Soil erosion early warning methods and devices

Abstract

This invention relates to the field of soil erosion monitoring technology, specifically to a soil erosion early warning method and device. The soil erosion early warning method includes: acquiring soil data corresponding to different locations and depths of a target slope; acquiring environmental data corresponding to the target slope; calculating soil mechanical data corresponding to different locations and depths of the target slope based on the soil data; determining the erosion level corresponding to the target slope based on the soil mechanical data and the soil data; and outputting early warning information according to the erosion level. This achieves a comprehensive assessment of the degree of slope erosion. The output of early warning information based on the erosion level is the implementation of the core value of the early warning system in "disaster prevention and mitigation," and its beneficial effects lie in achieving "advanced early warning" and "precise response."

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of soil erosion monitoring, in particular to a soil erosion early warning method and device. BACKGROUND

[0002] As an important land resource reserve area, the stability of the saline soil slope surface in the coastal reclamation area directly determines whether the land resources can be safely used, and is the core bottleneck restricting regional agricultural development, engineering construction and ecological protection. In the coastal environment, the saline soil slope surface is long-term affected by seawater infiltration, rainfall erosion and evaporation, and frequently experiences dry-wet cycles and desalination-re-salination processes, resulting in frequent slope instability events. Such instability not only causes land degradation and reduction of arable land, but also can damage surrounding water conservancy facilities, transportation lines and other infrastructure, posing a serious threat to regional economic development and ecological safety. Therefore, it is urgent to establish an effective monitoring and early warning means for the stability of the saline soil slope surface in the coastal area to ensure the sustainable use of land resources.

[0003] At present, the monitoring of slope stability in the industry relies on traditional technologies such as GNSS and inclinometer. The core capability of such technologies is to capture macroscopic deformation of the slope surface (such as displacement and settlement), which can only obtain the final result after the occurrence of instability and cannot provide early warning before the occurrence of instability. Therefore, how to provide early warning for soil erosion has become a problem to be solved. SUMMARY

[0004] The present application provides a soil erosion early warning method and device to solve the problem of how to provide early warning for soil erosion.

[0005] In a first aspect, the present application provides a soil erosion early warning method, which comprises:

[0006] obtaining soil data corresponding to soil at different positions and depths of a target slope body; the soil data includes at least one of volumetric water content, soil temperature, electrical conductivity, pH value and soil stress; obtaining environmental data corresponding to the target slope body; the environmental data includes rainfall data, wind speed data, wind direction data and evapotranspiration data; based on the soil data, calculating soil mechanical data corresponding to the soil at different positions and depths of the target slope body; the soil mechanical data includes at least one of electrostatic repulsion, van der Waals force, hydration repulsion and liquid bridge force; based on the soil mechanical data and the soil data, determining an erosion grade corresponding to the target slope body; and outputting early warning information according to the erosion grade.

[0007] In an optional implementation, based on the soil data, the soil mechanical data corresponding to different positions and depths of the target slope body are calculated, including: determining the cation concentration in the bulk solution based on the electrical conductivity for the soil at different positions and depths of the target slope body; converting the soil temperature to obtain the absolute temperature; calculating the electrostatic repulsion in the soil mechanical data based on the cation concentration, the absolute temperature, and the corresponding inter-particle distance of the soil; obtaining the initial Hamaker constant; correcting the initial Hamaker constant based on the electrical conductivity, the volumetric water content, and the soil temperature to obtain the target Hamaker constant; calculating the van der Waals force in the soil mechanical data based on the target Hamaker constant; calculating the hydration repulsion in the soil mechanical data according to the inter-particle distance; calculating the salt crystal precipitation degree based on the electrical conductivity and the volumetric water content; obtaining the initial contact angle and the initial liquid bridge volume; correcting the initial contact angle based on the salt crystal precipitation degree to obtain the target contact angle; correcting the initial liquid bridge volume based on the salt crystal precipitation degree to obtain the target liquid bridge volume; and calculating the liquid bridge force based on the target contact angle and the target liquid bridge volume.

[0008] In an optional implementation, based on the soil mechanical data and the soil data, the erosion grade corresponding to the target slope body is determined, including: calculating the soil action force based on the electrostatic repulsion, the van der Waals force, the hydration repulsion, and the liquid bridge force for the soil at different positions and depths of the target slope body; obtaining the target constitutive relation; the target constitutive relation is used to characterize the relationship between the soil action force and the soil cohesion; determining the soil cohesion corresponding to the soil at different positions and depths of the target slope body based on the target constitutive relation; and determining the erosion grade corresponding to the target slope body based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force, and the soil data.

[0009] In an optional implementation, the target constitutive relation is obtained, including: obtaining the initial cohesion, the critical repulsion, and the time decay coefficient corresponding to the soil; determining the action force influence coefficient based on the soil type corresponding to the soil; and constructing the target constitutive relation based on the initial cohesion, the critical repulsion, the time decay coefficient, and the action force influence coefficient.

[0010] In an optional implementation, based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force, and the soil data, the erosion grade corresponding to the target slope body is determined, including:

[0011] The soil at different positions and depths of the target slope body is determined as a node in a dynamic iterative graph structure; for each node, a target node feature vector corresponding to the node is constructed based on soil cohesion, soil force, and soil data corresponding to the node; the target node feature vector includes volume water content, soil temperature, electrical conductivity, pH value, soil stress, soil force, soil cohesion, salt migration trend, strain activity level, and node spatial coordinates; a base weight between any two nodes and a dynamic correction factor are calculated; based on the base weight and the dynamic correction factor, a target edge weight of a target structure edge between any two nodes is calculated, and a dynamic iterative graph structure is constructed; the dynamic iterative graph structure and environmental data are input into a preset particle erosion probability index model, and a particle erosion probability index corresponding to each node in the dynamic iterative graph structure is output; based on the particle erosion probability index corresponding to each node and the target node feature vector, an erosion level corresponding to the target slope body is determined.

[0012] In an optional implementation, calculating the base weight between any two nodes includes: for any two nodes, obtaining a spatial straight-line distance between the two nodes; calculating a physical distance weight between the two nodes according to the spatial straight-line distance; obtaining electrical conductivity and volume water content of the two nodes within a first preset time length before the current time; calculating an electrical conductivity correlation coefficient corresponding to the two nodes based on the electrical conductivity of the two nodes within the first preset time length before the current time; calculating a volume water content correlation coefficient corresponding to the two nodes based on the volume water content of the two nodes within the first preset time length before the current time; and calculating a correlation weight between the two nodes based on the electrical conductivity correlation coefficient and the volume water content correlation coefficient.

[0013] Obtaining a target slope soil texture distribution map corresponding to the target slope body; determining soil textures corresponding to the two nodes based on the target slope soil texture distribution map; determining a soil property weight between the two nodes according to the soil textures corresponding to the two nodes; and determining the base weight between the two nodes based on the physical distance weight, the correlation weight, and the soil property weight.

[0014] In an optional implementation, calculating the dynamic correction factor between any two nodes includes:

[0015] For any two nodes, calculating a salt migration trend and a strain activity level corresponding to the two nodes based on electrical conductivity and soil stress corresponding to the two nodes; calculating a soil state synergy factor corresponding to the two nodes based on the salt migration trend and the strain activity level corresponding to the two nodes; determining an environmental response synergy factor based on environmental data corresponding to the target slope body; and determining the dynamic correction factor based on the soil state synergy factor and the environmental response synergy factor.

[0016] In an optional embodiment, the dynamic iterative graph structure and the environmental data are input into a preset particle erosion probability index model, and a particle erosion probability index corresponding to each node in the dynamic iterative graph structure is output, including: based on the node spatial coordinates corresponding to each node and the environmental data, local environmental data is assigned to each node; a node feature matrix is generated by stacking the target node feature vectors corresponding to each node; an adjacency matrix is constructed based on the target edge weights of the target structure edges between any two nodes; spatial enhanced node features are obtained by performing feature extraction on the node feature matrix and the adjacency matrix; for each node, a time attention weight is calculated based on the relationship between the local environmental data corresponding to the node and the target node feature vector corresponding to the node; a spatial attention weight corresponding to each node is calculated based on the local environmental data corresponding to the node, the soil force and the strain activity level; for each node, each sub-feature included in the spatial enhanced node feature corresponding to the node and the local environmental data is divided into a soil mode, a mechanical mode, a strain mode and an environmental mode; a mode weight corresponding to each sub-feature is calculated; for each node, the sub-features in the node are weighted and fused based on the mode weight corresponding to each sub-feature, the spatial attention weight corresponding to the node and the time attention weight, to generate a multi-modal fusion feature vector corresponding to the node; for each node, the multi-modal fusion feature vector corresponding to the node is stacked in time sequence to form a time sequence feature; the time sequence feature is extracted by a multi-head attention layer and a feedforward neural network to output a final fusion feature; and the final fusion feature is input into a fully connected layer to output a particle erosion probability index corresponding to the node.

[0017] In an optional embodiment, based on the particle erosion probability index corresponding to each node and the target node feature vector, an erosion level corresponding to the target slope body is determined, including: for each node, soil force, soil cohesion, salt migration trend and strain activity level are extracted from the target node feature vector corresponding to the node; based on the particle erosion probability index corresponding to each node, the soil force, the soil cohesion, the salt migration trend and the strain activity level, a risk level corresponding to each node is determined; based on the risk level corresponding to each node, an erosion level corresponding to the target slope body is determined.

