic extreme erosion coefficient × 0.6 + trend erosion coefficient × 0.4. After fusion, the result is normalized to the 0~1 range to obtain the preliminary comprehensive erosion index.

[0129] Secondly, seawater salinity and pH value can chemically corrode the outer sheath of submarine cables, exacerbating the cumulative effect of mechanical erosion. Therefore, a chemical corrosion correction coefficient needs to be introduced to correct the preliminary fusion index. Based on the chemical corrosion resistance characteristics of the outer sheath materials of submarine cables (such as polyethylene and polypropylene), the calculation rules for correction coefficients are determined through experimental data: Salinity correction coefficient = 1 + 0.05 × (actual salinity - standard salinity), where the standard salinity is set at 35‰ (average salinity in shallow sea areas). For every 1‰ increase in actual salinity above the standard salinity, the correction coefficient increases by 0.05. The higher the salinity, the stronger the chemical corrosion. pH correction coefficient = 1 + 0.03 × (standard pH value - actual pH value), where the standard pH value is set at 8.2 (normal pH value of seawater). For every 0.1 decrease in actual pH value below the standard pH value, the correction coefficient increases by 0.03. The lower the pH value, the stronger the acidity and the more severe the chemical corrosion. Comprehensive chemical correction coefficient = salinity correction coefficient × pH correction coefficient. The correction coefficient ranges from 1.0 to 1.5 to ensure a reasonable correction range and avoid over-correction.

[0130] Finally, a periodic residual erosion index sequence was generated. The preliminary fusion index was multiplied by the comprehensive chemical correction coefficient to obtain the final periodic residual erosion index (0~1 range). The higher the value, the greater the comprehensive erosion risk of the marine environment to the outer sheath of the submarine cable. The periodic residual erosion index sequence was obtained by sorting by timestamp. This sequence includes the erosion characteristics of periodic extreme sea conditions, the cumulative effect of long-term environmental trends, and the influence of chemical corrosion, thus fully characterizing the comprehensive cumulative erosion risk of the marine environment to the outer sheath of the submarine cable.

[0131] Based on this, the method of obtaining long-term environmental risk trend characteristics by fusing the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion using evidence theory includes the following steps: S421. Spatial features are obtained by extending the heat map features of near-cable risk contact with the time series, and temporal features are obtained by encoding the periodic residual erosion index sequence through time series encoding. The heat map features of near-cable risk contact mainly reflect the spatial risk distribution and lack time dimension features. By combining the navigation pattern data of ships around the submarine cable (such as peak shipping periods, operation periods of working vessels, and low navigation periods), the heat map features are extended to a time series.

[0132] Specifically, time periods are divided according to time dimensions (daily, weekly, monthly), and spatial interpolation is used to supplement the risk values ​​of heat maps for different time periods, based on the ship activity patterns of the corresponding time periods. For example, during peak shipping periods (such as 8:00-18:00 daily), the risk value increment of high-risk grids is supplemented, and during off-peak shipping periods, the risk value decrement is supplemented to obtain spatial characteristics.

[0133] Considering that the periodic residual erosion index sequence is a continuous time series and cannot be directly used as input for evidence theory, it needs to be transformed into a structured and fusionable feature vector through time series encoding, while strengthening its periodic characteristics to adapt to the natural periodic patterns of the marine environment. First, a time series encoding method adapted to long time series and with low redundancy—Time Series Attention Encoding (TAE)—is selected. During the encoding process, the periodic residual erosion index sequence is divided into time windows, with the window size set to 12 hours. For each time window, three core statistical features—mean, peak, and trough—are extracted, transforming the continuous time series into a one-dimensional encoding vector. The vector length corresponds to the number of time windows, ensuring that the periodic patterns and trend characteristics of the original sequence are still preserved after encoding.

[0134] Then, a custom periodic activation function (PAF) is introduced, with the main period set to 12 hours (tidal semi-diurnal cycle) and the auxiliary period set to 3 months (sea state seasonal cycle). The amplitude range of the activation function is calibrated to 0~1. Through activation operation, the feature components in the encoding vector that match the marine environment cycle are amplified, the invalid features caused by random disturbances are suppressed, the periodic variation law of the periodic residual erosion index is strengthened, and the periodic effect of environmental erosion can be accurately captured during subsequent fusion.

[0135] Ultimately, the basic features of spatiotemporal fusion are obtained. Each time window corresponds to a set of spatial risk features (heatmap encoding vector) and temporal risk features (activated periodic residual erosion index encoding vector), achieving the initial alignment of spatial and temporal features.

