A method for determining landslide rainfall threshold curve based on machine learning

Abstract

The present invention relates to a method for determining a landslide rainfall threshold curve based on machine learning, constructing a slope model sample, determining a slope stability safety factor threshold, and an average annual rainfall in the current study area; calculating the seepage field of the sample under a given rainfall intensity, capturing the time point when the slope safety factor is equal to the slope stability safety factor threshold, and using the corresponding rainfall intensity and rainfall duration as the fitting data point of the I-D threshold curve; constructing an I-D threshold curve data set; using an artificial neural network to construct an I-D threshold curve prediction model, and training the prediction model; inputting slope parameters into the prediction model, obtaining the slope fitting data points, and fitting and drawing the I-D threshold curve on a coordinate system. The results obtained by the present invention are more accurate; the rapid prediction of the I-D threshold curve is achieved, the analysis efficiency is improved, and the I-D threshold curve has good practical value; and the method has generalizability.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster early warning technology, and in particular to a method for determining a landslide rainfall threshold curve based on machine learning in the field of landslide disaster early warning. Background Art

[0002] Landslide is one of the major geological disasters, most of which are triggered by rainfall. Rainfall-induced landslides are characterized by group and sudden occurrence, and are difficult to identify in advance. Therefore, establishing a practical rainfall-induced landslide early warning system has become an important task related to the safety of people's lives and property. The most commonly used early warning method is the rainfall intensity-rainfall duration threshold (ID threshold curve) based on historical landslide data statistics, which is generally expressed as I = aD -b , the threshold curve is plotted on the coordinate system. When the point corresponding to the rainfall event is above the curve, it is considered that the rainfall event will cause landslide disasters in this area.

[0003] The traditional ID threshold curve obtained based on statistical analysis of historical landslide data has the following problems:

[0004] 1. Its quality is greatly affected by objective statistical data. In areas where data is missing or incomplete, its accuracy will be greatly affected;

[0005] 2. The rainfall threshold based on empirical statistics does not consider the interaction mechanism between the landslide rock and soil and the external environment (rainwater, groundwater), and ignores the geological conditions of the slope itself, which often leads to many misjudgments;

[0006] 3. This type of ID threshold curve is usually a regional warning, which cannot accurately locate the time and space of landslide disasters, and cannot provide accurate guidance for relevant government departments to carry out disaster prevention and relief work.

[0007] In the prior art, many scholars have proposed a rainfall-type landslide early warning method based on a physical process model. The idea is to use a physical process model to simulate the hydrological process and water distribution of a single slope or basin under rainfall conditions, and then use mechanical analysis to evaluate the stability of the research object. The ID threshold curve of the single slope or basin is back-calculated using a hydraulic coupling model. Compared with the traditional statistical ID threshold curve based on historical landslide data, this method takes into account the geological characteristics of the research object and the hydrological process under rainfall conditions. Its analysis is more targeted, so the prediction results are more accurate; however, this method consumes a lot of computing power. For example, for slope stability evaluation of a large range of natural slope areas, the workload of slope modeling is extremely large, and the analysis efficiency is greatly limited, making it difficult to promote on a large scale. Summary of the invention

[0008] The present invention solves the problems existing in the prior art and provides a method for determining a landslide rainfall threshold curve based on machine learning.

[0009] The technical solution adopted by the present invention is a method for determining a landslide rainfall threshold curve based on machine learning, the method comprising the following steps:

[0010] Step 1: Obtain the slope shape parameters and geotechnical engineering parameter value ranges in the study area;

[0011] Step 2: Based on the range of slope parameter values, the uniform design method is used to construct a slope model sample and determine the slope stability safety factor threshold F st , the average annual rainfall in the current study area A;

[0012] Step 3: Based on the slope model sample, corresponding to the rainfall intensity I given by group L, the rainfall intensity I is input into the model and the seepage field is calculated, and the slope safety factor F is calculated along the time series. s , capturing the safety factor F s Equal to the slope stability safety factor threshold F st At a time point, record the rainfall intensity I and rainfall duration D corresponding to the time point as the fitting data point of the rainfall-type landslide ID threshold curve;