[0018] In a second aspect, the embodiments of the present application provide a soil erosion early warning device, which comprises:

[0019] A first acquisition module is configured to acquire soil data corresponding to soil at different positions and depths of a target slope body; the soil data includes at least one of volume moisture content, soil temperature, electrical conductivity, pH value and soil stress;

[0020] A second acquisition module is configured to acquire environmental data corresponding to the target slope body; the environmental data includes rainfall data, wind speed data, wind direction data and evapotranspiration data;

[0021] a third obtaining module, configured to calculate soil mechanics data corresponding to different positions and depths of the target slope body based on the soil data, wherein the soil mechanics data comprises at least one of electrostatic repulsion, van der Waals force, hydration repulsion and liquid bridge force;

[0022] a determining module, configured to determine an erosion grade of the target slope body based on the soil mechanics data and the soil data;

[0023] an outputting module, configured to output early warning information according to the erosion grade.

[0024] The soil erosion early warning method and device provided by the embodiments of the present application obtain soil data corresponding to soil at different positions and depths of a target slope body. Key soil data such as volume water content and soil temperature are collected to provide a basic data source for subsequent analysis, and the core value lies in realizing accurate two-dimensional description of the “space-depth” of the soil state. Environmental data corresponding to the target slope body is obtained, which can quantify the influence of external driving factors on the slope body and make up for the limitations of relying only on soil data. Based on the soil data, soil mechanics data corresponding to different positions and depths of the target slope body is calculated. Converting soil data into mechanics data such as electrostatic repulsion and van der Waals force is the key to analyzing the causes of erosion from the “microscopic mechanism” level. The microscopic mechanics state of soil particles tending to disperse or aggregate (such as particle dispersion when Fnet>0) is determined, which breaks through the limitations of traditional macroscopic deformation monitoring and provides physical mechanism support for subsequent erosion grade determination, making the determination result more scientific and interpretable. Based on the soil mechanics data and the soil data, the erosion grade of the target slope body is determined. Through multi-dimensional data fusion of “microscopic mechanics + macroscopic soil state”, comprehensive evaluation of the erosion degree of the slope body is realized. According to the erosion grade, early warning information is output. This step is the landing of the core value of the “disaster prevention and reduction” of the early warning system, and its beneficial effects lie in realizing “advance warning” and “precise response”. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0026] Figure 1 is the first flowchart of the soil erosion early warning method according to the embodiments of the present application;

[0027] Figure 2 is the second flowchart of the soil erosion early warning method according to the embodiments of the present application. DETAILED DESCRIPTION

[0028] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0029] It can be understood that, before using the technical solutions disclosed in the embodiments of the present application, the type of personal information, the use range, the use scenario and the like involved in the present application should be informed to the user and the authorization of the user should be obtained through appropriate means according to relevant laws and regulations.

[0030] The terms "first", "second" are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more than two, unless otherwise specifically limited.

[0031] According to the embodiments of the present application, a soil erosion early warning method embodiment is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0032] In the present embodiment, a soil erosion early warning method is provided, which can be used in electronic devices such as mobile phones, tablet computers and the like, Figure 1 is a flowchart of the soil erosion early warning method according to the embodiments of the present application, as Figure 1 shown, the flow includes the following steps:

[0033] Step S101, acquiring soil data corresponding to soil at different positions and depths of a target slope body.

[0034] Among them, the soil data includes at least one of volumetric water content, soil temperature, electrical conductivity, pH value, and soil stress.

[0035] Specifically, for the target slope body, the electronic device can select a typical monitoring section, each slope position as an independent monitoring unit, to ensure the spatial gradient of the slope soil state. Vertical monitoring holes are drilled at each monitoring section, and depth points are set according to the principle of "shallow layer-middle layer-deep layer", usually selecting three depths of 20cm, 50cm and 100cm, forming a three-dimensional monitoring network of "slope position x depth".

[0036] The electronic device can realize synchronous measurement of the corresponding volumetric water content, soil temperature, and electrical conductivity of the soil at different positions and depths of the target slope body through the FDR soil moisture / temperature / conductivity three-parameter sensor. The resolution can reach 0.1% (water content), 0.1°C (temperature), and 1 μS / cm (conductivity), respectively. The electrical conductivity data can be used to calculate the ion strength parameter for inter-particle force calculation.

[0037] The electronic device can collect pH values through a solid-state pH sensor. The measurement range is usually 2-12, and the accuracy is ±0.1 pH. The pH value reflects the acidity and alkalinity of the soil pore solution and affects the electrochemical properties of the particle surface (such as the midpoint potential).

[0038] The electronic device can indirectly perceive the stress of the soil body through a fiber Bragg grating (FBG) micro-strain sensor. The resolution can reach 1 με, and the soil stress can be calculated through the constitutive relationship between micro-strain and stress (previously calibrated through laboratory tests) to capture the mechanical response caused by the change of inter-particle force.

[0039] Step S102, obtaining the corresponding environmental data of the target slope body.

[0040] The environmental data includes rainfall data, wind speed data, wind direction data, and evapotranspiration data.

[0041] The rainfall data includes rainfall amount, rainfall intensity, raindrop size distribution, and kinetic energy. The rainfall amount and intensity reflect the infiltration amount and impact force of the rainfall on the slope surface, and the raindrop kinetic energy directly affects the risk of surface soil particle stripping.

[0042] The electronic device can monitor the rainfall amount and intensity based on a tipping bucket rain gauge (resolution 0.1 mm, accuracy ±2%), and monitor the raindrop size distribution and kinetic energy based on a laser raindrop spectrometer (measurement range 0.1-8 mm raindrop diameter, accuracy ±5%). The two can work together to comprehensively quantify the rainfall data.

[0043] The electronic device can measure the wind speed data and wind direction data using an ultrasonic wind speed and direction sensor. The measurement range is 0-60 m / s (wind speed) and 0-360° (wind direction), and the accuracy is ±0.1 m / s and ±3°, respectively. The dynamic changes of wind speed and direction can be captured in real time.

[0044] The electronic device can measure the evapotranspiration data using a small evapotranspiration sensor (such as a Penman-Monteith principle sensor). The measurement range is 0-10 mm / d, and the accuracy is ±0.1 mm / d. The actual evapotranspiration of the soil can be directly output.

[0045] Step S103, calculating the corresponding soil mechanical data of the target slope body at different positions and depths based on the soil data.

[0046] wherein the soil mechanics data comprises at least one of electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force.

[0047] Specifically, the electronic device calculates the soil mechanics data corresponding to different positions and depths of the target slope body based on the volumetric water content, soil temperature, and electrical conductivity in the soil data.

[0048] Details of this step will be described below.

[0049] In step S104, the erosion grade corresponding to the target slope body is determined based on the soil mechanics data and the soil data.

[0050] Specifically, the electronic device can determine the erosion grade corresponding to the target slope body according to the corresponding relationship between the soil mechanics data and the soil data and the erosion grade.

[0051] Details of this step will be described below.

[0052] In step S105, the warning information is output according to the erosion grade.

[0053] Specifically, the electronic device can compare the erosion grade with a preset erosion grade threshold, and output the warning information according to the erosion grade.

[0054] The soil erosion warning method provided in the embodiments of the present application acquires soil data corresponding to soil at different positions and depths of a target slope body. Key soil data such as volumetric water content and soil temperature are collected to provide a basic data source for subsequent analysis, and the core value lies in realizing accurate two-dimensional description of the "space-depth" of the soil state. Environmental data corresponding to the target slope body is acquired, which can quantify the influence of external driving factors on the slope body and make up for the limitations of relying only on soil data. Based on the soil data, soil mechanics data corresponding to different positions and depths of the target slope body is calculated. Converting soil data into mechanics data such as electrostatic repulsion and van der Waals force is the key to analyzing the causes of erosion from the "microscopic mechanism" level. The microscopic mechanics state of soil particles tending to disperse or aggregate (such as particle dispersion when Fnet>0) is determined, which breaks through the limitations of traditional macroscopic deformation monitoring and provides physical mechanism support for subsequent erosion grade determination, making the determination result more scientific and interpretable. Based on the soil mechanics data and the soil data, the erosion grade corresponding to the target slope body is determined. Through multi-dimensional data fusion of "microscopic mechanics + macroscopic soil state", comprehensive evaluation of the erosion degree of the slope body is realized. According to the erosion grade, the warning information is output. This step is the landing of the core value of the "disaster prevention and reduction" of the warning system, and its beneficial effects lie in realizing "early warning" and "precise response".

[0055] In the embodiments, a soil erosion warning method is provided, which can be used in electronic devices such as mobile phones, tablet computers, etc.Figure 2 is a flowchart of a soil erosion early warning method according to an embodiment of the present application, as shown in the figure, the flow includes the following steps: Figure 2

[0056] In step S201, soil data corresponding to soil at different positions and depths of a target slope body is acquired.

[0057] The soil data includes at least one of volumetric water content, soil temperature, electrical conductivity, pH value, and soil stress.

[0058] In step S202, environmental data corresponding to the target slope body is acquired.

[0059] The environmental data includes rainfall data, wind speed data, wind direction data, and evapotranspiration data.

