Submarine landslide risk prediction method and device

Abstract

The invention provides a submarine landslide risk prediction method and device, and the method comprises the steps: dividing a submarine slope of a first sea area into sub-sea areas, obtaining the sounding, boundary and geotechnical material strength parameters of the sub-sea areas, respectively selecting the sub-sea areas as a source area and a target area, constructing a three-dimensional search grid, and generating search points, the method comprises the steps of obtaining a reference deterministic model through a three-dimensional slope stability model, generating rock-soil random parameters conforming to logarithmic normal distribution, importing the reference model to calculate a safety coefficient, constructing a data set by using the safety coefficient of a three-dimensional search grid, sounding and the rock-soil random parameters, and training a prediction model based on domain adversarial neural network transfer learning. And finally, predicting a prediction safety coefficient of the second sea area seabed slope according to the prediction model, thereby rapidly completing risk assessment of the second sea area seabed slope.

Description

Technical Field

[0001] This invention relates to the field of geological disaster prediction technology, and in particular to a method and apparatus for predicting submarine landslide risks. Background Technology

[0002] Submarine landslides are serious geological hazards that occur when seabed sediments (such as mud, silt, and even rocks) move down a slope under the influence of gravity. This process can be a slow, creeping movement or a sudden, large-scale collapse. Submarine landslides can lead to a series of severe consequences, such as triggering tsunamis and damaging underwater infrastructure, making risk prediction crucial.

[0003] However, predicting submarine landslide risks involves a huge amount of computation and takes a lot of time. Therefore, a method for predicting submarine landslide risks is needed to reduce the amount of computation and improve the prediction speed. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for predicting submarine landslide risks, which significantly reduces the calculation time for submarine landslide risks and improves calculation efficiency.

[0005] To address the above problems, a first aspect of the present invention provides a method for predicting submarine landslide risks, comprising: Step S10000: Divide the seabed slope of the first sea area into several sub-sea areas, and obtain the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area. Select each sub-sea area as the source region in sequence, and take each remaining sub-sea area as the target region. Combine the target region with the source region to obtain several sets of training combinations. Step S20000: Construct a three-dimensional search grid based on the depth sounding, boundary conditions, and rock and soil material strength parameters of each sub-sea area and the grid elements of the depth sounding, and generate several search points within the three-dimensional search grid; use a three-dimensional slope stability evaluation model to perform stability analysis on the source area and the corresponding target area to obtain a baseline deterministic model; Step S30000: Generate several sets of random parameters for the rock and soil materials for the strength parameters of the rock and soil materials in each of the sub-sea areas, and make the random parameters of the rock and soil materials conform to a log-normal distribution; Step S40000: Import the random parameters of each set of the soil and rock materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of the soil and rock materials in each sub-sea area; Step S50000: Combine the safety factor of each three-dimensional search grid within the first seabed slope with the depth sounding and random parameters of the soil and rock materials of each sub-sea area to construct a dataset for predicting the safety factor of the seabed slope across regions. Train the dataset using a transfer learning model based on a domain adversarial neural network to obtain a prediction model. Step S60000: Obtain the depth measurement and material strength parameters of the seabed slope in the second sea area, and predict the predicted safety factor of the second sea area based on the prediction model.

[0006] Furthermore, step S10000 in the above-mentioned submarine landslide risk prediction method includes the following steps: Step S10100: Obtain the spatial coverage and depth sounding distribution of the seabed slope in the first sea area; Step S10200: Divide the sub-sea areas according to the spatial coverage and bathymetry distribution of the seabed slope of the first sea area, so that each sub-sea area has the same spatial coverage and similar bathymetry distribution, and the adjacent sub-sea areas have the greatest spatial correlation.

[0007] Furthermore, the rock and soil material strength parameters mentioned in the above-mentioned submarine landslide risk prediction method include: effective cohesion, effective internal friction angle, and equivalent buoyancy weight of the rock and soil mass.

[0008] Furthermore, step S20000 in the above-mentioned submarine landslide risk prediction method includes the following steps: S20100: Calculate the safety factor of the search point within each of the three-dimensional search grids; S20200: Records the minimum safety factor within each of the three-dimensional search grids; S20300: The overall stability of the seabed slope in the first sea area is determined based on the minimum safety factor within each of the three-dimensional search grids.

