Submarine cable fault detection method and system for external damage prevention early warning

Abstract

The invention provides a submarine cable fault detection method and system for external damage prevention early warning, and relates to the technical field of fault detection, and the method comprises the steps: obtaining monitoring sensing data and external cooperative sensing data through a multi-mode data sensing network port which is connected and arranged, and building a submarine cable sensing grid; extracting a multi-modal feature vector, and injecting the multi-modal feature vector into a pre-trained recognition model; according to the external event type identification result, the event risk level and the confidence coefficient, matching and calling an active early warning path, and executing external early warning operation; and tracking the submarine cable sensing grid, reconstructing an external event identification result, determining a submarine cable fault state, and generating a detection feedback result. According to the technical scheme, early warning is achieved through the multi-modal data sensing network, the hysteresis problem caused by post-repair in a traditional fault detection method is solved, early warning and active response to submarine cable faults are achieved through construction of the multi-modal data sensing network, and the safety of submarine cables is enhanced.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, specifically to a method and system for detecting submarine cable faults for early warning of external damage. Background Technology

[0002] Transoceanic power transmission lines refer to power transmission methods that carry out electricity from one location to another over oceans or large bodies of water, including submarine cables and offshore overhead transmission lines. Traditional methods for detecting submarine cable faults mainly rely on fault location and repair after an accident occurs, which has a significant lag for transoceanic cables. When a cable fault occurs, traditional methods usually require indirect signals such as power outages and equipment alarms to identify the fault, which cannot achieve real-time monitoring. This leads to delayed fault detection, resulting in long repair cycles and high repair costs. Especially in deep-sea environments, the repair process is complex, requires huge investments in equipment and personnel, and is extremely expensive, often taking several months to complete, further affecting the continuity and security of power supply.

[0003] In summary, existing technologies suffer from technical problems because traditional fault detection is usually done after the fact, resulting in a delay in fault detection, long repair time and high cost, which further affects the safety of submarine cables. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for detecting submarine cable faults for early warning of external damage, in order to solve the technical problems in the prior art where fault detection is usually done after the fact, resulting in a lag in fault detection, long repair time and high cost, which further affects the safety of submarine cables.

[0005] To achieve the above objectives, this application provides a method and system for detecting submarine cable faults for early warning of external damage.

[0006] Firstly, this application provides a method for detecting submarine cable faults for early warning of external damage. This method is implemented through a submarine cable fault detection system for early warning of external damage. The method includes: connecting to a deployed multimodal data sensing network port to acquire monitoring and sensing data and external collaborative sensing data; performing data fusion and collaborative establishment of a submarine cable sensing grid; extracting multimodal feature vectors based on the submarine cable sensing grid, injecting them into a pre-trained recognition model, and outputting external event type identification results, event risk levels, and their confidence levels; matching and invoking an active early warning path based on the external event type identification results, event risk levels, and their confidence levels, and executing external early warning operations; tracking the submarine cable sensing grid based on the external early warning operations, reconstructing the external event identification results, determining the submarine cable fault state, and generating detection feedback results.

[0007] Optionally, the multimodal data sensing network includes at least: distributed acoustic vibration sensing, distributed temperature sensing, distributed strain sensing, a ship identification system, and a radar system.

[0008] Optionally, the monitoring and sensing data includes at least vibration signals, temperature signals, and strain signals distributed along the cable. The vibration signals are demodulated into time-domain and frequency-domain data by a distributed acoustic vibration sensing system; the temperature signals are acquired by a distributed temperature sensing system; and the strain signals are acquired by a distributed strain sensing system. The external collaborative sensing data is obtained through a ship identification system and a radar system, acquiring ship identification data and radar data. Based on the sensing locations of the sensing sources deployed on the submarine cable, acquisition time and location tags are established for the monitoring and sensing data and the external collaborative sensing data. Based on the acquisition time and location tags of the monitoring and sensing data and the external collaborative sensing data, the submarine cable sensing grid is constructed.

[0009] Optionally, the monitoring and sensing data and the external collaborative sensing data are subjected to spatiotemporal synchronization processing, and all data are mapped to a unified spatiotemporal coordinate system centered on the submarine cable. The spatiotemporally synchronized data is associated and mapped with the geographic information system coordinates of the submarine cable according to its collection location label, so as to obtain a gridded data set indexed by the cable's geographical location. Each grid node in the gridded data set contains at least vibration spectrum data, temperature data, strain data, and ship identity information, ship track information, and radar reflection feature information that match the spatial location of the corresponding node within a preset tolerance.

[0010] Optionally, for each grid node, data trajectory identification is performed in time sequence to establish a data trajectory time sequence chain; based on the data trajectory time sequence chain, spatiotemporal grid coordinate mapping is performed using trajectory coordinates, and a perception grid of time sequence nodes is constructed based on the temporal change relationship, expanding the gridded data set from a three-dimensional spatial grid to a four-dimensional spatiotemporal grid, wherein each spatiotemporal grid unit stores a multimodal data sequence of the corresponding geographical location within a preset time window, and the dynamic whole process of external event perception data is described through the four-dimensional spatiotemporal grid.

[0011] Optionally, for each grid node, frequency domain and time domain features are extracted from vibration spectrum data, temperature change rate and gradient features are extracted from temperature data, and deformation intensity and trend features are extracted from strain data to form an internal sensing feature vector of the node state; speed, heading, and bow change rate features are extracted from the ship track information associated with the node, and ship type and tonnage features are extracted from the ship identity information to form an external target feature vector of the external threat source; the internal sensing feature vector and the external target feature vector are concatenated and normalized to generate the multimodal feature vector.

[0012] Optionally, multimodal feature vectors are injected into a pre-trained recognition model. The recognition model outputs the external event type recognition result and confidence level by fusing internal perception feature vectors. The event type includes at least anchoring incidents, trawling operations, and vessel anchoring. The external event type recognition result is spatiotemporally correlated with the external target feature vector to output the recognized event risk level and its confidence level.

[0013] Optionally, the pre-trained recognition model is a hybrid model combining a convolutional neural network and a long short-term memory network, wherein the convolutional neural network is used to extract spatial local features from the multimodal feature vector, and the long short-term memory network is used to learn the dependencies of features on the time series and identify dynamic patterns of event types.

[0014] Optionally, based on the submarine cable sensing grid, the intensity and trend of the change of multimodal feature vectors in the time series are calculated; according to the intensity and trend of the change, the external event type identification results and risk level are dynamically adjusted and updated to identify the occurrence intensity trajectory of external events; when the occurrence intensity trajectory of external events increases, cable signal monitoring is performed based on the multimodal data sensing network; when vibration, strain abrupt change and temperature anomaly that conform to the characteristics of cable breakage occur, the early warning is determined to be a failure; based on the early warning failure message, the fault point is located using distributed acoustic vibration sensing data and the time domain reflection method is used to generate detection feedback results.