[0060] In step S203, soil mechanical data corresponding to the target slope body at different positions and depths is calculated based on the soil data.

[0061] The soil mechanical data includes at least one of electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force.

[0062] Specifically, the above step S203 can include the following steps:

[0063] In step S2031, the cation concentration in the bulk solution is determined based on the electrical conductivity for the soil at different positions and depths of the target slope body.

[0064] Specifically, the user can collect soil samples at different positions and depths of the target slope body, prepare soil pore solutions of different concentrations in the laboratory (simulate the EC variation range in the field), and measure the EC value and corresponding cation concentration (c0, measured by ion chromatography) of each solution. For the main salt type of saline soil in the target area (such as mainly NaCl), a relationship model between electrical conductivity and cation concentration is established: if the data is linear, an empirical formula is fitted: c0=k×EC+b (k is the slope, b is the intercept, calibrated by the least squares method, for example, k≈0.01 mol / (L•mS / cm) and b≈0.02 mol / L in the NaCl system); if the data is nonlinear, an EC-c0 calibration curve is plotted, and the field measured EC value is converted to c by interpolation.

[0065] The electronic device can receive a user input relationship model between electrical conductivity and cation concentration corresponding to the target slope body at different positions and depths, then input the electrical conductivity into the relationship model between electrical conductivity and cation concentration, and determine the cation concentration in the bulk solution.

[0066] In step S2032, the soil temperature is converted to obtain an absolute temperature.

[0067] ​Specifically, the electronic device can convert the soil temperature to an absolute temperature based on the following formula: T abs = T + 273.15; where T is the soil temperature, and T abs is the absolute temperature.

[0068] In step S2033, the electronic device calculates the electrostatic repulsion in the soil mechanical data based on the cation concentration, the absolute temperature, and the inter-particle distance corresponding to the soil.

[0069] Specifically, the electronic device can receive a preset formula parameter input by a user. The preset formula parameter can include the following: R (ideal gas constant) is a fixed value, which is 8.314 J·mol-1·K-1 (a general constant in thermodynamics); F (Faraday constant) is a fixed value, which is 96485 C·mol -1 -1 (a general constant in electrochemistry); Z (valence of cations) is determined according to the main cation type of the target area saline soil (for example, Z = 1 if NaCl is the main component, Z = 2 if CaCl2 is the main component, and the value is determined through indoor ion analysis calibration); D (inter-particle distance) is determined by a soil particle size distribution test (such as a laser particle size analyzer) to measure the average particle diameter of the soil, and 1 / 10 of the average particle diameter is taken as the initial D (for example, if the particle diameter is 50 μm, then Dd = 5 μm), or directly measured by an atomic force microscope (AFM); φ(D / 2) (midpoint potential) is calculated in combination with the node pH data: first, the surface potential φ0 is obtained through the relationship between the pH value and the surface potential of the soil (such as the Nernst equation), and then φ(D / 2) is calculated by taking x = D / 2 in the double-layer potential distribution formula (such as the Gouy-Chapman model) φ(x) = φ0 x e-kx (κ is the Debye length, which is related to c0).

[0070] Then, the electronic device can calculate the electrostatic repulsion in the soil mechanical data based on the cation concentration, the absolute temperature, and the inter-particle distance corresponding to the soil based on the following preset formula.

[0071]

[0072] wherein F is the electrostatic repulsion, F is the Faraday constant, T is the absolute temperature, R is the ideal gas constant, Z is the valence of the cation, c0 is the cation concentration in the bulk solution in a single electrolyte solution system, D is the inter-particle distance, and φ(D / 2) is the midpoint potential. T0 = 298.15 K (25°C, a reference value in a normal temperature environment), and c0 is the cation concentration.

[0073] ​The midpoint potential φ(d / 2) in the double electric layer can be calculated by the soil pH value, the electrochemical properties of the particle surface, and the related model of the double electric layer theory. For example, the midpoint potential φ(d / 2) when the distance between particles is d is determined by using the relationship between the surface potential and the pH value and the potential distribution formula of the double electric layer. The cosh is the hyperbolic cosine function.

[0074] In step S2034, the initial Hamaker constant is obtained.

[0075] Specifically, the electronic device can receive the initial Hamaker constant input by the user, or receive the initial Hamaker constant input by the user.

[0076] In step S2035, the initial Hamaker constant is corrected based on the electrical conductivity, the volumetric water content, and the soil temperature to obtain the target Hamaker constant.

[0077] Specifically, the electronic device can correct the initial Hamaker constant based on the electrical conductivity, the volumetric water content, and the soil temperature to obtain the target Hamaker constant by using the following formula: A'=A0×[1+0.08×(EC / EC0)-0.12×(θ / θ sat )+0.05×(T / T0)].

[0078] Wherein, A' is the target Hamaker constant, A0 is the initial Hamaker constant, EC0 is the reference salt concentration, which is 1000 μS / cm, EC is the electrical conductivity, θ sat is the saturated water content, which is determined by the "soil saturated θ value" monitored by the sensor (such as clayey saline soil θ θsat ≈40%, sandy saline soil θ sat ≈30%); θ is the volumetric water content, T0 is the reference temperature T0=298.15 K (25℃, the reference value of normal temperature environment), T is the soil temperature.

[0079] In step S2036, the van der Waals force in the soil mechanical data is calculated based on the target Hamaker constant.

[0080] Specifically, the electronic device substitutes the target Hamaker constant A' into the following formula according to the van der Waals force formula to calculate the van der Waals force Pvdw.

[0081]

[0082] Wherein, is the distance between particles, is the target Hamaker constant.

[0083] In step S2037, the hydration repulsion in the soil mechanical data is calculated according to the distance between particles.

[0084] Specifically, the electronic device calculates the hydration repulsion force based on a hydration repulsion formula (exponential decay model, suitable for soil particle system):

[0085]

[0086] wherein D is the inter-particle distance, k h is the hydration characteristic constant, related to the hydrophilicity of the soil particle surface, determined by indoor test. For example, based on soil samples with different inter-particle distances, the hydration repulsion force is determined, and k h is fitted (e.g., k h ≈104Pa for clay-saline soil); λ h is the hydration layer thickness, reflecting the spatial range of the hydration layer, usually determined by dynamic light scattering instrument, and the difference between the hydration diameter and the dry diameter is 1 / 2 of λ h (e.g., λ h ≈2nm). Example calculation: if k h =104Pa, λ h =2nm, and d=3nm, then: Ph=104×e -3 / 2 ≈2231Pa.

[0087] Step S2038, based on the conductivity and the volumetric water content, the salt crystal precipitation degree is calculated.

[0088] Specifically, the electronic device can calculate the salt crystal precipitation degree based on the conductivity and the volumetric water content based on the following formula.

[0089]

[0090] wherein θsat is the saturated water content, is the volumetric water content, is the relative water content, reflecting the degree of soil drying (during the drying period, θ decreases, the value decreases, and β increases); is the conductivity change rate (during the drying period, water decreases, EC increases, ΔEC / Δt is positive, and β increases; during the wetting period, the opposite is true). For example, if , , and ΔEC / Δt=0.5μS / (cm•h), then: β=[1-20 / 35]×0.5≈0.214. Wherein β∈[0,1] (β=0 indicates no salt crystal precipitation, and β=1 indicates a large amount of salt crystal precipitation).

[0091] Step S2039, the initial contact angle and the initial liquid bridge volume are obtained.

[0092] Specifically, the electronic device can receive the initial contact angle and the initial liquid bridge volume input by the user.

[0093] Wherein, the initial contact angle can be determined by a contact angle measuring instrument to measure the contact angle of soil particles and pure water (such as α = 60°); the initial liquid bridge volume V is calculated by soil porosity and water content (such as V = 10 -15 m 3 )。

[0094] Step S20310, the initial contact angle is corrected based on the salt crystallization degree to obtain a target contact angle.

[0095] Specifically, the electronic device can correct the initial contact angle based on the salt crystallization degree to obtain a target contact angle by using the following formula.

[0096] α' = α × (1 + 0.2β), where α' is the target contact angle, and α is the initial contact angle. Example: if α = 60°, β = 0.214, then α' = 60 × (1 + 0.2 × 0.214) ≈ 62.57°.

[0097] Step S20311, the initial liquid bridge volume is corrected based on the salt crystallization degree to obtain a target liquid bridge volume.

[0098] Specifically, the electronic device can correct the initial liquid bridge volume based on the salt crystallization degree to obtain a target liquid bridge volume by using the following formula:

[0099] Liquid bridge volume correction: V' = V × (1 - 0.3 × β), salt crystals occupy pore space, resulting in a decrease in liquid bridge volume.

[0100] Wherein, V' is the target liquid bridge volume, and V is the initial liquid bridge volume.

[0101] Step S20312, based on the target contact angle and the target liquid bridge volume, the liquid bridge force is calculated.

[0102] Specifically, the electronic device can calculate the liquid bridge force based on the target contact angle and the target liquid bridge volume by using the following formula:

[0103] ;

[0104]

[0105] Wherein, is the liquid surface tension (such as σ = 0.073 N / m), N / m; represents the target contact angle between solid-liquid; η represents the viscosity coefficient of the liquid (the viscosity of the salt solution increases with the increase of the conductivity, and the viscosity of the pure water at 25°C ), Pa·s; D represents the distance between particles, mm; dD / dt is the separation speed; F cap represents the capillary force, N; represents the viscous force, N; F lipN, r is the radius of the soil particle, V' is the target liquid bridge volume, and R is the ideal gas constant.

