Construction method and device of slope creep deformation agent model and storage medium
By building a slope creep deformation agent model containing dynamic boundary conditions and time-related parameters, using Transformer physical mechanism to enhance modules and multi-layer stacked GRU network, the shortcomings of traditional models that cannot accurately reflect slope deformation are solved, and prediction accuracy is improved. Physical process capture capability, suitable for slope risk assessment.
Patent Information
- Application Number
- CN202411117380.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-14
AI Technical Summary
The traditional slope creep deformation agent model cannot accurately reflect the interaction between changing boundary conditions and intrinsic physical mechanisms, resulting in insufficient prediction accuracy for slope deformation.
A slope creep deformation agent model is constructed that includes dynamic external boundary conditions and time-dependent dynamic parameters. Through the Transformer physical mechanism enhances modules and multi-layer stacked GRU network, a slope creep deformation time series is generated that changes over time.
It improves the prediction accuracy of slope deformation, enhances the model's response sensitivity and prediction accuracy to physical processes, provides a more detailed time dynamic change capture capability, and is suitable for rapid inversion of slope parameters and risk assessment.
Smart Images

Figure CN118916966A_ABST
Abstract
Description
Technical Field
[0001] This article relates to slope digital twin technology, and in particular to a construction method, device and storage medium for a slope creep deformation proxy model. Background Art
[0002] The instability of reservoir bank slopes not only threatens the safety of people's lives and property, but also affects the normal operation of infrastructure such as hydropower stations. The use of numerical simulation to evaluate slope stability is crucial for timely discovering potential landslide risk areas, predicting possible landslide events, and implementing effective disaster prevention and mitigation measures.
[0003] The reliability of numerical simulation depends on whether the physical mechanism under consideration is consistent with the actual situation and whether the values of the constitutive parameters adopted are reasonable. Inversion of constitutive parameters based on field deformation monitoring data is an important method to improve the reliability of numerical simulation. Among them, the proxy model reflecting slope deformation, as an efficient simulation tool, plays the role of a bridge between constitutive parameters and creep deformation in the inversion process. However, with the complexity of field engineering conditions, traditional proxy models have shown deficiencies in dealing with complex physical processes and often cannot accurately reflect the interaction between variable boundary conditions and internal physical mechanisms. Summary of the invention
[0004] The embodiments of the present application provide a method, device and storage medium for constructing a slope creep deformation proxy model, which can improve the prediction accuracy of slope deformation.
[0005] The method for constructing a slope creep deformation proxy model provided in the embodiment of the present application includes:
[0006] A set of numerical samples for training the slope creep deformation proxy model is constructed, each numerical sample in the set includes a first constitutive parameter, an external boundary condition causing the slope creep deformation, a first intermediate implicit feature and a first slope creep deformation time series; the first constitutive parameter is determined by the original constitutive parameter of the creep constitutive model reflecting the creep deformation mechanism of the slope, including: a static first constitutive parameter that is not associated with the time dimension, a dynamic first constitutive parameter that is associated with the time dimension and a steady-state parameter; the first slope creep deformation time series is determined by the slope creep deformation response obtained by calculating the slope calculation grid through the slope creep constitutive model; the first intermediate implicit feature is a time dimension associated feature generated in the nonlinear calculation process of the slope creep constitutive model;
[0007] The following operations are performed on each numerical sample: the static first constitutive parameter that is not associated with the time dimension in the first constitutive parameter is input into the first learning network to obtain the predicted intermediate implicit feature; the dynamic first constitutive parameter that is associated with the time dimension in the first constitutive parameter, the predicted intermediate implicit feature, and the steady-state parameter in the first constitutive parameter are input into the second learning network to obtain the predicted slope creep deformation time series; based on the numerical sample, the predicted intermediate implicit feature and the predicted slope creep deformation time series, the combined learning network constructed by the first learning network and the second learning network is optimized in parameters;
[0008] The combined learning network is trained based on the numerical sample set and the final optimized parameters of the combined learning network, and the trained combined learning network is used as the slope creep deformation proxy model.
[0009] The non-transitory computer-readable storage medium provided in the embodiment of the present application stores a computer program, and the computer program can be executed by a processor to implement a method for constructing a slope creep deformation proxy model.
[0010] The device for constructing a slope creep deformation proxy model provided in an embodiment of the present application includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is read and executed by the processor, the method for constructing a slope creep deformation proxy model can be implemented.
[0011] Traditional slope creep deformation proxy models are usually static, that is, the parameters input into the slope creep deformation proxy model are static parameters that do not change with time, and the parameters output from the slope creep deformation proxy model are also static deformation increments. This slope creep deformation proxy model cannot accurately reflect the interaction between changing boundary conditions and internal physical mechanisms. The slope creep deformation proxy model constructed in the embodiment of the present application has input parameters that include dynamic external boundary conditions that cause slope creep deformation, as well as time-related dynamic parameters, and the output parameters are also time-varying slope creep deformation time series. Therefore, it can accurately reflect the time-dependent deformation of the slope, thereby improving the prediction accuracy of slope creep deformation.
[0012] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by implementing the present application. Other advantages of the present application can be realized and obtained by the schemes described in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The accompanying drawings are used to provide an understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.
[0014] Figure 1 A flow chart of a method for constructing a slope creep deformation proxy model provided in an embodiment of the present application;
[0015] Figure 2 An example diagram of constructing a slope creep deformation proxy model provided in this application;
[0016] Figure 3 An example diagram of a reservoir bank slope calculation grid provided for this application;
[0017] Figure 4 An example diagram of a full-process creep deformation curve provided in this application;
[0018] Figure 5 An example diagram of a Transformer physical mechanism enhancement module provided for this application;
[0019] Figure 6 An example diagram of a multi-layer stacked GRU network provided for this application;
[0020] Figure 7 An example diagram comparing the predicted value and actual value of the creep deformation time series of a typical GNSS monitoring point using a slope creep deformation proxy model provided in this application;
[0021] Figure 8 A module diagram of a device for constructing a slope creep deformation proxy model provided in an embodiment of the present application. DETAILED DESCRIPTION
[0022] The present application describes multiple embodiments, but the description is exemplary rather than restrictive, and it is obvious to those skilled in the art that there may be more embodiments and implementations within the scope of the embodiments described in the present application. Although many possible feature combinations are shown in the drawings and discussed in the specific embodiments, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.
[0023] The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features and elements disclosed in the present application may also be combined with any conventional features or elements to form a unique invention scheme. Any features or elements of any embodiment may also be combined with features or elements from other invention schemes to form another unique invention scheme. Therefore, it should be understood that any feature shown and / or discussed in the present application may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the attached claims and their equivalents, the embodiments are not subject to other restrictions. In addition, various modifications and changes may be made within the scope of protection of the attached claims.
