Submarine landslide multi-source data dynamic real-time fusion method

Abstract

The invention discloses a submarine landslide multi-source data dynamic real-time fusion method, which belongs to the technical field of multi-source data fusion, and comprises the steps of synchronously starting an edge preprocessing module and monitoring data quality in real time during data acquisition. In-region fast fusion is achieved through edge node dynamic fusion, shore-based center deep fusion adopts parallel computing, it is ensured that the total delay of data from collection to output is reduced through three-layer fusion collaboration, different ocean scenes and digital twinborn requirements can be met, fusion precision is improved, a dynamic weight self-adaption algorithm is adopted, and the fusion efficiency is improved. The weight is adjusted in real time in a three-dimensional mode based on requirements, scenes and data quality, and fusion deviation caused by a single weight is avoided.

Description

Technical Field

[0001] This invention belongs to the field of multi-source data fusion technology, specifically referring to a method for dynamic real-time fusion of multi-source data on submarine landslides. Background Technology

[0002] Submarine landslides, as a typical marine geological hazard, are widely distributed on continental slopes, submarine canyons, and areas with active oil and gas development. They are characterized by their suddenness, large scale, and high destructive power. They can not only directly destroy marine infrastructure such as submarine cables, pipelines, and platforms, but also trigger tsunamis, posing a serious threat to the safety of coastal areas. This places higher demands on the early identification, dynamic monitoring, and risk warning capabilities for submarine landslides.

[0003] However, existing methods for dynamic real-time fusion of source data still have certain shortcomings. Existing technologies rely on centralized processing architectures, requiring data to be directly transmitted from the acquisition end to the central server for fusion analysis. This results in high transmission latency and concentrated computational load, making it difficult to meet the dynamic needs of marine scenarios. They also lack edge-side preprocessing mechanisms and fail to design differentiated fusion strategies for different marine scenarios, leading to overall slow response speed and limited fusion accuracy. Furthermore, they employ fixed weights or simple averaging strategies for fusing multi-source data without dynamically adjusting weight allocation based on task requirements, real-time scenario status, and data quality. The lack of a unified three-dimensional spatiotemporal reference system results in significant deviations in timestamp synchronization and spatial coordinate alignment among multi-source data, leading to poor data consistency. Therefore, a dynamic real-time fusion method for multi-source data of submarine landslides is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic real-time fusion of multi-source data on submarine landslides, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic real-time fusion of multi-source data on submarine landslides, comprising the following steps:

[0006] S1. During data acquisition, the edge preprocessing module starts synchronously to monitor data quality in real time.

[0007] S2. After receiving real-time data transmitted from the acquisition terminal, the edge node completes the time alignment and spatial matching of the data through the spatiotemporal synchronization calibration module.

[0008] S3. After receiving dynamic data slices from all edge nodes, the shore-based central hub starts a multi-dimensional correlation and fusion engine for deep fusion.

[0009] S4. The feedback optimization scheduling module collects three types of data in real time and performs optimization analysis.

[0010] S5. The fusion system uses the twin adapter interface to perceive the simulation stage of the digital twin in real time and dynamically adjust the output strategy.

[0011] S6. During the digital twin simulation process, a simulation error report is generated and fed back to the fusion system.

[0012] Preferably, in step S1, the edge preprocessing module identifies data outside the preset normal data range, detects data mutations through first-order difference, marks them as abnormal data and stores them separately, and performs format standardization conversion on the normal data, converting the satellite remote sensing JPG image data into GeoTIFF format raster data, embedding coordinate information into image metadata; converting the sensor's CSV format time series data into a JSON format time series array; the edge preprocessing module extracts key features, and the feature data is transmitted synchronously with the original data, and the preprocessed normal data, feature data, and abnormal data marking information are transmitted in real time to the nearest edge computing node via 5G.

[0013] Preferably, in step S2, a lightweight sliding window weighted fusion algorithm is invoked through spatiotemporal synchronization calibration to perform preliminary fusion of multi-source data within the window. Within the time window [t-1, t], there are n data sources, and the observation value of each data source i at time t is... The data quality level is The data is then merged using a weighted average, with the weights representing the data quality level. The result is as follows:

[0014] ,

[0015] In the formula, This represents the merged data value.