[0106] In step S204, the soil mechanics data and the soil data are used to determine the erosion level of the target slope body.

[0107] Specifically, step S204 can include the following steps:

[0108] In step S2041, the electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force are used to calculate the soil action force for the soil at different positions and depths of the target slope body.

[0109] Specifically, for the soil at different positions and depths of the target slope body, the electronic device can use the following formula to calculate the soil action force based on the electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force.

[0110] F net =P vdw -P E -P h +F lip ;

[0111] where P vdw is the van der Waals force, which is positive in the formula and reflects the aggregation trend between particles; P E is the electrostatic repulsion, P h is the hydration repulsion, which are all repulsive forces, and are negative in the formula, reflecting the dispersion trend between particles; F lip is the liquid bridge force, reflecting the cohesive effect of the liquid bridge on the particles.

[0112] When F net > 0, the soil action force is repulsive, and the soil particles tend to disperse, which is a precursor to particle erosion and requires close attention to the subsequent risk evolution of this area; when F net ≤ 0, the soil action force is attractive or in a balanced state, and the soil particles are relatively stable, with low erosion risk.

[0113] In step S2042, the target constitutive relationship is obtained.

[0114] The target constitutive relationship is used to characterize the relationship between the soil action force and the soil cohesion.

[0115] Specifically, step S2042 can include the following steps:

[0116] In step a1, the initial cohesion, critical repulsion, and time decay coefficient corresponding to the soil are obtained.

[0117] Specifically, the electronic device can receive the initial cohesion, critical repulsion, and time decay coefficient corresponding to the soil input by the user.

[0118] The initial cohesion (c0'): The initial shear strength of the target soil is determined by a direct shear test in the laboratory, which reflects the shear resistance of the soil without long-term effects of water and salt, etc.

[0119] The critical repulsive force (F net0 ): The F net threshold value (F net0 > 0) is calibrated by a triaxial test in the laboratory, when the soil force exceeds the threshold value, the soil particles begin to show a dispersion trend.

[0120] The time decay coefficient (λt): It is determined by a long-term water and salt cycle test in the laboratory, simulating the long-term effect of field dry-wet cycles on cohesion. Select a typical soil sample and perform multiple dry-wet cycles (e.g. 10 cycles, each cycle including 7 days of saturated immersion + 3 days of drying) in the laboratory.

[0121] After each cycle, the cohesion ct' of the soil sample is measured, and the cohesion decay rate is calculated; with the cycle number (or time) as the horizontal axis and the cohesion decay rate as the vertical axis, an exponential decay curve is fitted, and the time decay coefficient (unit: 1 / day) is back calculated.

[0122] Step a2, determine the force influence coefficient based on the soil type corresponding to the soil.

[0123] Specifically, the electronic device can determine the force influence coefficient based on the soil type corresponding to the soil.

[0124] For example, sandy saline soil: the inter-particle pores are large, the effect of F net on cohesion is more significant, and the force influence coefficient k F = 1.2~1.5; loamy saline soil: the particle size distribution is uniform, the effect of F net is moderate, and the force influence coefficient k F = 1.0 (baseline value); clayey saline soil: the clay content is high, the inter-particle cementation is strong, and the effect of F net is weak, and the force influence coefficient k F = 0.7~0.9.

[0125] Step a3, based on the initial cohesion, the critical repulsive force, the time decay coefficient, and the force influence coefficient, construct the target constitutive relationship.

[0126] Specifically, the electronic device constructs the target constitutive relationship that can quantitatively calculate the cohesion (c'), as follows:

[0127]

[0128] wherein, is the soil cohesion force (kPa) at time t, i.e., the real-time cohesion force of the soil at different positions and depths of the target slope body; is the initial cohesion force (kPa), is the time decay coefficient (1 / day); t is the water salt action time (day); is the force influence coefficient; is the real-time soil force (N);

[0129] is the critical repulsive force (N). Reflects the decay effect of time on cohesion, the longer the action time, the lower the basic value of cohesion; : only when F net > F net0 , the additional decay of F net on cohesion is introduced, avoiding underestimating the cohesion in the stable state.

[0130] Step S2043, based on the target constitutive relation, determining the soil cohesion force corresponding to the soil at different positions and depths of the target slope body.

[0131] Specifically, the electronic device substitutes the matched parameters into the constructed target constitutive relation, and calculates the soil cohesion force of the soil at different positions and depths of the target slope body .

[0132] Step S2044, based on the soil cohesion force corresponding to the soil at different positions and depths of the target slope body, the soil force and the soil data, determining the erosion grade corresponding to the target slope body.

[0133] Specifically, the above step S2044 can include the following steps:

[0134] Step b1, determining the soil at different positions and depths of the target slope body as nodes in the dynamic iteration graph structure.

[0135] Specifically, the electronic device can correspond each "position-depth" combination to an independent node (such as a "middle of slope-50cm" node), and the node ID needs to be associated with the node number (such as node 2-50, representing the 2nd monitoring hole at a depth of 50cm), to ensure that the subsequent feature vector can be accurately matched with the measured data.

[0136] Step b2, for each node, based on the soil cohesion force corresponding to the soil, the soil force and the soil data, constructing a target node feature vector corresponding to each node.

[0137] The target node feature vector includes: volumetric water content, soil temperature, electrical conductivity, pH value, soil stress, soil force, soil cohesion force, salt migration trend, strain activity level and node spatial coordinates.

[0138] Specifically, the electronic device can use the linear regression method to calculate the time variation rate of EC with the conductivity corresponding to each node, so as to represent the salt migration trend, and the formula is as follows:

[0139]

[0140] Wherein, S is the salt migration trend (unit: mS / (cm•h), the time unit is adjusted according to the preset time length, such as hour h, day d); k is the linear regression slope of EC time series data, which directly reflects the salt change rate; n is the sample number of time series data; t is the time sequence (such as t=1 represents the first time point, t=2 represents the second time point, and so on); ECt is the measured value of conductivity at the t time point (unit: mS / cm).

[0141] When S>0, EC increases with time, which represents that the soil is in the salt return trend, the salt is migrated and gathered in the area where the node is located, which is easy to intensify the interparticle electrostatic repulsion and induce particle dispersion; when S=0: EC is basically stable, the salt migration is in the balance state, and the soil particles are relatively stable; when S<0: EC decreases with time, which represents that the soil is in the desalination trend, the salt is lost from the area where the node is located, the interparticle electrostatic repulsion is weakened, and the soil erosion resistance is relatively improved.

[0142] The electronic device can use the standard deviation method to calculate the fluctuation amplitude of micro-strain, so as to represent the strain activity degree, and the formula is as follows:

[0143]

[0144] Wherein, A is the strain activity degree (unit: με, consistent with the unit of micro-strain); std(·): standard deviation function, reflecting the data dispersion, the greater the value represents the more intense strain fluctuation; is the measured value of micro-strain at the t time point (unit: με); is the average value of micro-strain in the preset time length (unit: με); n is the sample number of time series data.

[0145] Then, the electronic device splices the volume water content, soil temperature, conductivity, pH value, soil stress, soil force, soil cohesion, salt migration trend, strain activity degree and node spatial coordinates corresponding to each node to generate the target node feature vector corresponding to each node.

[0146] Step b3, calculating the basic weight and dynamic correction factor between any two nodes.

[0147] Specifically, the above step b3 can include the following steps:

[0148] ​​Step b31, for any two nodes, obtain the spatial straight-line distance between the two nodes.

[0149] Specifically, for any two nodes, the electronic device can calculate the spatial straight-line distance between the two nodes according to the positions and depths between the two nodes.

[0150] Step b32, calculate the physical distance weight between the two nodes according to the spatial straight-line distance.

[0151] Specifically, the electronic device can calculate the physical distance weight between the two nodes according to the relationship between the spatial straight-line distance and the physical distance weight.

[0152] Wherein, the relationship between the spatial straight-line distance and the physical distance weight is: W1=1 / (1+D). Wherein, D is the spatial straight-line distance, and W1 is the physical distance weight.

[0153] Step b33, obtain the electrical conductivity and volume water content of the two nodes within a first preset time length before the current time.

[0154] Specifically, the electronic device can obtain the electrical conductivity time series data and volume water content time series data (time step 5 minutes, consistent with the node sampling frequency) of the two nodes within the first preset time length (such as 72 hours, covering 1 dry-wet cycle period).

[0155] Step b34, based on the electrical conductivity of the two nodes within the first preset time length before the current time, calculate the electrical conductivity correlation coefficient corresponding to the two nodes.

[0156] Specifically, the electronic device can calculate the electrical conductivity correlation coefficient corresponding to the two nodes based on the electrical conductivity of the two nodes within the first preset time length before the current time.

[0157] For example, the electronic device can use Pearson correlation coefficient to calculate the electrical conductivity correlation coefficient corresponding to the two nodes:

[0158] ;

[0159] Wherein, is the EC value of node i at time t, is the EC mean value of node i.

[0160] Step b35, based on the volume water content of the two nodes within the first preset time length before the current time, calculate the volume water content correlation coefficient corresponding to the two nodes.

