Landslide catastrophe risk probability intelligent prediction method based on two-phase substance point method

Abstract

The invention belongs to the technical field of geological disaster monitoring and early warning, and particularly relates to a landslide catastrophe risk probability intelligent prediction method based on a two-phase substance point method. A two-phase substance point method is used for achieving rainfall-induced landslide catastrophe whole-process numerical simulation, then probability distribution of response quantities such as the sliding distance, the slump range and the slip mass volume is obtained, and the risk probability is predicted in combination with disaster-bearing body characteristics; according to the prior distribution of each parameter variable, a random sample is extracted as an input data set, a response quantity sample obtained through analog calculation is used as an output data set, an XGBoost model is adopted for training, and an agent model of a two-phase substance point method is constructed; and carrying out parameter inversion on slope monitoring data by utilizing Bayesian probability inversion, dynamically updating probability distribution of response quantity of each model, and predicting risk probability in real time according to the response quantity distribution. The physical significance is clear, and the prediction efficiency and the prediction precision are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of geological disaster monitoring and early warning, and in particular relates to an intelligent prediction method for landslide disaster risk probability based on a two-phase material point method. Background Art

[0002] Landslide geological disasters are widely distributed, sudden and destructive. Under the condition of rainfall infiltration, the strength index of slope rock and soil deteriorates and the permeability increases, resulting in a decrease in slope stability, which is more likely to induce landslide geological disasters, threatening economic construction and the safety of life and property of the people. Predicting potential landslide disasters has become a key and difficult problem that needs to be solved urgently in the prevention and control of geological disasters.

[0003] At present, the main methods for predicting landslides include empirical prediction methods, statistical prediction methods, and nonlinear prediction methods. The empirical prediction method relies on the experience accumulation of landslide precursors, lacks theoretical basis, and has low reliability. The statistical prediction method predicts future trends through historical landslide monitoring data, which is difficult to reflect the changes of landslide response data with influencing factors and has certain limitations. The nonlinear method regards the landslide as a nonlinear dynamic system, and combines various theories such as nonlinear theory, system science theory, and neural network theory to achieve dynamic tracking and nonlinear prediction of landslides. However, due to the complexity of the landslide evolution process and the variability of external influences, the physical meaning of this method is not clear enough.

[0004] Most of the current landslide prediction methods lack complete physical meanings, and landslides involve many related factors and complex conditions. Therefore, in the absence of support from physical mechanics theory, most of the existing prediction methods have their own shortcomings and deficiencies. They do not truly reveal the internal mechanism of landslide deformation, destruction and evolution, and cannot be applied to all types of landslides. They can only predict landslides of a certain type or a certain evolution stage, resulting in unsatisfactory prediction results. Therefore, it is necessary to establish a real-time prediction method for landslide disaster risk probability based on full consideration of the physical evolution mechanism of landslides. Summary of the invention

[0005] In view of the above-mentioned defects of the prior art, the purpose of the present invention is to provide a new landslide disaster risk probability prediction method to solve the problems existing in traditional landslide prediction methods, such as unclear physical meaning, difficulty in real-time prediction and low prediction accuracy.

[0006] To achieve the above object, the present invention provides a method for intelligent prediction of landslide disaster risk probability based on a two-phase material point method, comprising the following steps:

[0007] 1) Based on the landslide investigation data and local rainfall monitoring data, the prior distribution of the landslide rock and soil parameters such as elastic modulus E, cohesion c, internal friction angle φ, permeability coefficient k and rainfall intensity q is obtained;

[0008] 2) Based on the prior distribution of various rock and soil parameters and rainfall intensity, Latin hypercube sampling (LSH) method is used to extract random samples to construct the input data set;

[0009] 3) According to the landslide stratum data, a two-phase material point numerical calculation model is constructed for each set of sample parameters in step 2), and the initial conditions, boundary conditions and load conditions of the numerical model are set according to the actual situation;