[0016] Preferably, in step S2, the fused data values ​​are obtained, and combined with the regional hydrological data, the correlation coefficient between displacement and flow velocity is calculated, which is achieved as follows:

[0017] ,

[0018] In the formula, A coefficient representing the relationship between displacement and flow velocity. This represents the merged displacement data. It represents the fused flow velocity data and generates regional-level dynamic data slices;

[0019] After receiving the data credibility feedback, the acquisition terminal automatically adjusts the acquisition strategy: when the data credibility is greater than or equal to the preset threshold, the current sampling frequency is maintained; when the credibility is within the preset threshold range, the sampling frequency is increased; when the credibility is less than the preset threshold, the sampling frequency is increased, and a data supplementation request is sent to the shore base.

[0020] Preferably, in step S3, the multi-dimensional correlation fusion engine deeply mines the coupling relationships between different regions and different types of data; the scene adaptive weight engine combines the current task instructions of the digital twin, real-time scene data, and data credibility to calculate the real-time weights of each type of data, thus achieving the following:

[0021] ,

[0022] In the formula, This represents the weight of data type i over time t. This indicates the weight of the task instruction on the data type. This indicates the impact of real-time scene data on data type i. This represents the data reliability of data type i. These represent coefficients that balance task instructions, real-time scenario data, and data reliability, respectively. This represents the credibility decay coefficient, enabling deep fusion across regions and dimensions, generating a dynamic fusion model across the entire domain, as follows:

[0023] ,

[0024] In the formula, This represents a global dynamic fusion model. This represents the weight of data type i at time t. This represents the historical mean of data type i. This represents the historical standard deviation of data type i.

[0025] Preferably, in step S3, the global dynamic fusion model is transmitted to the digital twin system through the twin adaptation interface. The interface reads the running status of the digital twin in real time. When the digital twin performs initial modeling, it outputs complete global fusion data; when the digital twin performs dynamic evolution simulation, it outputs incremental fusion data updated according to a preset time; when the digital twin triggers an early warning threshold, it outputs the core fusion data related to the early warning and marks key data items in red.

[0026] Preferably, in step S4, the feedback optimization scheduling module collects three types of data in real time: the deep fusion results of the shore-based central hub, the feedback information of the digital twin system, and the operating status of the collection equipment and edge nodes; it performs optimization analysis based on the collected data, generates optimization instructions and issues them; for unreasonable weights, it issues new weight parameters to the shore-based central hub and edge nodes; for missing data, it issues supplementary collection instructions to the collection end; for poor algorithm adaptability, it issues algorithm switching instructions to edge nodes and the shore-based central hub; and for excessive node load, it issues load sharing instructions to neighboring edge nodes.

[0027] Preferably, in step S5, the fusion system senses the simulation stage of the digital twin in real time through the twin adaptation interface and dynamically adjusts the output strategy: in the initial modeling stage, it outputs full-dimensional, high-precision fused data, and the digital twin constructs an initial virtual model. The adaptability in the simulation stage is achieved as follows:

[0028] ,

[0029] In the formula, Indicates the fitness during the simulation phase. This represents the weighting coefficient for each stage. This represents the priority score for the initial modeling phase. This indicates the priority score for the dynamic evolution stage. This indicates the priority score of the risk assessment node. This indicates the priority score for the early warning stage.

[0030] Preferably, in step S5, the dynamic evolution stage outputs high-frequency incremental fusion data, with the update frequency consistent with the simulation time step of the digital twin; the risk assessment stage outputs multi-scenario fusion data, generating corresponding fusion results based on different hypothetical scenarios; and the early warning stage outputs early warning indicator fusion data, determining the output strategy parameters based on the fitness of the simulation stage, thus achieving the following:

[0031] ,

[0032] In the formula, This represents the strategy parameter, and k represents the curve steepness parameter. This represents the curve threshold.

[0033] Preferably, in step S6, during the digital twin simulation process, a simulation error report is generated by comparing the calculation results of the virtual model with the actual monitoring data, and fed back to the fusion system through the twin adaptation interface. After receiving the error report, the fusion system automatically adjusts the fusion strategy; analyzes and optimizes the source of error; adjusts the position of the acquisition device for acquisition points with poor data quality, and applies the optimized strategy to the subsequent fusion process in real time.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] 1. This invention uses a distributed architecture design, edge preprocessing at the acquisition end to reduce data transmission latency, dynamic fusion of edge nodes to achieve rapid fusion within the region, and deep fusion of the shore-based central hub using parallel computing. The three-layer fusion collaboration ensures that the total latency from data acquisition to output is reduced, which can adapt to different marine scenarios and digital twin requirements, improve fusion accuracy, and designs a dynamic weight adaptive algorithm that adjusts the weights in real time based on three dimensions: requirements, scenario, and data quality, to avoid fusion deviation caused by a single weight.