[0015] Secondly, this application also provides a submarine cable fault detection system for early warning of external damage, used to execute the submarine cable fault detection method for early warning of external damage as described in the first aspect. The submarine cable fault detection system for early warning of external damage includes: a data sensing module, used to connect to a deployed multimodal data sensing network port to acquire monitoring sensing data and external collaborative sensing data, and to perform data fusion and collaborative establishment of a submarine cable sensing grid; an event recognition module, used to extract multimodal feature vectors based on the submarine cable sensing grid, inject them into a pre-trained recognition model, and output external event type recognition results, event risk levels, and their confidence levels; a path matching module, used to match and call active early warning paths according to the external event type recognition results, event risk levels, and their confidence levels, and to execute external early warning operations; and a fault recognition module, used to track the submarine cable sensing grid based on the external early warning operations, reconstruct the external event recognition results, determine the submarine cable fault status, and generate detection feedback results.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By acquiring monitoring and sensing data and external collaborative sensing data through the multimodal data sensing network ports, a submarine cable sensing grid is established through data fusion and collaboration. Multimodal feature vectors are extracted from this grid and injected into a pre-trained recognition model, outputting external event type identification results, event risk levels, and their confidence levels. Based on these external event type identification results, event risk levels, and their confidence levels, an active early warning path is invoked to execute external early warning operations. Based on these external early warning operations, the submarine cable sensing grid is tracked to reconstruct the external event identification results, determine the submarine cable fault status, and generate detection feedback results. In other words, by acquiring monitoring and sensing data and external collaborative sensing data through the multimodal data sensing network, a submarine cable sensing grid is collaboratively established. Combined with a pre-trained recognition model, external events are identified, and early warning paths are actively invoked. This allows for early warnings to be issued and interventions to be proactively implemented when dredging vessels are preparing to anchor or are engaged in trawling operations, thereby preventing faults and improving the safety of submarine cables.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the submarine cable fault detection method for early warning of external damage, as described in this application.

[0020] Figure 2 This is a schematic diagram of the submarine cable fault detection system for early warning of external damage, as described in this application.

[0021] Explanation of reference numerals in the attached diagram: Data sensing module 11, event recognition module 12, path matching module 13, fault recognition module 14. Detailed Implementation

[0022] This application provides a method and system for detecting submarine cable faults for early warning of external damage. It addresses the technical problems in existing technologies where fault detection is typically reactive, leading to delays, long repair times, and high costs, thus impacting submarine cable safety. By acquiring monitoring and external collaborative sensing data through a multimodal data sensing network, a submarine cable sensing grid is collaboratively established. Combined with a pre-trained recognition model, external events are identified, and early warning pathways are proactively invoked. This allows for early warning and intervention when dredging vessels are preparing to anchor or are engaged in trawling operations, preventing faults and improving submarine cable safety.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a method for detecting submarine cable faults for early warning of external damage. The method is implemented using a submarine cable fault detection system for early warning of external damage, and specifically includes the following steps: The system connects to the deployed multimodal data sensing network ports to acquire monitoring and sensing data and external collaborative sensing data, and then performs data fusion and collaboration to establish a submarine cable sensing grid.

[0025] Furthermore, this application also includes the following steps: the multimodal data sensing network includes at least: distributed acoustic vibration sensing, distributed temperature sensing, distributed strain sensing, ship identification system and radar system.

[0026] Furthermore, this application also includes the following steps: the monitoring and sensing data includes at least vibration signals, temperature signals, and strain signals distributed along the cable, wherein the vibration signals are demodulated into time-domain and frequency-domain data by a distributed acoustic vibration sensing system, the temperature signals are acquired by a distributed temperature sensing system, and the strain signals are acquired by a distributed strain sensing system; the external collaborative sensing data is obtained through a ship identification system and a radar system, namely, ship identification data and radar data; based on the sensing location of the sensing source deployed on the submarine cable, the acquisition time and acquisition location labels of the monitoring and sensing data and the external collaborative sensing data are established; based on the acquisition time and acquisition location labels of the monitoring and sensing data and the external collaborative sensing data, the submarine cable sensing grid is constructed.

[0027] Furthermore, this application also includes the following steps: performing spatiotemporal synchronization processing on the monitoring and sensing data and the external collaborative sensing data, mapping all data to a unified spatiotemporal coordinate system centered on the submarine cable; associating and mapping the spatiotemporally synchronized data with the geographic information system coordinates of the submarine cable according to its collection location label, to obtain a gridded data set indexed by the cable's geographical location, wherein each grid node data in the gridded data set includes at least vibration spectrum data, temperature data, strain data, and ship identity information, ship track information, and radar reflection feature information that match the spatial location of the corresponding node within a preset tolerance.

[0028] Furthermore, this application also includes the following steps: for each grid node, data trajectory identification is performed in time sequence to establish a data trajectory time sequence chain; based on the data trajectory time sequence chain, spatiotemporal grid coordinate mapping is performed using trajectory coordinates, and a perceptual grid of the time sequence node is constructed based on the temporal change relationship, thereby expanding the gridded data set from a three-dimensional spatial grid to a four-dimensional spatiotemporal grid, wherein each spatiotemporal grid unit stores a multimodal data sequence of the corresponding geographical location within a preset time window, and the fourth dimension is the time dimension, thereby comprehensively describing the evolution process of external events in the spatiotemporal dimension through the four-dimensional spatiotemporal grid.

[0029] Specifically, by deploying a multimodal data sensing network along submarine cables, multiple sensors are placed at key locations on the cables, continuously collecting information about the cables themselves and the external environment. The multimodal data sensing network acquires data from different physical dimensions using various types of sensors, enabling comprehensive monitoring of the health of submarine cables. This network includes distributed acoustic vibration sensors, distributed temperature sensors, distributed strain sensors, ship identification systems, and radar systems.

[0030] Information about the cable itself, such as vibration signals, temperature signals, and strain signals, is collected through distributed acoustic vibration sensing, distributed temperature sensing, and distributed strain sensing. Vibration signals are acquired through distributed acoustic vibration sensing, demodulating them into time and frequency domain data. For example, if a section of cable is affected by ship anchoring or fishing net dragging, the vibration signal will exhibit specific frequency patterns, which can be used to identify external threats. Temperature signals are acquired through distributed temperature sensing; submarine cables may experience temperature changes due to current overload or external pressure during operation, allowing for timely detection of overloads or potential faults. Strain signals are acquired through distributed strain sensing, capturing mechanical deformations such as tension and bending in the cable. If the cable is subjected to excessive tension or external force, these strain signals are captured, and cable faults can be predicted.