[0024] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps of the specific order described. As will be understood by those of ordinary skill in the art, other sequences of steps are also possible. Therefore, the specific sequence of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to the steps of performing them in the order written, and those skilled in the art can easily understand that these sequences can be changed and still remain within the spirit and scope of the embodiments of the present application.
[0025] The present application embodiment provides a method for constructing a slope creep deformation proxy model, such as Figure 1 As shown, the method includes:
[0026] Step S100 constructs a numerical sample set for training the slope creep deformation proxy model, each numerical sample in the set includes a first constitutive parameter, an external boundary condition causing the slope creep deformation, a first intermediate implicit feature and a first slope creep deformation time series;
[0027] The first constitutive parameter is determined by the original constitutive parameter of the slope creep constitutive model reflecting the creep deformation mechanism of the slope, including: a static first constitutive parameter not associated with the time dimension, a dynamic first constitutive parameter associated with the time dimension and a steady-state parameter;
[0028] The constitutive model is a mathematical model that describes the mechanical behavior of materials. It reflects the stress-strain relationship of materials when they are subjected to stress. The constitutive parameters are physical or empirical parameters in the constitutive model, which quantitatively describe the mechanical properties of materials, such as elastic modulus, yield strength, etc. The static first constitutive parameters in the constitutive parameters cannot be directly represented by a time series, which involves a highly nonlinear calculation process in the constitutive model (this process usually reflects a complex physical mechanism). The dynamic first constitutive parameters in the constitutive parameters can be directly represented by a time series. The steady parameters in the constitutive parameters are usually material parameters that do not change with time.
[0029] The external boundary conditions may include hydraulic boundary conditions (such as a real water storage process curve, a precipitation change process curve) and temperature boundary conditions;
[0030] The first slope creep deformation time series is determined by the slope creep deformation response obtained by calculating the slope calculation grid through the slope creep constitutive model; the slope calculation grid is a mathematical model that can discretize the continuous physical space into a large number of small units by meshing the reservoir bank slope; the interaction between each small unit is calculated according to the physical law, so as to simulate the mechanical behavior of the entire reservoir bank slope;
[0031] The first intermediate implicit feature is a time dimension correlation feature generated during the nonlinear calculation process of the slope creep constitutive model;
[0032] Step S101 performs the following operations on each numerical sample:
[0033] Inputting the time dimension-independent static first constitutive parameter in the first constitutive parameter into the first learning network to obtain the predicted intermediate implicit feature;
[0034] Inputting the dynamic first constitutive parameter associated with the time dimension in the first constitutive parameter, the predicted intermediate implicit feature, and the steady parameters in the first constitutive parameter into the second learning network to obtain the predicted slope creep deformation time series;
[0035] Optimizing parameters of a combined learning network constructed by the first learning network and the second learning network based on the numerical samples, the predicted intermediate implicit features and the predicted slope creep deformation time series;
[0036] Step S102 is to train the combined learning network based on the numerical sample set and the final optimization parameters of the combined learning network, and use the trained combined learning network as the slope creep deformation proxy model.
[0037] Traditional slope creep deformation proxy models are usually static, that is, the parameters input into the slope creep deformation proxy model are static parameters that do not change with time, and the parameters output from the slope creep deformation proxy model are also static deformation increments. This slope creep deformation proxy model cannot accurately reflect the interaction between changing boundary conditions and internal physical mechanisms. The slope creep deformation proxy model constructed in the embodiment of the present application has input parameters that include dynamic external boundary conditions that cause slope creep deformation, as well as time-related dynamic parameters, and the output parameters are also time-varying slope creep deformation time series. Therefore, it can accurately reflect the time-dependent deformation of the slope, thereby improving the prediction accuracy of slope creep deformation.
[0038] In an exemplary embodiment, the method further comprises:
[0039] Before constructing a numerical sample set for training the slope creep deformation proxy model, construct a slope calculation grid and construct the slope creep constitutive model; illustratively, a slope finite difference model calculation grid can be constructed based on engineering geology, hydrogeology and topographic exploration data, and the calculation grid includes geological structures such as bedrock, weathered rock mass, sliding zone, dumping fracture zone, and sliding body, and the bottom boundary adopts three-dimensional displacement constraint, and the lateral boundary adopts normal displacement constraint;
[0040] Correspondingly, the numerical sample set constructed for training the slope creep deformation proxy model includes:
[0041] Based on the slope calculation grid and the slope creep constitutive model, a numerical sample set for training the slope creep deformation proxy model is constructed.
[0042] Exemplarily, the step of constructing the slope creep constitutive model includes:
[0043] The whole process slope creep deformation data is obtained, wherein the whole process slope creep deformation data includes slope data before and after the slope creep occurs; and the slope creep constitutive model is determined based on the whole process slope creep deformation data.
[0044] The step of obtaining the whole process slope creep deformation data includes:
[0045] In the case where the monitored slope creep deformation data is data after the slope creep occurs, the slope data before the monitoring timeline and including the slope creep before the slope creep occurs are traced back; the monitored slope creep deformation data is spliced with the traced slope data to obtain the whole process slope creep deformation data. Exemplarily, the slope creep may include: one or more of slope creep, slope rear edge tensile creep, side edge tensile creep, slope surface tensile creep and shear creep.
[0046] As an example, the slope creep changes to slow slope creep. It is assumed that the deformation monitoring data obtained through the Global Navigation Satellite System (GNSS) is the data after the slow creep has occurred. For the historical missing data before the slow creep occurred, InSAR remote sensing monitoring backtracking can be used, and a linear correction model of the InSAR observation results can be established based on the deformation monitoring data obtained by the GNSS; the corrected InSAR observation results and the deformation monitoring data obtained based on GNSS are spliced to obtain the full process slope creep deformation data based on the slope GNSS monitoring points before and after the slope creep.
[0047] Exemplarily, determining the slope creep constitutive model based on the full-process slope creep deformation data includes:
[0048] Based on environmental data (such as engineering geology, hydrogeology, and topographic exploration data) and the whole process slope creep deformation data, determine the factors that cause slope creep deformation;
[0049] Determining the creep deformation mechanism of the slope based on the factors causing the creep deformation of the slope;
[0050] Based on the slope creep deformation mechanism, an elastic-viscoplastic slope creep constitutive model considering the slope creep deformation mechanism is established.
