A method and device for predicting the risk of submarine landslides

Abstract

The application provides a submarine landslide risk prediction method and device, wherein the method comprises the following steps: dividing a first submarine slope into subsea areas, obtaining the depth, boundary and rock-soil material strength parameters of the subsea areas, selecting the subsea areas as source areas and target areas respectively, constructing a three-dimensional search grid and generating search points, obtaining a benchmark deterministic model through a three-dimensional slope stability model, generating rock-soil random parameters conforming to a logarithmic normal distribution, importing the benchmark model to calculate a safety factor, constructing a data set by using the safety factor of the three-dimensional search grid, the depth and the rock-soil random parameters, training a prediction model based on domain adversarial neural network transfer learning, and finally predicting the prediction safety factor of a second submarine slope according to the prediction model, so that the risk evaluation of the second submarine slope can be quickly completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster prediction, and in particular to a submarine landslide risk prediction method and device. BACKGROUND

[0002] A submarine landslide is a serious geological disaster that occurs when sediments such as sand, silt, and even rock on the seabed move downward along a slope under the action of gravity. This process can be a slow creep or a sudden, large-scale collapse. A submarine landslide can also cause a series of serious consequences, such as triggering a tsunami and damaging submarine facilities, so it is necessary to predict the risk of a submarine landslide.

[0003] However, the calculation of submarine landslide risk is computationally intensive and time-consuming, so a submarine landslide risk prediction method is needed to reduce the computational load and improve the prediction speed. SUMMARY

[0004] The purpose of the present application is to provide a submarine landslide risk prediction method and device that greatly reduces the calculation time of submarine landslide risk and improves the calculation efficiency.

[0005] To solve the above problems, the first aspect of the present application provides a submarine landslide risk prediction method, comprising:

[0006] Step S10000: dividing a first sea area submarine slope into a plurality of subsea areas, and obtaining the depth, boundary conditions and rock-soil material strength parameters of each subsea area, selecting each subsea area as the source area in turn, and the remaining each subsea area as the target area, and combining the source area to obtain a plurality of training combinations;

[0007] Step S20000: constructing a three-dimensional search grid according to the depth, boundary conditions and rock-soil material strength parameters of each subsea area and the grid elements of the depth, and generating a plurality of search points in the three-dimensional search grid; using a three-dimensional slope stability evaluation model to analyze the stability of the source area and the corresponding target area, and obtaining a baseline deterministic model;

[0008] Step S30000: generating a plurality of sets of rock-soil material random parameters for each subsea area, and making the rock-soil material random parameters conform to a lognormal distribution;

[0009] Step S40000: importing each set of rock-soil material random parameters into the baseline deterministic model, and establishing a random sample calculation model to calculate the safety factor of each set of rock-soil material random parameters for each subsea area;

[0010] Step S50000: Construct a data set for cross-regional submarine slope safety factor prediction based on the bathymetry and the random parameters of the rock-soil material of each sub-sea area, and train a transfer learning model based on a domain adversarial neural network to obtain a prediction model.

[0011] Step S60000: Obtain the bathymetry and material strength parameters of the second submarine slope, and predict the predicted safety factor of the second sea area according to the prediction model.

[0012] Further, the step S10000 in the above submarine landslide risk prediction method comprises the following steps:

[0013] Step S10100: Obtain the spatial coverage and bathymetry distribution of the first submarine slope.

[0014] Step S10200: Divide the sub-sea area according to the spatial coverage and bathymetry distribution of the first submarine slope, so that each sub-sea area has equal spatial coverage and close bathymetry distribution, and the adjacent sub-sea areas have maximum spatial correlation.

[0015] Further, the rock-soil material strength parameters in the above submarine landslide risk prediction method include effective cohesion, effective internal friction angle and equivalent rock-soil body buoyant density.

[0016] Further, the step S20000 in the above submarine landslide risk prediction method comprises the following steps:

[0017] S20100: Calculate the safety factor of the search point in each three-dimensional search grid.

[0018] S20200: Record the minimum safety factor in each three-dimensional search grid.

[0019] S20300: Obtain the overall stability of the first submarine slope according to the minimum safety factor in each three-dimensional search grid.

[0020] Further, the step S40000 in the above submarine landslide risk prediction method comprises the following steps:

[0021] Step S40100: Screen the safety factor according to the preset threshold, and retain the safety factor of the three-dimensional search grid meeting the requirements.