[0009] Furthermore, step S40000 in the above-mentioned submarine landslide risk prediction method includes the following steps: Step S40100: Filter the safety coefficients according to the preset threshold and retain the safety coefficients of the three-dimensional search grids that meet the requirements.

[0010] Furthermore, step S50000 in the above-mentioned submarine landslide risk prediction method includes the following steps: Step S50100: Import the depth measurement of each sub-sea area, the random parameters of the soil and rock materials, and the safety factor into the transfer learning model based on the domain adversarial neural network; Step S50200: Select the depth measurement of the source region, the random parameters of the soil and rock materials, and the safety factor for training and supervised learning; select the depth measurement of the target region and the strength of the soil and rock materials for domain adversarial training. Step S50300: Using a transfer learning model based on a domain adversarial neural network, select a set of training combinations. Based on the depth sounding of the source region, the random parameters and safety factor of the soil and rock materials, and the depth sounding and soil and rock material strength of the target region, calculate the corresponding predicted safety factor of the target region. Compare the predicted safety factor with the safety factor of the target region, calculate the evaluation index, and calculate the loss function of the transfer learning model based on the domain adversarial neural network to automatically adjust the parameters and obtain the optimal parameter combination. Retrain the transfer learning model based on the domain adversarial neural network according to the optimal parameter combination until the evaluation index meets the requirements. Use the current transfer learning model based on the domain adversarial neural network as the prediction model.

[0011] Furthermore, the method for predicting submarine landslide risks described above, after step S60000, also includes: Step S70000: For each of the three-dimensional search grids of the seabed slope in the second sea area, perform step S60000 to repeat the prediction several times to obtain the corresponding predicted safety factor dataset. Count the number of failures of the safety factor in the predicted safety factor dataset of each three-dimensional search grid, calculate the predicted failure probability of each three-dimensional search grid, obtain the spatial distribution of the predicted failure probability of the seabed slope in the second sea area, and complete the stability assessment.

[0012] According to another aspect of the present invention, the present invention also provides a submarine landslide risk prediction device, comprising: The training combination module is used to divide the seabed slope of the first sea area into several sub-sea areas, and obtain the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area. Each sub-sea area is selected as the source region in sequence, and each of the remaining sub-sea areas is selected as the target region. The training combinations are then combined with the source regions to obtain several sets of training combinations. The module for obtaining the baseline deterministic model is used to construct a three-dimensional search grid based on the depth sounding, boundary conditions, and strength parameters of the soil and rock materials for each sub-sea area, and to generate several search points within the three-dimensional search grid; and to perform stability analysis on the source region and the corresponding target region using a three-dimensional slope stability evaluation model to obtain the baseline deterministic model. The module for generating random parameters of soil and rock materials is used to generate several sets of random parameters of soil and rock materials for the strength parameters of soil and rock materials in each sub-ocean area, and to make the random parameters of soil and rock materials conform to a log-normal distribution. Safety factor calculation module: used to import the random parameters of each set of the soil and rock materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of the soil and rock materials in each sub-sea area; The transfer learning model training module is used to construct a dataset for predicting the safety factor of the seabed slope across regions by combining the safety factor of each three-dimensional search grid within the first sea area with the depth sounding and random parameters of the soil and rock materials of each sub-sea area. The dataset is then trained using a transfer learning model based on a domain adversarial neural network to obtain a prediction model. Safety factor prediction module: used to obtain the depth and material strength parameters of the seabed slope in the second sea area, and predict the safety factor of the seabed slope in the second sea area based on the prediction model.

[0013] The above-mentioned technical solution of the present invention has the following beneficial technical effects: First, the safety factor of the seabed slope in the first sea area is calculated, and random parameters of the rock and soil materials are generated for the rock and soil material strength parameters of each sub-sea area of ​​the seabed slope in the first sea area, so that they conform to the log-normal distribution. Then, a prediction model is obtained by training a transfer learning model based on a domain adversarial neural network. Thus, the predicted safety factor of the seabed slope in the second sea area can be obtained through the prediction model, thereby quickly completing the risk assessment of the seabed slope in the second sea area. Attached Figure Description

[0014] Figure 1 This is a flowchart of the steps in the submarine landslide risk prediction method according to the present invention; Figure 2 This is a structural block diagram of the submarine landslide risk prediction device according to the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0016] In a first aspect, the present invention provides a method for predicting submarine landslide risk, comprising: Step S10000: Divide the seabed slope of the first sea area into several sub-sea areas, and obtain the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area. Select each sub-sea area as the source region in turn, and each of the remaining sub-sea areas as the target region. Combine them with the source region to obtain several training combinations.