[0051] The elastic-viscoplastic creep constitutive model combines three different mechanical behaviors, namely elasticity, viscosity and plasticity, and can more comprehensively reflect the creep characteristics of materials under different conditions.
[0052] In an exemplary embodiment, the constructing of a numerical sample set for training the slope creep deformation proxy model includes:
[0053] By sampling the original constitutive parameters of the slope creep constitutive model, N S The first constitutive parameter of the i-th group is expressed as represents the static first constitutive parameter; represents the first dynamic constitutive parameter; θ i represents the constant parameter; for example, Latin hypercube sampling can be performed on the original constitutive parameters to obtain N SThe first constitutive parameter of the group can be generated by Latin hypercube sampling to generate sample points uniformly distributed in the parameter space; the method for determining the original constitutive parameters can be divided into a theoretical analysis method and an experimental measurement method. The theoretical analysis method relies on the principles of physics and material science and existing data models, and infers the original constitutive parameters through the understanding of the deformation mechanism of the material; the experimental measurement method obtains the original constitutive parameters by performing a series of mechanical tests on the material sample, such as tensile test, compression test, torsion test, etc., and recording the stress-strain response of the material under different conditions;
[0054] For the first constitutive parameter Φ of the i-th group i , the Convert to time series j represents the number of the time series, a total of N X time series, T represents the length of the time series; based on the first constitutive parameter Φ of the i-th group i The corresponding slope creep deformation response is determined by determining the first slope creep deformation time series from the first monitoring point. and the first intermediate implicit feature from the second monitoring point Among them, u represents the serial number of the first monitoring point, a total of N Y The first monitoring point, v represents the sequence number of the second monitoring point, a total of N Z The second monitoring point; the first monitoring point usually corresponds to the on-site GNSS monitoring point, because the purpose of constructing the slope creep deformation proxy model is to achieve rapid inversion of slope parameters, and the goal of inversion is the actual monitoring deformation, so the position of the first monitoring point is usually selected from the position of the GNSS monitoring point; and the purpose of setting the second monitoring point is to give a time series representation (i.e., intermediate implicit features) that reflects the complex physical mechanism in the calculation process with strong nonlinearity in the creep constitutive model. The intermediate implicit features are used to construct physical losses. By considering the physical process in the training process of the slope creep deformation proxy model, the model accuracy is improved. Therefore, for convenience, in the numerical sample calculation process, the spatial coordinates of the second monitoring point and the first monitoring point can be made to coincide, that is, for the monitoring points on the same spatial coordinates, both the first slope creep deformation time series and the first intermediate implicit features are derived; however, this does not limit the first monitoring point and the second monitoring point to coincide in spatial coordinates, and the first monitoring point and the second monitoring point may not coincide in spatial coordinates;
[0055] Will as well as Combine to get the i-th numerical sample in, represents the external boundary condition that causes the creep deformation of the slope, where k represents the number of the external boundary condition, and there are N H external boundary conditions.
[0056] In an exemplary embodiment, the first learning network is a Transformer physical mechanism enhanced model with a position encoder, and the step of inputting the time dimension-independent static first constitutive parameter in the first constitutive parameter into the first learning network to obtain the predicted intermediate implicit feature includes:
[0057] The static first constitutive parameter is input into the constructed Transformer physical mechanism enhanced model with a position encoder to obtain the predicted intermediate implicit feature.
[0058] The Transformer physical mechanism model refers to a model with a self-attention mechanism, which enables the model to process all positions of the input data simultaneously, thereby capturing the long-range dependencies of the static first constitutive parameters. The embodiment of the present application introduces a position encoder into the Transformer physical mechanism model to enhance the functionality of the Transformer physical mechanism model. The Transformer physical mechanism enhanced model with a position encoder utilizes the self-attention mechanism and the position encoder to convert static constitutive parameters that are not associated with the time dimension into dynamic intermediate implicit features, thereby realizing the conversion of static constitutive parameters from other dimensions to the time dimension.
[0059] In an exemplary embodiment, the method for constructing the Transformer physical mechanism enhanced model with a position encoder includes:
[0060] Using a position encoder to positionally encode input data of the Transformer physical mechanism enhancement model;
[0061] Passing the position-encoded input data through multiple parallel self-attention layers;
[0062] The output data of the multiple parallel self-attention layers are concatenated through a multi-head self-attention layer;
[0063] The spliced data is passed through the feedforward neural network layer and the linear transformation layer in sequence to obtain the output data of the Transformer physical mechanism enhancement model.
[0064] Exemplarily, based on the Transformer physical mechanism enhanced model with a position encoder constructed by the method described in the above embodiment, the static first constitutive parameter is input into the Transformer physical mechanism enhanced model to obtain the predicted intermediate implicit feature, including:
[0065] The first constitutive parameter in the i-th group of numerical samples is As the input feature vector of the Transformer physical mechanism enhancement module, i = i = 1, 2, ..., N S ;
[0066] The input feature vector is converted into Assign location information;
[0067] The position-encoded input feature vector Through multiple parallel self-attention layers to calculate the correlation between different positions;
[0068] Pass the output data of multiple parallel self-attention layers through a multi-head self-attention layer to simultaneously focus on different parts of the input feature vector;
[0069] Inputting the output data of the multi-head self-attention layer into a feedforward neural network;
[0070] The data output by the feedforward neural network is input into the linear transformation layer again to obtain the predicted intermediate implicit features.
[0071] In an exemplary embodiment, the second learning network is a multi-layer stacked GRU network, and the dynamic first constitutive parameter associated with the time dimension in the first constitutive parameter, the predicted intermediate implicit feature, and the steady parameters in the first constitutive parameter are input into the second learning network to obtain the predicted slope creep deformation time series, including:
[0072] Input the predicted intermediate implicit features and the dynamic original features affecting the creep deformation of the slope determined by the dynamic first constitutive parameter into a multi-layer stacked GRU network; exemplarily, each GRU layer in the multi-layer stacked GRU network is followed by a Dropout operation, which is a regularization technique widely used in deep learning. Overfitting can be prevented by randomly setting the outputs of some neurons in the network to zero. The Dropout operation not only reduces the complex co-adaptation relationship between neurons, but also enhances the generalization ability of the network. In practical applications, especially when training data is limited, the use of the Dropout operation can significantly improve the performance of the multi-layer stacked GRU network;
[0073] Input the output data of the multi-layer stacked GRU network and the constant parameters into a fully connected layer;
[0074] The output data of the fully connected layer is used as the predicted slope creep deformation time series.
