A landslide evolution dynamic identification method and device

Abstract

This invention provides a method and apparatus for dynamic identification of landslide evolution, relating to the field of landslide identification technology. The method includes: constructing a spatiotemporal similarity matrix based on the directional consistency and spatial proximity terms of monitoring point data; constructing a matrix representing the total energy by performing eigenvalue decomposition on the spatiotemporal similarity matrix; obtaining a principal energy concentration index based on spatiotemporal similarity matrix analysis, and calculating the standard deviation of energy concentration over a sliding time window as a measure of volatility; determining the development stage of the slip zone based on the matrix representing the total energy and the principal energy concentration index and volatility measure, along with volatility constraints, to obtain the slip zone development stage results; training a deep feedforward neural network based on a backpropagation algorithm, inputting slope parameters, soil and rock parameters, and slip zone development stage results into the trained neural network to predict the slope safety factor, thus obtaining the landslide evolution results. This invention solves the problem of insufficient identification accuracy in the slip zone formation stage.

Description

Technical Field

[0001] This invention relates to the field of landslide identification technology, and more specifically, to a method and apparatus for dynamic identification of landslide evolution. Background Technology

[0002] Current landslide identification technologies typically monitor landslides using single techniques such as InSAR and displacement gauges. However, these technologies struggle to quantify the co-evolutionary patterns of displacement direction consistency and spatial proximity. Furthermore, they lack energy thresholds and dynamic criteria for assessing the entire process of landslide transition from disordered deformation to ordered connectivity. Under extreme weather conditions or engineering disturbances, this results in a significant delay in early warning of the critical state from localized damage to overall instability. It also fails to effectively capture the coupling relationship between the dynamic reorganization of displacement vector fields and energy concentration mechanisms during landslide evolution, leading to insufficient identification accuracy during the landslide formation stage.

[0003] Therefore, there is an urgent need for a method and device based on dynamic identification of landslide evolution, which solves the problem of insufficient identification accuracy in the landslide formation stage. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for dynamic identification of landslide evolution, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a method for dynamic identification of landslide evolution, including:

[0006] Acquire monitoring point data, slope parameters, and soil and rock parameters;

[0007] A spatiotemporal similarity matrix is ​​constructed based on the directional consistency term and spatial proximity term of the monitoring point data. By performing eigenvalue decomposition on the spatiotemporal similarity matrix, a matrix is ​​constructed to represent the total energy.

[0008] Based on the spatiotemporal similarity matrix, eigenvalue spectrum analysis is performed to obtain the principal energy concentration index. The standard deviation of the energy concentration is calculated over a sliding time window and used as a measure of volatility.

[0009] Based on the matrix representing the total energy, the development stage of the main energy concentration index and the volatility metric is determined according to the preset volatility constraint conditions to obtain the slip zone development stage results.

[0010] The deep feedforward neural network is trained based on the backpropagation algorithm. The slope parameters, soil and rock parameters and the results of the slip zone development stage are input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results.

[0011] Secondly, this application also provides a landslide evolution dynamic identification device, which includes:

[0012] The acquisition module is used to acquire monitoring point data, slope parameters, and soil and rock parameters;

[0013] The construction module is used to construct a spatiotemporal similarity matrix based on the directional consistency term and spatial proximity term of the monitoring point data, and to construct a matrix to represent the total energy by performing eigenvalue decomposition on the spatiotemporal similarity matrix;

[0014] The analysis module is used to perform eigenvalue spectrum analysis based on the spatiotemporal similarity matrix to obtain the principal energy concentration index, and to calculate the standard deviation of the energy concentration over a sliding time window as a measure of volatility.

[0015] The judgment module is used to judge the development stage of the main energy concentration index and the volatility metric and the preset volatility constraint conditions based on the matrix characterizing the total energy, and to obtain the slip zone development stage result.

[0016] The training module is used to train a deep feedforward neural network based on the backpropagation algorithm. The slope parameters, soil and rock parameters and the results of the slip zone development stage are input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention constructs a spatiotemporal similarity matrix using monitoring point data and performs eigenvalue decomposition to effectively extract key features from the data, thereby more accurately identifying potential landslide areas. Based on this, it further obtains a principal energy concentration index through eigenvalue spectrum analysis and combines it with volatility measurement to more accurately determine the development stage of the landslide, solving the problem of difficulty in determining the landslide development stage. Furthermore, this invention also reliably predicts the evolution process of landslides based on a deep feedforward neural network trained using the backpropagation algorithm, providing a scientific basis for landslide early warning. In summary, this invention solves the problem of insufficient accuracy in identifying the slip zone formation stage.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the landslide evolution-based dynamic identification method described in this embodiment of the invention;

[0022] Figure 2 This is a schematic diagram of the landslide evolution dynamic identification device described in an embodiment of the present invention.