[0013] Step 4: Repeat step 3 until the coordinates of L fitting data points are recorded;

[0014] Step 5: Repeat steps 3 to 4 for each slope sample until a set of fitting data point coordinates is corresponding to each slope sample, and an ID threshold curve data set is constructed;

[0015] Step 6: Use artificial neural network to build ID threshold curve prediction model, set the slope model sample slope parameters and geotechnical engineering parameters in step 2 as input variables, set the coordinates of ID threshold curve fitting data points as response variables, and use ID threshold curve data set as learning samples for training to obtain rainfall threshold prediction model;

[0016] Step 7: Input the slope parameters to be determined into the prediction model to obtain L fitting data points of the slope ID threshold curve, and obtain the slope ID threshold curve I = aD by least squares fitting. -b , draw the ID threshold curve on the coordinate system.

[0017] In the present invention, the main technical principles are:

[0018] First, based on the survey of slope parameter distribution intervals, several hypothetical slopes are constructed through the uniform design method. These hypothetical slopes will be used as data sets for the training of machine learning models.

[0019] Second, using the unsaturated soil mechanics theory as the basis for the analysis, the slope seepage simulation and stability analysis under rainfall conditions were carried out on the hypothetical slope, and the rainfall intensity (I) and rainfall duration (D) data points were captured in the time series of rainfall-landslide formation. These data points will be used as the output in the machine learning model training process;

[0020] 3. Using machine learning algorithms, such as artificial neural networks and support vector machines, the above slope data sets are trained, with slope shape and geotechnical parameters as input and ID curve fitting points as output. The final machine learning model can realize the rapid prediction of slope ID threshold curve.

[0021] Preferably, in step 1, the slope shape parameters include slope gradient and slope height, and the geotechnical engineering parameters include soil saturation density, saturated permeability coefficient, effective stress intensity index, and soil-water characteristic curve.

[0022] Preferably, in step 3, for each slope model sample, the seepage field simulation is started under the condition of the rainfall intensity I given by group L, the pore water pressure distribution in the slope is calculated by combining the two-dimensional unsaturated seepage model, and the two-dimensional slope stability safety factor F is calculated by the limit equilibrium method. s .

[0023] Preferably, the two-dimensional unsaturated seepage model is a two-dimensional unsaturated transient seepage model;

[0024]

[0025] Among them, k x , k y are the current horizontal and vertical permeability coefficients of the soil, H is the current total water head, γ w is the density of water, m w is the slope of the linear segment of the soil-water characteristic curve, t is the current time, Q is the boundary flow, which is converted according to the rainfall intensity; the Galerkin method and the post-difference method are applied to obtain the corresponding steady-state and transient discrete seepage control equations. Combined with the boundary conditions, the Gaussian integral method is used to solve the discrete domain to obtain the seepage field distribution.

[0026] Preferably, the boundary conditions of the model are:

[0027] Infiltration boundary: The slope surface is the seepage surface; when the pore water pressure is less than 0, the flow boundary is taken, and its value is equal to the rainfall intensity, otherwise it becomes the head boundary, and the total head is equal to its elevation; a unit gradient boundary is set at the bottom of the bedrock area, which is used for shallow soil landslides with deep groundwater.

[0028] Preferably, before the transient seepage analysis, the initial pore pressure distribution state inside the slope soil must be determined first. Based on historical statistical data, the annual average rainfall in the area where the slope is located is estimated, converted into the corresponding flow boundary and set at the infiltration boundary. The seepage analysis is performed to a steady state, and the result is used as the initial wet state of the slope. The initial pore pressure distribution of the transient seepage analysis is obtained through a two-dimensional unsaturated steady-state seepage analysis to meet the requirements.

[0029]

[0030] Among them, k x , k y is the current horizontal and vertical permeability of the soil, H is the current total water head, and Q0 is the boundary flow, which is converted according to the average annual rainfall.