[0161] Specifically, the electronic device can calculate the volume water content correlation coefficient corresponding to the two nodes based on the volume water content of the two nodes within the first preset time length before the current time. As shown in step b34 above.

[0162] Step b36, based on the conductivity correlation coefficient and the volumetric water content correlation coefficient, the correlation weight between the two nodes is calculated.

[0163] Specifically, the electronic device can calculate the average of the conductivity correlation coefficient and the volumetric water content correlation coefficient to obtain the correlation weight between the two nodes.

[0164] Step b37, obtaining the target slope soil texture distribution map corresponding to the target slope.

[0165] Specifically, the electronic device can receive the target slope soil texture distribution map corresponding to the target slope input by the user.

[0166] Step b38, based on the target slope soil texture distribution map, the soil texture corresponding to the two nodes respectively is determined.

[0167] Specifically, the electronic device can determine the soil texture of nodes i and j according to the target slope soil texture distribution map (for example, node i is loamy saline soil, and node j is clay saline soil).

[0168] Step b39, according to the soil texture corresponding to the two nodes respectively, the soil property weight between the two nodes is determined.

[0169] Specifically, the electronic device can determine the soil property weight between the two nodes according to the soil texture corresponding to the two nodes respectively.

[0170] For example, the value is assigned according to the texture similarity (1 for similar, 0.5 for partially similar, and 0 for dissimilar):

[0171] Sand saline soil-sand saline soil: W3=1; sandy saline soil-loamy saline soil (both containing sand particles): W3=0.5; sandy saline soil-clay saline soil (no common particle size): W3=0.

[0172] Step b310, based on the physical distance weight, the correlation weight and the soil property weight, the basic weight between the two nodes is determined.

[0173] Specifically, the electronic device can perform weighted calculation on the physical distance weight, the correlation weight and the soil property weight to obtain the basic weight between the two nodes.

[0174] For example, the physical distance weight accounts for 40%, the correlation weight accounts for 30%, and the soil property weight accounts for 30%. W base =0.4·W1+0.3·W2+0.3·W3. The basic weight ∈ [0,1], the larger the value represents the stronger the fixed association between the two nodes (for example, adjacent and same texture, water and salt cooperation nodes, W base ≈1).

[0175] Step b311, for any two nodes, based on the conductivity and soil stress corresponding to the two nodes respectively, the salt migration trend and the strain activity degree corresponding to the two nodes respectively are calculated.

[0176] Specifically, the specific process of calculating the salt migration trend and the strain activity degree can be referred to step b2, which will not be repeated here.

[0177] Step b312, based on the salt migration trend and the strain activity degree corresponding to the two nodes respectively, the soil state synergy factor corresponding to the two nodes is calculated.

[0178] Specifically, the synergy factor reflects the consistency of the real-time state of the two nodes.

[0179] ;

[0180] Wherein, is the salt migration trend of node i, is the salt migration trend of node j, is the strain activity degree of node i, is the strain activity degree of node j. If and are of the same sign (same salt or same desalination) and the difference is small, the first term tends to 1; if and are close (same stability or same activity), the second term tends to 1. ∈[0,1], the greater the value represents the more synergistic the real-time state.

[0181] Step b313, based on the environmental data corresponding to the target slope body, the environmental response synergy factor is determined.

[0182] Specifically, the electronic device can receive the response coefficient of each node to rainfall, the response coefficient of each node to evapotranspiration and the response coefficient of each node to wind speed input by the user. Among them, the response coefficient of each node to rainfall can be obtained by fitting based on the rainfall corresponding to the soil sample corresponding to each node and the volume water content change of the soil sample before and after the rainfall. For example, the electronic device can take rainfall P as the independent variable, and the increment of volume water content after rainfall (Δθ=θ 雨后 -θ 雨前 ) as the dependent variable, fit the linear regression model Δθ=k P •P+ε, to obtain the response coefficient k P of each node to rainfall.

[0183] Similarly, since evaporation leads to a decrease in moisture content, electronic devices can use evaporation ET as the independent variable and the decrease in volumetric moisture content (-Δθ, the negative sign turns to positive) as the dependent variable to fit a linear model -Δθ=k. ET •ET+ε, to obtain the response coefficient k of each node to evapotranspiration. ET The higher the wind speed, the stronger the evaporation of surface soil, the easier it is for salt to accumulate, and the higher the electrical conductivity. Therefore, the standardized wind speed (v) is used. std ) is the independent variable, and the standardized conductivity (EC) is the independent variable. std Using as the dependent variable, a linear regression is fitted: EC std =k v •v std +εk v Given the regression slope, we obtain the response coefficient k of each node to wind speed. v .

[0184] Then, based on the response coefficients of each node to rainfall, evapotranspiration, and wind speed, the electronic equipment determines the environmental response synergy factor using the following formula:

[0185] ;

[0186] in, Let be the response coefficient of node i to rainfall. Let be the response coefficient of node i to evapotranspiration. Let be the response coefficient of node i to wind speed. It is a synergistic factor for environmental response.

[0187] Step b314: Determine the dynamic correction factor based on the soil state synergy factor and the environmental response synergy factor.

[0188] Specifically, electronic devices can multiply soil state synergy factors and environmental response synergy factors to obtain dynamic correction factors.

[0189] Step b4: Based on the basic weights and dynamic correction factors, calculate the target edge weights of the target structure edges between any two nodes, and construct a dynamic iterative graph structure.

[0190] Specifically, electronic devices can calculate the target edge weight of the target structure edge between any two nodes by multiplying the base weight by a dynamic correction factor, and construct a dynamic iterative graph structure.

[0191] Step b5: Input the dynamic iterative graph structure and environmental data into the preset particle erosion probability index model, and output the particle erosion probability index corresponding to each node in the dynamic iterative graph structure.

[0192] Specifically, step b5 above includes the following steps:

[0193] Step b51: Based on the node spatial coordinates and environmental data corresponding to each node, allocate local environmental data to each node.

[0194] Specifically, the electronic device can determine the slope, aspect, and elevation data corresponding to each node based on the node's spatial coordinates. Then, based on the slope, aspect, and elevation data corresponding to each node, the rainfall, wind speed, and evapotranspiration in the environmental data are corrected to obtain the initial local environmental data after terrain adaptation. Then, based on the depth of each node, the rainfall and evapotranspiration are attenuated and corrected to obtain the final local environmental data for each node.

[0195] Step b52: Stack the target node feature vectors corresponding to each node to generate a node feature matrix.

[0196] Specifically, the electronic device can stack the target node feature vectors of all nodes row by row to form an N×D node feature matrix (N is the total number of nodes, D is the feature dimension, such as 11 dimensions: EC, θ, pH, T, με, F). net c′(t), salt migration trend, strain activity, slope position, and depth.

[0197] Step b53: Construct an adjacency matrix based on the target edge weights of the target structural edges between any two nodes.

[0198] Specifically, the electronic device can construct an N×N adjacency matrix based on the target edge weights of the target structural edges between any two nodes, where matrix element A i,j Let A be the target edge weight Wt between node i and node j (if the two nodes are not directly related or Wt < 0.1, then A i,j =0), and at the same time add a self-loop weight (A) to the diagonal of the adjacency matrix. i,i =1.0), ensuring that the node's own characteristics are not ignored.

[0199] Step b54: Extract features from the node feature matrix and adjacency matrix to obtain spatially enhanced node features.

[0200] Specifically, electronic devices can aggregate features from surrounding nodes through a GCN layer, integrating spatial correlation information into the node feature matrix to obtain spatially enhanced node features. GCN calculation formula:

[0201] ;

[0202] Where H(l) is the node feature matrix of the l-th layer (initially H(0)=X); W(l) is the trainable weight matrix of the l-th layer; b(l) is the bias term; The normalized adjacency matrix is ​​used, and σ is the activation function (ReLU is used to enhance the model's nonlinear fitting ability).

[0203] If node i is a core high-risk node (F net >F net0 F net0 The critical repulsive force calibrated indoors, such as 1.5 × 10⁻⁶. - 6 If node i has N and the edge weight Wt with node j is greater than 0.8 (high spatial correlation), then the high-risk characteristics of node i (such as Fnet exceeding the threshold, με activity) will be transmitted to node j through GCN aggregation.

[0204] Example: 100cm node at the toe of the slope (core high risk, F) net =2.1×10 -6 N) and Wt=0.85 at the 100cm node in the middle of the slope, GCN will change the "height F" of the node at the toe of the slope. net The features are integrated into the characteristics of nodes in the slope, reflecting the actual physical process of "salt infiltration leading to increased risk in the slope".

[0205] Step b55: For each node, calculate the temporal attention weight based on the relationship between the local environment data corresponding to the node and the feature vector of the target node corresponding to the node.

[0206] Specifically, for each node, the electronic device can extract local rainfall kinetic energy E from the local environmental data corresponding to the node. k,i (e.g., slope top node E) k,i =45J / m 2 Slope toe node E k,i =32J / m 2 ).

[0207] The electronic device can also calculate the change range of volumetric water content Δθ (the absolute value of the difference between two adjacent θ values) and the change range of soil stress Δμε (the absolute value of the difference between two adjacent με values) based on the volumetric water content and soil stress in the target node feature vector.