[0136] S422. Based on the DS evidence theory synthesis rules, the spatial features and the temporal features are treated as two independent sources of evidence for confidence allocation and fusion to obtain long-term environmental risk trend features.

[0137] First, two core sources of evidence are identified: spatial evidence (near-cable risk contact heatmap features, including one-dimensional ordered feature vectors and high-risk grid statistical features, representing the spatial risk distribution of ship contact) and temporal evidence (periodic residual erosion index features after time series encoding and periodic activation, representing the long-term erosion risk of the marine environment). The consistency of the two sources of evidence is verified, and abnormal data with obvious confidence conflicts are removed (such as contradictory data with extremely high spatial risk but extremely low temporal erosion risk, which are judged as data errors and corrected by neighborhood feature interpolation) to ensure the reliability of the evidence sources.

[0138] Secondly, the initial reliability allocation is calibrated: the initial reliability is based on the statistical analysis of historical wear data from the past 5 years of submarine cable operation and maintenance, clarifying the correlation between the two types of evidence sources and the wear of the outer sheath of the submarine cable, and setting the initial reliability as follows: initial reliability m1=0.7 for spatial evidence sources (ship contact) and initial reliability m2=0.3 for temporal evidence sources (environmental erosion); at the same time, in order to avoid the initial reliability being too absolute, a reliability fluctuation range (±0.1) is introduced, which is finely adjusted according to the specific scenario of the submarine cable laying area (e.g., in busy shipping areas, the initial reliability of spatial evidence sources is increased to 0.75 and that of environmental erosion is decreased to 0.25; in remote areas, the opposite is true), to ensure the adaptability of the initial reliability.

[0139] Next, the comprehensive real-time risk value of each submarine cable area (divided into 100-meter grids) is extracted, and a reliability correction rule is formulated: when the comprehensive real-time risk value of a certain area is >0.7 (high real-time risk), it indicates that the current risk of the area is significant, and the reliability of the corresponding evidence source is increased by 20% (the reliability of spatial evidence source is increased to 0.7×1.2=0.84, and the reliability of temporal evidence source is increased to 0.3×1.2=0.36); when the comprehensive real-time risk value is between 0.3 and 0.7 (medium real-time risk), the reliability remains unchanged from the initial value; when the comprehensive real-time risk value is <0.3 (low real-time risk), the reliability is reduced by 10% to avoid excessively high reliability in low-risk areas leading to fusion bias; after correction, it is necessary to ensure that the sum of the reliability of the two evidence sources is 1 (if it exceeds, normalization calibration is performed). For example, after correction, the sum of the reliability of the high real-time risk area is 0.84+0.36=1.2, and after normalization, m1=0.7 and m2=0.3. At the same time, the correction coefficient is retained for subsequent result retrospective analysis.

[0140] Finally, confidence functions for two evidence sources are constructed: the spatial evidence source confidence function focuses on the spatial distribution of high-risk grids, and the temporal evidence source confidence function focuses on the periodic peak of the erosion index. The final confidence allocation results are output, clarifying the confidence values ​​of the two types of evidence sources corresponding to each submarine cable area.

[0141] Furthermore, based on the DS evidence theory synthesis rules, the reliability allocation results of spatial evidence sources and temporal evidence sources are fused to solve the problem of heterogeneous fusion of the two types of features. At the same time, the synthesis rules are modified in conjunction with the submarine cable scenario to handle reliability conflicts, and finally the long-term environmental risk values ​​of each area of ​​the submarine cable are obtained and the future trend is predicted.

[0142] First, a conflict coefficient k is introduced to quantify the degree of confidence conflict between the two types of evidence sources, satisfying: Where Ai represents the risk level proposition (low risk, medium risk, high risk); a conflict threshold of k0=0.6 is set. When k≤k0 (low conflict), the traditional DS synthesis rule is directly used for fusion; when k>k0 (high conflict), a weighted conflict correction strategy is adopted, which allocates the conflict reliability according to the initial reliability ratio of the two evidence sources to avoid fusion bias caused by conflict. For example, the initial reliability of the spatial evidence source is 0.7 and that of the temporal evidence source is 0.3. 70% of the conflict reliability is allocated to the spatial evidence source and 30% to the temporal evidence source to ensure that the fusion result conforms to the actual risk pattern.