[0031] Preferably, the limit equilibrium method is used to carry out slope stability analysis. The limit equilibrium method is used to calculate the two-dimensional slope stability safety factor F s The most dangerous sliding surface is searched by the grid method. For this sliding surface, the shape of the sliding surface is optimized by segmentation technology. The optimization process is to divide the sliding surface into several straight line segments, move the endpoints of the line segments to explore a lower safety factor, and approach the global most dangerous sliding surface.

[0032] Preferably, the unsaturated soil shear strength τ satisfies,

[0033]

[0034] Where c′ is the effective cohesion of soil, in kPa. is the effective internal friction angle of soil, in degrees, and both are determined through investigation, sampling and testing; (σ-u a ) is the net normal stress of the soil, (u a -u w ) is the current soil matrix suction value, both of which are in kPa; θ w is the current soil volume water content, θ r is the residual volume moisture content of soil, θ s is the saturated volumetric water content of the soil; calculate the safety factor F s It is necessary to calculate the anti-sliding force and the sliding force, and the unsaturated shear strength reflects the anti-sliding force.

[0035] Preferably, in step 4, the value of rainfall intensity I is based on the soil saturated permeability coefficient k s Determined by proportion; if L is 5, the value of rainfall intensity I is based on the soil saturated permeability coefficient k s OK, and divide it into 0.3k s , 0.6k s , 0.9k s , 1.2ks , 1.5k s Five levels.

[0036] Preferably, in step 2, the slope stability safety factor threshold F st Determined according to the threshold level; for example, 1.0, 1.05, and 1.15 are taken for red warning, orange warning, and yellow warning respectively.

[0037] The present invention relates to a method for determining a landslide rainfall threshold curve based on machine learning, obtaining slope shape parameters and geotechnical engineering parameter value ranges in a study area, constructing a slope model sample, and determining a slope stability safety factor threshold F. st , the average annual rainfall A in the current study area; calculate the seepage field of the slope model sample under the given rainfall intensity I, and calculate the slope safety factor F along the time series s , capturing the safety factor F s Equal to the slope stability safety factor threshold F st At the time point, record the rainfall intensity I and rainfall duration D corresponding to the time point as the fitting data point of the rainfall-type landslide ID threshold curve; repeat until the coordinates of L fitting data points are recorded, and a set of fitting data point coordinates is corresponding to each slope sample to construct the ID threshold curve data set; use artificial neural network to build the ID threshold curve prediction model, set the slope shape parameters and geotechnical engineering parameters of the study area as input variables, set the coordinates of the ID threshold curve fitting data points as response variables, and use the ID threshold curve data set as learning samples to train the prediction model; input the slope parameters into the prediction model to obtain L fitting data points of the slope ID threshold curve, and obtain the slope ID threshold curve I=aD by least squares fitting. -b , draw the ID threshold curve on the coordinate system.

[0038] The beneficial effects of the present invention are:

[0039] (1) The rainfall-induced landslide ID threshold curve prediction method is based on a physical process model, taking into account the geological characteristics of the research object and the hydrological process under rainfall conditions. Compared with the traditional analysis method based on historical data analysis, the results obtained are more accurate;

[0040] (2) Compared with the traditional analysis method based on physical process model, the present invention introduces machine learning model to construct physical process model proxy model, thereby realizing rapid prediction of ID threshold curve, improving analysis efficiency and having good practical value;

[0041] (3) It is scalable. After the model is constructed, even non-professionals can use the model to predict the slope ID threshold curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the slope profile and boundary conditions in the present invention;

[0043] Figure 2 It is a schematic diagram of a flow chart in the present invention;

[0044] Figure 3 It is a schematic diagram of slope shape parameters in the embodiment;

[0045] Figure 4 It is the soil-water characteristic curve and relative permeability coefficient function diagram in the embodiment;

[0046] Figure 5 is a performance diagram of the artificial neural network after training in the embodiment;

[0047] Figure 6 It is the ID threshold curve prediction for a certain slope case in the embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] like Figure 1 to Figure 4 As shown, the method of the present invention is used to construct a rainfall-type soil shallow landslide ID threshold curve prediction model for a certain area.