[0017] Specifically, step S10000 includes the following steps: Step S10100: Obtain the spatial coverage and depth sounding distribution of the seabed slope in the first sea area; Step S10200: Divide the sea area into sub-sea areas based on the spatial coverage and bathymetry distribution of the seabed slope of the first sea area, so that each sub-sea area has the same spatial coverage and similar bathymetry distribution, and the adjacent sub-sea areas have the greatest spatial correlation.

[0018] The strength parameters of soil and rock materials include: effective cohesion, effective internal friction angle, and equivalent buoyancy of soil and rock mass.

[0019] In this embodiment, the entire sea area is divided into 4 sub-sea areas. One of the sub-sea areas is taken as the source region in turn, and the remaining sub-sea areas are combined with the source region as the target region, resulting in 12 combinations.

[0020] Step S20000: Construct a three-dimensional search grid based on the depth sounding, boundary conditions, and strength parameters of the soil and rock materials for each sub-sea area, and generate several search points within the three-dimensional search grid; use the three-dimensional slope stability evaluation model to perform stability analysis on the source area and the corresponding target area to obtain the baseline deterministic model.

[0021] In this embodiment, the three-dimensional slope stability evaluation model is Scoops3D software, but other commercial software can also be used.

[0022] Specifically, step S20000 includes the following steps: S20100: Calculate the safety factor of the search points within each 3D search grid; S20200: Records the minimum safety factor within each 3D search grid; S20300: The overall stability of the seabed slope in the first sea area is determined based on the minimum safety factor within each three-dimensional search grid.

[0023] Step S30000: Generate several sets of random parameters for the strength parameters of the soil and rock materials for each sub-sea area, and make the random parameters of the soil and rock materials conform to a log-normal distribution; Step S40000: Import the random parameters of each set of geotechnical materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of geotechnical materials in each sub-ocean area.

[0024] Step S40000 includes the following steps: Step S40100: Filter the safety factors according to the preset threshold and retain the safety factors of the 3D search grids that meet the requirements.

[0025] In this embodiment, the preset threshold is typically set to 3, meaning that grid cells with a safety factor less than 3 are retained. Calculating the safety factor for each sub-ocean area would be time-consuming. Therefore, the safety factors are filtered. Since slopes within an ocean area typically remain stable when the safety factor is between 1 and 3, and are considered unstable when the safety factor is less than 1, data with a safety factor less than 3 are retained, thus not affecting prediction accuracy or training speed.

[0026] Step S50000: Combine the safety factor of each three-dimensional search grid within the first seabed slope with the depth sounding and random parameters of the soil and rock materials of each sub-sea area to construct a dataset for predicting the safety factor of the seabed slope across regions. Train the dataset using a transfer learning model based on a domain adversarial neural network to obtain a prediction model.

[0027] Specifically, step S50000 includes the following steps: Step S50100: Import the depth measurement, random parameters of soil and rock materials, and safety factor of each sub-sea area into the transfer learning model based on the domain adversarial neural network; Step S50200: Select the depth sounding, random parameters of soil and rock materials, and safety factor of the source region for training and supervised learning; select the depth sounding and soil and rock material strength of the target region for domain adversarial training. Step S50300: Using a transfer learning model based on a domain adversarial neural network, select a set of training combinations. Based on the depth sounding, random parameters of soil and rock materials, and safety factor of the source region, as well as the depth sounding and soil and rock material strength of the target region, calculate the corresponding predicted safety factor of the target region. Compare the predicted safety factor with the safety factor of the target region, calculate the evaluation index, and calculate the loss function of the transfer learning model based on the domain adversarial neural network to automatically adjust the parameters and obtain the optimal parameter combination. Retrain the transfer learning model based on the domain adversarial neural network according to the optimal parameter combination until the evaluation index meets the requirements. Use the current transfer learning model based on the domain adversarial neural network as the prediction model.