[0208] Then, the correlation strength between rainfall kinetic energy and soil response (including the change in water content Δθ and the change in soil stress Δμε) is calculated based on the Pearson coefficient:

[0209] ;

[0210] in, For correlation strength, Let be the local rainfall kinetic energy corresponding to node i at time t. Let i be the average local rainfall kinetic energy corresponding to node i. Let be the change in accumulated moisture content at time t. Let t be the magnitude of the change in soil stress at time t.

[0211] Electronic devices can determine critical moments and stationary moments based on the correlation strength. Then, the temporal attention weights corresponding to critical moments and stationary moments are calculated based on the correlation strength.

[0212] For example, the weighting of key time periods: , This represents the largest rainfall momentum in history, amplifying the impact of data during periods of heavy rainfall.

[0213] Weighting during stable periods: We downplay environmental data and focus on long-term changes within the soil (such as salt enrichment).

[0214] Step b56: Calculate the spatial attention weight for each node based on the local environmental data, soil forces, and strain activity corresponding to the node.

[0215] Specifically, the electronic device can calculate the mechanical risk, strain risk, and environmental risk corresponding to each node. Among these, the mechanical risk (S1) is: (F) net0 The critical repulsive force is 1.5 × 10⁻⁶. -6 N), S1>1.0 indicates high mechanical risk; strain risk (S2): (με0=10με, S2>1.0 indicates high strain risk; environmental risk (S3):) (I represents the rainfall intensity associated with the node, and Imax represents the historical maximum rainfall intensity, such as 100 mm / h). S3>0.8 indicates a high environmental risk.

[0216] The electronic device can sum the mechanical risks, strain risks, and environmental risks corresponding to each node to obtain the total risk. If the total risk corresponding to a node is greater than or equal to a first preset risk threshold, then the node is identified as a core high-risk node. If the total risk corresponding to a node is less than the first preset risk threshold but greater than or equal to a second preset risk threshold, then the node is identified as a potential high-risk node; if the total risk corresponding to a node is less than the second preset risk threshold, then the node is identified as a potential ordinary node.

[0217] The electronic device determines the spatial attention weight of each node based on its risk level. For example, a core high-risk node has a spatial attention weight of 1.2 and a spatial attention weight of 1.1 for nodes within a 5m radius (based on spatial coordinates determined by a dynamic iterative graph structure); a potential high-risk node has a spatial attention weight of 1.1 and a spatial attention weight of 1.0 for nodes within a 3m radius; and a normal node has a spatial attention weight of 1.0.

[0218] Step b57: For each node, the spatially enhanced node features and the various sub-features included in the local environmental data corresponding to the node are divided into soil mode, mechanical mode, strain mode, and environmental mode. For example, the soil mode (M1) includes: EC (electrical conductivity), θ (volume water content); the mechanical mode (M2) includes: F... net (Soil forces), c′(t) (soil cohesion); Strain mode (M3) includes: με (soil stress); Environmental mode (M4) includes: Ek,i (local rainfall kinetic energy), I (rainfall intensity).

[0219] Step b58: Calculate the modal weights corresponding to each sub-feature.

[0220] Specifically, electronic devices can determine the modal weights corresponding to each sub-feature based on the correspondence between modalities and modal weights.

[0221] Step b59: For each node, based on the modal weights corresponding to each sub-feature, the spatial attention weights corresponding to the node, and the temporal attention weights, the sub-features in the node are weighted and fused to generate the multimodal fusion feature vector corresponding to the node.

[0222] Specifically, the electronic device can use the modal weights corresponding to each sub-feature, the spatial attention weights and temporal attention weights corresponding to the nodes, and the total attention weights corresponding to each sub-feature in the node. The total attention weights are summed as follows: Wtotal = Wmodal × Wt × Ws (Wmodal is the modal weight, Wt is the temporal attention weight, and Ws is the spatial attention weight). Then, the electronic device multiplies the total attention weights with the corresponding modal sub-features to obtain the multimodal fusion feature vector corresponding to the node.

[0223] Step b510: For each node, stack the multimodal fusion feature vectors corresponding to the node according to the time series to form a time series feature.

[0224] Specifically, electronic devices can stack the multimodal fused feature vectors of each node according to a time series (e.g., the past 24 hours, 10 minutes / time) to form a T×D time series feature (T is the time step, such as 144; D is the fused feature dimension, such as 11). For example, the fused features of a node over 24 hours at 144 time steps are stacked into a 144×11 matrix to capture the dynamic changes of features "before rainfall - during rainfall - after rainfall".

[0225] Step b511 involves extracting features from the time series data using a multi-head attention layer and a feedforward neural network, and outputting the final fused features.

[0226] Specifically, the multi-head attention layer uses three trainable linear projection matrices (W... Q ∈R D×dk W K ∈R D×dk W V ∈R D×dv The input time series features X are mapped to queries (Query, Q=XW). Q Key (K=XW) K Value (V=XW) V ), where dk is the dimension of the query / key, dv is the dimension of the value, and h×dk=D (h is the number of attention heads, usually taken as 8).

[0227] For each attention head, the correlation weights between time steps are calculated using "scaled dot product attention": In the formula, This is a scaling factor to prevent the softmax function's gradient from vanishing due to excessively large calculation results; the softmax function normalizes the weights to [0,1], where a larger weight indicates a stronger influence of the corresponding time step's feature on the current time step. Example: When identifying the correlation between "a sharp increase in Fnet 2 hours after rainfall" and "an increase in Pd 6 hours after rainfall," t=2 hours (F... net The key K corresponding to the sudden increase i With t=6 hours (P) d The query Q corresponding to the increase) i The high similarity results in an attention weight close to 0.8, which is much higher than other time steps (such as 0.1~0.3), thus strengthening this lagging association.

[0228] Then, the output features of h attention heads (each with a dimension of T×dv) are concatenated column-wise, and then projected through a linear projection matrix W. O ∈R h×dv×D Mapping back to the original dimension D, we obtain the output H of the multi-head attention layer. attn ∈R T×D .

[0229] Feed-Forward Neural Network (FFN) is a further optimization of the output features of multi-head attention layers. Through "nonlinear transformation + feature scaling", it strengthens the "key time node features" (such as the peak rainfall period and the Fnet exceeding the threshold period) that are crucial for early warning of particle erosion, while suppressing the interference of irrelevant features, and adapting to the monitoring and early warning logic of "focusing on the key factors driving particle erosion".

[0230] The feedforward neural network employs a simple and efficient structure of "two-layer linear transformation + ReLU activation," with the input being the output H of the multi-head attention layer.attn ∈R T×D The output is the characteristic matrix H after nonlinear transformation. ffn ∈R T×D The specific structure is as follows: H ffn =max(0,H attn W1+b1)W2+b2.

[0231] First-level linear transformation: through the weight matrix W1∈R D×4D Expanding the feature dimension from D to 4D increases the model's non-linear fitting capability; b1∈R 4D This is a bias term used to adjust the feature distribution.

[0232] ReLU activation function: max(0,x), sets negative eigenvalues ​​to 0, filtering out invalid negative interference (such as physically meaningless negative strain changes, negative F-values, etc.). net (deviation), while retaining key positive features (such as positive F) net Incremental, positive P d The changes conform to the physical law that "soil parameters are non-negative".

[0233] Second-level linear transformation: through the weight matrix W2∈R 4D×D The feature dimension is compressed from 4D back to D to ensure that the output dimension is consistent with the input dimension, which facilitates subsequent feature processing.

[0234] After processing by the multi-head attention layer and the feedforward neural network, the T×D time series features need to be compressed into the final 1×D fusion features to achieve the integration of "space-time-multimodal" full-dimensional information, providing a unified feature input for the subsequent fully connected layer to output the particle erosion probability index (Pd).

[0235] Feature compression is performed using Global Average Pooling (GAP). The specific process is as follows: For each feature dimension (D in total) of the time series feature matrix Hffn∈RT×D, the average of the feature values ​​at all time steps (T) is calculated to obtain the final fused feature H×D. final ∈R1×D, the formula is: .

[0236] To avoid information loss caused by selecting features at a single time step, features from all time periods, such as "before rainfall - during rainfall - after rainfall", should be integrated simultaneously. High-dimensional features of T×D should be compressed into low-dimensional features to reduce the number of parameters in subsequent fully connected layers and avoid overfitting.

[0237] Step b512: Input the final fused features into the fully connected layer and output the particle erosion probability index corresponding to the node.

[0238] Specifically, the electronic device can input the final fused features into two fully connected layers. The first layer compresses the feature dimension to 32 dimensions, and the second layer compresses it to 1 dimension. Then, the fused features are processed by the Sigmoid function (…). Mapping the 1-dimensional feature to the [0,1] interval yields the particle erosion probability index of the node.

[0239] Each node corresponds to a particle erosion probability index value. For example, the Pd value for the 100cm node at the toe of the slope (core high risk) is 0.92, and the Pd value for the 20cm node at the top of the slope (ordinary node) is 0.25. The closer Pd is to 1, the higher the risk of soil particles at that node being eroded by water, salt, mechanical, and environmental factors; the closer it is to 0, the more stable the particles are.

[0240] Step b6: Based on the particle erosion probability index corresponding to each node and the feature vector of the target node, determine the erosion level corresponding to the target slope.

[0241] Specifically, step b6 above may include the following steps:

[0242] Step b61: For each node, extract soil forces, soil cohesion, salt migration trends, and strain activity from the target node feature vector corresponding to the node.