[0022] Further, the step S50000 in the above submarine landslide risk prediction method comprises the following steps:

[0023] Step S50100: the depth of each sub-sea area, the random parameters of the rock-soil material, and the safety factor are introduced into the domain-adversarial neural network-based transfer learning model;

[0024] Step S50200: the depth of the source area, the random parameters of the rock-soil material, and the safety factor are selected for training and supervised learning, and the depth and the strength of the rock-soil material of the target area are selected for domain-adversarial training;

[0025] Step S50300: the domain-adversarial neural network-based transfer learning model is used to select a group of the training combinations, calculate the predicted safety factor of the target area according to the depth of the source area, the random parameters of the rock-soil material, and the safety factor, and the depth and the strength of the rock-soil material of the target area, compare the predicted safety factor with the safety factor of the target area, calculate the evaluation index, and calculate the loss function of the domain-adversarial neural network-based transfer learning model to automatically optimize the parameters, obtain the best parameter combination, retrain the domain-adversarial neural network-based transfer learning model according to the best parameter combination until the evaluation index meets the requirement, and take the current domain-adversarial neural network-based transfer learning model as the prediction model.

[0026] Further, the submarine landslide risk prediction method further comprises the following steps after step S60000:

[0027] Step S70000: for each three-dimensional search grid of the second-sea-area submarine slope, step S60000 is repeatedly predicted for several times to obtain a corresponding predicted safety factor dataset, the number of failures of the safety factor in the predicted safety factor dataset of each three-dimensional search grid is counted, the predicted failure probability of each three-dimensional search grid is calculated, the spatial distribution of the predicted failure probability of the second-sea-area submarine slope is obtained, and the stability evaluation is completed.

[0028] According to another aspect of the present application, the present application also provides a submarine landslide risk prediction device, comprising:

[0029] The training combination acquisition module is used to divide the first-sea-area submarine slope into a plurality of sub-sea areas, acquire the depth, boundary conditions, and strength parameters of the rock-soil material of each sub-sea area, and sequentially select each sub-sea area as the source area, and the remaining each sub-sea area as the target area to obtain a plurality of training combinations.

[0030] The reference deterministic model acquisition module is configured to construct a three-dimensional search grid according to the depth, the boundary condition and the rock-soil material strength parameter of each sub-ocean area and a grid element of the depth, and generate a plurality of search points in the three-dimensional search grid; and perform stability analysis on the source area and the target area corresponding to the source area by using a three-dimensional slope stability evaluation model to obtain a reference deterministic model.

[0031] The rock-soil material random parameter generation module is configured to generate a plurality of sets of rock-soil material random parameters for the rock-soil material strength parameter of each sub-ocean area, and make the rock-soil material random parameters conform to a lognormal distribution.

[0032] The safety factor calculation module is configured to import each set of rock-soil material random parameters into the reference deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of rock-soil material random parameters of each sub-ocean area.

[0033] The transfer learning model training module is configured to construct a data set for cross-regional submarine slope safety factor prediction by using the safety factor of each three-dimensional search grid in the first sea area submarine slope, the depth and the rock-soil material random parameter of each sub-ocean area, and train a transfer learning model based on a domain adversarial neural network to obtain a prediction model.

[0034] The safety factor prediction module is configured to acquire the depth and the material strength parameter of a second sea area submarine slope, and predict the predicted safety factor of the second sea area submarine slope according to the prediction model.

[0035] The above technical solutions of the present application have the following beneficial technical effects: the safety factor of the first sea area submarine slope is calculated first, and the rock-soil material random parameters are generated for the rock-soil material strength parameter of each sub-ocean area of the first sea area submarine slope, so that the rock-soil material random parameters conform to a lognormal distribution, then a prediction model is obtained by training a transfer learning model based on a domain adversarial neural network, and thus the predicted safety factor of the second sea area submarine slope can be obtained through the prediction model, so that the risk assessment of the second sea area submarine slope can be quickly completed. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a step flow chart of the submarine landslide risk prediction method according to the present application;

[0037] Figure 2 is a structural block diagram of the submarine landslide risk prediction device according to the present application. DETAILED DESCRIPTION

[0038] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application with reference to the specific embodiments and the accompanying drawings. It should be understood that the description is only exemplary and is not intended to limit the scope of the present application. In addition, in the following description, the description of the known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present application.

[0039] In a first aspect, the present application provides a submarine landslide risk prediction method, comprising:

[0040] Step S10000: dividing the submarine slope of the first sea area into a plurality of sub-sea areas, and obtaining the depth, boundary condition and rock-soil material strength parameter of each sub-sea area, sequentially selecting each sub-sea area as a source area, and the remaining each sub-sea area as a target area, and combining the source area to obtain a plurality of training combinations.