[0143] Secondly, the reliability allocation results of the two evidence sources are substituted into the modified synthesis rules, and reliability fusion is performed region by region. The synthesis reliability of the three propositions corresponding to low risk (0~0.3), medium risk (0.3~0.7), and high risk (0.7~1.0) for each region is calculated. For example, if the high risk reliability of the spatial evidence source in a certain region is 0.7 and the high risk reliability of the temporal evidence source is 0.3, under low conflict conditions, the synthesis high risk reliability = (0.7×0.3) / (1-k). The risk level corresponding to the proposition with the highest synthesis reliability is the initial long-term environmental risk level of the region, and the synthesis reliability is the initial risk value.

[0144] Subsequently, the composite confidence level was converted into a long-term environmental risk value in the range of 0 to 1, and the normalization threshold was calibrated based on the maximum value of the periodic residual erosion index and the maximum value of the spatial risk. Simultaneously, calibration was performed using historical long-term risk data from submarine cable maintenance, and the goodness of fit between the fused risk value and the historical actual long-term risk value was calculated to ensure R... 2 If the fitting deviation of a certain region is too large (error > 10%), the fusion results of the neighboring regions are used for interpolation correction to improve the reliability of the risk value.

[0145] Finally, risk trend prediction for the next 6 months: Based on the time series of the fused long-term environmental risk values, a method of linear regression and periodic overlay is used to predict the long-term environmental risk change trend of each region in the next 6 months. Specifically, the fused risk values ​​are decomposed into periodic components and trend components. Combined with the current fused risk values, the average risk value and fluctuation range of each month in the future are predicted, and the trend judgment result (rising, falling, stable) is output to provide long-term trend support for subsequent wear and tear prediction.

[0146] In this embodiment, the fused long-term environmental risk value only reflects the risk level of each region. Further extraction of multi-dimensional trend features is needed to transform it into a standardized feature vector that comprehensively represents the spatiotemporal variation pattern of long-term environmental risk. Specifically, three core sub-features are extracted, and each sub-feature is quantified: First, the time-varying trend characteristics are analyzed using a sliding window trend analysis algorithm (with the window size set to 1 month to match the prediction period). The trend slope k of the long-term environmental risk value for each region over the next 6 months is calculated. A slope k > 0.05 / month is considered an upward trend, k < -0.05 / month is considered a downward trend, and the rest are considered a stable trend. At the same time, the volatility (standard deviation) of the risk value is calculated. A volatility > 0.1 is considered a high volatility risk, and a volatility < 0.05 is considered a low volatility risk. The trend type (coded as 0 = stable, 1 = upward, 2 = downward) and the volatility (normalized to 0~1) are integrated as a sub-feature of the time-varying trend.

[0147] Second, spatial distribution trend characteristics: Based on the long-term environmental risk values ​​of each region, a spatial clustering algorithm is used to screen out long-term high-risk areas (risk value ≥ 0.7, and the future trend is upward), long-term medium-risk areas (0.3 ≤ risk value < 0.7), and long-term low-risk areas (risk value < 0.3). The core features of high-risk areas are extracted, including the number of high-risk areas, total length (converted according to the number of submarine cable micro-elements, accurate to 1 meter), central route coordinates, and distance from submarine cable inflection points / slope change points. At the same time, the spatial change trend of high-risk areas (such as expansion, contraction, and translation) is tracked. These features are quantified into spatial distribution trend sub-features to provide long-term spatial clues for wear location.

[0148] Third, the periodic variation trend characteristics are analyzed. Combining the periodic residual erosion index cycle and the periodic characteristics of the merged risk value, the Fourier transform method is used to calculate the main period length of the long-term environmental risk value (to verify whether it matches the 12-hour tidal cycle and the 3-month sea state cycle) and the period amplitude (the larger the amplitude, the more intense the risk fluctuation within the cycle). At the same time, the frequency of the period peak is calculated (such as the number of times the high-risk peak occurs each month). The period length, amplitude (normalized to 0~1), and peak frequency are integrated as sub-features of the periodic variation trend to strengthen the characterization of the periodic effect of the marine environment.

[0149] Finally, the three sub-features are standardized and integrated. Using the submarine cable micro-element number as an index, each micro-element corresponds to a set of feature values ​​containing time, space, and periodic trends, constructing a one-dimensional long-term environmental risk trend feature vector. The length of the vector is consistent with the number of submarine cable micro-elements, and each element contains the quantified value of the three sub-features, ensuring that the feature vector is structured and standardized.

[0150] S5. Combining the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, hierarchical decision-making is used to predict the location and extent of wear on the outer sheath of the submarine cable.