[0243] Specifically, for each node, the electronic device extracts soil forces, soil cohesion, salt migration trends, and strain activity from the target node feature vector corresponding to the node.

[0244] Step b62: Based on the particle erosion probability index, soil interaction force, soil cohesion, salt migration trend and strain activity of each node, determine the risk level of each node.

[0245] Specifically, electronic devices can determine the risk level of each node based on the relationship between the particle erosion probability index, soil action force, soil cohesion, salt migration trend, and strain activity level.

[0246] For example, if the particle erosion probability index, soil interaction force, soil cohesion, salt migration trend, and strain activity level corresponding to a node meet the following first condition, it is determined to be at the first risk level. The first condition may include P... d (Particle erosion probability index) ≤ 0.6, F net ≤F net0 At least one of the following: c′(t)≥0.7c0′, S≤0, or A≤A0.

[0247] If the particle erosion probability index, soil interaction force, soil cohesion, salt migration trend, and strain activity level corresponding to a node meet the following second condition, then it is determined to be at the second risk level. The second condition may include 0.6. <P d ≤0.9; F net0 <F net ≤1.5F net0 At least one of the following: 0.5c0′≤c′(t)<0.7c0′; S>0 and A>A0.

[0248] If the particle erosion probability index, soil interaction force, soil cohesion, salt migration trend, and strain activity level corresponding to a node meet the following third condition, then it is determined to be at the third risk level. The third condition may include P... d >0.9; F net >1.5F net0 At least one of the following: c′(t)<0.5c0′; S>0 and A≥2A0 (sudden jump).

[0249] Step b63: Determine the erosion level of the target slope based on the risk level corresponding to each node.

[0250] Specifically, if each node corresponds to the first risk level, a first-level warning message is issued; if each node corresponds to the second risk level, a second-level warning message is issued; and if each node corresponds to the third risk level, a third-level warning message is issued. The third-level warning message is more urgent than the second-level warning message, and the second-level warning message is more urgent than the first-level warning message.

[0251] Step S205: Output early warning information based on the erosion level.

[0252] Please refer to the above description of step S105 for details on this step, which will not be repeated here.

[0253] The soil erosion early warning method provided in this application determines the cation concentration in the bulk solution based on conductivity for soil at different locations and depths on a target slope; it converts the soil temperature to obtain absolute temperature. Based on the cation concentration, absolute temperature, and corresponding interparticle distances, it calculates the electrostatic repulsion force in the soil mechanics data. This method can accurately quantify the influence of the electrochemical properties of soil particle surfaces on microscopic forces, avoiding mechanical calculation deviations caused by neglecting key parameters such as ion concentration and temperature, and providing a micromechanical basis for subsequent erosion resistance analysis. An initial Hamac constant is obtained; based on conductivity, volumetric water content, and soil temperature, the initial Hamac constant is corrected to obtain a target Hamac constant; based on the target Hamac constant, van der Waals forces in the soil mechanics data are calculated. The van der Waals forces calculated in this way more closely resemble the real soil micromechanical environment, improving the accuracy of van der Waals force calculations and supporting subsequent net force analysis. Based on the interparticle distance, the hydration repulsion force in the soil mechanics data is calculated. The salt crystal precipitation degree is calculated based on electrical conductivity and volumetric water content. The initial contact angle and initial liquid bridge volume are obtained. The initial contact angle is corrected based on the salt crystal precipitation degree to obtain the target contact angle. The initial liquid bridge volume is also corrected based on the salt crystal precipitation degree to obtain the target liquid bridge volume. The liquid bridge force is calculated based on the target contact angle and target liquid bridge volume. This reflects the influence of salt on the mechanical properties of liquid bridges, making the liquid bridge force calculation more consistent with the "water-salt-mechanical coupling" characteristics of coastal saline soils, and improving the microscopic force calculation system. For soil at different locations and depths on the target slope, soil forces are calculated based on electrostatic repulsion, van der Waals forces, hydration repulsion, and liquid bridge forces. This can directly determine whether particles tend to disperse or aggregate (Fnet>0 indicates a dispersion trend), providing core microscopic indicators for subsequent risk assessment.

[0254] Next, the target constitutive relation is obtained. Based on the target constitutive relation, the soil cohesion corresponding to soil at different locations and depths on the target slope is determined. The microscopic mechanical indicators are transformed into macroscopic shear strength indicators (cohesion), realizing the correlation between "microscopic" and "macroscopic" mechanical indicators, providing a macroscopic basis for risk level determination based on erosion resistance. Soil at different locations and depths on the target slope is identified as nodes in the dynamic iterative graph structure. For each node, based on the corresponding soil cohesion, soil forces, and soil data, a target node feature vector is constructed. The basic weights and dynamic correction factors between any two nodes are calculated; based on the basic weights and dynamic correction factors, the target edge weights of the target structure edges between any two nodes are calculated, constructing the dynamic iterative graph structure. By setting soil at different locations / depths as nodes, and the feature vectors covering multi-dimensional data such as water and salt, mechanics, and space, and combining the basic weights and dynamic correction factors to calculate edge weights, the spatial correlation and dynamic influence between nodes (such as the coordinated changes in node states caused by water and salt migration) can be characterized, breaking through the limitations of traditional isolated node analysis and making the graph structure more closely reflect the spatial correlation of slope soil states. Based on the spatial coordinates of each node and environmental data, local environmental data is assigned to each node. The target node feature vectors of each node are stacked to generate a node feature matrix. An adjacency matrix is ​​constructed based on the target edge weights of the target structural edges between any two nodes. Feature extraction is performed on the node feature matrix and the adjacency matrix to obtain spatially enhanced node features. This strengthens the understanding of the impact of spatial association information on node risk, avoiding feature bias caused by neglecting spatial synergy effects.

[0255] For each node, a temporal attention weight is calculated based on the relationship between the local environmental data corresponding to the node and the feature vector of the target node. A spatial attention weight is calculated based on the local environmental data, soil forces, and strain activity of each node. For each node, the spatially enhanced node features and the sub-features included in the local environmental data are divided into soil, mechanical, strain, and environmental modes. Modal weights are calculated for each sub-feature. For each node, based on the modal weights, spatial attention weights, and temporal attention weights, the sub-features are weighted and fused to generate a multimodal fused feature vector for the node. Temporal weights focus on critical periods (such as periods of heavy rainfall), spatial weights highlight high-risk nodes, and modal weights quantify the contribution of different data types (soil, mechanical, etc.). The weighted fusion of these three factors achieves "strengthening of key information and weakening of secondary information," enabling multimodal features to more accurately reflect the node's risk status and improving the efficiency and accuracy of subsequent feature processing.

[0256] Finally, for each node, the multimodal fusion feature vectors corresponding to the node are stacked in time series to form time series features. Features are extracted from the time series features using a multi-head attention layer and a feedforward neural network to output the final fused features. The final fused features are then input into a fully connected layer, which outputs the particle erosion probability index corresponding to the node. Stacking features in time series and processing them through a Transformer can capture long-term dependencies in the time dimension (such as the correlation between a sudden increase in Fnet after rainfall and subsequent increases in Pd). The Pd index (0~1) output by the fully connected layer can quantify the erosion probability, providing an intuitive and comparable quantitative indicator for risk level assessment.

[0257] For each node, soil forces, soil cohesion, salt migration trends, and strain activity are extracted from the feature vector of the target node. Based on the particle erosion probability index, soil forces, soil cohesion, salt migration trends, and strain activity of each node, the risk level of each node is determined. Based on the risk level of each node, the erosion level of the target slope is determined. By extracting key parameters such as soil forces and cohesion from the feature vector and combining them with Pd to determine the node risk level, and then statistically determining the slope erosion level based on the node risk level, a layer-by-layer correlation between "node micro-risk" and "slope macro-erosion" is achieved. This ensures that the erosion level determination is supported by micro-data and reflects the overall state of the slope, providing a scientific and accurate macro-based basis for early warning issuance.

[0258] This application embodiment also provides a soil erosion early warning device, the device comprising:

[0259] The first acquisition module is used to acquire soil data corresponding to soil at different locations and depths of the target slope; the soil data includes at least one of volumetric water content, soil temperature, electrical conductivity, pH value, and soil stress.

[0260] The second acquisition module is used to acquire environmental data corresponding to the target slope; the environmental data includes rainfall data, wind speed data, wind direction data, and evapotranspiration data;

[0261] The third acquisition module is used to calculate soil mechanical data at different locations and depths of the target slope based on soil data; the soil mechanical data includes at least one of electrostatic repulsion, van der Waals force, hydration repulsion and liquid bridge force.

[0262] The determination module is used to determine the erosion level of the target slope based on soil mechanics data and soil data;

[0263] The output module is used to output early warning information based on the erosion level.