[0041] Specifically, step S10000 comprises the following steps:

[0042] Step S10100: obtaining the spatial coverage and depth distribution of the submarine slope of the first sea area;

[0043] Step S10200: dividing the sub-sea areas according to the spatial coverage and depth distribution of the submarine slope of the first sea area, so that each sub-sea area has equal spatial coverage and close depth distribution, and the adjacent sub-sea areas have maximum spatial correlation.

[0044] The rock-soil material strength parameters include effective cohesion, effective internal friction angle and equivalent rock-soil body buoyant density.

[0045] In this embodiment, the entire sea area is divided into 4 sub-sea areas, one of which is sequentially taken as a source area, and the remaining sub-sea areas are respectively combined with the source area to obtain 12 combinations.

[0046] Step S20000: constructing a three-dimensional search grid according to the depth, boundary condition and rock-soil material strength parameter of each sub-sea area and the grid element of the depth, and generating a plurality of search points in the three-dimensional search grid; using a three-dimensional slope stability evaluation model to analyze the stability of the source area and the corresponding target area to obtain a baseline certainty model.

[0047] In this embodiment, the three-dimensional slope stability evaluation model is Scoops3D software, and other commercial software can also be used.

[0048] Specifically, step S20000 comprises the following steps:

[0049] S20100: calculating the safety factor of the search point in each three-dimensional search grid;

[0050] S20200: record the minimum safety factor in each three-dimensional search grid;

[0051] S20300: derive the overall stability of the seabed slope in the first sea area according to the minimum safety factor in each three-dimensional search grid.

[0052] Step S30000: generate a plurality of sets of random parameters of rock and soil materials for the strength parameters of rock and soil materials of each sub-sea area, and make the random parameters of rock and soil materials conform to a lognormal distribution;

[0053] Step S40000: import each set of random parameters of rock and soil materials into a reference deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random parameters of rock and soil materials of each sub-sea area.

[0054] Step S40000 includes the following steps:

[0055] Step S40100: screen the safety factor according to a preset threshold, and retain the safety factor of the three-dimensional search grid that meets the requirements.

[0056] In this embodiment, the preset threshold is usually set to 3, that is, the safety factor of the grid unit with a safety factor less than 3 is retained. If the safety factor in each sub-sea area is calculated, a large amount of time will be consumed, so the safety factor is filtered. Since the safety factor is 1 to 3, the slope in the sea area usually remains stable, and when the safety factor is less than 1, the slope is considered to be unstable, so the data with a safety factor less than 3 is retained, that is, the prediction accuracy and the training speed will not be affected.

[0057] Step S50000: construct a data set for cross-regional seabed slope safety factor prediction by using the safety factor of each three-dimensional search grid in the first sea area seabed slope, the depth of each sub-sea area, and the random parameters of rock and soil materials, and train a transfer learning model based on a domain adversarial neural network to obtain a prediction model.

[0058] Specifically, step S50000 includes the following steps:

[0059] Step S50100: import the depth of each sub-sea area, the random parameters of rock and soil materials, and the safety factor into the transfer learning model based on the domain adversarial neural network;

[0060] Step S50200: select the depth, random parameters of rock and soil materials, and safety factor of the source area for training and supervised learning, and select the depth and strength of rock and soil materials of the target area for domain adversarial training;

[0061] Step S50300: using the domain-adversarial neural network-based transfer learning model, selecting a set of training combinations, calculating the predicted safety factor of the target area according to the depth of the source area, the random parameters of the rock-soil material and the safety factor, and the depth and rock-soil material strength of the target area, comparing the predicted safety factor with the safety factor of the target area, calculating the evaluation index, and calculating the loss function of the domain-adversarial neural network-based transfer learning model to automatically optimize the parameters to obtain the best parameter combination, retraining the domain-adversarial neural network-based transfer learning model according to the best parameter combination until the evaluation index meets the requirements, and taking the current domain-adversarial neural network-based transfer learning model as the prediction model.

[0062] Step S60000: obtaining the depth and material strength parameters of the second sea area submarine slope, and predicting the predicted safety factor of the second sea area submarine slope according to the prediction model.

[0063] Step S70000: for each three-dimensional search grid of the second sea area submarine slope, performing step S60000 repeatedly for several times to obtain a corresponding predicted safety factor dataset, counting the failure times of the safety factor in each three-dimensional search grid predicted safety factor dataset, calculating the predicted failure probability of each three-dimensional search grid, obtaining the spatial distribution of the predicted failure probability of the second sea area submarine slope, and completing the stability evaluation.

[0064] Generally, when the safety factor is less than 1, the three-dimensional search grid is considered to be invalid. The safety factor of each three-dimensional search grid is predicted m times, and then the failure times n are counted. Therefore, the predicted failure probability is n / m. After the statistics of each three-dimensional search grid, the spatial distribution of the predicted failure probability of the second sea area submarine slope can be obtained, and the stability evaluation is completed.