[0023] The markings in the diagram are: 800, landslide evolution dynamic identification device; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a method for dynamic identification of landslide evolution.

[0028] See Figure 1 The figure shows that the method includes steps S1 to S5, including:

[0029] S1: Acquire monitoring point data, slope parameters, and soil and rock parameters;

[0030] S2: Construct a spatiotemporal similarity matrix based on the directional consistency term and spatial proximity term of the monitoring point data. By performing eigenvalue decomposition on the spatiotemporal similarity matrix, construct a matrix to represent the total energy.

[0031] In this step, a spatiotemporal similarity matrix is ​​constructed and eigenvalue decomposition is performed using the monitoring point data to effectively extract key features from the data and accurately identify potential landslide areas.

[0032] To clarify the specific method for obtaining the total energy represented by the matrix, step S2 includes S21 to S25, specifically:

[0033] S21: Based on the first and second displacement vectors of the monitoring point data, the angles between them and the slope parameters along the x-axis are used to construct the first monitoring angle and the second monitoring angle.

[0034] In this step, the development stage of the landslide is assessed by the size and changing trend of the first and second monitoring angles, which reflects the changes in the stability of the landslide and provides a scientific basis for the formulation of landslide early warning and control measures.

[0035] S22: Calculate the consistency difference based on the first monitoring angle and the second monitoring angle to obtain the direction consistency term;

[0036] In this step, the expression for the direction consistency term is:

[0037] Δφ ij =|φ i ′ -φ j ′ |(1)

[0038] In equation (1) above, Δφ ij For the direction consistency term, φ i ′ Let φ be the angle between the displacement vector of monitoring point i and the slope surface. j ′ Let be the angle between the displacement vector of monitoring point j and the slope surface;

[0039] Wherein, the angle between the displacement vector of monitoring point i and the slope surface is the first monitoring angle, and the angle between the displacement vector of monitoring point j and the slope surface is the second monitoring angle.

[0040] S23: Calculate the Euclidean distance between two points based on the coordinates of the monitoring point data to obtain the spatial proximity term;

[0041] In this step, the expression for the spatial proximity term is:

[0042] Δx ij=|x i -x j |(2)

[0043] In equation (2) above, Δx ij For spatial proximity terms, x i Let x be the coordinates of point i in the monitoring data. j Let j be the coordinates of the monitoring point data.

[0044] S24: Based on the directional consistency term and the spatial proximity term, fuse similarity coefficients to construct a spatiotemporal similarity matrix;

[0045] In this step, the expression for the spatiotemporal similarity matrix M is:

[0046]

[0047] In equation (3) above, M ij M is the matrix element of the spatiotemporal similarity matrix M. ij , φ i ′ Let φ be the angle between the displacement vector of monitoring point i and the slope surface, exp be an exponential function, and Δφ be the angle between the displacement vector of monitoring point i and the slope surface. ij For the direction consistency term, Δx ij For spatial proximity terms, d max σ represents the maximum distance between monitoring points, and σ is the Gaussian kernel bandwidth.

[0048] S25: Based on the spectral theorem, the spatiotemporal similarity matrix is ​​decomposed into eigenvalues ​​to obtain an orthogonal eigenvector matrix. The total energy is obtained by fusing the orthogonal eigenvector matrix with the sum of the diagonal elements.

[0049] In this step, the spectral theorem specifically refers to the linear algebraic spectral theorem, which guarantees that any real symmetric matrix can be diagonalized, and its eigenvectors form an orthogonal basis.

[0050] The mathematical expression for eigenvalue decomposition is:

[0051] M=Q∧Q T (4)

[0052] In equation (4) above, Q is the orthogonal eigenvector matrix, ∧ is the diagonal matrix, and Q T Let Q be the transpose of matrix Q.

[0053] Among them, ∧=diag(λ1,λ2,...,λ n ), the λ1,λ2,...,λ n For eigenvalues;

[0054] The physical meaning of the eigenvalue decomposition is that the spatiotemporal correlation patterns implicit in the spatiotemporal similarity matrix M can be decomposed into mutually orthogonal motion modes, and the energy weight of each mode is quantified by its corresponding eigenvalue. Mathematically, the spectral theorem guarantees that the trace of the matrix is ​​equal to the sum of the eigenvalues. Physically, the normalization design of the diagonal elements gives it a clear potential energy meaning: the zero value corresponds to the ideal slip direction with minimum potential energy, and the increase in absolute value indicates an increase in potential energy.

[0055] Off-diagonal elements regulate the eigenvalue distribution through spatiotemporal cooperative terms, significantly increasing λ1 in spatially adjacent and coordinated node groups, thereby implicitly accumulating cooperative potential energy during energy redistribution. Specifically, the largest eigenvalue λ1 dominates the energy in the slip direction, while secondary eigenvalues ​​reflect local dissipation mechanisms, and the overall system satisfies the law of energy conservation.