[0264] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method of soil erosion early warning, characterized by, The method comprises: obtaining soil data corresponding to soil at different positions and depths of a target slope body; the soil data comprises at least one of volumetric water content, soil temperature, electrical conductivity, pH value, and soil stress; obtaining environmental data corresponding to the target slope body; the environmental data comprises rainfall data, wind speed data, wind direction data, and evapotranspiration data; based on the soil data, calculating soil mechanics data corresponding to the target slope body at different positions and depths; the soil mechanics data comprises at least one of electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force; based on the soil mechanics data and the soil data, determining the erosion grade of the target slope body; according to the erosion grade, outputting a warning information; wherein, based on the soil mechanics data and the soil data, determining the erosion grade of the target slope body, comprising: for the soil at different positions and depths of the target slope body, based on the electrostatic repulsion, the van der Waals force, the hydration repulsion and the liquid bridge force, calculating the soil action force; obtaining a target constitutive relation; the target constitutive relation is used to represent the relationship between the soil action force and the soil cohesion; based on the target constitutive relation, determining the soil cohesion corresponding to the soil at different positions and depths of the target slope body; based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force and the soil data, determining the erosion grade of the target slope body; wherein, based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force and the soil data, determining the erosion grade of the target slope body, comprising: determining the soil at different positions and depths of the target slope body as nodes in a dynamic iterative graph structure; for each node, based on the soil corresponding to the soil cohesion, the soil action force and the soil data, constructing a target node feature vector corresponding to each node; the target node feature vector comprises: volumetric water content, soil temperature, electrical conductivity, pH value, soil stress, soil action force, soil cohesion, salt migration trend, strain activity level and node spatial coordinates; calculating the basic weight and dynamic correction factor between any two nodes; based on the basic weight and the dynamic correction factor, calculating the target edge weight of the target structure edge between any two nodes, and constructing the dynamic iterative graph structure; inputting the dynamic iterative graph structure and the environmental data into a preset particle erosion probability index model, and outputting the particle erosion probability index corresponding to each node in the dynamic iterative graph structure; based on the particle erosion probability index corresponding to each node and the target node feature vector, determining the erosion grade of the target slope body.

2. The method of claim 1, wherein, based on the soil data, calculating the soil mechanics data corresponding to the target slope body at different positions and depths, comprising: for the soil at different positions and depths of the target slope body, determining the cation concentration in the bulk solution based on the electrical conductivity; Converting the soil temperature to obtain an absolute temperature; Calculating the electrostatic repulsion in the soil mechanical data based on the cation concentration, the absolute temperature, and the inter-particle distance corresponding to the soil; Obtaining an initial Hamaker constant; Correcting the initial Hamaker constant based on the electrical conductivity, the volumetric water content, and the soil temperature to obtain a target Hamaker constant; Calculating the van der Waals force in the soil mechanical data based on the target Hamaker constant; Calculating the hydration repulsion in the soil mechanical data according to the inter-particle distance; Calculating the degree of salt crystallization based on the electrical conductivity and the volumetric water content; Obtaining an initial contact angle and an initial liquid bridge volume; Correcting the initial contact angle based on the degree of salt crystallization to obtain a target contact angle; Correcting the initial liquid bridge volume based on the degree of salt crystallization to obtain a target liquid bridge volume; Calculating the liquid bridge force based on the target contact angle and the target liquid bridge volume.

3. The method of claim 1, wherein, The obtaining of the target constitutive relation comprises: Obtaining an initial cohesion, a critical repulsion, and a time decay coefficient corresponding to the soil; Determining a force influence coefficient based on the soil type corresponding to the soil; Constructing the target constitutive relation based on the initial cohesion, the critical repulsion, the time decay coefficient, and the force influence coefficient.

4. The method of claim 1, wherein, The calculation of the base weight between any two nodes comprises: For any two nodes, obtaining a spatial straight-line distance between the two nodes; According to the spatial straight-line distance, calculating a physical distance weight between the two nodes; Obtaining the electrical conductivity and the volumetric water content of the two nodes within a first preset time length before the current time; Based on the electrical conductivity of the two nodes within the first preset time length before the current time, calculating an electrical conductivity correlation coefficient corresponding to the two nodes; Based on the volumetric water content of the two nodes within the first preset time length before the current time, calculating a volumetric water content correlation coefficient corresponding to the two nodes; Based on the electrical conductivity correlation coefficient and the volumetric water content correlation coefficient, calculating a correlation weight between the two nodes; Obtaining a target slope soil texture distribution map corresponding to the target slope body; Based on the target slope soil texture distribution map, determining the soil texture corresponding to the two nodes respectively; According to the soil texture corresponding to the two nodes respectively, determining a soil property weight between the two nodes; Based on the physical distance weight, the correlation weight, and the soil property weight, determining the base weight between the two nodes.

5. The method of claim 1, wherein, The calculation of the dynamic correction factor between any two nodes comprises: For any two nodes, based on the electrical conductivity and the soil stress corresponding to the two nodes respectively, calculating the salt migration trend and the strain activity degree corresponding to the two nodes respectively; Based on the salt migration trend and the strain activity degree corresponding to the two nodes respectively, calculating a soil state synergy factor corresponding to the two nodes; Based on the environmental data corresponding to the target slope body, determining an environmental response synergy factor; Determine the dynamic correction factor based on the soil state synergy factor and the environmental response synergy factor.

6. The method of claim 1, wherein, The dynamic iterative graph structure and the environmental data are input into a preset particle erosion probability index model to output a particle erosion probability index corresponding to each node in the dynamic iterative graph structure, including: Assigning local environmental data to each node based on the node spatial coordinates corresponding to each node and the environmental data; Stacking the target node feature vectors corresponding to each node to generate a node feature matrix; Constructing an adjacency matrix based on the target edge weights of the target structure edges between any two nodes; Performing feature extraction on the node feature matrix and the adjacency matrix to obtain spatial enhanced node features; For each node, calculating a time attention weight based on the relationship between the local environmental data corresponding to the node and the target node feature vector corresponding to the node; Calculating a spatial attention weight corresponding to each node based on the local environmental data corresponding to the node, the soil action force, and the strain activity level; For each node, dividing each sub-feature included in the spatial enhanced node features and the local environmental data into a soil modality, a mechanical modality, a strain modality, and an environmental modality; Calculating a modality weight corresponding to each sub-feature; For each node, performing weighted fusion on each sub-feature in the node based on the modality weight corresponding to each sub-feature, the spatial attention weight corresponding to the node, and the time attention weight to generate a multi-modal fusion feature vector corresponding to the node; For each node, stacking the multi-modal fusion feature vector corresponding to the node in a time sequence to form a time sequence feature; Performing feature extraction on the time sequence feature through a multi-head attention layer and a feedforward neural network to output a final fusion feature; Inputting the final fusion feature into a fully connected layer to output a particle erosion probability index corresponding to the node.

7. The method of claim 1, wherein, The determination of the erosion level corresponding to the target slope body based on the particle erosion probability index corresponding to each node and the target node feature vector includes: For each node, extracting the soil action force, the soil cohesion, the salt migration trend, and the strain activity level from the target node feature vector corresponding to the node; Determining a risk level corresponding to each node based on the particle erosion probability index corresponding to each node, the soil action force, the soil cohesion, the salt migration trend, and the strain activity level; Determining the erosion level corresponding to the target slope body based on the risk level corresponding to each node.

8. A soil erosion early warning device, characterized by, The device includes: A first acquisition module configured to acquire soil data corresponding to soil at different positions and depths of a target slope body; the soil data includes at least one of volumetric water content, soil temperature, electrical conductivity, pH value, and soil stress; The second acquisition module is configured to acquire environmental data corresponding to the target slope body; the environmental data comprises rainfall data, wind speed data, wind direction data, and evapotranspiration data; The third acquisition module is configured to calculate soil mechanics data corresponding to different positions and depths of the target slope body based on the soil data; the soil mechanics data comprises at least one of electrostatic repulsion, van der Waals force, hydration repulsion, and liquid bridge force; The determination module is configured to determine an erosion grade corresponding to the target slope body based on the soil mechanics data and the soil data; wherein, the determination of the erosion grade corresponding to the target slope body based on the soil mechanics data and the soil data comprises: calculating soil action force based on the electrostatic repulsion, the van der Waals force, the hydration repulsion, and the liquid bridge force for the soil at different positions and depths of the target slope body; acquiring a target constitutive relation; the target constitutive relation is used to represent the relationship between the soil action force and soil cohesion; determining the soil cohesion corresponding to the soil at different positions and depths of the target slope body based on the target constitutive relation; determining the erosion grade corresponding to the target slope body based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force, and the soil data; wherein, the determination of the erosion grade corresponding to the target slope body based on the soil cohesion corresponding to the soil at different positions and depths of the target slope body, the soil action force, and the soil data comprises: determining the soil at different positions and depths of the target slope body as nodes in a dynamic iterative graph structure; constructing a target node feature vector corresponding to each node based on the soil cohesion, the soil action force, and the soil data corresponding to the soil of each node; the target node feature vector comprises: volumetric water content, soil temperature, electrical conductivity, pH value, soil stress, soil action force, soil cohesion, salt migration trend, strain activity level, and node spatial coordinates; calculating a basic weight and a dynamic correction factor between any two nodes; calculating a target edge weight of a target structure edge between any two nodes based on the basic weight and the dynamic correction factor, and constructing the dynamic iterative graph structure; inputting the dynamic iterative graph structure and the environmental data into a preset particle erosion probability index model to output a particle erosion probability index corresponding to each node in the dynamic iterative graph structure; determining the erosion grade corresponding to the target slope body based on the particle erosion probability index corresponding to each node and the target node feature vector; The output module is configured to output early warning information according to the erosion grade.

Images