[0031] External environmental data, or external collaborative sensing data, is collected through ship identification systems and radar systems. The ship identification system monitors the surrounding vessel activity in real time, analyzing vessel tracks and anchoring positions to determine potential threats to the cables. The radar system provides more precise information on surface and seabed obstacles, enabling real-time monitoring and prevention of collisions or interference between vessels and cables.

[0032] The multimodal data sensing network is based on the sensing locations of the sensor sources deployed on the submarine cable. Different types of sensors are placed at key locations along the cable to monitor its real-time condition, including data on vibration, temperature, and strain. It also monitors external environmental factors such as ships and radar reflection characteristics. Each sensor data point is appended with a timestamp and location identifier to ensure that each data point clearly indicates the time and geographical location of its acquisition.

[0033] Through spatiotemporal synchronization processing, all collected data is transformed into the same time and space coordinate system. This involves spatiotemporally synchronizing monitoring and sensing data with external collaborative sensing data, mapping all data to a unified spatiotemporal coordinate system centered on the submarine cable. Different data sources may have different acquisition times; time synchronization unifies all data from monitoring and sensing data and external collaborative sensing data into a single spatiotemporal coordinate system. The acquisition locations of various data types are not entirely the same, and sensors are installed at different cable locations. Based on the actual cable deployment location and the location information of external equipment, all data are mapped to the same spatial coordinate system. For example, if a vibration sensor is located at a specific location on the cable, such as 5 kilometers from the cable's starting point, and a ship is sailing nearby, the spatial positions of these two data points are aligned to ensure they are in the same coordinate system.

[0034] All data is mapped to a unified spatiotemporal coordinate system centered on the submarine cable, ensuring that both monitoring and sensing data, as well as external collaborative sensing data, have clear location identifiers and timestamps. Each spatiotemporally synchronized data point has a clear time and location label, specifying the time and location of data acquisition. The spatiotemporally synchronized monitoring and sensing data and external collaborative sensing data are correlated based on their acquisition location labels and the geographic information system coordinates of the submarine cable. Each grid node contains multimodal data such as vibration spectrum, temperature, and strain, and is matched with corresponding external information, such as vessel identification, track information, and radar reflection characteristics, based on the node's spatial location. Specifically, the spatial location of each sensor is pre-recorded during cable deployment, and the data from each acquisition point is correlated with GIS coordinates through its acquisition location label, accurately mapping all acquired data to the cable's geographical location. All correlated data is organized into a gridded dataset divided into multiple small grids, each corresponding to a specific geographical location. Each grid node stores data including vibration spectrum, temperature, and strain data. External collaborative sensing data is also correlated to each grid node. For example, if a ship is anchoring near a certain location, its ship identity information, ship track information, and radar reflection characteristics will be associated with that node, such as ship name, ID, route, speed, and radar reflection signal strength.

[0035] Each grid node in the gridded dataset contains multimodal data, including both cable monitoring and sensing data and external collaborative sensing data. For example, suppose a submarine cable is equipped with 50 vibration sensors, 30 temperature sensors, and 20 strain sensors, collecting data every minute. Each sensor has an accurate geographic location tag, and the GIS coordinates of the submarine cable have been pre-recorded. A vibration sensor located 10 kilometers from the cable's starting point records a signal with a vibration frequency of 250Hz, a temperature sensor records a temperature of 75°C, and a strain sensor in that area shows a strain of 2%. At the same time, a radar system detects a vessel 11 kilometers from the cable's starting point, traveling at 5 knots, with a strong anchoring signal. Through spatiotemporal synchronization processing, this data is accurately mapped to a unified spatiotemporal coordinate system, and the data is associated with the specific location of the cable according to the GIS coordinate system. All data is organized into a gridded dataset. Each grid node includes vibration spectrum data (250Hz), temperature data (75°C), strain data (2%), and external vessel information.

[0036] The monitoring data of each grid node changes continuously over time. By recording the time stamp of each data acquisition point, a time-series chain is established for each grid node. For example, a vibration sensor at a certain location may record vibration frequency data once per minute, forming a time-series chain of vibration data when sorted by time. For each grid node, all the monitoring and sensing data of that node that changes over time are connected into a time-series chain. For example, assuming that a node records different vibration frequency data at time points t1, t2, and t3, this constitutes the vibration data trajectory time-series chain for that node. Similarly, time-series chains are created for temperature data, strain data, etc. The data trajectory time-series chain represents the data sequence of a certain location changing over time, and can show the dynamic change process of the node over time.

[0037] Based on the time-series chain of the data trajectory, a spatiotemporal grid coordinate mapping is performed using the trajectory coordinates. This combines the data at each time point with its corresponding spatial location, forming a four-dimensional spatiotemporal grid. The four-dimensional spatiotemporal grid expands the original three-dimensional spatial network into a four-dimensional spatiotemporal grid, introducing a time dimension. In addition to conventional three-dimensional spatial coordinates, it incorporates a time dimension into the data grid, used to represent the dynamic changes of the cable and its surrounding environment within a specific time window. In the four-dimensional spatiotemporal grid, each spatiotemporal grid cell stores the multimodal data sequence of the corresponding geographical location within a preset time window. For example, if the preset time window is set to 1 hour, then each spatiotemporal grid cell will contain all vibration signals, temperature, strain, ship identification information, and radar reflection characteristics collected within 1 hour. Assuming vibration data at a distance of 50km along the cable, the following grid data is established based on the mapping of time and location: Time 10:00, vibration intensity 6dB, location 50km; Time 10:10, vibration intensity 7dB, location 50km; Time 10:20, vibration intensity 8dB, location 50km; Time 10:30, vibration intensity 10dB, location 50km. This four-dimensional spatiotemporal grid is used to describe the entire process of the impact of external events on the cable, including its spatial distribution and temporal variation.

[0038] Multimodal feature vectors are extracted based on the submarine cable sensing grid and injected into a pre-trained recognition model to output the external event type recognition result, event risk level and its confidence level.

[0039] Furthermore, this application also includes the following steps: for each grid node, extracting frequency domain and time domain features from vibration spectrum data, extracting temperature change rate and gradient features from temperature data, and extracting deformation intensity and trend features from strain data to form an internal sensing feature vector of the node state; extracting speed, heading, and bow change rate features from the ship track information associated with the node, and extracting ship type and tonnage features from ship identity information to form an external target feature vector of the external threat source; and concatenating and normalizing the internal sensing feature vector with the external target feature vector to generate the multimodal feature vector.