[0056] Through eigenvalue decomposition, we can obtain an expression for the matrix representing the total energy:

[0057]

[0058] In equation (5) above, Tr(M) is a matrix representing the total energy. Let λ be all eigenvalues ​​from k=1 to k=n k The sum of λ k For eigenvalues, φ i ′ Let be the angle between the displacement vector of monitoring point i and the slope surface.

[0059] S3: Perform eigenvalue spectrum analysis based on the spatiotemporal similarity matrix to obtain the principal energy concentration index, and combine it with the standard deviation of energy concentration calculated over the sliding time window as a measure of volatility;

[0060] During landslide development, many factors influence the process, and relying solely on the energy concentration index R to determine the stage of landslide development may lead to errors due to occasional extreme conditions. To overcome the limitations of this static threshold method, the standard deviation of the energy concentration R(t) within the sliding time window is introduced as a measure of volatility to accurately determine the development stage of the landslide, thus solving the problem of difficulty in determining the landslide development stage.

[0061] The expression for the main energy concentration index is:

[0062]

[0063] In equation (6) above, R is the main energy concentration index, and λ1 is the maximum eigenvalue. Let λ be all eigenvalues ​​from k=1 to k=n k The sum of λ k These are the eigenvalues.

[0064] The expression for the volatility measure is:

[0065]

[0066] In equation (7) above, σ R R(t) is the volatility index at time t, where T is the time window length, k is the index variable for summation, and R(k) is the energy concentration value at time step k. Let R be the mean value of R within the window.

[0067] The volatility index is used to measure the volatility of energy concentration.

[0068] S4: Based on the matrix representing the total energy, determine the development stage of the main energy concentration index and the volatility metric and the preset volatility constraint conditions to obtain the slip zone development stage results;

[0069] To clarify the specific method for obtaining the results of the spondylolisthesis development stage, step S4 includes S41 to S46, specifically:

[0070] S41: Obtain the constitutive model of the soil and rock mass;

[0071] S42: Based on the distribution law of the eigenvalues ​​of the total energy characterized by the matrix and the constitutive model of the soil and rock mass, construct the peak strength and residual strength in the parameters of the soil and rock mass to obtain the energy concentration threshold;

[0072] In this step, the determination of the energy concentration threshold should take into account the peak strength and residual strength of the soil and rock mass as key parameters. The expression for the energy concentration threshold is:

[0073]

[0074] In equation (8) above, R crit Here, f is the energy concentration threshold, f is a parameter related to the constitutive model of the soil and rock mass, and c is... r c represents the residual cohesion of the soil and rock mass. p φ is the peak cohesion of the soil and rock mass. r φ is the residual internal friction angle of the rock and soil mass. p is the peak internal friction angle of the rock and soil mass.

[0075] The value of parameter f is related to the selected constitutive model of the soil and rock mass. When the constitutive model of the soil and rock mass is an ideal elastic-plastic model, parameter f is 0.4 to 0.6. When there is strain softening, parameter f is 1.

[0076] S43: Based on the constitutive model of the soil and rock mass and the dimensions in the slope parameters, determine the critical value of fluctuation.

[0077] S44: Based on the energy concentration threshold and the main energy concentration index, the energy concentration condition is obtained;

[0078] S45: Based on the volatility metric within the sliding time window and the volatility critical value, the volatility stability condition is obtained;

[0079] In this step, the volatility stability condition is:

[0080]

[0081] In equation (9) above, R(t) represents the energy concentration within the sliding time window. crit σ is the energy concentration threshold. R (t) is a function of volatility over time, σ crit This is the critical value for volatility.

[0082] S46: Based on the energy concentration condition and the fluctuation stability condition, the development stage is determined to obtain the slip zone development stage result, which includes the undeveloped stage, the initial development stage, and the through stage.

[0083] S5: Based on the backpropagation algorithm, a deep feedforward neural network is trained. The slope parameters, soil and rock parameters, and the results of the slip zone development stage are input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results.

[0084] In this step, the slope safety factor is predicted to provide a scientific basis for landslide early warning.

[0085] To clarify the specific methods for obtaining landslide evolution results, step S5 includes S51 to S54, specifically:

[0086] S51: Based on the backpropagation algorithm, the energy characteristics and mechanical parameters in the soil and rock mass parameters are input into a deep feedforward neural network for initial training to obtain the initial neural network;

[0087] To clarify the specific method for obtaining the initial neural network, step S51 includes S511 to S514, specifically:

[0088] S511: Input the soil and rock mechanical parameters from the soil and rock parameters into the input layer of the deep feedforward neural network, and pass them through the input layer to the first hidden layer to obtain the initial neural network output;

[0089] In this step, the soil and rock mechanical parameters, energy concentration index, and fluctuation index are input into the deep feedforward neural network, and the initial neural network is obtained through the first hidden layer.

