Potential landslide random motion feature prediction and evaluation method

Abstract

The invention discloses a potential landslide random motion feature prediction and evaluation method comprising the following steps: 1) determining an analysis object, inputting related data, and establishing a landslide motion numerical model; 2) determining an initial random variable, and generating a random variable sample combination; 3) performing numerical simulation analysis on the sample; 4) extracting window motion features in a sample simulation process to form a data set; 5) establishing a prediction model, and performing training and optimization; 6) predicting the motion characteristics of the landslide in the target area through the prediction model; and 7) evaluating the influence condition of the landslide on the target area. According to the method, the uncertainty of the potential landslide in a project can be fully considered, the landslide movement process can be efficiently and accurately simulated, the landslide movement random features are analyzed, and the method is more suitable for analysis of the potential landslide compared with a deterministic method.

Description

Technical Field

[0001] This invention relates to a method for predicting and evaluating the random motion characteristics of potential landslides, belonging to the field of engineering landslide disaster prevention and mitigation technology. Background Technology

[0002] Landslides pose a serious threat to people's lives and property, causing enormous losses to social and economic development. Modeling the movement process of potential landslides using numerical simulation methods to predict the stochastic characteristics of the landslide mass at the target location, such as depth, velocity, and flow rate, is crucial. Because natural rock and soil materials are uncertain and non-homogeneous, and easily affected by the external environment, while experimental testing and expert analysis can provide the distribution range of some parameters, it is difficult to obtain definite values. Therefore, deterministic numerical simulations for analyzing potential landslides cannot fully account for the stochastic characteristics of landslide movement; thus, it is necessary to analyze the stochastic features of the landslide movement process.

[0003] In the uncertainty analysis of landslides, the randomness of initial conditions needs to be considered. Some researchers have conducted related work. Chen Li et al. published an article in the *Journal of Wuhan University* entitled "A Small Sample Prediction Method for Wide-Area Landslide Susceptibility Based on Meta-Learning," which established a prediction model for landslide susceptibility based on disaster-causing factors. However, this prediction model can only predict the susceptibility of potential landslides and cannot predict the movement characteristics and disaster-causing situation after a landslide occurs. Zhang Weijie et al. published an article in *Engineering Geology* entitled "Numerical analysis of the effect of strength parameters on the large-deformation flow process of earthquake-induced landslides," which analyzed the influence of strength parameters on landslide movement characteristics through simulation of eight sets of conceptual landslide samples. However, due to the long computation time required to perform a single numerical simulation, the number of sample simulations analyzed was small, and only one random variable was considered, which could not provide a quantitative analysis of the randomness of landslide movement characteristics. Existing analyses of landslide uncertainty mainly focus on assessing landslide susceptibility. Studies on the uncertainty of landslide movement processes are limited by the computational efficiency of numerical simulations, typically employing only small sample analyses and failing to extract and analyze landslide movement characteristics at specific locations. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting and evaluating the random motion characteristics of potential landslides, so as to overcome the shortcomings of the prior art. This method can efficiently and accurately simulate the landslide motion process and analyze the random characteristics of landslide motion, and is more suitable for the analysis of potential landslides than deterministic methods.

[0005] To address the aforementioned technical problems, this invention provides a method for predicting and evaluating the random motion characteristics of potential landslides, comprising the following steps:

[0006] Step 1: Identify potential landslide research objects, obtain digital topographic information and landslide mass information of the landslide analysis area, and establish a numerical model of landslide movement;

[0007] Step 2: Determine the initial random variables, determine the sampling method, and generate a sample combination of random variables; the sampling method adopts either deterministic distribution type sampling or uncertain distribution type sampling, and uses sampling methods to generate a sample combination of random variables;

[0008] Step 3: Introduce the random variable sample combination into the landslide motion numerical model, and calculate the landslide motion numerical model to simulate the landslide motion process;

[0009] Step 4: Set up a measurement window in the landslide motion numerical model, extract landslide motion features, and generate a dataset;

[0010] Step 5: Establish a neural network mapping model, train and optimize it to form a prediction model;

[0011] Step 6: Input the initial random variables or the motion characteristics during the landslide sliding process into the prediction model to predict the motion characteristics of the landslide in the target area;

[0012] Step 7: Based on the dataset generated in Step 4 or the prediction model in Step 5, assess the impact of the landslide on the target area according to the probability distribution of the random variables.