[0075] In an exemplary embodiment, the parameter optimization of the combined learning network constructed by the first learning network and the second learning network based on the numerical samples, the predicted intermediate implicit features and the predicted slope creep deformation time series includes:
[0076] Determining a prediction loss of an intermediate implicit feature based on the first intermediate implicit feature of the numerical sample and the predicted intermediate implicit feature;
[0077] Determining the predicted loss of the slope creep deformation time series based on the first slope creep deformation time series of the numerical sample and the predicted slope creep deformation time series;
[0078] Determine a combined loss by combining the predicted loss of the intermediate implicit feature and the predicted loss of the slope creep deformation time series;
[0079] Parameters of the combined learning network are optimized based on the numerical samples and the combined loss.
[0080] In an exemplary embodiment, a time-dependent weight adjustment mechanism can be introduced when determining the prediction loss of the slope creep deformation time series, so that the prediction loss of the slope creep deformation time series pays more attention to the accuracy of predicting the slope creep deformation time series. Exemplarily, the prediction loss function of the slope creep deformation time series with the time-dependent weight adjustment mechanism is It can be expressed as:
[0081]
[0082] Where T represents the length of the time series; ω(t) represents the weight function associated with time step t; u represents the serial number of the first monitoring point (such as a GNSS monitoring point), with a total of N Y The first monitoring point; is the Huber loss function, which is used to control the sensitivity to large errors. represents the predicted deformation value corresponding to the i-th numerical sample obtained by the u-th first monitoring point at time step t (i.e., the predicted slope creep deformation time series obtained by the multi-layer stacked GRU network), represents the actual deformation value corresponding to the ith numerical sample obtained through the uth first monitoring point at time step t (i.e., the first slope creep deformation time series contained in the numerical sample);
[0083] Exemplarily, the weight function ω(t) associated with the time step t may be expressed as follows:
[0084]
[0085] Among them, t is the time step, T represents the length of the time series, a and b are weight coefficients, and γ is the nonlinear rate of weight growth.
[0086] In an exemplary embodiment, the combining the predicted loss of the intermediate implicit feature and the predicted loss of the slope creep deformation time series to determine the combined loss includes:
[0087] The combined loss is determined by:
[0088]
[0089] in, represents the prediction loss function of the slope creep deformation time series corresponding to the i-th numerical sample; represents the prediction loss function of the intermediate implicit feature corresponding to the i-th numerical sample, Represents the predicted value of the intermediate implicit feature corresponding to the i-th numerical sample, Z i represents the actual value of the intermediate implicit feature corresponding to the i-th numerical sample, represents the combined loss function corresponding to the i-th numerical sample; α is a hyperparameter between 0 and 1, which is used to adjust the proportion of the above two loss functions in the combined loss function. If α is closer to 1, the combined loss function will pay more attention to the accuracy of slope creep deformation time series prediction; if α is closer to 0, the combined loss function will pay more attention to the accuracy of intermediate implicit feature prediction.
[0090] In an exemplary embodiment, the performing parameter optimization on the combined learning network based on the numerical sample and the combined loss includes:
[0091] Based on the numerical samples and the combined loss, a Bayesian optimizer is used to optimize the parameters of the combined learning network. The Bayesian optimizer can model the dependencies between parameters through Gaussian processes, which makes it suitable for combined learning networks with large computational complexity, such as a combined learning network composed of a Transformer physical mechanism enhancement model and a multi-layer stacked GRU network.
[0092] Exemplarily, when the combined learning network includes a Transformer physical mechanism enhancement model and a multi-layer stacked GRU network, the parameters to be optimized may include: the feature dimension d_model of the Transformer physical mechanism enhancement module, the number of heads nhead of the multi-head self-attention, and the number of encoders num_encoder_layers; the number of layers num_GRU of the multi-layer stacked GRU network, the number of hidden units in each layer hidden_sizes, and the Dropout ratio dropout_rates after each layer; the weight coefficients a and b of the weight function ω(t), the nonlinear rate γ of the weight growth, the threshold δ in the Huber loss function, and the combined loss function The hyperparameter α of the training model; the batch size batch_size of the training model, and the number of iterations num_epochs.
[0093] It can be seen from the technical solutions of the above embodiments that the embodiments of the present application have the following beneficial effects:
[0094] (1) Through the Transformer physical mechanism enhancement module, the static constitutive parameters associated with non-time dimensions can be effectively converted into intermediate implicit features related to time dynamics, thereby capturing the complex dynamic changes driven by the physical mechanism, and improving the response sensitivity and prediction accuracy of the slope creep deformation proxy model constructed in the embodiment of the present application to the actual physical process; this can not only enhance the understanding of the input data structure of the slope creep deformation proxy model constructed in the embodiment of the present application, but also improve the prediction results of the slope creep deformation time series;
[0095] (2) Through the multi-layer stacked GRU network, multiple types of input data can be processed simultaneously. This design can not only adapt to time-independent constant parameters, but also process time-varying parameters. In addition, the complex intermediate implicit feature time series extracted by the Transformer physical mechanism enhancement module can be fused. This fusion provides a comprehensive data processing framework, allowing the slope creep deformation proxy model constructed in the embodiment of the present application to capture more detailed time dynamic changes and their relationship with slope creep deformation;
[0096] (3) By constructing a combined loss function, the slope creep deformation proxy model constructed in the embodiment of the present application not only pays attention to the accuracy of creep deformation prediction, but also emphasizes the accurate extraction of intermediate implicit features, ensuring that the slope creep deformation proxy model can maintain sensitivity to physical processes while learning data, thereby improving the reliability and generalization ability of the prediction; in addition, the introduction of the time-dependent weight adjustment mechanism can enable the slope creep deformation proxy model to focus on the accuracy of the prediction of the key deformation stage;
[0097] (4) The slope creep deformation proxy model constructed in the embodiment of the present application can be used to quickly invert slope parameters, provide engineers and decision makers with instant slope stability risk assessment, reduce potential safety hazards, and has high application value.