[0040] Specifically, multimodal feature vectors are extracted from each grid node in the submarine cable sensing grid, including internal sensing feature vectors and external target feature vectors. Frequency and time domain features are extracted from the vibration spectrum data. Frequency domain analysis extracts features such as peak frequency, spectral mean, and spectral variance, reflecting whether the cable is subjected to external interference. For example, a sudden increase in the amplitude of the vibration frequency may be the result of an external object colliding with or pulling on the cable. The temporal fluctuations of the vibration spectrum data are analyzed to extract features such as the signal mean, standard deviation, and peak value, reflecting the vibration intensity and stability of the cable at different time points. For example, a sudden increase in the standard deviation of the vibration signal may indicate that the cable is in an unstable operating state and has potential damage.

[0041] Temperature change rate and gradient features are extracted from temperature data. The temperature change rate is the rate at which temperature changes per unit time, measured in °C / minute. A large temperature change rate in the cable, such as greater than 1 °C / minute, may indicate that the cable is operating under overload or abnormal ambient temperature conditions. The temperature gradient represents the rate of temperature change with spatial location, reflecting the temperature difference at different locations on the cable, such as whether a certain part of the cable is overheating. The temperature gradient is obtained by calculating the temperature difference between different locations on the cable. If the temperature of a certain part of the cable is significantly higher than other parts, such as a temperature gradient greater than 0.5 °C / meter, it indicates that this part is affected by excessive current or external factors.

[0042] Deformation intensity and trend characteristics are extracted from strain data. Deformation intensity refers to the magnitude of the strain data, usually the maximum strain value or amplitude, reflecting the degree of stress on the cable. Trend characteristics are the changes in strain data over time or location, such as increasing, decreasing, or stable trends, used to determine whether the cable is gradually failing. The maximum value of the strain signal is calculated; if the strain value exceeds a certain threshold, it means the cable has experienced strong tension or compression. For example, a strain value greater than 2% indicates that the cable has been subjected to significant external tension or pressure, posing a risk of damage. Analyzing the trend of strain data over time or location reveals that if the strain value continues to increase, it means the cable is experiencing increasingly greater external forces, potentially leading to breakage.

[0043] The feature values ​​extracted from vibration spectrum data, temperature data, and strain data are merged to form a complete internal sensing feature vector for the node state. For example, the maximum peak value of the vibration frequency in the vibration spectrum data is 250Hz, the mean is 120Hz, and the variance is 15Hz. The extracted frequency domain feature vector is [250Hz, 120Hz, 15Hz]. The mean of the vibration signal is 0.1g, the standard deviation is 0.02g, and the peak value is 0.3g. The extracted time domain feature vector is [0.1g, 0.02g, 0.3g]. The temperature change rate is 0.3℃ / min, and the temperature gradient is 0.1℃ / m. The extracted temperature feature vector is [0.3℃ / min, 0.1℃ / m]. The maximum strain value is 1.2%, and the strain trend is gradually increasing. The extracted strain feature vector is [1.2%, increasing trend].

[0044] Speed, heading, and rate of change of heading are extracted from ship track information. Speed ​​is calculated using the distance and time difference between two points. For example, if a ship moves from point A to point B in one hour, and the distance between points A and B is 10 kilometers, then the speed is 10 kilometers per hour, or approximately 5.4 knots. Heading indicates the direction of the ship's movement, usually an angle relative to north. For example, a heading of 90° at a specific time indicates that the ship is heading east. The rate of change of heading is the rate at which the ship's heading angle changes over time, i.e., the speed of change of the ship's direction, used to assess whether the ship is performing a turning or changing course maneuver. For example, assuming the ship's heading is 80° and 100° over a period of time, then the rate of change of heading is (100° - 80°) / time difference.

[0045] The ship type and tonnage features are extracted from the ship's identity information. Ship type refers to the classification of the ship, such as tanker, cargo ship, fishing vessel, etc. Different types of ships have different operating methods and potential impacts. The ship's tonnage refers to the ship's carrying capacity, usually measured in tons, and is used to assess the ship's weight and size, thereby determining the physical pressure and threat it may pose to submarine cables. The speed, heading, and rate of change of heading features extracted from the ship's track information are combined with the ship type and tonnage features extracted from the ship's identity information to form an external target feature vector of the external threat source. For example, assuming the ship's speed is 10 knots, its heading is 90° (east), its rate of change of heading is 2° / min, its ship type is a tanker, and its tonnage is 50,000 tons, then the generated external target feature vector is [10 knots, 90°, 2° / min, tanker, 50,000 tons].

[0046] The internal sensing feature vector and the external target feature vector are concatenated, and the concatenated feature vector is normalized to eliminate the dimensional differences between different feature values, ensuring that each feature plays an equal role in the model. After concatenation and normalization, the resulting multimodal feature vector includes both the internal sensing feature vector of the cable's own state and the external target feature vector of the external environment, comprehensively considering the cable's health status and the influence of the surrounding environment. By combining internal and external data, the operating status of the cable in a specific environment can be accurately determined, improving the accuracy of fault prediction.

[0047] Furthermore, this application also includes the following steps: injecting multimodal feature vectors into a pre-trained recognition model, wherein the recognition model outputs external event type recognition results and confidence levels by fusing internal perception feature vectors, wherein the event types include at least anchoring incidents, trawling operations, and vessel anchoring; and spatiotemporally associating the external event type recognition results with the external target feature vectors to output the recognized event risk level and its confidence level.

[0048] Furthermore, this application also includes the following steps: the pre-trained recognition model is a hybrid model combining a convolutional neural network and a long short-term memory network, wherein the convolutional neural network is used to extract spatial local features from the multimodal feature vector, and the long short-term memory network is used to learn the dependencies of features on the time series and identify dynamic patterns of event types.

[0049] Specifically, the pre-trained recognition model is a hybrid model combining a convolutional neural network (CNN) and a long short-term memory (LSM) network. The CNN extracts spatial local features from multimodal feature vectors through multiple convolutional and pooling layers, that is, extracting potential local patterns from vibration spectra, temperature data, and strain data, capturing important features in these data. The extracted feature sequences are then input into the LSM network to learn the dependencies of features in the time series and identify dynamic patterns that change over time with events. The LSM network is suitable for capturing the time series features of ship activities. Through the combined action of the CNN and LSM networks, the recognition model extracts spatial features, while the LSM network captures temporal trends, outputting the recognition result of the external event type, including but not limited to anchor strikes, trawling operations, and ship anchoring. Each event type is accompanied by a confidence value, representing the model's confidence in recognizing that event type. For example, the recognition model outputs an anchor strike event recognition result with a confidence level of 80%.

[0050] Multimodal feature vectors include vibration spectrum features, temperature change rate, deformation intensity, and features such as ship speed and heading. These are input into a pre-trained recognition model. A convolutional neural network extracts specific frequency components, temperature change rate, deformation intensity, and other spatial local features from the vibration spectrum. A long short-term memory network learns the patterns of these features over time to identify the type of external event, such as anchor strike, trawling, or ship anchoring. The recognition model outputs the identification result of the external event type and its confidence level. For example, if the model outputs: event type is anchor strike, confidence level is 92%, it means the model is 92% confident that this is a high-risk event.