[0013] Furthermore, in step 1, the potential landslide research object belongs to the high-speed long-distance landslide type; the landslide analysis area is a rectangular area, including the area where the landslide body was located before sliding and the area that the landslide movement may pass through, and the area of ​​irrelevant areas is minimized as much as possible; the digital terrain information consists of the spatial coordinates of the terrain surface nodes of the analysis area; the landslide body shape is determined by the planar coordinates of the nodes on the sliding surface of the landslide body and the vertical depth of the landslide body at that point.

[0014] Furthermore, in step 2, the initial random variables are selected from the landslide porosity, friction coefficient, solid phase density, liquid phase density, permeability coefficient, initial height scaling factor, and initial range scaling factor, with the number of variables being m;

[0015] To determine the distribution type, the sampling method used is the Monte Carlo method, with a specified sample size of n. s Based on the value ranges and probability distributions of each random variable, a probability distribution of size n is generated using the Monte Carlo method. sAn array of size ×m; for uncertain distribution types, the sampling method employs random number generation or orthogonal experimental design, specifying the k-th variable p. k The number of samples is n pk For variable p k Perform the number of times n pk If the sampling is done at equal intervals or by random number sampling, then the sample size is...

[0016] Furthermore, in step 3, the calculation and analysis adopts the landslide motion depth integral equation as the control model of landslide motion and uses the smooth particle hydrodynamics method for discrete solution; when performing numerical analysis on the samples, a multi-threaded parallel computing mode is adopted. According to the number L of idle threads of the computer executing the analysis task, the generated samples are divided into L sub-sample combinations, and each thread independently completes the solution calculation of one sub-sample combination.

[0017] Furthermore, in step 4, the measurement window is a rectangle, the position, size and orientation of which are determined by the given fixed point coordinates, side lengths a and b, and the angle α between it and the X-axis of the landslide analysis area; the measurement window is of two types, one is a process window for measuring during the landslide movement, and the other is a target window set in the target area.

[0018] The landslide movement characteristics include the time it takes for the landslide to reach the measurement window, the time it takes for the landslide to leave the measurement window, the flow rate, normal velocity, and depth of the crest in the measurement window, the total volume flowing through the measurement window, and the minimum distance between the landslide and the measurement window when the landslide has not passed through the measurement window.

[0019] For any given time, measure the depth h of the landslide within the measurement window. w Measure the velocity v of the landslide body within the window. w The landslide flow rate q within the measurement window w and the total flow volume Q w Calculated using the following expression:

[0020]

[0021] q w =v w h w a (3)

[0022]

[0023] Where n is the total number of particles in the measurement window, ρ i v represents the density of particle i located within the measurement window. iLet Δt be the normal velocity of particle i within the measurement window with respect to the measurement window. j Let T be the time step size of the j-th calculation step, and T be the number of calculation steps; the peak characteristic value is obtained by taking the maximum value in the time domain.

[0024] Furthermore, in step 4, the dataset includes data A, data B, and data C, where data A is the initial random variable, data B is the motion feature of the process window, and data C is the motion feature of the target window.

[0025] Furthermore, in step 5, the neural network mapping model includes two types of mapping: one is to predict the movement characteristics of the landslide in the target area based on the initial random variables, and the other is to predict the movement characteristics of the landslide in the target area based on the movement characteristics of the landslide in the process window; the neural network adopts a fully connected layer structure; and the optimization process adopts the Adam optimizer.

[0026] Furthermore, in step 6, the demand is predicted for two scenarios. The first scenario is before the landslide occurs, in which case the initial random variable is input to predict the motion characteristics of the target window. The second scenario is when the landslide has just occurred, the process monitoring station has measured the landslide motion characteristics, but the landslide has not yet reached the target area, in which case the motion characteristics of the process window are input to predict the motion characteristics of the target window.

[0027] The predicted movement characteristics of a landslide within a target window are first determined to determine whether the landslide will move to the target window. If the landslide will reach the target window, the movement characteristics of the target window further include the time of arrival and departure from the target window, the speed, depth, flow rate, and other characteristics within the target window. If the landslide will not reach the target window, the movement characteristics of the target window further include the minimum distance between the landslide and the target window.