[0151] In this embodiment, a finite element simulation model of a submarine cable is used as the physical framework, and measured data from submarine cable operation and maintenance is incorporated for data assimilation. This achieves trend determination by the physical model and precision determination by the measured data, accurately quantifying the actual mechanical strain distribution of each local micro-element of the submarine cable, characterizing the actual mechanical deformation of the submarine cable caused by flow field and ship disturbance, and providing core physical characteristics for predicting wear location and degree. Specifically, obtaining the actual mechanical strain distribution includes the following steps: S511. Load the flow-induced dynamic curvature sequence into the submarine cable finite element simulation model to generate the theoretical strain distribution sequence of each micro-element.

[0152] First, the mechanical parameters (elastic modulus, moment of inertia of section, bending strength of outer sheath, etc.) of each micro-element in the finite element simulation model are all based on measured parameters (e.g., elastic modulus of polyethylene outer sheath E=0.8GPa). Boundary constraints are set in combination with the type of seabed sediment (flexible constraints are added to silty sediment micro-elements, and rigid constraints are added to rock sediment micro-elements) to avoid the model from being disconnected from the previous analysis.

[0153] Secondly, the flow-induced dynamic curvature sequence consists of the normalized curvature values ​​of each micro-element over time (0~1 interval, 10 minutes / data point). During loading, the normalized values ​​must first be converted back to the actual curvature (unit: rad / m), satisfying the following: , The allowable curvature of the outer sheath is then used. Subsequently, a time-point-by-time, micro-element-by-micro-element loading method is adopted. The actual curvature of the micro-element corresponding to each time point is used as a dynamic load boundary condition to be loaded onto the corresponding micro-element of the model. The loading sequence follows the submarine cable route (from the start end to the end end) to ensure that the load distribution is consistent with the actual flow field action law.

[0154] Finally, based on the loaded finite element model, according to the material mechanics bending strain calculation formula and combined with the micro-element bending moment distribution, the theoretical mechanical strain value is calculated for each micro-element and for each time point, and finally the theoretical strain distribution sequence of each micro-element is generated. Each micro-element corresponds to a one-dimensional time series array, and the array elements are the theoretical strain values ​​of each time stamp.

[0155] S512. The simulation model is corrected through data assimilation, and the actual mechanical strain distribution is extracted using the corrected simulation model.

[0156] Measured strain distribution sequence of each micro-element First, the Ensemble Kalman Filter (EnKF) algorithm, which is adapted to nonlinear models and has high computational accuracy, was selected as the basic assimilation algorithm. The parameters were optimized for the micro-element analysis scenario of submarine cables: the number of ensemble samples was set to 50, and the assimilation step size was set to 10 minutes.

[0157] Secondly, two core physical constraints are set to ensure the rationality of the assimilation results: First, there is a strain-stress linear constraint. Based on Hooke's law, the strain value and stress value after assimilation must have a linear relationship, and the strain value must not exceed the allowable strain of the outer sheath. If the assimilation result exceeds this range, it will be automatically corrected to the allowable strain value and marked as a high-risk strain. Second, there is a strain continuity constraint, which limits the difference between the strain values ​​of adjacent micro-elements to no more than 200 micro-strains. If the difference exceeds the threshold, local smoothing is used for correction.

[0158] Subsequently, assimilation and model correction are implemented. The obtained measured strain distribution sequence (observed values) and the theoretical strain distribution sequence (predicted values) output by the finite element model are simultaneously input into the EnKF assimilation algorithm. Combined with physical constraints, assimilation calculations are performed on a micro-element and time-point basis: the deviation of the predicted values ​​is corrected by the observed values, and the key parameters of the finite element model (boundary constraint stiffness, mechanical parameter errors) are corrected in reverse. The correction rules are as follows: If the measured strain value is consistently greater than the theoretical strain value, it indicates that the model boundary constraints are too strong, and the constraint stiffness should be appropriately reduced; if the measured strain value is consistently less than the theoretical strain value, it indicates that the model's mechanical parameters are set incorrectly, and the elastic modulus should be adjusted to a reasonable range.

[0159] Finally, the assimilation results are verified and output. The accuracy of the assimilated strain calculation results is verified, and the goodness of fit R between the assimilation results and the measured strain values ​​is calculated. 2 Ensure R 2 ≥0.92 (fitting error) After verification, the corrected strain calculation results of each micro-element are output. These results retain the physical rationality of the finite element model and are consistent with the actual monitoring data.