[0051] The external event type identification results are spatiotemporally correlated with the external target feature vector, which includes information such as the ship's speed, heading, and tonnage. For example, if the identification model identifies an anchor strike event and associates it with a fast-moving, large-tonnage ship, the event is judged to have a high risk level because an anchor strike from a large ship could put significant stress on the cables. Spatiotemporal correlation combines the identified event type with the external target feature vector to estimate the event's risk level. The risk level of the event and its corresponding confidence level are output.

[0052] For example, suppose that in one instance, the submarine cable is laid over an area from 0 km to 100 km, and the following data are monitored by a multimodal sensing network: at 0 km, the vibration signal is 5 dB, the temperature is 25 °C, the strain is 0.1%, the ship information is cargo ship, the speed is 12 km / s, the heading is 180°, and the radar data shows a relative distance of 5 km to the cable, with a heading of 180°; at 50 km, the vibration signal is 12 dB, the temperature is 30 °C, the strain is 0.2%, the ship information is cargo ship, the speed is 15 km / s, the heading is 190°, and the radar data shows a relative distance of 3 km to the cable, with a heading of 190°; at 100 km, the vibration signal is 15 dB, the temperature is 33 °C, the strain is 0.3%, the ship information is tugboat, the speed is 8 km / s, the heading is 200°, and the radar data shows a relative distance of 1 km to the cable, with a heading of 200°. At the cable node at 50km, the temperature data varied over time as follows: 25℃ at 10:00, 26℃ at 10:10, 28℃ at 10:20, and 30℃ at 10:30. The vibration signal data varied over time as follows: vibration intensity 6dB at 10:00, 7dB at 10:10, 8dB at 10:20, and 10dB at 10:30. A four-dimensional spatiotemporal grid was constructed, yielding a vibration spectrum of 6-10dB, a temperature of 25℃-30℃, and a strain of 0.1%-0.2% for the time window 10:00-10:30. Multimodal feature vectors were extracted from the data at each grid node. For a given grid node, the vibration signal's mid-frequency domain characteristic was a dominant frequency of 10Hz, the time domain characteristic was a maximum amplitude of 8dB, the temperature change rate was 5℃ / min, the temperature gradient was 0.1℃ / km, and the strain intensity was 0.5%, with a strain change rate of 0.1% / min. The following data was obtained through the ship identification system: ship speed 15 knots, ship heading 180°, ship type oil tanker, ship tonnage 50,000 tons. The multimodal feature vector of the cable node at 50km is as follows: the principal frequency component in the frequency domain feature of the vibration data is 10Hz, the amplitude is 6dB, and the maximum vibration value in the time domain feature is 0.8m / s². 2 The average value is 0.5 m / s 2The vibration duration was 2 seconds. The temperature change rate was 5℃ / min, indicating a rapid temperature change per unit time; the temperature gradient was 0.1℃ / km, indicating a relatively uniform temperature distribution along the seabed. The strain intensity was 0.5%, representing the cable's deformation amplitude under stress; the strain change rate was 0.1% / min, representing the rate of change of stress on the cable. Ship information included a speed of 15 knots, indicating a relative speed of approximately 27.8 km / h; a heading of 180°, indicating the ship was heading south; the ship type was an oil tanker; and the tonnage was 50,000 tons. Multimodal feature vectors were injected into the pre-trained recognition model. Given the ship type was an oil tanker, the tonnage was 50,000 tons, and it was anchoring, combined with the cable location at 50km and other signals, the output event type was "anchor strike," with a confidence level of 95%. The event occurred at a cable node 50km away; the event time was 10:15 AM on May 10th, at which time the cable was threatened by an anchor strike.

[0053] By combining convolutional neural networks and long short-term memory networks, and simultaneously extracting both spatial and temporal features of the data, the accuracy of event type identification is significantly improved. Real-time analysis of external events around the cable identifies different ship activities, provides a confidence level for each event, and offers timely fault warnings.

[0054] Based on the external event type identification results, event risk level and confidence level, the active early warning channel is matched and invoked to execute the external early warning operation.

[0055] Specifically, based on the identification results of the external event type, the event risk level, and its confidence level, a decision is made as to whether proactive early warning measures need to be activated. If the confidence level and risk level are both high, proactive early warning channels need to be activated. Proactive early warning channels refer to automatically taken proactive measures to warn and deter target vessels when an event occurs. This involves proactively invoking at least one proactive response method to warn and deter target vessels; proactive response methods include sending short message warning information compliant with IMO standards to target vessels via AIS base stations, broadcasting synthesized voice warnings via VHF radio, automatically sending SMS messages or making phone calls to the target vessel's registered number via communication modules, and controlling high-power laser devices to project warning light spots onto the target vessel. For example, if an anchor strike event is identified and the risk level is high, AIS short message broadcasting is prioritized according to a preset rule base, followed by VHF voice alarms, and finally automatic SMS / phone calls. AIS short messages are brief messages sent via AIS base stations to issue warnings or instructions to vessels. VHF radio is a commonly used frequency band for inter-ship communication. Sending voice warnings via VHF radio is a traditional and direct method of ship communication used to alert vessels to potential risks.

[0056] A pre-set early warning rule base defines the corresponding handling strategies for different event types, risk levels, and confidence combinations. Each handling strategy specifies a priority sequence for activating at least one proactive early warning channel. Based on the current identification results, the early warning rule base is queried to generate a handling strategy for the current external event. According to the priority sequence in the handling strategy, the corresponding proactive early warning channels are invoked sequentially to execute the operation. Proactive early warning channels include AIS short message broadcasting, VHF voice alarms, automatic SMS / telephone calls, and laser projection deterrence. In other words, according to the invoked handling strategy, the corresponding proactive early warning channels are invoked sequentially according to their priority sequence to execute the operation. Specifically, AIS short message broadcasting sends IMO-compliant short messages to the target vessel via AIS base stations to alert the crew of potential cable risks; VHF voice alarms send synthesized voice warnings to the target vessel via VHF radios to alert the crew of risks in cable areas; automatic SMS / telephone calls automatically send SMS messages to the crew's mobile phones or make phone calls to notify the crew of the existence of risks; and laser projection deterrence refers to projecting warning light spots onto the target vessel using a high-power laser device if the aforementioned methods fail to achieve effective communication, thereby attracting the crew's attention. For example, trawling incidents are high-risk and have a 90% confidence level, triggering early warning measures such as AIS short message broadcasts, VHF voice alarms, and laser warnings; shipwrecks / falling objects incidents are medium-risk and have an 80% confidence level, which may only trigger AIS short message broadcasts and VHF voice alarms.