[0090] The mechanical parameters of the rock and soil include the internal friction angle, cohesion, and unit weight.

[0091] The first hidden layer (512 neurons) uses the GELU activation function, with the formula GELU(x) = xΦ(x), where Φ(x) is the cumulative distribution function of the standard normal distribution, expressed as:

[0092]

[0093] In equation (10) above, Φ(x) is the cumulative distribution function of the standard normal distribution, and π is the mathematical constant pi. Let dt be the integral from negative infinity (-∞) to x, e be the base of the natural logarithm, g be the variable of integration, and dt be the differential element of the integral.

[0094] The cumulative distribution function of the standard normal distribution adaptively adjusts the activation level of neurons based on the probability distribution of the input data, which can efficiently capture complex features in the data.

[0095] S512: Based on the initial neural network output, an activation function is used to perform a nonlinear transformation on multiple hidden layers to obtain intermediate feature representations;

[0096] In this step, the multi-layer hidden layer comprises six layers: a second hidden layer, a third hidden layer, a fourth hidden layer, a fifth hidden layer, a sixth hidden layer, and a seventh hidden layer. Each layer employs a specific activation function to enhance the network's non-linear expressive power and feature selection ability. Specifically:

[0097] The second hidden layer (256 neurons) uses the Swish-Max activation function, with the formula: Swish-Max(x)=max(0,x·σ(x)), where σ(x) is the activation part of the Swish function;

[0098] The expression for the activation part of the Swish function is:

[0099]

[0100] In equation (11) above, σ(x) is the activation part of the Swish function, e is the base of the natural logarithm, and x is the input value.

[0101] The Swish function combines the adaptability of the Swish function with the non-linear filtering characteristics of the ReLU function, further enhancing the network's ability to filter and combine important features.

[0102] The third hidden layer (128 neurons): uses the Mish-Leaky activation function, the expression of which is:

[0103]

[0104] In equation (12) above, x is the input value, tanh is the hyperbolic tangent function, ln is the natural logarithm function, e is the base of the natural logarithm, and α is the negative slope parameter of the LeakyReLU function.

[0105] In this case, α = 0.01. The Mish-Leaky activation function combines the smoothness of the Mish function with the ability of the LeakyReLU function to handle negative data, effectively avoiding the problem of neuron "death" and improving the network's learning effect on data features of different polarities.

[0106] The fourth hidden layer (64 neurons): uses the ELU (Exponential Linear Unit) activation function, the expression of which is:

[0107]

[0108] In equation (13) above, x is the input value, ln is the natural logarithm function, e is the base of the natural logarithm, and α is the negative slope parameter of the LeakyReLU function.

[0109] Typically, a = 1 is chosen. This ensures the network has a non-zero gradient on the negative half-axis, accelerating training convergence, while maintaining linearity on the positive half-axis to ensure effective transfer of feature information.

[0110] Fifth hidden layer (32 neurons): Introduces the Softplus activation function, the expression of which is:

[0111]

[0112] In equation (14) above, x is the input value, ln is the natural logarithm function, e is the base of the natural logarithm, and β is the parameter of the Softplus function.

[0113] Among them, taking β=1 smoothly maps the input data to the positive range, performs well when processing data features with non-negative characteristics, and approximates the ReLU function near the origin, which helps to alleviate the gradient vanishing problem.

[0114] The sixth hidden layer (16 neurons): uses SiLU (Sigmoid-1234 ...

[0115]

[0116] In equation (15) above, x is the input value, ln is the natural logarithm function, and e is the base of the natural logarithm.

[0117] By using the Sigmoid function to weight the input, the nonlinear transformation capability of the network is enhanced, improving the model's fitting effect on complex data.

[0118] The seventh hidden layer (8 neurons): uses the ReLU6 activation function, the expression of which is:

[0119] ReLU6(x)=min(max(0,x),6)(16)

[0120] In equation (16) above, x is the input value.

[0121] S513: Adaptively adjust the neuron activation level according to the probability distribution of the energy characteristics to obtain the adjusted neuron activation state;

[0122] In this step, the probability distribution is estimated using kernel density based on the probability distribution of the energy characteristics. For each neuron, the activation level is adjusted according to its corresponding energy characteristic value, and each element in the intermediate feature representation is updated to obtain the adjusted neuron activation state.

[0123] S514: Based on the intermediate feature representation and the adjusted neuron activation state, the output of the initial neural network is iteratively optimized to obtain the initial neural network.