[0028] Furthermore, in step 7, the assessment of the landslide's impact on the target area is divided into two methods:

[0029] Method 1 is that when the dataset is generated in step 4, the initial random variables already conform to a certain probability distribution, so the relevant motion characteristics can be statistically analyzed directly based on the dataset to calculate the mean and standard deviation of the motion characteristics;

[0030] Method 2 uses the Monte Carlo method to generate a large-scale sample based on the probability distribution of the initial random variables. The sample size is recommended to be controlled at 1×10⁻⁶. 6 ~1×10 7 The appropriate method can be selected based on the complexity of the problem. After generating the sample, input it into the prediction model in step 5 to calculate the corresponding motion feature sample. Perform statistical analysis on the output motion feature sample to calculate the mean and standard deviation of the motion feature.

[0031] This invention generates and analyzes sample combinations using a specific sampling method. The analysis employs a depth integral equation as the control model for landslide movement and uses a smoothed particle hydrodynamics method for discrete solution, enabling efficient and accurate simulation of the landslide movement process and ensuring the efficiency of landslide numerical simulation sample analysis. By setting a measurement window in the computational domain, landslide movement characteristics at specific locations can be extracted. A neural network method is used to form an efficient and accurate landslide movement characteristic prediction model. This invention fully considers the uncertainties of potential landslides, providing accurate and efficient prediction and stochastic analysis of landslide movement characteristics. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention.

[0033] Figure 2 This is a numerical model of potential landslides in an embodiment of the present invention;

[0034] Figure 3 This is a simulation process of landslide movement according to an embodiment of the present invention;

[0035] Figure 4 This is a schematic diagram of the dataset structure;

[0036] Figure 5 It is the training process of a neural network prediction model. Detailed Implementation

[0037] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0038] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0039] This invention provides a method for predicting and evaluating the random motion characteristics of potential landslides. Combining a landslide motion depth integral model and a neural network model, this invention proposes a method for predicting and evaluating the random motion characteristics of potential landslides. It performs a sample analysis of potential landslides with uncertainties, extracts landslide motion characteristics at local locations by setting a measurement window, establishes a landslide motion characteristic prediction model using a neural network model, and performs random feature analysis. A detailed implementation flowchart is attached. Figure 1 As shown. It includes 7 steps:

[0040] Step 1: Identify potential landslide research objects, obtain digital topographic information and landslide mass information of the landslide analysis area, and establish a numerical model of landslide movement;

[0041] Step 2: Determine the initial random variables, determine the sampling method, and generate a sample combination of random variables; the sampling method adopts either deterministic distribution type sampling or uncertain distribution type sampling, and uses sampling methods to generate a sample combination of random variables;

[0042] Step 3: Introduce the random variable sample combination into the landslide motion numerical model, and calculate the landslide motion numerical model to simulate the landslide motion process;

[0043] Step 4: Set up a measurement window in the landslide motion numerical model, extract landslide motion features, and generate a dataset;

[0044] Step 5: Establish a neural network mapping model, train and optimize it to form a prediction model;

[0045] Step 6: Input the initial random variables or the motion characteristics during the landslide sliding process into the prediction model to predict the motion characteristics of the landslide in the target area;

[0046] Step 7: Based on the dataset generated in Step 4 or the prediction model in Step 5, assess the impact of the landslide on the target area according to the probability distribution of the random variables.

[0047] The following is a detailed explanation using a landslide case as an example, following a flowchart:

[0048] The first step is to select a specific landslide as the research object, obtain digital topographic information and landslide mass information of the landslide analysis area, and establish a numerical model of landslide movement, as shown in the attached figure. Figure 2 As shown.

[0049] The computational domain is 270m × 330m, the terrain grid spacing is 2m, the landslide particle spacing is 0.75m, and the initial height of the landslide mass varies from approximately 10cm to 5m at different locations. The densities of the solid component, pore fluid, and mixture are 2400kg / m³. 3 1000kg / m 3 and 2000kg / m 3 .

[0050] The second step is to select the coefficient of kinetic friction. Initial height scaling factor F h Initial range scaling factor F m As the initial random variable, F h The depth of the landslide can be magnified or reduced, F m The initial extent of the landslide can be expanded or reduced, and the three variables satisfy the following conditions respectively: F h ∈[1,1.5], F m ∈[1,1.5]. 8000 samples were generated using a random number sampling method.