[0036] 2. This invention reduces data consistency error and missing data completion error by accurately matching and completing multi-source data in time and space. It establishes a unified three-dimensional time and space benchmark mechanism and an association-driven missing data completion algorithm, which solves the problems of data asynchrony, inconsistency and incompleteness. It adopts a distributed architecture and edge preprocessing technology to filter invalid data in advance and process regional data in a decentralized manner, avoiding the centralized transmission and calculation of a large amount of data.

[0037] 3. This invention improves the simulation accuracy of digital twins by seamlessly integrating and co-optimizing fused data with digital twins. It designs a twin adaptive interface and a simulation error feedback closed loop, and the fused data can accurately match the needs of each stage of digital twins. At the same time, it continuously optimizes the fusion strategy based on simulation errors. Attached Figure Description

[0038] Figure 1 The operation flow of the multi-source dynamic real-time fusion method for submarine landslides according to the present invention Figure 1 ;

[0039] Figure 2 The operation flow of the multi-source dynamic real-time fusion method for submarine landslides according to the present invention Figure 2 ;

[0040] Figure 3 The operation flow of the multi-source dynamic real-time fusion method for submarine landslides according to the present invention Figure 3 ;

[0041] Figure 4 The operation flow of the multi-source dynamic real-time fusion method for submarine landslides according to the present invention Figure 4 . Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example

[0044] Please see Figures 1-4 As shown, the present invention provides a technical solution comprising the following steps:

[0045] S1. During data acquisition, the edge preprocessing module starts synchronously to monitor data quality in real time.

[0046] S2. After receiving real-time data transmitted from the acquisition terminal, the edge node completes the time alignment and spatial matching of the data through the spatiotemporal synchronization calibration module.

[0047] S3. After receiving dynamic data slices from all edge nodes, the shore-based central hub starts a multi-dimensional correlation and fusion engine for deep fusion.

[0048] S4. The feedback optimization scheduling module collects three types of data in real time and performs optimization analysis.

[0049] S5. The fusion system uses the twin adapter interface to perceive the simulation stage of the digital twin in real time and dynamically adjust the output strategy.

[0050] S6. During the digital twin simulation process, a simulation error report is generated and fed back to the fusion system.

[0051] In this embodiment, in step S1, the edge preprocessing module identifies data that exceeds the preset normal data range, detects data abrupt changes through first-order difference, marks them as abnormal data and stores them separately, and performs format standardization conversion on the normal data, converting the satellite remote sensing JPG image data into GeoTIFF format raster data, embedding coordinate information into image metadata; and converting the sensor's CSV format time series data into a JSON format time series array, including acquisition timestamp, three-dimensional coordinates, data values, and data quality level.

[0052] Specifically, the edge preprocessing module extracts key features, including but not limited to extracting core features such as slope, elevation difference, and curvature from topographic data; extracting core features such as flow velocity, flow direction, and average temperature from hydrological data; and extracting core features such as shear strength and peak pore water pressure from geomechanical data.

[0053] Specifically, feature data is transmitted synchronously with the original data. Through 5G or acoustic communication links, the preprocessed normal data, feature data, and abnormal data labeling information are transmitted in real time to the nearest edge computing node.

[0054] In this embodiment, in step S2, a lightweight sliding window weighted fusion algorithm is invoked through spatiotemporal synchronization calibration. With a sliding window length of 1 second, preliminary fusion of multi-source data within the window is performed. Within the time window [t-1, t], there are n data sources, and the observation value of each data source i at time t is... The data quality level is , 1 indicates the best quality.

[0055] Specifically, the data is merged using a weighted average, with the weights representing the data quality level.

[0056] ,

[0057] In the formula, This represents the merged data value.