[0028] Step S60000: Obtain the depth measurement and material strength parameters of the seabed slope in the second sea area, and predict the predicted safety factor of the seabed slope in the second sea area based on the prediction model.

[0029] Step S70000: For each three-dimensional search grid of the seabed slope in the second sea area, repeat step S60000 several times to obtain the corresponding predicted safety factor dataset. Count the number of failures of the safety factor in the predicted safety factor dataset of each three-dimensional search grid, calculate the predicted failure probability of each three-dimensional search grid, obtain the spatial distribution of the predicted failure probability of the seabed slope in the second sea area, and complete the stability assessment.

[0030] Typically, when the safety factor is less than 1, the three-dimensional search grid is considered to be in failure. The safety factor of each three-dimensional search grid is predicted m times, and the number of failures is counted as n times. Thus, the predicted failure probability is n / m. After counting each three-dimensional search grid, the spatial distribution of the predicted failure probability of the seabed slope in the second sea area can be obtained, and the stability assessment can be completed.

[0031] Secondly, the present invention also provides a submarine landslide risk prediction device, comprising: The training combination module is used to divide the seabed slope of the first sea area into several sub-sea areas, and obtain the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area. Each sub-sea area is selected as the source area in turn, and each of the remaining sub-sea areas is selected as the target area. The training combination is then combined with the source area to obtain several sets of training combinations. The module for obtaining the baseline deterministic model is used to construct a three-dimensional search grid based on the depth sounding, boundary conditions, and strength parameters of the soil and rock materials for each sub-sea area, and to generate several search points within the three-dimensional search grid. The three-dimensional slope stability evaluation model is used to perform stability analysis on the source area and the corresponding target area to obtain the baseline deterministic model. Safety factor calculation module: used to generate several sets of random parameters of soil and rock materials for the strength parameters of soil and rock materials in each sub-sea area, and to make the random parameters of soil and rock materials conform to a log-normal distribution; Safety factor calculation module: Used to import the random parameters of each set of geotechnical materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of geotechnical materials in each sub-sea area; The transfer learning model training module is used to construct a dataset for predicting the safety factor of the seabed slope across regions by combining the safety factor of each three-dimensional search grid within the first sea area with the bathymetry and random parameters of the soil and rock materials of each sub-sea area. The prediction model is then trained using a transfer learning model based on a domain adversarial neural network. Safety factor prediction module: used to obtain the depth and material strength parameters of the seabed slope in the second sea area, and predict the safety factor of the seabed slope in the second sea area based on the prediction model.

[0032] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A method for predicting submarine landslide risk, characterized in that, include: Step S10000: Divide the seabed slope of the first sea area into several sub-sea areas, and obtain the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area. Select each sub-sea area as the source region in sequence, and take each of the remaining sub-sea areas as the target region. Combine them with the source region to obtain several training combinations. Step S20000: Construct a three-dimensional search grid based on the depth sounding, boundary conditions, and soil and rock material strength parameters and depth sounding grid elements for each sub-sea area, and generate several search points within the three-dimensional search grid; A three-dimensional slope stability evaluation model is used to perform stability analysis on the source region and the corresponding target region to obtain a baseline deterministic model. Step S30000: Generate several sets of random parameters for the rock and soil materials for the strength parameters of the rock and soil materials in each of the sub-sea areas, and make the random parameters of the rock and soil materials conform to a log-normal distribution; Step S40000: Import the random parameters of each set of the soil and rock materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of the soil and rock materials in each sub-sea area; Step S50000: Combine the safety factor of each three-dimensional search grid within the first seabed slope with the depth sounding and random parameters of the soil and rock materials of each sub-sea area to construct a dataset for predicting the safety factor of the seabed slope across regions. Train the dataset using a transfer learning model based on a domain adversarial neural network to obtain a prediction model. Step S60000: Obtain the depth measurement and material strength parameters of the seabed slope in the second sea area, and predict the predicted safety factor of the seabed slope in the second sea area based on the prediction model.

2. The method for predicting submarine landslide risk according to claim 1, characterized in that: Specifically, step S10000 includes the following steps: Step S10100: Obtain the spatial coverage and depth sounding distribution of the seabed slope in the first sea area; Step S10200: Divide the sub-sea areas according to the spatial coverage and bathymetry distribution of the seabed slope of the first sea area, so that each sub-sea area has the same spatial coverage and similar bathymetry distribution, and the adjacent sub-sea areas have the greatest spatial correlation.