[0050] The studied area is a third-level watershed in Zhejiang Province. The geological disasters in this area are mainly shallow soil landslides, and they mainly occur on convex slopes ( Figure 1 ). Every rainy season, this type of landslide causes huge loss of life and property in the local area. It is urgent to build an accurate rainfall-type landslide ID threshold curve for each geological disaster risk point. The following is an explanation of this solution in conjunction with this case.

[0051] Figure 2 It is a schematic diagram of the technical route.

[0052] Step 1: Collect and compile statistics on the geological environment of the slopes in the study area to determine the main value ranges of the slope shape and geotechnical engineering parameters in the area.

[0053] The terrain in this area is complex, mainly mountainous and hilly, with an average annual rainfall of about 1500mm. Landslides mainly occur in the loose accumulation layer of the Quaternary, and the thickness of the sliding body is basically within 6m. It is a shallow landslide. The underlying bedrock is mainly medium-weathered granite or tuff with weak permeability. The statistics of the slope terrain parameters in this area are mainly based on geological surveys, satellite image analysis, and DEM data extraction. The statistics of rock and soil mechanics parameters are mainly based on geological surveys, soil sample test determination, and literature research. The convex slope is generalized as a broken line slope. The required slope parameters are shown in Figure 3 In order to construct a slope model sample database, the main value ranges of the slope shape and geotechnical parameters in the area were statistically analyzed. The statistical results are shown in Table 1. The slope height ratio, slope gradient ratio, and relative burial depth are indirect parameters that reflect the slope shape. The slope shape of the slope sample can be inversely calculated through the indirect parameters. Among them, the definition of relative burial depth satisfies

[0054] In the formula, z max satisfy,

[0055]

[0056] It reflects the maximum value of the thickness of the soil layer of the slope. If the thickness exceeds this value, the slope will be unstable in its natural state and cannot exist.

[0057] z min satisfy,

[0058]

[0059] It reflects the minimum thickness of the soil layer of the slope. Below this value, the slope will not become unstable in a fully saturated state and no early warning is required.

[0060] By z r The soil thickness of the slope sample can be back-calculated;

[0061] Among them, γ s —Soil saturation density (kN·m -3 ), c′—effective cohesion of soil (kPa), —Effective internal friction angle (°), α—Slope of the steeper section of the broken line slope (°), γ w — Weight of water (kN·m -3 ).

[0062] Table 1: Slope shape and geotechnical parameters range

[0063]

[0064]

[0065] Step 2: Based on the parameter statistics results, construct the ID threshold curve data set.

[0066] Based on the range of values ​​of each parameter listed in Table 1, the slope sample data set is constructed by the uniform design method. 101 (101 9 ), and based on this table, 101 slope models were constructed. The boundary conditions of the models are shown in Figure 3 ,include:

[0067] (1) Infiltration boundary: The slope surface is the seepage surface. When the pore water pressure is less than 0, the flow boundary is taken, and its value is equal to the rainfall intensity. Otherwise, it becomes the head boundary, and the total head is equal to its elevation.

[0068] (2) A unit gradient boundary is set at the bottom of the bedrock area to simulate the drainage of slope water;

[0069] (3) Both sides of the model are set as impermeable interfaces.

[0070] In terms of material properties, the variability of the soil-water characteristic curve and relative permeability function of the soil is not considered, and the curve used is shown in Figure 4 As a simplification, the soil-water characteristic curve and relative permeability function of the underlying bedrock are consistent with those of the soil, and the saturated permeability of the bedrock is 1 / 10 of that of the overlying soil.

[0071] This example adopts a two-dimensional unsaturated soil seepage model It is used for seepage simulation under rainfall conditions. The Morgenstern-Price strip method is used for slope stability analysis, and the shear strength of unsaturated soil is calculated using the Vanapalli formula.