[0058] In this embodiment, in step S2, the fused data values ​​are obtained, and combined with the regional hydrological data, the correlation coefficient between displacement and flow velocity is calculated, which is implemented as follows:

[0059] ,

[0060] In the formula, A coefficient representing the relationship between displacement and flow velocity. This represents the merged displacement data. It represents the fused flow velocity data and generates regional dynamic data slices; the dynamic data slices contain core data items and metadata items. The core data items include the fusion results such as the current displacement, flow velocity, slope, and pore water pressure of the region, while the metadata items include data update time, data coverage, data reliability, and fusion algorithm type.

[0061] After receiving the data credibility feedback, the acquisition terminal automatically adjusts the acquisition strategy: when the data credibility is greater than or equal to the preset threshold, the current sampling frequency is maintained; when the credibility is within the preset threshold range, the sampling frequency is increased; when the credibility is less than the preset threshold, the sampling frequency is increased, and a data supplementation request is sent to the shore base.

[0062] In this embodiment, in step S3, the multi-dimensional correlation fusion engine deeply mines the coupling relationships between different regions and different types of data; the scene adaptive weight engine combines the current task instructions of the digital twin, real-time scene data, and data credibility to calculate the real-time weights of each type of data, which is implemented as follows:

[0063] ,

[0064] In the formula, This represents the weight of data type i over time t. This indicates the weight of the task instruction on the data type. This indicates the impact of real-time scene data on data type i. This represents the data reliability of data type i. These represent coefficients that balance task instructions, real-time scenario data, and data reliability, respectively. It represents the credibility decay coefficient, enabling deep integration across regions and dimensions.

[0065] Specifically, the generation of a global dynamic fusion model is achieved as follows:

[0066] ,

[0067] In the formula, This represents a global dynamic fusion model. This represents the weight of data type i at time t. This represents the historical mean of data type i. This represents the historical standard deviation of data type i.

[0068] In this embodiment, in step S3, the global dynamic fusion model is transmitted to the digital twin system through the twin adaptation interface. The interface reads the running status of the digital twin in real time. When the digital twin performs initial modeling, it outputs complete global fusion data. When the digital twin performs dynamic evolution simulation, it outputs incremental fusion data updated according to a preset time. When the digital twin triggers an early warning threshold, it outputs the core fusion data related to the early warning and marks key data items in red.

[0069] In this embodiment, in step S4, the feedback optimization scheduling module collects three types of data in real time: the deep fusion results of the shore-based central hub, the feedback information of the digital twin system, and the operating status of the collection equipment and edge nodes; it performs optimization analysis based on the collected data, generates optimization instructions and issues them; for unreasonable weights, it issues new weight parameters to the shore-based central hub and edge nodes; for missing data, it issues supplementary collection instructions to the collection end.

[0070] Specifically, to address the issue of poor algorithm adaptability, algorithm switching instructions are sent to edge nodes and shore-based central nodes; to address the issue of excessive node load, load-sharing instructions are sent to neighboring edge nodes.

[0071] In this embodiment, in step S5, the fusion system senses the simulation stage of the digital twin in real time through the twin adaptation interface and dynamically adjusts the output strategy: In the initial modeling stage, it outputs full-dimensional, high-precision fused data, including 3D terrain point cloud data, stratigraphic mechanical parameters, and initial hydrological field data. The digital twin constructs a 1:1 scale initial virtual model. The adaptability in the simulation stage is achieved as follows:

[0072] ,

[0073] In the formula, Indicates the fitness during the simulation phase. This represents the weighting coefficient for each stage. This represents the priority score for the initial modeling phase. This indicates the priority score for the dynamic evolution stage. This indicates the priority score of the risk assessment node. This indicates the priority score for the early warning stage.

[0074] In this embodiment, in step S5, the dynamic evolution stage outputs high-frequency incremental fusion data, with the update frequency consistent with the simulation time step of the digital twin; the risk assessment stage outputs multi-scenario fusion data, generating corresponding fusion results based on different hypothetical scenarios; and the early warning stage outputs early warning indicator fusion data, focusing on providing core early warning parameters such as stability coefficient, displacement acceleration rate, and pore water pressure growth rate.

[0075] Specifically, the output strategy parameters are determined based on the fitness during the simulation phase, as follows:

[0076] ,

[0077] In the formula, This represents the strategy parameter, ranging from [0-1]. A larger value indicates that the output strategy is more biased towards the early warning stage. k represents the curve steepness parameter. This represents the curve threshold.