[0098] The following is a detailed description of the solution recorded in the above embodiment, taking a certain reservoir bank slope in my country as the engineering background:
[0099] like Figure 2 As shown in Figure 1, the steps for building the slope creep deformation proxy model include:
[0100] Step S1, constructing a reservoir bank slope calculation grid based on engineering geology, hydrogeology and topographic exploration data; analyzing and determining the reservoir bank slope deformation mechanism in combination with on-site slope creep deformation monitoring data, and establishing a slope creep constitutive model that takes into account the reservoir bank slope deformation mechanism;
[0101] Step S2, based on the calculation grid of the reservoir bank slope and the slope creep constitutive model obtained in step S1, construct a set of numerical samples, each numerical sample in the set includes a first constitutive parameter, an external boundary condition, a first intermediate implicit feature and a first slope creep deformation time series; the first constitutive parameter includes: a static first constitutive parameter that is not associated with the time dimension, a dynamic first constitutive parameter that is associated with the time dimension and a steady-state parameter;
[0102] Step S3, constructing a Transformer physical mechanism enhancement module, using a self-attention mechanism and a position encoder to convert the static first constitutive parameter that is not associated with the time dimension into a dynamic predictive intermediate implicit feature;
[0103] Step S4, building a multi-layer stacked GRU (Gate Recurrent Unit) network, each GRU layer is followed by a Dropout operation, and a fully connected layer is externally connected after the last GRU layer; the time series features (such as water-induced parameter degradation curve) (i.e., the original features in the figure) obtained by substituting the dynamic first constitutive parameter associated with the time dimension into the mathematical expression and the predicted intermediate implicit features obtained based on step S3 are input as time series into the multi-layer stacked GRU network; the external fully connected layer is used as the input layer of the constant parameter; the output of the multi-layer stacked GRU network is the predicted slope creep deformation time series;
[0104] Step S5, constructing a combined loss function that considers the prediction loss of the prediction intermediate implicit features of step S3 and the prediction loss of the predicted slope creep deformation time series of step S4, and introducing a time-dependent weight adjustment mechanism in the prediction loss calculation process of the predicted slope creep deformation time series of step S4;
[0105] Step S6, based on the numerical sample set obtained in step S2 and the combined loss function constructed in step S5, a Bayesian optimizer is used to optimize the hyperparameters of the Transformer-multi-layer stacked GRU network jointly composed of step S3 and step S4;
[0106] Step S7, based on the numerical sample set obtained in step S2 and the optimal hyperparameters obtained in step S6, training the Transformer-multi-layer stacked GRU network jointly formed in step S3 and step S4 to obtain a slope creep deformation proxy model;
[0107] The input characteristics of the slope creep deformation proxy model are constitutive parameters and external boundary conditions, and the output characteristics of the slope creep deformation proxy model are slope creep deformation time series.
[0108] The following is an explanation of steps S1-S7:
[0109] Step S1 includes:
[0110] Step S1-1:
[0111] Based on engineering geology, hydrogeology and topographic exploration data, a computational grid of the finite difference model of the reservoir bank slope is constructed. The computational grid includes geological structures such as bedrock, weathered rock mass, sliding zone, dumping fracture zone, and sliding body. The bottom boundary adopts three-dimensional displacement constraints, and the lateral boundary adopts normal displacement constraints. The computational grid is as follows: Figure 3 As shown;
[0112] Step S1-2:
[0113] The bank slope of the reservoir began to creep slowly due to the water storage of the hydropower station at the end of May 2019. The on-site GNSS emergency monitoring system was completed and put into use on July 7, 2019;
[0114] For the missing slope deformation data before July 7, 2019, InSAR remote sensing monitoring was used for backtracking, and a linear correction model of InSAR observation results was established based on the deformation data monitored by on-site GNSS;
[0115] The corrected InSAR observation results (from December 10, 2018 to July 6, 2019) and the deformation data monitored by GNSS (from July 7, 2019 to September 15, 2019) were spliced to obtain the creep deformation curve of the GNSS monitoring points on the reservoir bank slope before and after the slope sliding (from December 10, 2018 to September 15, 2019);
[0116] The creep deformation curve of the GNSS monitoring point before and after slope sliding is as follows: Figure 4As shown in the figure, G01-G21 represent 21 GNSS monitoring points. Affected by various factors such as slope geometry, geological structure, and setting parameters of monitoring points, the deformation data obtained from different monitoring points are different.
[0117] Step S1-3:
[0118] Based on the engineering geology, hydrogeology, and topographic exploration data in step S1-1 and the creep deformation curve of the GNSS monitoring point before and after slope sliding obtained in step S1-2, the deformation mechanism of the reservoir bank slope can be analyzed and determined:
[0119] After many historical evolutions, the slope is in a critical stability state. After the reservoir is filled with water, the water level rises, which reduces the mechanical properties of the deposits at the front edge of the slope, breaking the critical stability of the slope in its natural state. Under the action of gravity, the slope body causes the existing slip surface to undergo overall deformation adjustment, resulting in cracks at the rear edge and side edge of the slope body, and the formation of multiple tension cracks and shear cracks on the slope surface, causing the slope to become unstable. After experiencing continuous large deformation, the slope gradually stabilizes without any external force constraints. From a microscopic perspective, this can be explained as the sliding belt being destroyed and the friction strength being reduced after the slope triggers sliding. In the process of continuous sliding and self-adjustment, the sliding belt is gradually squeezed and compacted, and the friction strength increases, even exceeding the initial friction strength during sliding, thus leading to the self-stability of the slope.
[0120] Step S1-4:
[0121] Based on the deformation mechanism of the reservoir bank slope determined in step S1-3, an elastic-viscoplastic creep constitutive model considering the deformation mechanism of the reservoir bank slope is established under the framework of internal variable thermodynamics;
[0122] The complexity of the elastic-viscoplastic creep constitutive model is as follows:
[0123] On the one hand, the evolution equation f(ε) of the friction coefficient f() of the elastic-viscoplastic creep constitutive model is controlled by static parameters velo_a, velo_b, velo_c, velo_d, and velo_e and expressed in sections, as shown in the following formula:
[0124]
[0125] It can be seen that f(ε) is related to the shear strain coefficient ε, rather than to the time t, and the evolution equation f(ε) needs to be converted to f(t), which involves complex dimensional conversion in the creep calculation process;
[0126] On the other hand, the elastic-viscoplastic creep constitutive model of the reservoir bank slope is generally not applicable to the Mohr-Coulomb (abbreviated as MC) strength criterion, but the Drucker-Prager (abbreviated as DP) strength criterion that considers the plastic flow and shear failure characteristics of rocks under three-dimensional stress state. Since the friction coefficient f belongs to the MC strength criterion parameter, it is necessary to convert f(t) into the parameters a(t) and R(t) applicable to the DP strength criterion. The a(t) and R(t) are intermediate variables of the elastic-viscoplastic creep constitutive model, involving a strong nonlinear calculation process, and are intermediate implicit features.