[0072] In order to conduct seepage analysis, the initial wetting conditions of the model must be determined. The principle of determining the initial wetting conditions in this embodiment is as follows: -8 A rainfall condition of 1500 mm / s (converted from the annual average rainfall of 1500 mm) is applied to the infiltration boundary, and the seepage analysis is performed to a steady state to obtain the initial pore pressure profile.

[0073] For each slope model, the value of the applied rainfall intensity I is determined according to the soil saturated permeability coefficient k s Consider the rainfall intensity of 0.3k s , 0.6k s , 0.9k s , 1.2k s , 1.5k s Five rainfall conditions, safety factor threshold F st Take 1.0. Calculate the slope safety factor F along the time series s , capturing the safety factor Fs Equal to the slope stability safety factor threshold F st The time point of the rainfall-type landslide is used as the fitting data point of the ID threshold curve. Five fitting data points are obtained for each slope model. After the calculation of 101 slope models is completed, the ID threshold curve data set is obtained.

[0074] Step 3: Construct an ID threshold curve prediction model through machine learning algorithm.

[0075] Based on the ID threshold curve data set, this example uses a three-layer BP neural network to construct a rainfall-type landslide ID threshold curve prediction model. The specific steps are as follows:

[0076] First, the ID threshold curve dataset is divided into a training set and a test set in a ratio of 8:2. The training set is used for model training and hyperparameter optimization. After the training and parameter adjustment are completed, the final model is tested on the test set to determine the model performance.

[0077] Secondly, the 9 parameters described in Table 1 were set as predictor variables, and the horizontal coordinates of the ID threshold curve fitting data points were set as the response variables (the vertical coordinates were known to be 0.3k s , 0.6k s , 0.9k s , 1.2k s , 1.5k s ). In this embodiment, only the number of hidden layer neurons of the BP neural network is considered as the hyperparameter that needs to be adjusted. The 10-fold cross-validation method is adopted, that is, the training group is divided into 10 equal-sized parts, 9 of which are used as training groups and 1 as validation groups in turn, and 10 tests are carried out. The average mean square error under repeated validation tests is used as the evaluation standard of model performance. The value of the number of hidden layer neurons is continuously changed to determine the most suitable BP neural network structure to minimize the average mean square error. In this embodiment, the number of hidden layer neurons is 16.

[0078] Again, evaluate whether the model meets the requirements on the test set. If not, return to step 2 and increase the amount of data in the ID threshold curve data set. If it does, the performance of the model is finally verified. Figure 5 , indicating that the neural network training results are good and no additional data or adjustments are needed.

[0079] Finally, the slope shape and geotechnical parameters of the slope to be studied are input into the prediction model as prediction variables, and the output values ​​are the rainfall intensity of 0.3k s , 0.6k s , 0.9k s , 1.2k s , 1.5ks The rainfall duration required to issue a landslide warning is 1 / 2000. These five rainfall intensity-rainfall duration combinations constitute five fitting data points. The slope ID threshold curve I = aD is obtained by least squares fitting. -b , thus drawing the ID threshold curve on the coordinate system. The parameters of the verification case are shown in Table 2, and the predicted ID threshold curve is shown in Figure 6 . It has been verified that the model has good prediction performance.

[0080] Table 2: Slope shape and geotechnical parameters of the verification case

[0081]

[0082] The above embodiments are only used to explain the concept of the present invention, rather than to limit the protection of the rights of the present invention. Any non-substantial changes to the present invention using this concept shall fall within the protection scope of the present invention.

[0083] In order to achieve the above content, the present invention also relates to a computer-readable storage medium, on which a program for determining a landslide rainfall threshold curve based on machine learning is stored. When the program is executed by a processor, the method for determining a landslide rainfall threshold curve based on machine learning is implemented.

[0084] In order to achieve the above content, the present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for determining the landslide rainfall threshold curve based on machine learning is implemented.