[0010] 4) Based on the two-phase material point method, the motion equations of landslide rock and soil and pore fluid are discretized in material point space and solved by explicit time integration to simulate the whole process of destruction evolution of unsaturated rock and soil of landslide under the action of hydrodynamics;

[0011] 5) Based on the two-phase material point method, the displacement and pore pressure values ​​of the monitoring points with time series are calculated for parameter inversion, and the final response of each sample such as sliding distance s, collapse range S, sliding body thickness h, sliding body volume V, sliding velocity v, etc. are obtained to construct the output data set;

[0012] 6) Based on the data sets obtained in step 2) and step 5), the XGBoost model is used for training and testing to construct a proxy model of the numerical model to improve computational efficiency;

[0013] 7) Based on the prior distribution of each parameter, the Monte Carlo simulation algorithm is used in combination with the proxy model to statistically calculate the probability of landslide failure and the probability distribution characteristics of each response quantity, and combined with the characteristics of the hazard-bearing body, the initial landslide disaster risk probability prediction results are calculated;

[0014] 8) Monitoring instruments are arranged on the slope to collect real-time monitoring data such as displacement and pore pressure. The Bayesian probability inversion method is used to perform parameter inversion based on the monitoring data to calculate the posterior distribution of rock and soil parameters. The rainfall intensity q is determined based on the measured data;

[0015] 9) Based on the posterior distribution of each rock and soil parameter obtained by inversion, the Monte Carlo simulation algorithm is used again to statistically calculate the failure probability and the probability distribution characteristics of each model response, and the risk probability prediction results are updated;

[0016] 10) Using the posterior distribution of the rock and soil parameters obtained in step 8) as the prior distribution, repeating steps 8) to 9) through real-time collected landslide displacement, pore pressure monitoring data and rainfall data, inversely updating the rock and soil parameters, and realizing real-time dynamic update of risk probability prediction results.

[0017] Furthermore, in step 2), the model parameter variable sample can be expressed as X = [x1, x2, ..., x n ], where x i represents parameters such as elastic modulus E, cohesion c, internal friction angle φ, etc., n represents the dimension of model parameter variables, and the number of samples required to be extracted in each dimension is m. When using Latin hypercube sampling, if m n-dimensional samples that conform to the prior distribution of model parameters are to be extracted in the sample space, then the [0,1] interval needs to be divided into m intervals, and a sample is randomly selected in each interval, and the extracted values ​​are mapped to the sample space to obtain an n×m-order sample matrix A, thereby constructing the model input data set.

[0018] Furthermore, in step 3), a three-dimensional geological model of the landslide body is constructed based on the three-dimensional stratigraphic data of the landslide area, and the model is discretized into spatial material points. At the same time, a background grid covering the entire landslide sliding area is set behind the material points, and rock and soil parameters are assigned to each material point. Gravity and other external load conditions are input, displacement boundary conditions and seepage boundary conditions are set, an initial groundwater level is set, and initial ground stress balance is performed.

[0019] Furthermore, in step 4), the rock mass is described by the DP elastic-plastic constitutive model, and the pore fluid is described by Darcy's law. The Euler-Cromer explicit integration algorithm is used to solve the problem by combining the effective stress principle, soil-water characteristic curve and permeability curve. The constitutive equations of pore fluid and solid are shown in equations (1) and (2), respectively:

[0020]

[0021]

[0022] Among them, ρ L is the liquid density, p L is the liquid pressure, v L 、v S are the absolute velocities of liquid and solid respectively, n is the porosity of the solid skeleton, S L is saturation, D ep is the tangential stiffness matrix, and h' is the constitutive tensor.

[0023] The relationship between pore water pressure and saturation is given by the soil water characteristic curve (SWRC), as shown in formula (3). The actual permeability k and the saturated permeability k sat The relationship between is obtained from the permeability curve (HCC), as shown in formula (4).

[0024]

[0025]

[0026] Where S max , S min are the maximum saturation and the residual saturation, respectively, p ref and λ are fitting parameters.