[0160] Based on this, the strain calculation results after data assimilation and correction are digitized and characterized to extract core features that can characterize the actual mechanical deformation of each micro-element of the submarine cable, thus achieving a refined characterization of the strain distribution. The specific implementation process consists of three steps: First, using the allowable strain of the submarine cable's outer sheath as a threshold, the corrected strain calculation results are normalized a second time (to eliminate possible numerical deviations during the initial assimilation process): The normalization calculation formula satisfies: ,in Allowable strain for outer sheath (calibrated as) ),like The normalized value is taken as 1.0 and marked as plastic strain (the outer sheath has undergone irreversible deformation, and the risk of wear is extremely high); if (Negative strain caused by theoretical calculation deviation), the normalized value is set to 0 to ensure that the normalized values ​​are all in the range of 0~1. The higher the value, the more severe the mechanical deformation of the micro-element and the higher the risk of wear of the outer sheath.

[0161] Secondly, three key sub-features are extracted from the normalized strain data to comprehensively characterize the strain distribution pattern: First, there are micro-element level strain values, which are the normalized strain values ​​of each micro-element and each timestamp, organized into a micro-element level strain time series, accurately reflecting the dynamic deformation process of a single micro-element. Second, the location of high strain value areas is determined by establishing a high-value judgment standard: if the normalized strain value of three or more consecutive adjacent micro-elements is >0.7 and the duration is ≥1 hour, it is judged as a high strain value area. The center coordinates and micro-element numbers of all micro-elements in this area are extracted and marked as potential high-risk wear areas, providing direct physical clues for subsequent wear location. Thirdly, the strain-time change rate is calculated as the ratio of the strain value difference between two adjacent timestamps of each micro-element to the time interval, which characterizes the rate of strain change. Micro-elements with a change rate > 0.05 / hour are marked as micro-elements with rapid strain growth (the outer sheath wears faster and requires special attention).

[0162] Finally, the three extracted sub-features are integrated to form the actual mechanical strain distribution features of each local micro-element of the submarine cable. Using the micro-element number as an index, the micro-element-level strain time series, strain high value area markers, and strain time change rate are integrated. At the same time, all high-risk micro-elements with strain values ​​> 0.7 are individually marked, and their corresponding timestamps, strain peak values, and locations are labeled to form a complete actual mechanical strain distribution.

[0163] Based on this, the method of predicting the location and extent of wear on the outer sheath of submarine cables through hierarchical decision-making, combining the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, includes the following steps: S521. Standardize the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics to obtain a standardized feature matrix.

[0164] First, the min-max normalization method is used to normalize the actual mechanical strain distribution, the comprehensive immediate risk vector, and the long-term environmental risk trend characteristics to the 0~1 range. Then, following the submarine cable micro-element division standard, all normalized feature values ​​are integrated into a micro-element-level standardized feature matrix using the micro-element number as an index. The matrix dimension is the number of micro-elements × the total number of features, where each micro-element corresponds to a high-dimensional feature vector. Specifically, the vector includes: physical features (micro-element-level normalized strain value, strain time change rate, and high-value strain region markers), immediate risk features (comprehensive immediate risk value, risk value time fluctuation amplitude), and long-term risk features (long-term environmental risk value, time trend slope, spatial distribution markers, and periodic amplitude).

[0165] S522. The uncertainty of the standardized feature matrix is ​​quantified, and an error correction coefficient is introduced to obtain the corrected feature matrix.

[0166] Specifically, through multi-dimensional analysis, three types of errors are accurately identified and quantified: the first is the measurement error of the measured data, which mainly comes from the strain sensor, and the magnitude of the error is based on the sensor's factory precision ( The calculations are based on three main factors: 1) the measurement error of each strain-related feature; 2) the model calculation error, encompassing fluid-structure interaction models, finite element simulation models, and attention fusion models. The model's output value is compared with the measured value, and the residuals between these residuals are used to quantify the calculation error of each model's features. A larger residual indicates a higher calculation error. 3) the data assimilation fitting error, derived from the EnKF data assimilation process. This is quantified by comparing the deviation between the assimilated strain value and the measured strain value. A deviation exceeding this threshold indicates a higher error. The points marked as high-error feature points.