3. The method for predicting submarine landslide risk according to claim 2, characterized in that: The strength parameters of the soil and rock materials include: effective cohesion, effective internal friction angle, and equivalent buoyancy of the soil and rock mass.

4. The method for predicting submarine landslide risk according to claim 1, characterized in that: Step S20000 includes the following steps: S20100: Calculate the safety factor of the search point within each of the three-dimensional search grids; S20200: Records the minimum safety factor within each of the three-dimensional search grids; S20300: The overall stability of the seabed slope in the first sea area is determined based on the minimum safety factor within each of the three-dimensional search grids.

5. The method for predicting submarine landslide risk according to claim 1, characterized in that: Step S40000 includes the following steps: Step S40100: Filter the safety coefficients according to the preset threshold and retain the safety coefficients of the three-dimensional search grids that meet the requirements.

6. The method for predicting submarine landslide risk according to claim 1, characterized in that: Step S50000 includes the following steps: Step S50100: Import the depth measurement of each sub-sea area, the random parameters of the soil and rock materials, and the safety factor into the transfer learning model based on the domain adversarial neural network; Step S50200: Select the depth measurement of the source region, the random parameters of the soil and rock materials, and the safety factor for training and supervised learning; select the depth measurement of the target region and the strength of the soil and rock materials for domain adversarial training. Step S50300: Using a transfer learning model based on a domain adversarial neural network, select a set of training combinations. Based on the depth sounding of the source region, the random parameters and safety factor of the soil and rock materials, and the depth sounding and soil and rock material strength of the target region, calculate the corresponding predicted safety factor of the target region. Compare the predicted safety factor with the safety factor of the target region, calculate the evaluation index, and calculate the loss function of the transfer learning model based on the domain adversarial neural network to automatically adjust the parameters and obtain the optimal parameter combination. Retrain the transfer learning model based on the domain adversarial neural network according to the optimal parameter combination until the evaluation index meets the requirements. Use the current transfer learning model based on the domain adversarial neural network as the prediction model.

7. The method for predicting submarine landslide risk according to claim 1, characterized in that: Following step S60000, the following is also included: Step S70000: For each of the three-dimensional search grids of the seabed slope in the second sea area, perform step S60000 to repeat the prediction several times to obtain the corresponding predicted safety factor dataset. Count the number of failures of the safety factor in the predicted safety factor dataset of each three-dimensional search grid, calculate the predicted failure probability of each three-dimensional search grid, obtain the spatial distribution of the predicted failure probability of the seabed slope in the second sea area, and complete the stability assessment.

8. A device for predicting the risk of submarine landslides, characterized in that, include: Training combination acquisition module: used to divide the seabed slope of the first sea area into several sub-sea areas, and acquire the depth sounding, boundary conditions and rock and soil material strength parameters of each sub-sea area, sequentially select each sub-sea area as the source region, and each of the remaining sub-sea areas as the target region, and combine them with the source region to obtain several training combinations; The module for obtaining the baseline deterministic model is used to construct a three-dimensional search grid based on the depth sounding, boundary conditions, and strength parameters of the soil and rock materials for each sub-sea area, and to generate several search points within the three-dimensional search grid. A three-dimensional slope stability evaluation model is used to perform stability analysis on the source region and the corresponding target region to obtain a baseline deterministic model. The module for generating random parameters of soil and rock materials is used to generate several sets of random parameters of soil and rock materials for the strength parameters of soil and rock materials in each sub-ocean area, and to make the random parameters of soil and rock materials conform to a log-normal distribution. Safety factor calculation module: used to import the random parameters of each set of the soil and rock materials into the benchmark deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of the soil and rock materials in each sub-sea area; The transfer learning model training module is used to construct a dataset for predicting the safety factor of the seabed slope across regions by combining the safety factor of each three-dimensional search grid within the first sea area with the depth sounding and random parameters of the soil and rock materials of each sub-sea area. The dataset is then trained using a transfer learning model based on a domain adversarial neural network to obtain a prediction model. Safety factor prediction module: used to obtain the depth and material strength parameters of the seabed slope in the second sea area, and predict the safety factor of the seabed slope in the second sea area based on the prediction model.

Images