[0065] In a second aspect, the present application also provides a submarine landslide risk prediction device, comprising:

[0066] The training combination acquisition module is used to divide the first sea area submarine slope into a plurality of sub-sea areas, and obtain the depth, boundary condition and rock-soil material strength parameters of each sub-sea area, and sequentially select each sub-sea area as a source area, and the remaining each sub-sea area as a target area, and combine the source area to obtain a plurality of training combinations;

[0067] The reference deterministic model acquisition module is used to construct a three-dimensional search grid according to the depth, boundary condition and rock-soil material strength parameters of each sub-sea area and the grid elements of the depth, and generate a plurality of search points in the three-dimensional search grid; using a three-dimensional slope stability evaluation model to analyze the stability of the source area and the corresponding target area, and obtaining a reference deterministic model;

[0068] The safety factor calculation module is configured to generate a plurality of sets of random rock-soil material parameters for each sub-sea area, and make the random rock-soil material parameters conform to a lognormal distribution.

[0069] The safety factor calculation module is configured to import each set of random rock-soil material parameters into a reference deterministic model, and establish a random sample calculation model to calculate the safety factor of each set of random rock-soil material parameters for each sub-sea area.

[0070] The transfer learning model training module is configured to construct a data set for cross-regional submarine slope safety factor prediction by using the safety factor of each three-dimensional search grid in the first sea area submarine slope, the depth of each sub-sea area, and the random rock-soil material parameters, and train a transfer learning model based on a domain adversarial neural network to obtain a prediction model.

[0071] The safety factor prediction module is configured to obtain the depth and material strength parameters of the second sea area submarine slope, and predict the predicted safety factor of the second sea area submarine slope according to the prediction model.

[0072] It should be understood that the above specific embodiments of the present application are only used for illustrative or explanatory purposes of the principles of the present application, and do not constitute a limitation on the present application. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present application shall be included in the protection scope of the present application. In addition, the appended claims of the present application are intended to cover all variations and modifications falling within the scope and boundary of the appended claims, or the equivalent forms of such scope and boundary.

Claims

1. A method of predicting the risk of submarine landslides, characterized in that, The method comprises the following steps: Step S10000: dividing a first submarine slope of a sea area into a plurality of sub-sea areas, and obtaining the depth, boundary condition and rock-soil material strength parameter of each sub-sea area; selecting each sub-sea area as a source area in turn, and selecting each sub-sea area as a target area; and combining the source area and the target area to obtain a plurality of training combinations; Step S20000: constructing a three-dimensional search grid according to the depth, boundary condition and rock-soil material strength parameter of each sub-sea area and the grid element of the depth; and generating a plurality of search points in the three-dimensional search grid; using a three-dimensional slope stability evaluation model to analyze the stability of the source area and the corresponding target area, and obtaining a benchmark deterministic model; Step S30000: generating a plurality of sets of rock-soil material random parameters for the rock-soil material strength parameter of each sub-sea area, and making the rock-soil material random parameters conform to a lognormal distribution; Step S40000: importing each set of rock-soil material random parameters into the benchmark deterministic model, and establishing a random sample calculation model to calculate the safety factor of each set of rock-soil material random parameters of each sub-sea area; Step S50000: constructing a data set for cross-regional submarine slope safety factor prediction by using the safety factor of each three-dimensional search grid in the first submarine slope of the sea area, the depth and the rock-soil material random parameter of each sub-sea area, and training a domain-adversarial neural network-based transfer learning model to obtain a prediction model; The step S50000 comprises the following steps: Step S50100: importing the depth, rock-soil material random parameter and safety factor of each sub-sea area into the domain-adversarial neural network-based transfer learning model; Step S50200: selecting the depth, rock-soil material random parameter and safety factor of the source area for training and supervised learning, and selecting the depth and rock-soil material strength of the target area for domain-adversarial training; Step S50300: using the domain-adversarial neural network-based transfer learning model, selecting a set of training combinations, calculating the predicted safety factor of the corresponding target area according to the depth, rock-soil material random parameter and safety factor of the source area and the depth and rock-soil material strength of the target area, comparing the predicted safety factor with the safety factor of the target area, calculating an evaluation index, calculating the loss function of the domain-adversarial neural network-based transfer learning model to automatically optimize the parameters, obtaining the best parameter combination, retraining the domain-adversarial neural network-based transfer learning model according to the best parameter combination until the evaluation index meets the requirements, and taking the current domain-adversarial neural network-based transfer learning model as the prediction model; Step S60000: obtaining the depth and material strength parameter of a second submarine slope of a sea area, and predicting the predicted safety factor of the second submarine slope of the sea area according to the prediction model.