[0027] The nodal momentum balance equations for the solid and fluid that need to be solved are shown in Equations (5) and (6).

[0028]

[0029]

[0030] Among them, a S and a L are the nodal acceleration vectors of the solid and fluid, respectively, and τ is the boundary The pulling vector at For the border The liquid pressure at the location, N is the node shape function matrix, B is the gradient matrix of the node shape function at the local material point MP, V MP is the volume of the material point, ρ m is the mixture density, n MP is the number of material points, n MP is the porosity of the solid skeleton of the material point, g is the acceleration vector, is the material point permeability.

[0031] Further, in step 6), the specific steps of using the XGBoost model are as follows:

[0032] ① Take the input and output data obtained in step 2) and step 5) as the data set;

[0033] ② Divide the data set into training set and test set in a ratio of 70% and 30%;

[0034] ③Define XGBoost model parameters, including the maximum depth of the tree, learning rate, number of iterations, objective function, etc.;

[0035] ④ Use the training set data to train the XGBoost model;

[0036] ⑤ Use the cross-validation method to adjust the model parameters and find the optimal parameter combination;

[0037] ⑥Use the test set to evaluate the performance of the model.

[0038] Furthermore, in step 7), based on the sliding distance s, the landslide range S, etc., combined with the distribution of buildings, infrastructure, and personnel, the hazard-bearing body that may be affected by the landslide is evaluated, and then based on the sliding body thickness h, sliding body volume V, sliding speed v, etc., combined with the vulnerability of buildings, infrastructure, and personnel, the loss caused by the landslide to the hazard-bearing body is estimated, and finally the risk prediction is performed based on the loss probability distribution obtained by statistical calculation.

[0039] Further, in step 8), according to the Bayesian principle, the posterior probability density function can be calculated by formula (15):

[0040] f(x|y)=af(x)f(y|x) (15)

[0041] Where a is the normalization coefficient, f(x) is the prior distribution of parameter x, y is the monitoring data, and f(y|x) is the likelihood function, which can be calculated by formula (16).

[0042]

[0043] Among them, M is the number of samples, F is the theoretical value corresponding to y, ε is the residual vector, and ε i =y i -F i , i = 1, 2, ... M. Assume that ε follows an independent equal variance Gaussian distribution, that is,

[0044] MH sampling is used to implement Markov chain Monte Carlo simulation to solve the posterior probability density function. The process is as follows Figure 6 As shown, where α(i,j) is the acceptance rate, Π(x) is the stationary distribution of the Markov chain, and C is the state transfer matrix of the Markov chain.

[0045] Furthermore, in step 10), the displacement, pore pressure data and rainfall data of the landslide monitoring points are collected in real time, the monitoring data are updated every 3 hours, and then the rock and soil parameters are updated by inversion, the probability distribution characteristics of the landslide failure probability and the model response quantity are recalculated, and the landslide risk probability prediction results are dynamically updated.

[0046] The technical solution provided by the present invention has the following effects:

[0047] 1. The material point numerical simulation method is applied to the large deformation simulation of landslide rock and soil, taking into account the unsaturated hydrodynamic effect, using the two-phase material point method to simulate the entire landslide destruction process, fully considering the catastrophic mechanism of the landslide, and the physical meaning is clear;

[0048] 2. Use the XGBoost algorithm to build a proxy model for the numerical model, greatly improving the efficiency of numerical calculations, thereby improving the timeliness of risk probability prediction;

[0049] 3. Using the Bayesian probability inversion algorithm, combined with real-time monitoring data, the posterior distribution of uncertainty parameters is continuously inverted and calculated, the uncertainty of each parameter variable is reduced, and the calculation model is continuously approached to the actual situation. In this way, the landslide probability prediction results are updated and corrected in real time, thereby improving the accuracy of risk probability prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of the intelligent prediction method for landslide disaster risk probability based on the two-phase material point method;

[0051] Figure 2 Calculation model diagram for two-phase material points of landslide;