[0167] Subsequently, an error correction coefficient k is introduced. This coefficient is calibrated based on historical labeled data, and a gray-scale correlation analysis method is used to calculate the correlation between the error magnitude and the prediction deviation. The larger the error, the higher the weight of the correction coefficient. The correction coefficient ranges from 0.8 to 1.0; the smaller the error, the closer the correction coefficient is to 1.0 (no over-correction is needed). The specific implementation of the correction is as follows: for each feature value x, the corrected feature value x = x × k + (1-k) × x_mean, where x_mean is the mean of that feature in the same type of micro-elements (such as the same laying method or the same route segment). This weighted correction formula effectively counteracts the interference of various errors. After the correction is completed, a highly reliable corrected feature matrix is ​​constructed, and the goodness-of-fit verification (R²) is performed. 2 (≥0.93) ensures that the corrected features fit the historical labeled data, significantly improves the input accuracy of subsequent prediction models, and lays a solid foundation for location and degree prediction.

[0168] S523. Construct a hierarchical decision fusion model, and combine the hierarchical decision fusion model and the corrected feature matrix to predict the location and degree of wear on the outer sheath of the submarine cable.

[0169] The core objective of the location prediction layer is to achieve precise positioning of wear at the micro-element level. Using the corrected feature matrix as input, the improved DBSCAN algorithm of this invention accurately identifies irregularly distributed high-risk micro-element clusters, closely matching the actual characteristics of localized concentrated wear in submarine cables. Clustering parameters are calibrated based on historical wear data: the neighborhood radius is set to 5 meters (corresponding to 5 micro-elements, fitting the typical range of the wear area), and the minimum number of micro-elements is set to 3 (ensuring that clusters are continuous micro-elements, excluding interference from isolated high-risk micro-elements). Through the clustering algorithm, micro-elements with high similarity in feature values ​​(such as strain values ​​and risk values ​​both being in the high-value range) are grouped together, forming multiple micro-element clusters.

[0170] Subsequently, based on the preset wear assessment rules, high-risk micro-element clusters were selected. The assessment rules strictly followed the previously mentioned high-risk thresholds (strain value > 0.7, comprehensive immediate risk > 0.7, long-term environmental risk > 0.7), while supplementing the cluster selection criteria: the proportion of high-risk micro-elements within the cluster ≥ 70%, and the continuous length of the cluster ≥ 3 meters (corresponding to 3 micro-elements), excluding isolated high-risk micro-element clusters caused by temporary fluctuations. For the selected high-risk micro-element clusters, the geometric center method was used to calculate the cluster center position. The average coordinates of all micro-elements within the cluster were taken as the cluster center. The specific route coordinates of the submarine cable and the micro-element number corresponded to this to determine the predicted location of wear on the outer sheath of the submarine cable. The positioning accuracy was accurate to 1 meter, fully meeting the positioning requirements for submarine cable operation and maintenance.

[0171] At the same time, the high-risk area around the wear location is marked. A range of 10 meters with the cluster center as the core is set as the high-risk area around the wear location. This range is marked based on the historical wear diffusion pattern. Marking this area can provide maintenance personnel with an extended inspection range and prevent wear diffusion in advance.

[0172] Based on precise location of wear, the core of the degree prediction layer is to achieve quantitative classification of wear degree, enabling accurate quantification of mild, moderate, and severe wear. Simultaneously, it incorporates the cumulative effect of wear to improve prediction accuracy. Specifically, firstly, the classification standards for wear degree are clearly defined. Combining submarine cable operation and maintenance specifications with historical wear data, wear degree is divided into three categories, precisely matched with characteristic thresholds: Mild wear (0~0.3), corresponding to an outer sheath wear depth <0.3mm, with no obvious structural damage, only slight surface friction, requiring no immediate repair; Moderate wear (0.3~0.7), corresponding to an outer sheath wear depth of 0.3~0.7mm, with obvious scratches on the surface, localized material thinning, and a slight structural risk, requiring regular inspection and monitoring; Severe wear (0.7~1), corresponding to an outer sheath wear depth ≥0.7mm, with severely damaged material, even approaching the cable core, posing a risk of mechanical failure, requiring immediate shutdown and repair.

[0173] Subsequently, a classification prediction model was constructed using the infinitesimal feature vector of the wear prediction location as input, and a random forest classification model was selected. During model training, historical labeled data of wear on the outer sheath of the submarine cable was used as the training set. The training set included wear locations, corresponding feature values ​​(corrected multimodal features), and actual wear degrees from the past 5 years. Five-fold cross-validation was used during training to ensure the model's generalization ability.