[0127] Step S2 includes:
[0128] Step S2-1:
[0129] Based on the elastic-viscoplastic creep constitutive model considering the creep deformation mechanism established in step S1-4, Latin hypercube sampling is performed on the original constitutive parameters of the model within a reasonable value range to obtain N S =10000 sets of the first constitutive parameters;
[0130] Each set of first constitutive parameters can be expressed as Where i represents the parameter group number, there are N S Group parameters;
[0131] represents a static parameter array that implies a complex physical mechanism in each set of the first constitutive parameters and cannot be directly represented by a time series, such as the control parameters [velo_a, velo_b, velo_c, velo_d, velo_e] of the evolution equation f(ε);
[0132] represents the array of equation parameters in each set of the first constitutive parameters that can be directly represented by a time series, such as the control parameters of the water-induced degradation effect equation [A f ,B f ,D c ,A E ,B E ];
[0133] θ i represents the constant parameters in each group of the first constitutive parameters that are independent of the time dimension, such as the flow coefficient k2vis;
[0134] Step S2-2:
[0135] Each set of first constitutive parameters generated in step S2-1 Integrate into the command flow of the finite difference calculation software, where the equation parameters need to be Convert the equation expression into a time series format in advance. For example, For [A f ,B f ,D c ,A E ,B E ]hour, The conversion expression is as follows:
[0136]
[0137]
[0138]
[0139] in, are the water-induced degradation effect equations for the friction coefficient, cohesion, and elastic modulus, respectively; t represents the cumulative number of days since water storage (December 10, 2018); A f ,B f D is the control parameter that affects the friction coefficient and cohesion water-induced degradation effect equation. c A is the control parameter affecting the equation of water-induced degradation effect of cohesion. E ,B E is the control parameter that affects the equation of water-induced degradation effect of elastic modulus;
[0140] Based on the above conversion expression, we can get Where j represents the number of the time series, a total of N X = 3 time series, T = 280 represents the length of the time series (from December 10, 2018 to September 15, 2019, a total of 280 days);
[0141] The command flow also needs to consider external boundary conditions Where k represents the number of the external boundary condition, a total of N H external boundary conditions, T represents the length of the time series; in this example, only the external boundary condition of water level change is considered, so N H =1, T=280 (from December 10, 2018 to September 15, 2019, a total of 280 days);
[0142] Step S2-3:
[0143] Calculate each group of first constitutive parameters based on the command stream constructed in step S2-2 The corresponding creep deformation response of the reservoir bank slope;
[0144] For each set of the first constitutive parameters The corresponding creep deformation response of the reservoir bank slope is:
[0145] Recording creep deformation time series of GNSS monitoring points Where u represents the serial number of the GNSS monitoring point, a total of N Y = 21 GNSS monitoring points, T = 280 represents the length of the time series (from December 10, 2018 to September 15, 2019, a total of 280 days);
[0146] Record and Intermediate implicit characteristic time series that are relevant and can reflect complex physical mechanisms In this example represents the parameter a(t) satisfying the DP intensity criterion described in step S1-4 for the 12 monitoring points, The parameter R(t) that satisfies the DP intensity criterion described in step S1-4 of the 24 monitoring points, where v represents the serial number of the monitoring point, a total of N Z = 24 monitoring points, T = 280 represents the length of the time series (from December 10, 2018 to September 15, 2019, a total of 280 days);
[0147] Step S2-4:
[0148] Each group of first constitutive parameters obtained in step S2-1 and the one obtained in step S2-2 And the step S2-3 obtained The numerical samples including the first constitutive parameters, boundary conditions, intermediate implicit features and creep deformation time series are obtained by combining
[0149] Step S3 includes:
[0150] Step S3-1:
[0151] Step S2-1 As the input feature vector of the Transformer physical mechanism enhancement module;
[0152] The position encoder is used to convert the above input feature vector Assign location information;
[0153] For example, the position encoder is calculated using the following formula:
[0154] PE(pos,2n)=sin(pos / 10000 2n / dmodel )
[0155] PE(pos,2n+1)=cos(pos / 10000 2n / dmodel )
[0156] Among them, pos represents the input feature vector The position of an element in the feature vector, n represents the dimension index, d model is the dimension of the embedding vector;
[0157] Step S3-2:
[0158] The input feature vector after position encoding in step S3-1 Compute the association between different positions through multiple parallel self-attention layers;
[0159] For example, the calculation of each self-attention layer can be expressed as:
[0160]
[0161] Among them, Q m , K m , V m are the query, key, and value matrices of the mth self-attention layer, d k is the dimension of the key, represents the transposed matrix of the key matrix, softmax() represents the softmax function, V m The input feature vector after position encoding Obtained by linear transformation;
[0162] Step S3-3:
[0163] Perform a splicing operation on the outputs of multiple parallel self-attention layers in step S3-2 to obtain a multi-head self-attention layer to achieve simultaneous attention to information in different parts of the sequence;
[0164] For example, the multi-head self-attention layer can be expressed as:
[0165] MultiHead(Q,K,V)=Concat(head1,…,head m ,…,head h )W O
[0166] Where Q, K, and V are the complete query, key, and value matrices respectively, and Concat() is a concatenation operation that merges all the head outputs. m represents the output of the mth self-attention layer, there are h self-attention layers in total; W O is the weight matrix of the output linear transformation;
[0167] Step S3-4:
[0168] After the multi-head self-attention layer in step S3-3, a feedforward neural network (FNN) is constructed, which can be expressed as:
[0169] FNN(x)=max(0,xW1+b1)W2+b2
[0170] Where x represents the activation output from the self-attention mechanism or the previous encoder layer; max(0,z) represents the ReLU activation function, which applies nonlinear activation to each element z; W1 and W2 are weight matrices in the feedforward neural network; b1 and b2 are bias terms in the feedforward neural network;
[0171] Step S3-5:
[0172] The data processed by the feedforward neural network in step S3-4 is passed through the last linear transformation layer to map the output of the encoder to the intermediate implicit feature time series. the expected dimensions of
[0173] For example, the last linear transformation layer can be expressed as:
[0174]
[0175] in, is the prediction sequence of the intermediate implicit characteristic time series that can reflect the complex physical mechanism described in step S2-3; encoder is the output of the Transformer physical mechanism enhancement module, that is, the input feature vector after self-attention mechanism and position encoding processing High-level abstract representation of W latent It is the output from the Transformer physical mechanism enhancement module to the intermediate implicit feature time series The mapping weight of the dimension; b latent is the bias term in the mapping process.
[0176] The schematic diagram of the Transformer physical mechanism enhancement module in this example is as follows Figure 5 shown.