[0085] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0090] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for determining a landslide rainfall threshold curve based on machine learning, characterized in that: The method comprises the following steps: Step 1: Obtain the slope shape parameters and geotechnical engineering parameter value ranges in the study area; Step 2: Based on the range of slope parameter values, construct a slope model sample and determine the slope stability safety factor threshold F st , the average annual rainfall in the current study area A; Step 3: Based on the slope model sample, corresponding to the given rainfall intensity I of group L, calculate the slope safety factor F along the time series s For each slope model sample, the seepage field simulation is started under the condition of rainfall intensity I given by group L, the pore water pressure distribution in the slope is calculated by combining the two-dimensional unsaturated seepage model, and the two-dimensional slope stability safety factor F is calculated by the limit equilibrium method. s ; The two-dimensional unsaturated seepage model is a two-dimensional unsaturated transient seepage model; ， in, , are the current horizontal and vertical permeability coefficients of the soil, is the current total water head, is the weight of water, is the slope of the linear segment of the soil-water characteristic curve, is the current time, is the boundary flow rate; the Galerkin method and post-difference method are used to obtain the corresponding steady-state and transient discrete seepage control equations. Combined with the boundary conditions, the Gaussian integral method is used to solve the discrete domain to obtain the seepage field distribution; The initial pore pressure distribution of transient seepage analysis is obtained by two-dimensional unsaturated steady-state seepage analysis, satisfying ， in, , is the current horizontal and vertical permeability coefficient of the soil, is the current total water head, is the boundary flow, converted according to the average annual rainfall; Capture safety factor F s Equal to the slope stability safety factor threshold F st At a time point, record the rainfall intensity I and rainfall duration D corresponding to the time point as the fitting data point of the rainfall-type landslide ID threshold curve; Step 4: Repeat step 3 until the coordinates of L fitting data points are recorded; Step 5: Repeat steps 3 to 4 for each slope sample until a set of fitting data point coordinates is corresponding to each slope sample, and an ID threshold curve data set is constructed; Step 6: Use artificial neural network to build ID threshold curve prediction model, set the slope model sample slope parameters and geotechnical engineering parameters in step 2 as input variables, set the coordinates of ID threshold curve fitting data points as response variables, and use ID threshold curve data set as learning samples for training to obtain rainfall threshold prediction model; Step 7: Input the slope parameters to be determined into the prediction model to obtain L fitting data points of the slope ID threshold curve, and obtain the slope ID threshold curve by least squares fitting. , draw the ID threshold curve on the coordinate system.

2. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 1, characterized in that: In step 1, the slope shape parameters include slope gradient and slope height, and the geotechnical engineering parameters include soil saturation density, saturated permeability coefficient, effective stress intensity index, and soil-water characteristic curve.

3. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 1, characterized in that: The boundary conditions of the model are: Infiltration boundary: The slope surface is the seepage surface; when the pore water pressure is less than 0, the flow boundary is taken, and its value is equal to the rainfall intensity; otherwise, it becomes the head boundary, and the total head is equal to its elevation; A unit gradient boundary is set at the bottom of the bedrock area for shallow soil landslides with deep groundwater.

4. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 1, characterized in that: The safety factor F of two-dimensional slope stability is obtained by limit equilibrium method s The most dangerous sliding surface is searched by the grid method. For this sliding surface, the shape of the sliding surface is optimized by segmentation technology. The optimization process is to divide the sliding surface into several straight line segments, move the endpoints of the line segments to explore a lower safety factor, and approach the global most dangerous sliding surface.

5. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 4 is characterized in that: Calculate the safety factor F s It is necessary to calculate the anti-sliding force and the sliding force. The shear strength of unsaturated soil reflects the anti-sliding force. satisfy, ， in, is the effective cohesion of soil, is the effective internal friction angle of soil, is the net normal stress of soil, is the current soil matrix suction value, is the current soil volume water content, is the residual volume moisture content of the soil, is the saturated volumetric water content of soil.

6. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 1, characterized in that: In step 3, the rainfall intensity I is based on the soil saturated permeability coefficient Determined in proportion.

7. The method for determining a landslide rainfall threshold curve based on machine learning according to claim 1, characterized in that: In step 2, the slope stability safety factor threshold F st Determined by threshold level.

Images