[0052] Figure 3 This is the final simulation result of the two-phase material point method under a set of sample parameters;

[0053] Figure 4 The simulation results of the displacement and pore pressure of the monitoring point at different times under a set of sample parameters;

[0054] Figure 5 This is the regression analysis chart of the sliding distance test set of the proxy model;

[0055] Figure 6 Solve the flow chart for the Markov chain;

[0056] Figure 7 are the prior and posterior distributions of the elastic modulus;

[0057] Figure 8 To update the probability distribution map of landslide sliding distance before and after;

[0058] Fig. 9 To update the probability distribution map of landslide disaster risk before and after; DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0060] See also Figure 1 This embodiment discloses a method for intelligent prediction of landslide disaster risk probability based on a two-phase material point method, comprising the following steps:

[0061] 1) Taking an actual slope as an example, the prior distribution of the slope rock and soil parameters and local rainfall intensity is obtained through exploration and monitoring, as shown in Table 1.

[0062] Table 1 Prior distribution characteristics of rock and soil parameters and rainfall intensity

[0063]

[0064] Where E is the elastic modulus (kPa), c is the cohesion c (kPa), φ is the internal friction angle (o), k is the permeability coefficient (m / h), and q is the rainfall intensity (m / h).

[0065] 2) Based on the prior distribution of rock and soil parameters and rainfall intensity, the Latin hypercube sampling method (LSH) was used to extract a 5×2000 sample matrix to construct the input data set;

[0066] 3) Select a typical slope section, and establish a two-phase material point model of the slope based on the stratum data and the rock and soil parameters and rainfall intensity samples in step 2). The model diagram is as follows: Figure 2 As shown in the figure, the model is 600m wide, 175m high, and has a slope of 45°. A total of 122,750 material points are used for discretization. The initial material point spacing is 1m, the background grid size is 2m*2m, and each grid unit contains 4 material points, totaling 27,391 grid nodes. The rock and soil parameters are assigned to each material point, gravity is set as the external load, the bottom edge is fixed with xz direction displacement, the two sides are fixed with x direction displacement, the bottom edge is an impermeable interface, and the head boundaries are set on both sides according to the groundwater level. Monitoring points are arranged on the slope, and the initial ground stress balance calculation is performed.

[0067] 4) Based on the two-phase material point method, the motion equations of landslide rock and soil and pore fluid are discretized in material point space and solved by explicit time integration. The solution time step is 1s, and the total number of steps is 80,000. The whole process of the failure evolution of unsaturated rock and soil in the landslide under 2,000 groups of different sample parameters is simulated. Figure 3 The simulation results at the final moment under one set of sample parameters show that the sliding range is 110.3m, the sliding distance is 158.6m, the sliding body thickness is 55.2m, and the sliding body volume is 14398.9m 3 , maximum sliding speed 12.6m / s, Figure 4 The displacement and pore pressure simulation results of the monitoring points at different times under this working condition;

[0068] 5) constructing an output data set based on 2000 sets of pore pressures, displacement values ​​and model response quantities with time series such as sliding distance s, collapse range S, sliding body thickness h, sliding body volume V, sliding velocity v, etc. calculated in step 4);

[0069] 6) Based on the data set obtained in step 2) and step 5), the training set and the test set were divided into a ratio of 7:3. The XGBoost model was used to train and test the data set. The Optuna library was used to optimize the hyperparameters. The optimal combination of hyperparameters was obtained as follows: maximum tree depth 7, learning rate 0.06, number of decision trees 257, sample random sampling ratio 0.7, minimum child node weight sum 6, regularization parameter 0.5, and down sampling rate 0.9. Based on this hyperparameter combination, a proxy model of the numerical model was constructed. Figure 5 Taking sliding distance as an example, this is the regression analysis chart of the proxy model test set;