[0177] Step S4 includes:
[0178] Step S4-1:
[0179] Create a multi-layer stacked GRU (Gate Recurrent Unit) network and apply Dropout operation after each GRU layer to prevent overfitting;
[0180] For example, each GRU unit uses the following formula to update its state:
[0181] R t =σ(W r ·[H t-1 ,X t ]+U r ·X t )
[0182] Z t =σ(W z ·[H t-1 ,X t ]+U z ·X t )
[0183]
[0184]
[0185] Among them, R t , Z t Represent the reset gate and update gate of the current time step t respectively, H t They represent the candidate hidden state and final hidden state at the current time step t respectively; σ represents the sigmoid function, which is responsible for activating the reset gate and the update gate; tanh is the hyperbolic tangent function, which is used to activate the candidate hidden state; ⊙ represents the Hadamard product; X t represents the input feature vector of the current time step t; H t-1 represents the final hidden state of the previous time step t-1; W r , W z , W h They represent the reset gate, update gate, and candidate hidden state corresponding to H at the current time step t. t-1 The weight matrix of r , U z , U h Respectively represent the reset gate, update gate, and candidate hidden state corresponding to X at the current time step t t The weight matrix of
[0186] Step S4-2:
[0187] The original features (in step S2-2) ) and predict the intermediate implicit features (obtained in step S3-5 ) performs feature merging as the input of the multi-layer stacked GRU network in step S4-1;
[0188] Step S4-3:
[0189] A fully connected layer is added at the end of the multi-layer stacked GRU network constructed in step S4-1 to fuse the constant parameter θ that is independent of the time dimension described in the input step S2-1. i , and the final output of the multi-layer stacked GRU network for dimensional transformation is the creep deformation time series of the GNSS monitoring points described in step S2-2 The prediction sequence of
[0190] The schematic diagram of the multi-layer stacked GRU network in this example is as follows Figure 6 shown.
[0191] Step S5 includes:
[0192] Step S5-1:
[0193] Construct the output of the linear transformation layer described in step S3-5 (predicting the intermediate implicit feature time series )
[0194] The loss function of the output of the linear transformation layer is It can be expressed as:
[0195]
[0196] Among them, T = 280 represents the length of the time series (from December 10, 2018 to September 15, 2019, a total of 280 days); v represents the serial number of the intermediate implicit feature monitoring point, a total of N Z = 24 intermediate implicit feature monitoring points; It represents the feature prediction value corresponding to the i-th numerical sample obtained at the v-th intermediate implicit feature monitoring point at time step t, It represents the actual value of the feature corresponding to the i-th numerical sample obtained at the v-th intermediate implicit feature monitoring point at time step t;
[0197] Step S5-2:
[0198] The final output of the multi-layer stacked GRU network described in step S4-3 (GNSS monitoring point creep deformation time series) At the same time, a time-dependent weight adjustment mechanism is introduced to make the model pay more attention to the accuracy of the prediction sequence at a specific stage;
[0199] The loss function that introduces the time-dependent weight adjustment mechanism It can be expressed as:
[0200]
[0201] Wherein, T = 280 represents the length of the time series (from December 10, 2018 to September 15, 2019, a total of 280 days); ω(t) represents the weight function associated with the time step t, which is used to guide the slope creep deformation proxy model to focus on the accuracy of the prediction of the key deformation stage; u represents the serial number of the GNSS monitoring point, a total of N Y =21 GNSS monitoring points; is the Huber loss function, which is used to control the sensitivity to large errors. represents the deformation prediction value corresponding to the i-th numerical sample obtained at the u-th GNSS monitoring point at time step t, represents the actual value of deformation corresponding to the i-th numerical sample obtained at the u-th GNSS monitoring point at time step t;
[0202] The weight function ω(t) associated with the time step t is shown as follows:
[0203]
[0204] Among them, t is the time step, T represents the length of the time series, a and b are weight coefficients, and γ is the nonlinear rate of weight growth;
[0205] The Huber loss function As shown below:
[0206]
[0207] Where δ is the threshold in the Huber loss function;
[0208] Step S5-3:
[0209] Using step S5-1 and step S5-2 Constructing the combined loss function To balance the two losses of the intermediate implicit features of the slope creep deformation proxy model and the creep deformation of the GNSS monitoring points, the expression is as follows:
[0210]
[0211] Among them, α is a hyperparameter between 0 and 1, which is used to adjust the proportion of the above two losses in the total loss. If α is closer to 1, the constructed slope creep deformation proxy model will pay more attention to the accuracy of the creep deformation time series prediction of the GNSS monitoring points; if α is closer to 0, the constructed slope creep deformation proxy model will pay more attention to the accuracy of the intermediate implicit feature time series prediction.
[0212] Step S6 includes:
[0213] Based on the combined loss function constructed in step S5-3 and the numerical sample set obtained in step S2-4 The Bayesian optimizer is used to optimize the hyperparameters of the slope creep deformation proxy model (Transformer-multi-layer stacked GRU network) composed of step S3 and step S4;
[0214] The final optimization results of the hyperparameters in this example are as follows:
[0215] The feature dimension of the Transformer physical mechanism enhancement module is d_model = 128, the number of heads of multi-head self-attention is nhead = 16, and the number of encoders is num_encoder_layers = 6;
[0216] The number of layers of the multi-layer stacked GRU network is num_GRU=4, the number of hidden units in each layer is hidden_sizes=[128,128,128,128], and the Dropout ratio after each layer is dropout_rates=[0.3,0.3,0.2,0.2];
[0217] The weight coefficients of the weight function ω(t) are a=1, b=3, the nonlinear rate of weight growth γ=3, the threshold δ in the Huber loss function is 0.4, and the combined loss function The hyperparameter α=0.6;
[0218] The batch size of the training model is batch_size=96, and the number of iterations is num_epochs=20.
[0219] Step S7 includes:
[0220] Based on the combined loss function constructed in step S5-3 and the numerical sample set obtained in step S2-4 The proxy model (Transformer-multi-layer stacked GRU network) composed of steps S3 and S4 is trained in combination with the optimal hyperparameters obtained in step S6 to obtain a slope creep deformation proxy model between the slope constitutive parameters, boundary conditions and creep deformation time series.
[0221] The comparison between the predicted value and the actual value of the creep deformation time series of typical GNSS monitoring points by the slope creep deformation proxy model in this example is as follows: Figure 7 As shown in the figure, it can be seen that the predicted value of the creep deformation time series is basically consistent with the actual value.
[0222] The embodiment of the present application also provides a non-transitory computer-readable storage medium storing a computer program, which can be executed by a processor to implement the method for constructing a slope creep deformation proxy model as described in any of the previous embodiments.
[0223] The present application also provides a device for constructing a slope creep deformation proxy model. Figure 8 As shown, it includes a memory 801 and a processor 802, wherein the memory 801 stores a computer program, and when the computer program is read and executed by the processor 802, it can implement the method for constructing a slope creep deformation proxy model as described in any of the previous embodiments.