[0124] In this step, the output of the initial neural network is iteratively optimized multiple times based on the intermediate feature representation and the adjusted neuron activation state, adjusting the network weights and biases to obtain the final initial neural network.

[0125] S52: Calculate the error between the predicted output and the actual output of the initial neural network, and train and optimize the initial neural network by combining it with a regularized elastic network loss function to obtain an optimized neural network;

[0126] To clarify the specific method for obtaining the optimized neural network, step S52 includes S521 to S523, specifically:

[0127] S521: Calculate the error between the predicted output and the actual output of the initial neural network to obtain the output error value;

[0128] This step is used to quantify the difference between the model predictions and the actual values.

[0129] S522: Based on the output error value, adjust the network parameters of the initial neural network using the elastic network loss function to obtain the adjusted network parameters;

[0130] In this step, based on the output error value, an elastic network loss function is obtained by combining L1 regularization and L2 regularization loss functions. The elastic network loss function is then used to adjust the network parameters of the initial neural network to obtain adjusted network parameters, thereby reducing errors and improving the generalization ability of the model.

[0131] The expression for the elastic network loss function is:

[0132]

[0133] In equation (17) above, L is the elastic network loss function, n is the number of samples, and y i For the true value, Here, p represents the predicted value, and w represents the number of network parameters. j Let λ1 and λ2 be the j-th parameter, and let λ1 and λ2 be the weight coefficients for L1 regularization and L2 regularization, respectively.

[0134] S523: The network parameters are updated based on an adaptive optimization algorithm until the preset iteration conditions are met, and the optimized neural network is output.

[0135] In this step, the network parameters are updated based on the adaptive optimization algorithm. The adaptive optimization algorithm dynamically adjusts the learning rate according to the historical information of the gradient until the preset iteration conditions (maximum number of iterations or convergence of the loss function) are met, and then outputs the optimized neural network.

[0136] The adaptive optimization algorithm (Adafactor optimization algorithm) is updated as follows:

[0137] Calculate the gradient: Where θ represents the network parameters and J(θ) represents the loss function.

[0138] Calculate the second moment estimate: V t Let V be the second moment estimate at time step t, ρ be the decay rate, and V be the second moment estimate at time step t. t-1 This is the estimate of the second moment at time step t-1. For gradient.

[0139] Calculate the adaptive learning rate: η t Let η be the adaptive learning rate at time step t, η be the initial learning rate, and ∈ be a small constant to prevent the denominator from being zero.

[0140] Update parameters: θ t+1 Let θ be the network parameters at time step t+1. t Here are the network parameters at time step t.

[0141] S53: Input the slope parameters, soil and rock parameters and the results of the slip zone development stage into the optimization neural network for iterative training to obtain the trained neural network model;

[0142] To clarify the specific method for obtaining the training neural network model, step S53 includes steps S531 to S533, specifically:

[0143] S531: Preprocess the slope parameters, soil and rock parameters and the results of the slip zone development stage to obtain standardized input data;

[0144] S532: Input the standardized input data into the optimized neural network, calculate the standardized input data through forward propagation, and obtain the network output result;

[0145] In this step, the standardized input data is fed into the optimized neural network, and forward propagation is performed through linear transformations and activation functions at each layer to obtain the network output.

[0146] S533: Calculate the output difference value based on the network output result and the target output result, and iteratively train the optimized neural network using the backpropagation algorithm and the output difference value until the preset convergence condition is met, and output the trained neural network model.

[0147] In this step, the difference between the network output and the target output is calculated, the error is calculated using a loss function, the gradient is calculated using a backpropagation algorithm, the network parameters are updated using an optimization algorithm, and the training is repeated iteratively until a preset convergence condition is met. The convergence condition is that the loss function is less than a certain threshold or the maximum number of iterations is reached, and the trained neural network model is output.

[0148] S54: Based on the trained neural network model, the slope safety factor is predicted to obtain the landslide evolution result.

[0149] In this step, slope parameters, soil and rock parameters, and slip zone development stage results are input into the trained neural network model to predict the slope's safety factor. Based on the predicted safety factor, the slope's stability is further evaluated; a higher safety factor indicates a more stable slope, and vice versa. Through this prediction and evaluation, the slope's safety status can be understood in advance, providing a scientific basis for landslide prevention and helping to formulate reasonable prevention measures.

[0150] Example 2:

[0151] This embodiment provides a landslide evolution dynamic identification device, the device comprising:

[0152] The acquisition module is used to acquire monitoring point data, slope parameters, and soil and rock parameters;

[0153] The construction module is used to construct a spatiotemporal similarity matrix based on the directional consistency term and spatial proximity term of the monitoring point data, and to construct a matrix to represent the total energy by performing eigenvalue decomposition on the spatiotemporal similarity matrix;

[0154] To clarify the specific methods for obtaining the building modules, the following are included:

[0155] The first construction unit is used to construct the first monitoring angle and the second monitoring angle based on the angles between the first and second displacement vectors of the monitoring point data and the slope parameters along the x-axis, respectively.