[0057] When performing proactive early warning operations, actions are executed sequentially according to the priority sequence in the early warning rule base. If the first step fails to effectively attract the crew's attention, the second, third, and even fourth steps are continued until a sufficient response is obtained. The appropriate early warning pathways are automatically matched and prioritized based on the event type, risk level, and confidence level to ensure that the crew receives timely and effective warnings, reducing the probability of accidents. For example, if the identification result is a shipwreck / falling object, a high risk level, and a 90% confidence level, the matching early warning strategy according to the early warning rule base is as follows: Priority 1: Send a short message warning via AIS short message broadcast; Priority 2: Issue a voice warning via VHF voice broadcast; Priority 3: Deter the target vessel using a laser warning light spot.

[0058] Based on the external early warning operation tracking submarine cable sensing grid, the external event identification results are reconstructed, the submarine cable fault status is determined, and the detection feedback results are generated.

[0059] Furthermore, this application also includes the following steps: based on the submarine cable sensing grid, calculating the intensity and trend of the change of multimodal feature vectors in the time series; dynamically adjusting and updating the external event type identification results and risk levels according to the intensity and trend of the change, and identifying the occurrence intensity trajectory of external events; when the occurrence intensity trajectory of external events increases, monitoring cable signals according to the multimodal data sensing network, and determining the early warning failure when vibration, strain abrupt change and temperature anomaly that conform to the characteristics of cable breakage occur; based on the early warning failure message, using distributed acoustic vibration sensing data, using the time domain reflection method to locate the fault point, and generating detection feedback results.

[0060] Specifically, after an external warning is issued, the occurrence of external events continues to be monitored. Data from the submarine cable sensing grid is tracked to analyze whether the event effectively triggered a cable response. In other words, after issuing a warning, its effectiveness is determined by conducting full-cycle monitoring and analyzing the temporal changes in the intensity of the grid data to determine if the signal has weakened. The intensity changes of external events are judged by monitoring the time-series data of the grid nodes. If the intensity trajectory of the external event decreases or the signal weakens, the warning is considered valid; if the signal strength continues to increase, it indicates that the event may escalate further, requiring a reassessment of the cable's condition.

[0061] The intensity and trend of multimodal eigenvector changes over time are calculated. By analyzing changes in data such as vibration, strain, and temperature, the magnitude and trend of data changes are determined. For example, a sudden increase in vibration signals indicates that the ship has conducted a strong anchoring operation on the cable, while an increase in temperature may indicate cable overheating. Based on the intensity and trend of multimodal data changes, the identification results and risk levels of external events are dynamically adjusted. For example, if a continuous increase in vibration signals and an abnormally high temperature are detected, the event risk level is adjusted to high, and the external event is reassessed. The occurrence intensity trajectory of an external event refers to the path of intensity change of the external event over time. By monitoring and analyzing the occurrence intensity trajectory of external events, it is determined whether the impact of the event on the cable is increasing.

[0062] When the intensity of external events increases, cable signals are monitored using a multimodal data sensing network. If a signal matching cable fracture characteristics is detected, such as sudden vibration changes, rapid strain changes, or abnormal temperature, the warning is deemed to have failed, indicating a cable fault. In the event of a warning failure, the time-domain reflectometry method is used to locate the cable fault. Distributed acoustic vibration sensing data is used to detect the signal reflection time to determine the specific location of the cable fault. Based on the time delay of the reflected signal, the exact point of the cable fault is located, and a detection feedback result is generated.

[0063] Cable fracture characteristics refer to the specific changes in sensor data such as vibration, strain, and temperature when a cable breaks. For example, when a cable breaks, the vibration signal shows a drastic change, the temperature changes abnormally, and the strain signal also shows large fluctuations. Time-domain reflectometry (TD-RESP) is a technique used for fault location. It locates the fault point by transmitting a signal and measuring the reflection time, and is commonly used for cable fault detection. By transmitting an electrical signal to the cable and measuring the signal return time, the specific location of the cable break can be determined. The signal reflection time is directly proportional to the location of the fault point. Using T-RESP, an electrical signal is sent and the signal return time is measured; the signal reflection delay time is calculated, thereby determining the location of the cable fault. Fault point localization refers to determining the location of a cable fault by detecting data such as signal reflection time and vibration changes. Based on the fault location results from T-RESP, a detection feedback result is generated, which includes the specific location of the fault point and related status information.

[0064] For example, suppose a vibration signal with a characteristic frequency matching trawl operations is detected near an area where fishing vessel activity is reported. The initial signal strength is 50 dB, and after 10 minutes, the vibration signal strength continues to increase to 70 dB, with more pronounced vibration spectrum characteristics, indicating that the trawl equipment is continuously approaching and may come into contact with the cable. The initial temperature change rate is less than 0.1℃ / h, and there is still no significant change after 10 minutes (<0.1℃ / h), ruling out overheating as the cause. It is confirmed that the fishing vessel's speed is stable at 3 knots, and its course intersects with the cable route. The risk level is raised to high, the confidence level is increased to 95%, and an audible and visual warning is activated, alerting the fishing vessel via VHF. Micro-strain is detected in the cable, with a change trend of +5µε / min, initially judged to be trawl operations, with a medium risk level and an 80% confidence level. After 10 minutes, the strain change rate accelerates to +20µε / min, indicating that the cable is undergoing continuous mechanical tension and increased stress. After 20 minutes of continued monitoring, a severe, transient vibration spike was detected, with an intensity >90 dB and a duration of approximately 100 milliseconds, consistent with the characteristics of a broken cable armor steel wire. Following the vibration spike, a permanent step change occurred in the strain value, suddenly increasing by 500 µε, followed by a zero strain rate, indicating that the cable had fractured at this point and stress was released. A rapid temperature spike was detected at the fault point, indicating that the early warning had failed and the cable had failed. An automatic time-domain reflectometer was triggered to send a light pulse into the cable, and a strong Fresnel reflection peak was detected after 10.0 µs. Given that the propagation speed of light in this type of cable is approximately 200,000 km / s, the fault location was calculated to be 1000 m by multiplying the return time and the speed of light by 2. Fresnel reflection is a phenomenon of light signal reflection at points of change in the medium. The generated fault detection feedback results indicate that the cable fault point is located at 1000 meters, and the cause of the fault is severe external mechanical damage, namely trawling operation. Vibration signals and abnormal temperature data indicate that the fault occurred at this location.

[0065] By dynamically tracking external events and monitoring multimodal data in real time, risk assessments are adjusted accordingly to ensure accurate identification of the intensity and trend of events. When a cable fault occurs, high-precision fault location is achieved using time-domain reflectometry, quickly pinpointing the damaged area, reducing downtime, and improving repair efficiency. Real-time tracking of the intensity and trend of external events allows for timely assessment of changes in their impact, preventing false alarms or missed alarms, thereby improving the effectiveness of early warnings and response speed.