2. The submarine landslide risk prediction method according to claim 1, wherein: Specifically, the step S10000 comprises the following steps: Step S10100: Obtain the spatial coverage and bathymetry distribution of the first submarine slope. Step S10200: Divide the first submarine slope according to the spatial coverage and bathymetry distribution, so that each sub-sea area has equal spatial coverage and close bathymetry distribution, and the adjacent sub-sea areas have maximum spatial correlation.

3. The submarine landslide risk prediction method according to claim 2, wherein: The rock-soil material strength parameters include effective cohesion, effective internal friction angle, and equivalent rock-soil body buoyant density.

4. The submarine landslide risk prediction method according to claim 1, wherein: The step S20000 includes the following steps: S20100: Calculate the safety factor of the search point in each three-dimensional search grid; S20200: Record the minimum safety factor in each three-dimensional search grid; S20300: Obtain the overall stability of the first submarine slope according to the minimum safety factor in each three-dimensional search grid.

5. The submarine landslide risk prediction method according to claim 1, wherein: The step S40000 includes the following steps: Step S40100: Screen the safety factor according to a preset threshold, and retain the safety factor of the three-dimensional search grid that meets the requirements.

6. The submarine landslide risk prediction method according to claim 1, wherein: After the step S60000, the method further includes: Step S70000: For each three-dimensional search grid of the second submarine slope, execute the step S60000 repeatedly for several times to obtain a corresponding predicted safety factor data set, count the failure times of the safety factor in the predicted safety factor data set of each three-dimensional search grid, calculate the prediction failure probability of each three-dimensional search grid, obtain the spatial distribution of the prediction failure probability of the second submarine slope, and complete the stability evaluation.

7. A submarine landslide risk prediction apparatus characterized by comprising: including: An acquisition training combination module: used for dividing the first submarine slope into several sub-sea areas, and acquiring the bathymetry, boundary conditions, and rock-soil material strength parameters of each sub-sea area, sequentially selecting each sub-sea area as a source area, and each remaining sub-sea area as a target area, and combining the source area and the target area to obtain several training combinations; An acquisition reference deterministic model module: used for constructing a three-dimensional search grid according to the bathymetry, the boundary conditions, and the rock-soil material strength parameters of each sub-sea area and the grid elements of the bathymetry, and generating several search points in the three-dimensional search grid; Using a three-dimensional slope stability evaluation model to perform stability analysis on the source area and the corresponding target area to obtain a reference deterministic model; A rock-soil material random parameter generation module: used for generating several groups of rock-soil material random parameters for the rock-soil material strength parameters of each sub-sea area, and making the rock-soil material random parameters conform to a lognormal distribution; The safety factor calculation module is configured to input each set of the random parameters of the rock-soil material into the benchmark deterministic model, and to establish a random sample calculation model to calculate the safety factor of each set of the random parameters of the rock-soil material of each of the sub-sea areas; The safety factor calculation module comprises the following modules: The input migration learning model module is configured to input the depth, the random parameters of the rock-soil material, and the safety factor of each of the sub-sea areas into the domain-adversarial neural network-based migration learning model; The domain-adversarial training module is configured to select the depth, the random parameters of the rock-soil material, and the safety factor of the source area for training and supervised learning, and to select the depth and the strength of the rock-soil material of the target area for domain-adversarial training; The prediction model module is configured to select a set of the training combinations, to calculate the predicted safety factor of the target area according to the depth, the random parameters of the rock-soil material, and the safety factor of the source area and the depth and the strength of the rock-soil material of the target area, to compare the predicted safety factor with the safety factor of the target area, to calculate an evaluation index, to calculate the loss function of the domain-adversarial neural network-based migration learning model to automatically optimize the parameters, to obtain the best parameter combination, to retrain the domain-adversarial neural network-based migration learning model according to the best parameter combination until the evaluation index meets the requirement, and to take the current domain-adversarial neural network-based migration learning model as the prediction model; The migration learning model training module is configured to construct a data set for cross-area seabed slope safety factor prediction by using the safety factor of each of the three-dimensional search grids in the seabed slope of the first sea area, the depth, and the random parameters of the rock-soil material of each of the sub-sea areas, and to train the domain-adversarial neural network-based migration learning model to obtain a prediction model; The safety factor prediction module is configured to obtain the depth and the material strength parameters of the seabed slope of the second sea area, and to predict the predicted safety factor of the seabed slope of the second sea area according to the prediction model.

Images