[0156] The first calculation unit is used to calculate the consistency difference based on the first monitoring angle and the second monitoring angle to obtain the direction consistency term.

[0157] The second calculation unit is used to calculate the Euclidean distance between two points based on the coordinates of the monitoring point data, and obtain the spatial proximity term.

[0158] A fusion unit is used to fuse similarity coefficients based on the directional consistency term and the spatial proximity term to construct a spatiotemporal similarity matrix;

[0159] The decomposition unit is used to perform eigenvalue decomposition on the spatiotemporal similarity matrix based on the spectral theorem to obtain an orthogonal eigenvector matrix. The total energy is obtained by fusing the orthogonal eigenvector matrix with the sum of the diagonal elements.

[0160] The analysis module is used to perform eigenvalue spectrum analysis based on the spatiotemporal similarity matrix to obtain the principal energy concentration index, and to calculate the standard deviation of the energy concentration over a sliding time window as a measure of volatility.

[0161] The judgment module is used to judge the development stage of the main energy concentration index and the volatility metric and the preset volatility constraint conditions based on the matrix characterizing the total energy, and to obtain the slip zone development stage result.

[0162] The training module is used to train a deep feedforward neural network based on the backpropagation algorithm. The slope parameters, soil and rock parameters and the results of the slip zone development stage are input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results.

[0163] To clarify the specific methods for obtaining the training modules, the following are included:

[0164] An initial training unit is used to input the energy characteristics and mechanical parameters of the soil and rock mass parameters into a deep feedforward neural network for initial training based on the backpropagation algorithm, thereby obtaining an initial neural network.

[0165] To clarify the specific methods for obtaining the initial training units, the following are included:

[0166] The first training unit is used to input the soil and rock mechanical parameters from the soil and rock parameters into the input layer of the deep feedforward neural network, and pass them through the input layer to the first hidden layer to obtain the initial neural network output.

[0167] The second training unit is used to perform nonlinear transformations on multiple hidden layers through activation functions based on the output of the initial neural network to obtain intermediate feature representations.

[0168] The third training unit is used to adaptively adjust the activation level of neurons according to the probability distribution of the energy characteristics, so as to obtain the adjusted neuron activation state.

[0169] The fourth training unit is used to iteratively optimize the output of the initial neural network based on the intermediate feature representation and the adjusted neuron activation state to obtain the initial neural network.

[0170] An error calculation unit is used to calculate the error between the predicted output and the actual output of the initial neural network, and to train and optimize the initial neural network by combining it with a regularized elastic network loss function to obtain an optimized neural network.

[0171] An iterative training unit is used to input the slope parameters, soil and rock parameters and the results of the slip zone development stage into the optimization neural network for iterative training to obtain a trained neural network model.

[0172] The prediction unit is used to predict the slope safety factor based on the trained neural network model and obtain the landslide evolution result.

[0173] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0174] Example 3:

[0175] Corresponding to the above method embodiments, this embodiment also provides a landslide evolution dynamic identification device. The landslide evolution dynamic identification device described below and the landslide evolution dynamic identification method described above can be referred to each other.

[0176] Figure 2 This is a block diagram illustrating a landslide evolution dynamic identification device 800 according to an exemplary embodiment. Figure 2As shown, the landslide evolution-based dynamic identification device 800 may include: a processor 801 and a memory 802. The landslide evolution-based dynamic identification device 800 may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.

[0177] The processor 801 controls the overall operation of the landslide evolution-based dynamic identification device 800 to complete all or part of the steps in the landslide evolution-based dynamic identification method described above. The memory 802 stores various types of data to support the operation of the landslide evolution-based dynamic identification device 800. This data may include, for example, instructions for any application or method operating on the landslide evolution-based dynamic identification device 800, as well as application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the landslide evolution-based dynamic identification device 800 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0178] In an exemplary embodiment, the landslide evolution-based dynamic identification device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the landslide evolution-based dynamic identification method described above.

[0179] Example 4:

[0180] Corresponding to the above method embodiments, this embodiment also provides a medium. The medium described below can be referred to in relation to the landslide evolution dynamic identification method described above.

[0181] A medium storing a computer program, which, when executed by a processor, implements the steps of the landslide evolution-based dynamic identification method described in the above method embodiments.