[0066] In summary, the submarine cable fault detection method for early warning of external damage provided in this application has the following technical effects: It acquires monitoring and sensing data and external collaborative sensing data through the connection of a deployed multimodal data sensing network port, and performs data fusion to collaboratively establish a submarine cable sensing grid; it extracts multimodal feature vectors based on the submarine cable sensing grid, injects them into a pre-trained recognition model, and outputs the external event type identification result, event risk level, and its confidence level; it matches and calls the active early warning path according to the external event type identification result, event risk level, and its confidence level, and executes external early warning operations; it tracks the submarine cable sensing grid based on the external early warning operations, reconstructs the external event identification result, determines the submarine cable fault state, and generates detection feedback results. In other words, by acquiring monitoring and sensing data and external collaborative sensing data through a multimodal data sensing network, collaboratively establishing a submarine cable sensing grid, and combining it with a pre-trained recognition model to identify external events, it actively calls the early warning path, issuing an early warning when the dredging vessel is preparing to anchor or is engaged in trawling operations, and actively intervening in its behavior, thereby preventing faults and improving the safety of submarine cables.

[0067] Example 2: Based on the same inventive concept as the submarine cable fault detection method for early warning of external damage in Example 1, this application also provides a submarine cable fault detection system for early warning of external damage. Please refer to the appendix. Figure 2 The submarine cable fault detection system for early warning of external damage includes: The data sensing module 11 is used to connect to the deployed multimodal data sensing network port to acquire monitoring and sensing data and external collaborative sensing data, and to perform data fusion and collaborative establishment of a submarine cable sensing grid; the event recognition module 12 is used to extract multimodal feature vectors based on the submarine cable sensing grid, inject them into a pre-trained recognition model, and output the external event type recognition result, event risk level and its confidence level; the path matching module 13 is used to match and call the active early warning path according to the external event type recognition result, event risk level and its confidence level, and to execute the external early warning operation; the fault recognition module 14 is used to track the submarine cable sensing grid based on the external early warning operation, reconstruct the external event recognition result, determine the submarine cable fault status, and generate detection feedback results.

[0068] Furthermore, the data sensing module 11 in the submarine cable fault detection system for early warning of external damage is also used for: the multimodal data sensing network includes at least: distributed acoustic vibration sensing, distributed temperature sensing, distributed strain sensing, ship identification system and radar system.

[0069] Furthermore, the data sensing module 11 in the submarine cable fault detection system for early warning of external damage is also used for: the monitoring and sensing data includes at least vibration signals, temperature signals, and strain signals distributed along the cable, wherein the vibration signals are demodulated into time-domain and frequency-domain data by a distributed acoustic vibration sensing system, the temperature signals are acquired by a distributed temperature sensing system, and the strain signals are acquired by a distributed strain sensing system; the external collaborative sensing data is obtained through a ship identification system and a radar system, acquiring ship identification data and radar data; establishing acquisition time and acquisition location labels for the monitoring and sensing data and the external collaborative sensing data based on the sensing location of the sensing source deployed on the submarine cable; and constructing the submarine cable sensing grid based on the acquisition time and acquisition location labels of the monitoring and sensing data and the external collaborative sensing data.

[0070] Furthermore, the data sensing module 11 in the submarine cable fault detection system for early warning of external damage is also used to: perform spatiotemporal synchronization processing on the monitoring sensing data and external collaborative sensing data, and map all data to a unified spatiotemporal coordinate system centered on the submarine cable; and associate the spatiotemporally synchronized data with the geographic information system coordinates of the submarine cable according to its collection location label to obtain a gridded data set indexed by the cable's geographical location. Each grid node in the gridded data set contains at least vibration spectrum data, temperature data, strain data, and ship identity information, ship track information, and radar reflection feature information that match the spatial location of the corresponding node within a preset tolerance.

[0071] Furthermore, the data sensing module 11 in the submarine cable fault detection system for early warning of external damage is also used to: identify the data trajectory of each grid node in time sequence and establish a data trajectory time sequence chain; based on the data trajectory time sequence chain, perform spatiotemporal grid coordinate mapping with trajectory coordinates, construct the sensing grid of the time sequence node with the time sequence change relationship, and expand the gridded data set from a three-dimensional spatial grid to a four-dimensional spatiotemporal grid, wherein each spatiotemporal grid unit stores the multimodal data sequence of the corresponding geographical location within a preset time window, and describes the dynamic whole process of external event sensing data through the four-dimensional spatiotemporal grid.

[0072] Furthermore, the event recognition module 12 in the submarine cable fault detection system for early warning of external damage is also used to: extract frequency domain and time domain features from vibration spectrum data, extract temperature change rate and gradient features from temperature data, and extract deformation intensity and trend features from strain data for each grid node, thereby forming an internal perception feature vector of the node state; extract speed, heading, and bow change rate features from the ship track information associated with the node, and extract ship type and tonnage features from ship identity information, thereby forming an external target feature vector of the external threat source; and concatenate and normalize the internal perception feature vector and the external target feature vector to generate the multimodal feature vector.

[0073] Furthermore, the event recognition module 12 in the submarine cable fault detection system for early warning of external damage is also used to: inject multimodal feature vectors into a pre-trained recognition model, wherein the recognition model outputs the external event type recognition result and confidence level by fusing internal perception feature vectors, wherein the event type includes at least anchor strike, trawling operation, and ship anchoring; and spatiotemporally correlate the external event type recognition result with the external target feature vector to output the identified event risk level and its confidence level.

[0074] Furthermore, the event recognition module 12 in the submarine cable fault detection system for early warning of external damage is also used for: the pre-trained recognition model is a hybrid model combining a convolutional neural network and a long short-term memory network, wherein the convolutional neural network is used to extract spatial local features from the multimodal feature vector, and the long short-term memory network is used to learn the dependence of features on the time series and identify the dynamic pattern of event types.