[0051] The third step involves numerical simulation of the samples using the smoothed particle hydrodynamics method. Sixteen threads were used, thus dividing the samples into 16 subsets, each with a capacity of 500. The landslide motion process calculated from one sample is shown in the attached figure. Figure 3 As shown.

[0052] Fourth step: Set up measurement window 1 and measurement window 2 in the numerical model of the sample. Measurement window 1 is the process window, and measurement window 2 is the target window. The fixed point coordinates of measurement window 1 are (148, 220), a and b are both 8, and the included angle in radians is -0.588. The fixed point coordinates of measurement window 2 are (94.1, 92.2), a and b are both 10, and the included angle in radians is -0.23. (See attached diagram) Figure 3 As shown; the landslide movement characteristics are captured through the measurement window, including the time when the landslide body arrives at the measurement window, the time when the landslide body leaves the measurement window, the flow rate, normal velocity, and depth of the peak in the measurement window, the total volume flowing through the window, and the minimum distance between the landslide body and the measurement window 2 when the landslide body does not pass through the measurement window 2.

[0053] At any given time, measure the depth h of the landslide within the measurement window. w Measure the velocity v of the landslide body within the window. w The landslide flow rate q within the measurement window w and the total flow volume Q w It can be calculated using the following expression:

[0054]

[0055]

[0056] q w =v w h w a (3)

[0057]

[0058] Where n is the total number of particles in the measurement window, ρ i v represents the density of particle i located within the measurement window. i Let Δt be the normal velocity of particle i within the measurement window with respect to the measurement window. j Let T be the time step size of the j-th computation step, and T be the number of computation steps. The peak eigenvalue is obtained by taking the maximum value in the time domain.

[0059] The structure of the generated dataset is shown in the attached figure. Figure 4 As shown.

[0060] The fifth step is to build a neural network model, train and optimize it to form a predictive model. The training process is shown in the attached figure. Figure 5 As shown.

[0061] Step 6: Input the initial random variables and the motion characteristics of the process window into the prediction model. The model determines whether the landslide has reached the target window. If the landslide will reach the target window, it will further output the motion characteristics of the landslide within the target window. If the landslide does not reach the target window, it will further output the minimum distance of the landslide from the target window. The prediction error is shown in the table below.

[0062]

[0063] Step 7: Determine that the initial random variable is normally distributed. Based on the mean and variance, use the Monte Carlo method to generate a 1×10⁻⁶ random variable. 6 A set of samples is input into the prediction model to obtain 1×10n 6 The landslide movement characteristics were analyzed, and then statistical analysis was performed on these characteristics to obtain statistical features that can be used for landslide risk analysis. The computation time required for a single landslide numerical simulation is approximately 20 seconds. Based on a neural network prediction model, a 1×10⁻⁶ simulation was performed. 6 The prediction of the second feature took a total of 3.6 seconds, which shows that the method proposed in this invention can effectively improve the analysis efficiency of random motion features of landslides.

[0064] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalent elements of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0065] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting and evaluating the stochastic motion characteristics of potential landslides, characterized in that, Includes the following steps: Step 1: Identify potential landslide research objects, obtain digital topographic information and landslide mass information of the landslide analysis area, and establish a numerical model of landslide movement; Step 2: Determine the initial random variables, determine the sampling method, and generate a sample combination of random variables; the sampling method adopts either deterministic distribution type sampling or uncertain distribution type sampling, and uses sampling methods to generate a sample combination of random variables; Step 3: Introduce the random variable sample combination into the landslide motion numerical model, and calculate the landslide motion numerical model to simulate the landslide motion process; Step 4: Set up a measurement window in the landslide motion numerical model, extract landslide motion features, and generate a dataset; Step 5: Establish a neural network mapping model, train and optimize it to form a prediction model; Step 6: Input the initial random variables or the motion characteristics during the landslide sliding process into the prediction model to predict the motion characteristics of the landslide in the target area; Step 7: Based on the dataset generated in Step 4 or the prediction model in Step 5, assess the impact of the landslide on the target area according to the probability distribution of the random variables.

2. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 1, the potential landslide research object belongs to the high-speed long-distance landslide type; the landslide analysis area is a rectangular area, including the area where the landslide body was located before sliding and the area that the landslide movement may pass through, and the area of ​​irrelevant areas is minimized as much as possible; the digital terrain information consists of the spatial coordinates of the terrain surface nodes of the analysis area; the landslide body shape is determined by the planar coordinates of the nodes on the sliding surface of the landslide body and the vertical depth of the landslide body at that point.

3. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 2, the initial random variables are selected from multiple factors including landslide porosity, friction coefficient, solid phase density, liquid phase density, permeability coefficient, initial height scaling factor, and initial range scaling factor, with the number of variables being m. To determine the distribution type, the sampling method used is the Monte Carlo method, with a specified sample size of n. s Based on the value ranges and probability distributions of each random variable, a probability distribution of size n is generated using the Monte Carlo method. s An array of size ×m; for uncertain distribution types, the sampling method employs random number generation or orthogonal experimental design, specifying the k-th variable p. k The number of samples is n pk For variable p k Perform the number of times n pk If the sampling is done at equal intervals or by random number sampling, then the sample size is...

4. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 3, the calculation and analysis adopts the landslide motion depth integral equation as the control model of landslide motion and uses the smooth particle hydrodynamics method for discrete solution. When performing numerical analysis on the samples, a multi-threaded parallel computing mode is adopted. According to the number L of idle threads of the computer executing the analysis task, the generated samples are divided into L sub-sample combinations, and each thread independently completes the solution calculation of one sub-sample combination.

5. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 4, the measurement window is a rectangle, the position, size and orientation of which are determined by the given fixed point coordinates, side lengths a and b, and the angle α between it and the X-axis of the landslide analysis area; the measurement window is of two types, one is a process window for measuring during the landslide movement, and the other is a target window set in the target area; The landslide movement characteristics include the time it takes for the landslide to reach the measurement window, the time it takes for the landslide to leave the measurement window, the flow rate, normal velocity, and depth of the crest in the measurement window, the total volume flowing through the measurement window, and the minimum distance between the landslide and the measurement window when the landslide has not passed through the measurement window. For any given time, measure the depth h of the landslide within the measurement window. w Measure the velocity v of the landslide body within the window. w The landslide flow rate q within the measurement window w and the total flow volume Q w Calculated using the following expression: Where n is the total number of particles within the measurement window, ρ i v represents the density of particle i located within the measurement window. i Let Δt be the normal velocity of particle i within the measurement window with respect to the measurement window. j Let t be the time step size of the j-th calculation step, and t be the number of calculation steps; the peak characteristic value is obtained by taking the maximum value in the time domain.

6. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 4, the dataset includes data A, data B, and data C, where data A is the initial random variable, data B is the motion feature of the process window, and data C is the motion feature of the target window.

7. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 5, the neural network mapping model includes two types of mapping: one is to predict the movement characteristics of the landslide in the target area based on the initial random variables, and the other is to predict the movement characteristics of the landslide in the target area based on the movement characteristics of the landslide in the process window; the neural network adopts a fully connected layer structure; the optimization process adopts the Adam optimizer.

8. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 6, the demand is predicted for two scenarios. The first scenario is before the landslide occurs, in which case the initial random variable is input to predict the motion characteristics of the target window. The second scenario is when the landslide has just occurred, the process monitoring station has measured the landslide motion characteristics, but the landslide has not yet reached the target area, in which case the motion characteristics of the process window are input to predict the motion characteristics of the target window. The predicted movement characteristics of a landslide within a target window are first determined to determine whether the landslide will move to the target window. If the landslide will reach the target window, the movement characteristics of the target window further include the time of arrival and departure from the target window, the speed, depth, flow rate, and other characteristics within the target window. If the landslide will not reach the target window, the movement characteristics of the target window further include the minimum distance between the landslide and the target window.

9. The method for predicting and evaluating the random motion characteristics of potential landslides according to claim 1, characterized in that, In step 7, the assessment of the landslide's impact on the target area is divided into two methods: Method 1 involves directly performing statistical analysis on the relevant motion characteristics based on the dataset when the dataset is generated in step 4, since the initial random variables already conform to a certain probability distribution. The mean and standard deviation of the motion characteristics are then calculated. Method 2 uses the Monte Carlo method to generate a large-scale sample based on the probability distribution of the initial random variables. The sample size is recommended to be controlled at 1×10⁻⁶. 6 ~1×10 7 After generating the samples, input them into the prediction model in step 5, calculate the corresponding motion feature samples, perform statistical analysis on the output motion feature samples, and calculate the mean and standard deviation of the motion features.

Images