[0182] The medium can specifically be any medium capable of storing program code, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0183] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0184] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for dynamic identification of landslide evolution, characterized in that, include: Acquire monitoring point data, slope parameters, and soil and rock parameters; A spatiotemporal similarity matrix is ​​constructed based on the directional consistency term and spatial proximity term of the monitoring point data. By performing eigenvalue decomposition on the spatiotemporal similarity matrix, a matrix is ​​constructed to represent the total energy. Specifically, the matrix characterizing the total energy is as follows: Based on the first and second displacement vectors of the monitoring point data, the angles between them and the slope parameters along the x-axis are used to construct the first monitoring angle and the second monitoring angle. Based on the first monitoring angle and the second monitoring angle, a consistency difference is calculated to obtain a direction consistency term; Calculate the Euclidean distance between two points based on the coordinates of the monitoring point data to obtain the spatial proximity term; Based on the directional consistency term and the spatial proximity term, a spatiotemporal similarity matrix is ​​constructed by fusing similarity coefficients. Based on the spectral theorem, the spatiotemporal similarity matrix is ​​decomposed into eigenvalues ​​to obtain an orthogonal eigenvector matrix. The total energy is obtained by fusing the orthogonal eigenvector matrix with the sum of the diagonal elements. Based on the spatiotemporal similarity matrix, eigenvalue spectrum analysis is performed to obtain the principal energy concentration index. The standard deviation of the energy concentration is calculated over a sliding time window and used as a measure of volatility. Based on the matrix characterizing the total energy, the development stage of the main energy concentration index and the volatility metric is determined according to the preset volatility constraint conditions to obtain the slip zone development stage results. Specifically, the results of the slip band development stage are as follows: Obtain the constitutive model of the soil and rock mass; Based on the distribution law of the eigenvalues ​​of the total energy characterized by the matrix and the constitutive model of the soil and rock mass, the peak strength and residual strength in the parameters of the soil and rock mass are constructed to obtain the energy concentration threshold. Based on the constitutive model of the soil and rock mass and the dimensions in the slope parameters, the critical value of fluctuation is determined. Based on the energy concentration threshold and the main energy concentration index, the energy concentration conditions are obtained; Based on the volatility metric within the sliding time window and the volatility threshold, the volatility stability condition is obtained. The development stage is determined based on the energy concentration condition and the fluctuation stability condition to obtain the slip zone development stage result, which includes the undeveloped stage, the initial development stage, and the through stage. Based on the backpropagation algorithm, the energy characteristics and mechanical parameters in the soil and rock parameters are input into a deep feedforward neural network for initial training to obtain an initial neural network. The slope parameters, soil and rock parameters and the results of the slip zone development stage are input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results. The obtained initial neural network is specifically: The soil and rock mechanical parameters in the soil and rock parameters are input into the input layer of the deep feedforward neural network, and then passed to the first hidden layer through the input layer to obtain the initial neural network output. Based on the initial neural network output, an activation function is used to perform a nonlinear transformation on multiple hidden layers to obtain intermediate feature representations; The activation level of neurons is adaptively adjusted based on the probability distribution of the energy characteristics to obtain the adjusted neuron activation state; The initial neural network output is iteratively optimized based on the intermediate feature representation and the adjusted neuron activation state to obtain the initial neural network.

2. The landslide evolution-based dynamic identification method according to claim 1, characterized in that, Based on the backpropagation algorithm, the energy characteristics and mechanical parameters of the soil and rock mass are input into a deep feedforward neural network for initial training to obtain an initial neural network. The slope parameters, soil and rock mass parameters, and the results of the slip zone development stage are then input into the trained neural network to predict the slope safety factor, yielding landslide evolution results, including: Based on the backpropagation algorithm, the energy characteristics and mechanical parameters of the soil and rock mass are input into a deep feedforward neural network for initial training to obtain the initial neural network; The error between the predicted output and the actual output of the initial neural network is calculated, and the initial neural network is trained and optimized by combining the regularized elastic network loss function to obtain an optimized neural network. The slope parameters, soil and rock parameters, and the results of the slip zone development stage are input into the optimized neural network for iterative training to obtain the trained neural network model. The slope safety factor is predicted based on the trained neural network model, and the landslide evolution results are obtained.

3. The landslide evolution-based dynamic identification method according to claim 1, characterized in that, The error between the predicted output and the actual output of the initial neural network is calculated. The initial neural network is then trained and optimized using a sparsely regularized elastic network loss function to obtain an optimized neural network, including: The error between the predicted output and the actual output of the initial neural network is calculated to obtain the output error value; Based on the output error value, the network parameters of the initial neural network are adjusted using the elastic network loss function to obtain the adjusted network parameters; The network parameters are updated based on an adaptive optimization algorithm until the preset iteration conditions are met, and an optimized neural network is output.