[0075] Furthermore, the fault identification module 14 in the submarine cable fault detection system for external damage early warning is also used for: calculating the intensity and trend of the change of multimodal feature vectors in the time series based on the submarine cable sensing grid; dynamically adjusting and updating the external event type identification results and risk levels according to the intensity and trend of the change, and identifying the occurrence intensity trajectory of external events; when the occurrence intensity trajectory of external events increases, monitoring cable signals according to the multimodal data sensing network, and determining that the early warning has failed when vibration, strain abrupt change and temperature anomaly that conform to the characteristics of cable breakage occur; based on the early warning failure information, relying on distributed acoustic vibration sensing data, using the time domain reflection method to locate the fault point and generate detection feedback results.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1The submarine cable fault detection method and specific examples for external damage early warning in Example 1 are also applicable to the submarine cable fault detection system for external damage early warning in this embodiment. Through the foregoing detailed description of the submarine cable fault detection method for external damage early warning, those skilled in the art can clearly understand the submarine cable fault detection system for external damage early warning in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0077] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0078] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for detecting submarine cable faults for early warning of external damage, characterized in that, include: The system connects to the deployed multimodal data sensing network ports to acquire monitoring and sensing data and external collaborative sensing data, and performs data fusion and collaboration to establish a submarine cable sensing grid. Multimodal feature vectors are extracted based on the submarine cable sensing grid and injected into a pre-trained recognition model to output the external event type recognition result, event risk level and its confidence level; Based on the external event type identification results, event risk level and confidence level, the active early warning channel is invoked to execute the external early warning operation; Based on the external early warning operation tracking submarine cable sensing grid, the external event identification results are reconstructed, the submarine cable fault status is determined, and the detection feedback results are generated.

2. The submarine cable fault detection method for early warning of external damage according to claim 1, characterized in that, The multimodal data sensing network includes at least: distributed acoustic vibration sensing, distributed temperature sensing, distributed strain sensing, a ship identification system, and a radar system.

3. The submarine cable fault detection method for early warning of external damage according to claim 2, characterized in that, The process of acquiring monitoring and sensing data and external collaborative sensing data, and then fusing and collaboratively establishing a submarine cable sensing grid, includes: The monitoring and sensing data includes at least vibration signals, temperature signals, and strain signals distributed along the cable. The vibration signals are demodulated into time-domain and frequency-domain data by a distributed acoustic vibration sensing system, the temperature signals are acquired by a distributed temperature sensing system, and the strain signals are acquired by a distributed strain sensing system. The external collaborative sensing data refers to ship identification data and radar data acquired through ship identification systems and radar systems; Based on the sensing location of the sensor source deployed on the submarine cable, establish the acquisition time and acquisition location tags for the monitoring sensing data and external collaborative sensing data; The submarine cable sensing grid is constructed based on the collection time and location labels of the monitoring and sensing data and the external collaborative sensing data.

4. The submarine cable fault detection method for early warning of external damage according to claim 3, characterized in that, Constructing the submarine cable sensing grid includes: The monitoring and sensing data and the external collaborative sensing data are processed in a spatiotemporal synchronization manner, and all data are mapped to a unified spatiotemporal coordinate system centered on the submarine cable. The spatiotemporally synchronized data is associated and mapped with the geographic information system coordinates of the submarine cable based on its collection location label, to obtain a gridded data set indexed by the cable's geographical location. Each grid node in the gridded data set contains at least vibration spectrum data, temperature data, strain data, and ship identity information, ship track information, and radar reflection feature information that match the spatial location of the corresponding node within a preset tolerance.

5. The submarine cable fault detection method for early warning of external damage according to claim 4, characterized in that, Constructing the submarine cable sensing grid also includes: For each grid node, data trajectory identification is performed in time sequence to establish a data trajectory time sequence chain; Based on the data trajectory time-series chain, the trajectory coordinates are used to perform spatiotemporal grid coordinate mapping, and the temporal change relationship is used to construct the perception grid of the time-series nodes. The gridded data set is expanded from a three-dimensional spatial grid to a four-dimensional spatiotemporal grid, where each spatiotemporal grid unit stores the multimodal data sequence of the corresponding geographical location within a preset time window. The four-dimensional spatiotemporal grid describes the dynamic whole process of external event perception data.

6. The submarine cable fault detection method for early warning of external damage according to claim 4, characterized in that, Extracting multimodal feature vectors based on the submarine cable sensing grid includes: For each grid node, frequency domain and time domain features are extracted from vibration spectrum data, temperature change rate and gradient features are extracted from temperature data, and deformation intensity and trend features are extracted from strain data to form an internal sensing feature vector of the node state. The features of speed, heading, and rate of change of bow direction are extracted from the ship track information associated with the nodes, and the features of ship type and tonnage are extracted from the ship identity information to form the external target feature vector of the external threat source. The internal perception feature vector and the external target feature vector are concatenated and normalized to generate the multimodal feature vector.

7. The submarine cable fault detection method for early warning of external damage according to claim 6, characterized in that, The pre-trained identification model outputs external event type identification results, event risk level, and their confidence level, including: Multimodal feature vectors are injected into a pre-trained recognition model, which outputs external event type recognition results and confidence scores by fusing internal perception feature vectors. The event types include at least anchor strikes, trawling operations, and vessel anchoring. The external event type identification result is spatiotemporally correlated with the external target feature vector to output the identified event risk level and its confidence level.

8. The submarine cable fault detection method for early warning of external damage according to claim 7, characterized in that, The pre-trained recognition model is a hybrid model combining a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract spatial local features from the multimodal feature vectors, and the long short-term memory network is used to learn the dependencies of features in the time series and identify dynamic patterns of event types.

9. The submarine cable fault detection method for early warning of external damage according to claim 5, characterized in that, Based on the aforementioned external early warning operation tracking submarine cable sensing grid, the external event identification results are reconstructed to determine the submarine cable fault status and generate detection feedback results, including: Based on the submarine cable sensing grid, the intensity and trend of the changes in multimodal feature vectors over time are calculated; Based on the intensity and trend of change, the identification results and risk levels of the external event types are dynamically adjusted and updated to identify the trajectory of the occurrence intensity of external events; When the intensity trajectory of external events increases, cable signals are monitored based on a multimodal data sensing network. If vibrations, sudden strain changes, and temperature anomalies that match the characteristics of cable breakage occur, the early warning is deemed to have failed. Based on the early warning failure message, the fault point is located using distributed acoustic vibration sensing data and the time-domain reflectometry method, and the detection feedback result is generated.

10. A submarine cable fault detection system for early warning of external damage, characterized in that, The steps for implementing the submarine cable fault detection method for early warning of external damage as described in any one of claims 1 to 9, wherein the submarine cable fault detection system for early warning of external damage includes: The data sensing module is used to connect to the deployed multimodal data sensing network ports to acquire monitoring and sensing data and external collaborative sensing data, and to perform data fusion and collaborative establishment of a submarine cable sensing grid. The event recognition module is used to extract multimodal feature vectors based on the submarine cable sensing grid, inject them into the pre-trained recognition model, and output the external event type recognition result, event risk level and its confidence level. The path matching module is used to match and call the active early warning path based on the external event type identification result, event risk level and its confidence level, and to execute the external early warning operation; The fault identification module is used to track the submarine cable sensing grid based on the external early warning operation, reconstruct the external event identification results, determine the fault status of the submarine cable, and generate detection feedback results.

Images