[0070] 7) Based on the prior distribution of each parameter, random sampling was performed through the Monte Carlo simulation algorithm, and the proxy model was used for calculation. For each group of samples, the failure criterion was the plastic zone penetration. The final statistical result was that the failure probability P of the landslide was 80.3%, and the initial probability distribution of the response (taking the sliding distance as an example) was 80.3%. Figure 8 For each group of samples, the hazard-bearing body that may be affected by the landslide is evaluated based on the sliding distance s, the sliding range S, etc., combined with the distribution of buildings, infrastructure, and personnel. Then, based on the sliding body thickness h, sliding body volume V, sliding speed v, etc., combined with the vulnerability of buildings, infrastructure, and personnel, the loss caused by the landslide to the hazard-bearing body is estimated. Finally, the initial probability distribution of personnel loss and economic loss is obtained by statistics (as shown in Fig. 9 shown);

[0071] 8) Monitoring instruments were arranged at the corresponding positions on the slope. After the rainfall started, the displacement and pore pressure data of the monitoring points were collected every 3 hours. The Bayesian parameter probability inversion method was used to construct the likelihood function to carry out parameter inversion calculation and solve the Markov chain of the parameters. The number of Markov chains was initially set to 10, and the maximum sample size on each chain was 5000. The posterior distribution of the rock and soil parameters was obtained. Figure 8 The comparison of the probability density curves of the prior distribution and the posterior distribution of the elastic modulus shows that the uncertainty of the rock and soil parameters is significantly reduced after the inversion;

[0072] 9) Based on the posterior distribution of each parameter obtained by inversion, the Monte Carlo simulation algorithm is used again to update the landslide failure probability P to 82.6%, and the probability distribution characteristics of each model response are recalculated. Figure 8 Compare the initial probability distribution of the model response sliding distance with the updated probability distribution, and update the probability distribution of personnel loss and economic loss (such as Fig. 9 shown);

[0073] 10) Using the posterior distribution of the rock and soil parameters obtained in step 8) as the prior distribution, repeating steps 8) to 9) through real-time collected landslide displacement, pore pressure monitoring data and rainfall data, inversely updating the rock and soil parameters, recalculating the probability distribution characteristics of the landslide failure probability and the model response quantity, and realizing real-time dynamic update of the risk probability prediction results.

[0074] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. Similarly, in order to simplify the present invention and help understand one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. Wherein, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself is a separate embodiment of the present invention.

[0075] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim that lists a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

Claims

1. A landslide disaster risk probability intelligent prediction method based on two-phase material point method, characterized in that: The following steps are involved: 1) According to the landslide investigation data and local rainfall monitoring data, the parameters of the landslide rock and soil mass and the prior distribution of the rainfall intensity q are obtained, wherein the parameters of the landslide rock and soil mass include elastic modulus E, cohesion c, internal friction angle φ, and permeability coefficient k; 2) Based on the prior distribution of various rock and soil parameters and rainfall intensity, Latin hypercube sampling method is used to extract random samples to construct the input data set; 3) According to the landslide stratum data, a two-phase material point numerical calculation model is constructed for each set of sample parameters in step 2), and the initial conditions, boundary conditions and load conditions of the numerical model are set according to the actual situation; 4) Based on the two-phase material point method, the motion equations of landslide rock and soil and pore fluid are discretized in material point space and solved by explicit time integration to simulate the whole process of destruction evolution of unsaturated rock and soil of landslide under the action of hydrodynamics; 5) Based on the two-phase material point method, the displacement and pore pressure values ​​of the monitoring points with time series are calculated for parameter inversion, and the final response of each sample is obtained to construct an output data set. The final response includes sliding distance s, collapse range S, sliding body thickness h, sliding body volume V, and sliding velocity v; 6) Based on the data sets obtained in step 2) and step 5), the XGBoost model is used for training and testing to construct a proxy model of the numerical model; 7) Based on the prior distribution of each parameter, the Monte Carlo simulation algorithm is used in combination with the proxy model to statistically calculate the probability of landslide failure and the probability distribution characteristics of each response quantity, and combined with the characteristics of the hazard-bearing body, the initial landslide disaster risk probability prediction results are calculated; 8) Monitoring instruments are arranged on the slope to collect monitoring data in real time. The detection data includes displacement and pore pressure. Parameter inversion is performed based on the monitoring data using the Bayesian probability inversion method to calculate the posterior distribution of rock and soil parameters. The rainfall intensity q is determined based on the measured data; 9) Based on the posterior distribution of each rock and soil parameter obtained by inversion, the Monte Carlo simulation algorithm is used again to statistically calculate the failure probability and the probability distribution characteristics of each model response, and the risk probability prediction results are updated; 10) Using the posterior distribution of the rock and soil parameters obtained in step 8) as the prior distribution, repeating steps 8) to 9) through real-time collected landslide displacement, pore pressure monitoring data and rainfall data, inversely updating the rock and soil parameters, and dynamically updating the risk probability prediction results in real time.

2. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 2), the model parameter variable sample is represented by X = [x1, x2, ..., x n ], where x i represents the elastic modulus E, cohesion c, and internal friction angle φ parameters, n represents the dimension of the model parameter variable, and the number of samples required to be extracted in each dimension is m. When using the Latin hypercube sampling method, if m n-dimensional samples that conform to the prior distribution of the model parameters are to be extracted in the sample space, then the [0,1] interval needs to be divided into m intervals, and a sample is randomly selected in each interval, and the extracted values ​​are mapped to the sample space to obtain an n×m-order sample matrix A, thereby constructing the model input data set.

3. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 3), a three-dimensional geological model of the landslide body is constructed based on the three-dimensional stratigraphic data of the landslide area, and the model is discretized into spatial material points. At the same time, a background grid covering the entire landslide sliding area is set behind the material points, and the rock and soil parameters are assigned to each material point. The gravity and external load conditions are input, the displacement boundary conditions and the seepage boundary conditions are set, the initial groundwater level is set, and the initial ground stress balance is performed.

4. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 4), the rock mass is described by the DP elastoplastic constitutive model, and the pore fluid is described by Darcy's law. Combined with the effective stress principle, soil-water characteristic curve and permeability curve, the Euler-Cromer explicit integration algorithm is used to solve the problem.

5. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 6), the specific steps of using the XGBoost model are as follows: ① Take the input and output data obtained in step 2) and step 5) as the data set; ② Divide the data set into training set and test set; ③Define the XGBoost model parameters, including the maximum depth of the tree, learning rate, number of iterations, and objective function; ④ Use the training set data to train the XGBoost model; ⑤ Use the cross-validation method to adjust the model parameters and find the optimal parameter combination; ⑥Use the test set to evaluate the performance of the model.

6. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 7), based on the sliding distance s, the collapse range S, and the distribution of buildings, infrastructure, and personnel, the hazard-bearing body that may be affected by the landslide is evaluated. Then, based on the sliding body thickness h, sliding body volume V, and sliding speed v, and the vulnerability of buildings, infrastructure, and personnel, the loss caused by the landslide to the hazard-bearing body is estimated. Finally, risk prediction is performed based on the loss probability distribution obtained by statistical calculation.

7. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 8), according to the prior distribution of rock and soil parameters, the Bayesian parameter probability inversion method is used to construct the likelihood function to carry out parameter inversion calculation, and the Kalkov chain Monte Carlo simulation is used to solve the posterior probability density function, and the posterior distribution of the rock and soil parameters is calculated to reduce the uncertainty of the rock and soil parameters.

8. The method for intelligent prediction of landslide disaster risk probability based on two-phase material point method according to claim 1, characterized in that: In step 10), the displacement, pore pressure data and rainfall data of the landslide monitoring point are collected in real time, the monitoring data are updated every 3 hours, and then the rock and soil parameters are updated by inversion, the probability distribution characteristics of the landslide failure probability and the model response quantity are recalculated, and the landslide risk probability prediction results are dynamically updated.

Images