4. The landslide evolution-based dynamic identification method according to claim 1, characterized in that, The slope parameters, soil and rock parameters, and the results of the slip zone development stage are input into an optimized neural network for iterative training to obtain a trained neural network model, including: The slope parameters, soil and rock parameters, and the results of the slip zone development stage are preprocessed to obtain standardized input data. The standardized input data is input into the optimized neural network, and the standardized input data is calculated through forward propagation to obtain the network output result; The difference between the network output and the target output is calculated, and the optimized neural network is iteratively trained using the backpropagation algorithm and the output difference until the preset convergence condition is met, and the trained neural network model is output.

5. A landslide evolution dynamic identification device, characterized in that, The method for dynamic identification based on landslide evolution as described in any one of claims 1-4 is used, including: The acquisition module is used to acquire monitoring point data, slope parameters, and soil and rock parameters; The construction module is used to construct a spatiotemporal similarity matrix based on the directional consistency term and spatial proximity term of the monitoring point data, and to construct a matrix to represent the total energy by performing eigenvalue decomposition on the spatiotemporal similarity matrix; The building module includes: The first construction unit is used to construct the first monitoring angle and the second monitoring angle based on the angles between the first and second displacement vectors of the monitoring point data and the slope parameters along the x-axis, respectively. The first calculation unit is used to calculate the consistency difference based on the first monitoring angle and the second monitoring angle to obtain the direction consistency term. The second calculation unit is used to calculate the Euclidean distance between two points based on the coordinates of the monitoring point data, and obtain the spatial proximity term. A fusion unit is used to fuse similarity coefficients based on the directional consistency term and the spatial proximity term to construct a spatiotemporal similarity matrix; The decomposition unit is used to perform eigenvalue decomposition on the spatiotemporal similarity matrix based on the spectral theorem to obtain an orthogonal eigenvector matrix. The total energy is obtained by fusing the orthogonal eigenvector matrix with the sum of the diagonal elements. The analysis module is used to perform eigenvalue spectrum analysis based on the spatiotemporal similarity matrix to obtain the principal energy concentration index, and to calculate the standard deviation of the energy concentration over a sliding time window as a measure of volatility. The judgment module is used to judge the development stage of the main energy concentration index and the volatility metric and the preset volatility constraint conditions based on the matrix characterizing the total energy, and to obtain the slip zone development stage result. The results of the slip band development stage include: Obtain the constitutive model of the soil and rock mass; Based on the distribution law of the eigenvalues ​​of the total energy characterized by the matrix and the constitutive model of the soil and rock mass, the peak strength and residual strength in the parameters of the soil and rock mass are constructed to obtain the energy concentration threshold. Based on the constitutive model of the soil and rock mass and the dimensions in the slope parameters, the critical value of fluctuation is determined. Based on the energy concentration threshold and the main energy concentration index, the energy concentration conditions are obtained; Based on the volatility metric within the sliding time window and the volatility threshold, the volatility stability condition is obtained. The development stage is determined based on the energy concentration condition and the fluctuation stability condition to obtain the slip zone development stage result, which includes the undeveloped stage, the initial development stage, and the through stage. The training module is used to input the energy characteristics and mechanical parameters of the soil and rock mass parameters into a deep feedforward neural network for initial training based on the backpropagation algorithm to obtain an initial neural network. The slope parameters, soil and rock mass parameters and the results of the slip zone development stage are then input into the trained neural network to predict the slope safety factor and obtain the landslide evolution results. The initial neural network includes: The first training unit is used to input the soil and rock mechanical parameters from the soil and rock parameters into the input layer of the deep feedforward neural network, and pass them through the input layer to the first hidden layer to obtain the initial neural network output. The second training unit is used to perform nonlinear transformations on multiple hidden layers through activation functions based on the output of the initial neural network to obtain intermediate feature representations. The third training unit is used to adaptively adjust the activation level of neurons according to the probability distribution of the energy characteristics, so as to obtain the adjusted neuron activation state. The fourth training unit is used to iteratively optimize the output of the initial neural network based on the intermediate feature representation and the adjusted neuron activation state to obtain the initial neural network.

6. The landslide evolution dynamic identification device according to claim 5, characterized in that, The training module includes: An initial training unit is used to input the energy characteristics and mechanical parameters of the soil and rock mass parameters into a deep feedforward neural network for initial training based on the backpropagation algorithm, thereby obtaining an initial neural network. An error calculation unit is used to calculate the error between the predicted output and the actual output of the initial neural network, and to train and optimize the initial neural network by combining it with a regularized elastic network loss function to obtain an optimized neural network. An iterative training unit is used to input the slope parameters, soil and rock parameters and the results of the slip zone development stage into the optimization neural network for iterative training to obtain a trained neural network model. The prediction unit is used to predict the slope safety factor based on the trained neural network model and obtain the landslide evolution result.

Images