[0174] To further improve prediction accuracy, a cumulative effect coefficient is introduced. The core of this coefficient is to quantify the cumulative effect of wear—the wear of the submarine cable outer sheath is the result of long-term accumulation. Current characteristic values ​​alone cannot fully reflect the degree of wear; it is necessary to combine the time accumulation of immediate risk with the trend accumulation of long-term risk. The specific calculation of the cumulative effect coefficient is: Cumulative Effect Coefficient = 0.6 × Immediate Risk Cumulative Value + 0.4 × Long-Term Risk Trend Cumulative Value. The immediate risk cumulative value is the sum of the comprehensive immediate risk values ​​of this micro-element over the past 3 months, and the long-term risk trend cumulative value is the product of the slope of the long-term environmental risk trend of this micro-element over the next 6 months and time (an upward trend results in a positive cumulative value, and a downward trend results in a negative one).

[0175] The correction strategy is as follows: if the cumulative effect coefficient is >0.5 and the current prediction level is mild, then adjust it to moderate; if the cumulative effect coefficient is >0.8 and the current prediction level is moderate, then adjust it to severe; if the cumulative effect coefficient is <0.2 and the current prediction level is moderate, then adjust it to mild. This correction strategy effectively makes up for the limitations of single feature prediction and makes the wear level prediction more consistent with the actual cumulative wear pattern.

[0176] Finally, the results of location and severity predictions are transformed into standardized outputs that submarine cable maintenance personnel can directly use, ensuring the accuracy, comprehensiveness, and practicality of the outputs, while providing confidence assessments to offer a scientific basis for maintenance decisions. Specifically, the outputs include three core components: First, the wear location at the micro-element level is clearly marked with specific information for each wear prediction location, including the micro-element number, the route coordinates in the WGS84 coordinate system, the corresponding submarine cable route mileage, and the location on the seabed topography. Second, the wear level is clearly marked for each predicted location (light / moderate / severe), and the specific wear depth prediction value corresponding to the level is supplemented, as well as the basis for judging the level (e.g., strain value 0.8, comprehensive immediate risk 0.75, long-term environmental risk 0.72, judged as severe wear). Third, the confidence level of the prediction results. Each prediction location and degree level corresponds to a confidence level value (0~1). The confidence level is calculated based on the reliability of the output probability of the prediction model and the feature error correction. A confidence level ≥ 0.8 is high confidence (the prediction result is reliable and maintenance should be arranged first), 0.5~0.8 is medium confidence (regular inspection and verification are required), and < 0.5 is low confidence (feature data needs to be re-checked to eliminate error interference).

[0177] In addition, the output results support visualization, which can overlay the wear location, severity level, and confidence level onto the submarine cable route map and seabed topography map, clearly showing the spatial distribution and risk level of wear. At the same time, it generates a standardized prediction report, including a summary of prediction results, key points of high-risk locations, and operation and maintenance suggestions (such as immediate shutdown and repair for severely worn locations, and monthly inspection for moderately worn locations). This provides accurate and actionable decision-making basis for the preventive operation and maintenance of submarine cables, and significantly reduces the probability of submarine cable failures and operation and maintenance costs.

[0178] Please see Figure 2 In this embodiment, to efficiently execute the multi-source data fusion method for predicting wear of submarine cable outer sheaths provided by this invention, the present invention also provides a multi-source data fusion system for predicting wear of submarine cable outer sheaths, comprising: an input device 1, an output device 2, a processor 3, and a memory 4. The input device 1, output device 2, processor 3, and memory 4 are interconnected. The memory 4 stores program instructions for executing the steps of the multi-source data fusion method for predicting wear of submarine cable outer sheaths. The multi-source data fusion system for predicting wear of submarine cable outer sheaths of this invention has a compact structure and stable performance, and can stably execute the multi-source data fusion method for predicting wear of submarine cable outer sheaths of this invention, further improving the overall applicability and practical application capability of this invention.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for predicting wear of submarine cable outer sheaths through multi-source data fusion, characterized in that, Includes the following steps: A spatiotemporal joint analysis of ship trajectories is conducted to obtain the threat index of ships near the mooring line; Based on a simplified fluid-structure interaction model, the periodic bending stress load on the local micro-element of the submarine cable is calculated by combining the known spatial coordinates and mechanical parameters of the submarine cable, and the flow-induced dynamic curvature sequence varies with time. Based on the cable's inherent properties and historical maintenance data, time-varying baseline vulnerability is extracted. The near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability are then integrated to obtain a comprehensive real-time risk vector. Based on the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion, long-term environmental risk trend characteristics are obtained by integrating evidence theory. By combining the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, hierarchical decision-making is used to predict the location and extent of wear on the outer sheath of submarine cables. The method of performing spatiotemporal joint analysis of ship trajectories to obtain the near-lined vessel threat index includes the following steps: The spatial density characteristics are obtained by quantifying the spatial distribution density of ship trajectories along the submarine cable using kernel density estimation. The DBSCAN clustering algorithm is optimized by introducing a time dimension to distinguish ship behaviors of different risk levels and obtain spatiotemporal clustering risk features; Trajectory anomaly features are extracted based on trajectory anomaly detection using short-time energy zero-crossing rate; By integrating the spatial density characteristics, the spatiotemporal clustering risk characteristics, and the trajectory anomaly characteristics, a near-cable vessel threat index is obtained.

2. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, The method, based on a simplified fluid-structure interaction model and combining the known spatial coordinates and mechanical parameters of the submarine cable, calculates the periodic bending stress load on local micro-elements of the submarine cable to obtain a time-varying fluid-induced dynamic curvature sequence, including the following steps: Based on a simplified fluid-structure interaction model, the hydrodynamic effect of the ocean current field on the submarine cable is transformed into periodic bending stress of local micro-elements. Based on the physical relationship between stress and curvature, the dynamic bending deformation of the submarine cable under the action of the flow field is quantified, and the flow-induced dynamic curvature sequence changing with time is obtained.

3. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, The extraction of time-varying baseline vulnerability based on the cable's inherent properties and historical maintenance data includes the following steps: Set the variables and parameters for the survival analysis algorithm; The prior knowledge in historical maintenance data is quantified and embedded into the survival analysis algorithm to correct the innate vulnerability coefficient, thereby obtaining the basic vulnerability coefficient that takes into account both innate attributes and the cumulative effects of acquired factors. The time-varying baseline vulnerability is calculated using the aforementioned basic vulnerability coefficient.

4. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, The process of fusing the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability to obtain a comprehensive real-time risk vector includes the following steps: Dynamically calculate the risk weights of the attention mechanism; By combining the risk weights, the near-cable vessel threat index, the current-induced dynamic curvature sequence, and the time-varying baseline vulnerability, a comprehensive real-time risk vector is obtained.

5. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, The method, which focuses on the spatial characteristics of near-cable risk contact and the temporal characteristics of periodic residual erosion, utilizes evidence theory to obtain long-term environmental risk trend characteristics, including the following steps: Spatial features are obtained by extending the heat map characteristics of near-cable risk contact with time series, and temporal features are obtained by encoding the periodic residual erosion index sequence through time series. Based on the DS evidence theory synthesis rules, the spatial features and the temporal features are treated as two independent sources of evidence for confidence allocation and fusion to obtain long-term environmental risk trend features.

6. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 5, characterized in that, Extracting the near-cable risk contact heatmap features includes the following steps: Register multi-source trajectories to obtain a preliminary fused ship trajectory dataset; Based on the aforementioned ship trajectory dataset, high-precision fused trajectory data is obtained by optimizing acoustic positioning accuracy. Based on the high-precision fused trajectory data, the spatial risk near the cable is quantified to obtain a spatial risk distribution matrix; A heat map of near-cable risk contact was constructed using a spatial risk distribution matrix to obtain the characteristics of the near-cable risk contact heat map.

7. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, The method of predicting the location and extent of wear on the outer sheath of submarine cables through hierarchical decision-making, combining the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, includes the following steps: By standardizing the actual mechanical strain distribution, the comprehensive instantaneous risk vector, and the long-term environmental risk trend characteristics, a standardized feature matrix is ​​obtained; Uncertainty quantification is performed on the standardized feature matrix, and an error correction coefficient is introduced to obtain the corrected feature matrix; A hierarchical decision fusion model is constructed, and the location and extent of wear on the outer sheath of the submarine cable are predicted by combining the hierarchical decision fusion model with the modified feature matrix.

8. The method for predicting wear of submarine cable outer sheath by multi-source data fusion according to claim 1, characterized in that, Obtaining the actual mechanical strain distribution includes the following steps: The flow-induced dynamic curvature sequence is loaded into the finite element simulation model of the submarine cable to generate the theoretical strain distribution sequence of each micro-element. The simulation model is corrected by data assimilation, and the actual mechanical strain distribution is extracted using the corrected simulation model.

9. A multi-source data fusion system for predicting wear of submarine cable outer sheath, characterized in that, The multi-source data fusion submarine cable outer sheath wear prediction system includes: an input device, an output device, a processor, and a memory, wherein the input device, output device, processor, and memory are interconnected, and the memory stores program instructions, which are used to execute the multi-source data fusion submarine cable outer sheath wear prediction method according to any one of claims 1-8.

Images