[0078] In this embodiment, during the digital twin simulation process (S6), a simulation error report is generated by comparing the calculation results of the virtual model with the actual monitoring data. This report is then fed back to the fusion system through the twin adaptation interface. After receiving the error report, the fusion system automatically adjusts the fusion strategy, including increasing the data acquisition density and increasing the weight of regional data in areas with large errors.

[0079] Specifically, the sources of error are analyzed and optimized; for collection points with poor data quality, the location or parameters of the collection equipment are adjusted, and the optimized strategy is applied in real time to the subsequent fusion process.

[0080] Working principle: During the data acquisition phase, the edge preprocessing module monitors the quality of the raw data in real time, identifies outliers by setting a normal data range, detects data mutations through a first-order difference algorithm, marks outlier data, performs standardized format conversion on normal data, extracts key features such as topography, hydrology, and geomechanics, and transmits the preprocessed data to the edge computing node in real time through the communication link.

[0081] After receiving multi-source data from the acquisition end, the edge nodes eliminate timestamp deviations and spatial location differences through a spatiotemporal synchronization calibration module. A lightweight sliding window weighted fusion algorithm, combined with data quality levels, is used to weight and average the multi-source observations, generating preliminary fusion results. The correlation between displacement and flow velocity is calculated using regional hydrological data, forming a regional-level dynamic data slice containing core parameters and metadata. The shore-based central hub integrates the dynamic data slices from all edge nodes, using a multi-dimensional correlation fusion engine to uncover the coupling relationships between different regions and data types. A scene-adaptive weight engine dynamically adjusts the weights of various data types based on the digital twin's task requirements, real-time scene status, and data credibility, achieving deep cross-regional and multi-dimensional fusion, ultimately generating a full-domain dynamic fusion model. The feedback optimization scheduling module continuously collects the fusion results from the shore-based central hub and the digital twin system. The system monitors the operational feedback and edge node status; based on this information, it analyzes system performance and generates and issues optimization instructions to address issues such as unreasonable weight allocation, missing data, poor algorithm adaptability, or node overload; the fusion system uses a twin adaptation interface to perceive the simulation stage of the digital twin in real time and dynamically adjusts the data output strategy accordingly; in the initial modeling stage, it outputs high-precision data across all dimensions to support the construction of the virtual model; in the dynamic evolution stage, it updates incremental data frequently to match the simulation step size; in the risk assessment stage, it generates multi-scenario fusion data to simulate different conditions; in the early warning stage, it focuses on core early warning parameters and highlights key information; during the simulation process, the digital twin system compares the virtual model with the actual monitoring data, generates an error report, and feeds it back to the fusion system; the fusion system identifies problem areas based on the error report and automatically optimizes the data acquisition strategy or adjusts the fusion parameters.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

[0083] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for dynamic real-time fusion of multi-source data on submarine landslides, characterized in that, Includes the following steps: S1. During data acquisition, the edge preprocessing module starts synchronously to monitor data quality in real time. S2. After receiving real-time data transmitted from the acquisition terminal, the edge node completes the time alignment and spatial matching of the data through the spatiotemporal synchronization calibration module. S3. After receiving dynamic data slices from all edge nodes, the shore-based central hub starts a multi-dimensional correlation and fusion engine for deep fusion. S4. The feedback optimization scheduling module collects three types of data in real time and performs optimization analysis. S5. The fusion system uses the twin adapter interface to perceive the simulation stage of the digital twin in real time and dynamically adjust the output strategy. S6. During the digital twin simulation process, a simulation error report is generated and fed back to the fusion system.

2. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In step S1, the edge preprocessing module identifies data outside the preset normal data range, detects data mutations through first-order difference, marks them as abnormal data and stores them separately, and performs format standardization conversion on the normal data, converting the satellite remote sensing JPG image data into GeoTIFF format raster data, embedding coordinate information into image metadata; converting the sensor's CSV format time series data into a JSON format time series array; the edge preprocessing module extracts key features, and the feature data is transmitted synchronously with the original data, and the preprocessed normal data, feature data, and abnormal data marking information are transmitted in real time to the nearest edge computing node via 5G.

3. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In step S2, a lightweight sliding window weighted fusion method is invoked through spatiotemporal synchronization calibration to perform preliminary fusion of multi-source data within the window. Within the time window [t-1, t], there are n data sources, and the observation value of each data source i at time t is... The preset data quality level is The data is then merged using a weighted average, with the weights representing the data quality level. The result is as follows: ， In the formula, This represents the merged data value.

4. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In step S2, the merged data values ​​are obtained, and combined with the regional hydrological data, the correlation coefficient between displacement and flow velocity is calculated, which is implemented as follows: ， In the formula, A coefficient representing the relationship between displacement and flow velocity. This represents the merged displacement data. It represents the fused flow velocity data and generates regional-level dynamic data slices; After receiving data reliability feedback, the acquisition terminal automatically adjusts the acquisition strategy: if the data reliability is greater than or equal to the preset threshold, the current sampling frequency is maintained; When the reliability is within a preset threshold range, the sampling frequency is increased; If the reliability is less than the preset threshold, the sampling frequency is increased, and a data supplementation request is sent to the shore base.

5. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In S3, the multi-dimensional correlation and fusion engine deeply mines the coupling relationships between different regions and different types of data; the scene adaptive weight engine combines the current task instructions of the digital twin, real-time scene data, and data credibility to calculate the real-time weights of each type of data, as follows: ， In the formula, This represents the weight of data type i over time t. This indicates the weight of the task instruction on the data type. This indicates the impact of real-time scene data on data type i. This represents the data reliability of data type i. These represent coefficients that balance task instructions, real-time scenario data, and data reliability, respectively. This represents the credibility decay coefficient, enabling deep fusion across regions and dimensions, generating a dynamic fusion model across the entire domain, as follows: ， In the formula, This represents a global dynamic fusion model. This represents the weight of data type i at time t. This represents the historical mean of data type i. This represents the historical standard deviation of data type i.

6. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In S3, the global dynamic fusion model is transmitted to the digital twin system through the twin adaptation interface. The interface reads the running status of the digital twin in real time. When the digital twin performs initial modeling, it outputs complete global fusion data. When the digital twin performs dynamic evolution simulation, it outputs incremental fused data that is updated at preset times; When the digital twin triggers the warning threshold, it outputs the core fused data related to the warning and marks the key data items in red.

7. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In S4, the feedback optimization scheduling module collects three types of data in real time: the deep fusion results of the shore-based central hub, the feedback information of the digital twin system, and the operating status of the collection equipment and edge nodes; it performs optimization analysis based on the collected data, generates optimization instructions and issues them; for unreasonable weights, it issues new weight parameters to the shore-based central hub and edge nodes; for missing data, it issues supplementary collection instructions to the collection end. To address the issue of poor algorithm adaptability, algorithm switching instructions are sent to edge nodes and shore-based central nodes; to address the issue of excessive node load, load-sharing instructions are sent to neighboring edge nodes.

8. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In S5, the fusion system senses the simulation stage of the digital twin in real time through the twin adaptation interface and dynamically adjusts the output strategy: in the initial modeling stage, it outputs full-dimensional, high-precision fused data, and the digital twin constructs an initial virtual model. The adaptability in the simulation stage is achieved as follows: ， In the formula, Indicates the fitness during the simulation phase. This represents the weighting coefficient for each stage. This represents the priority score for the initial modeling phase. This represents the priority score for the dynamic evolution stage. This indicates the priority score of the risk assessment node. This indicates the priority score for the early warning stage.

9. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In S5, the dynamic evolution stage outputs high-frequency incremental fusion data, and the update frequency is consistent with the simulation time step of the digital twin. The risk assessment phase outputs multi-scenario fusion data and generates corresponding fusion results based on different hypothetical scenarios. During the early warning phase, the fused data of early warning indicators is output, and the output strategy parameters are determined based on the fitness of the simulation phase, as follows: ， In the formula, This represents the strategy parameter, and k represents the curve steepness parameter. This represents the curve threshold.

10. The method for dynamic real-time fusion of multi-source data on submarine landslides according to claim 1, characterized in that: In the S6 process of digital twin simulation, a simulation error report is generated by comparing the calculation results of the virtual model with the actual monitoring data, and then fed back to the fusion system through the twin adaptation interface. After receiving the error report, the fusion system automatically adjusts the fusion strategy. The sources of error are analyzed and optimized; for collection points with poor data quality, the parameters of the collection equipment are adjusted, and the optimized strategy is applied in real time to the subsequent fusion process.

Images