[0224] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.
Claims
1. A method for constructing a slope creep deformation proxy model, the method comprising: A set of numerical samples for training the slope creep deformation proxy model is constructed, each numerical sample in the set includes a first constitutive parameter, an external boundary condition causing the slope creep deformation, a first intermediate implicit feature and a first slope creep deformation time series; the first constitutive parameter is determined by the original constitutive parameter of the creep constitutive model reflecting the creep deformation mechanism of the slope, including: a static first constitutive parameter that is not associated with the time dimension, a dynamic first constitutive parameter that is associated with the time dimension and a steady-state parameter; the first slope creep deformation time series is determined by the slope creep deformation response obtained by calculating the slope calculation grid through the slope creep constitutive model; the first intermediate implicit feature is a time dimension associated feature generated in the nonlinear calculation process of the slope creep constitutive model; The following operations are performed on each numerical sample: the static first constitutive parameter that is not associated with the time dimension in the first constitutive parameter is input into the first learning network to obtain the predicted intermediate implicit feature; the dynamic first constitutive parameter that is associated with the time dimension in the first constitutive parameter, the predicted intermediate implicit feature, and the steady-state parameter in the first constitutive parameter are input into the second learning network to obtain the predicted slope creep deformation time series; based on the numerical sample, the predicted intermediate implicit feature and the predicted slope creep deformation time series, the combined learning network constructed by the first learning network and the second learning network is optimized in parameters; The combined learning network is trained based on the numerical sample set and the final optimized parameters of the combined learning network, and the trained combined learning network is used as the slope creep deformation proxy model.
2. The construction method according to claim 1, characterized in that: The method further comprises: Before constructing a numerical sample set for training the slope creep deformation proxy model, constructing a slope calculation grid and constructing the slope creep constitutive model; Correspondingly, the numerical sample set constructed for training the slope creep deformation proxy model includes: Based on the slope calculation grid and the slope creep constitutive model, construct a numerical sample set for training the slope creep deformation proxy model; The step of constructing the slope creep constitutive model comprises: Acquiring full-process slope creep deformation data, wherein the full-process slope creep deformation data includes slope data before and after slope creep occurs; The slope creep constitutive model is determined based on the full-process slope creep deformation data.
3. The construction method according to claim 2, characterized in that: The whole process of obtaining slope creep deformation data includes: In the case where the monitored slope creep deformation data is data after the slope creep occurs, backtracking to the slope data before the monitoring timeline and including the slope data before the slope creep occurs; The monitored slope creep deformation data is spliced with the backtracked slope data to obtain the whole process slope creep deformation data.
4. The construction method according to claim 1, characterized in that: The constructing of a numerical sample set for training the slope creep deformation proxy model includes: By sampling the original constitutive parameters of the slope creep constitutive model, N S The first constitutive parameter of the i-th group is expressed as represents the static first constitutive parameter; represents the first dynamic constitutive parameter; θ i represents a constant parameter; For the first constitutive parameter Φ of the i-th group i , the Convert to time series j represents the number of the time series, a total of N X time series, T represents the length of the time series; based on the first constitutive parameter Φ of the i-th group i The corresponding slope creep deformation response is used to determine the first slope creep deformation time series from the first monitoring point. and the first intermediate latent feature from the second monitoring point Among them, u represents the serial number of the first monitoring point, a total of N Y The first monitoring point, v represents the sequence number of the second monitoring point, a total of N Z A second monitoring point; Will as well as Combine to get the i-th numerical sample in, represents the external boundary condition that causes the creep deformation of the slope, where k represents the number of the external boundary condition, and there are N H external boundary conditions.
5. The construction method according to claim 1, characterized in that: The step of inputting the time dimension-independent static first constitutive parameter in the first constitutive parameter into the first learning network to obtain the predicted intermediate implicit feature includes: The static first constitutive parameter is input into the constructed Transformer physical mechanism enhanced model with a position encoder to obtain the predicted intermediate implicit feature.
6. The construction method according to claim 5, characterized in that: The method for constructing the Transformer physical mechanism enhanced model with a position encoder comprises: Using a position encoder to positionally encode input data of the Transformer physical mechanism enhancement model; Passing the position-encoded input data through multiple parallel self-attention layers; The output data of the multiple parallel self-attention layers are concatenated through a multi-head self-attention layer; The spliced data is passed through the feedforward neural network layer and the linear transformation layer in sequence to obtain the output data of the Transformer physical mechanism enhancement model.
7. The construction method according to claim 1, characterized in that: The step of inputting the dynamic first constitutive parameter associated with the time dimension in the first constitutive parameter, the predicted intermediate implicit feature, and the steady parameters in the first constitutive parameter into the second learning network to obtain the predicted slope creep deformation time series includes: Inputting the predicted intermediate implicit features and the dynamic original features affecting the creep deformation of the slope determined by the dynamic first constitutive parameter into a multi-layer stacked GRU network; Input the output data of the multi-layer stacked GRU network and the constant parameters into a fully connected layer; The output data of the fully connected layer is used as the predicted slope creep deformation time series.
8. The construction method according to claim 7, characterized in that: Each GRU layer in the multi-layer stacked GRU network is followed by a Dropout operation.
9. The construction method according to claim 1, characterized in that: The parameter optimization of the combined learning network constructed by the first learning network and the second learning network based on the numerical sample, the predicted intermediate implicit features and the predicted slope creep deformation time series includes: Determining a prediction loss of an intermediate implicit feature based on the first intermediate implicit feature of the numerical sample and the predicted intermediate implicit feature; Determining the predicted loss of the slope creep deformation time series based on the first slope creep deformation time series of the numerical sample and the predicted slope creep deformation time series; Determine a combined loss by combining the predicted loss of the intermediate implicit feature and the predicted loss of the slope creep deformation time series; Parameters of the combined learning network are optimized based on the numerical samples and the combined loss.
10. The construction method according to claim 9, characterized in that: The prediction loss of the slope creep deformation time series is determined based on the first slope creep deformation time series of the numerical samples and the predicted slope creep deformation time series, which is achieved by the following formula: Where T represents the length of the time series; ω(t) represents the weight function associated with time step t; u represents the sequence number of the first monitoring point, a total of N Y The first monitoring point; is the Huber loss function; represents the predicted slope creep deformation time series corresponding to the ith numerical sample obtained at the uth first monitoring point at time step t, represents the first slope creep deformation time series corresponding to the ith numerical sample obtained at the uth first monitoring point at time step t; Represents the prediction loss function of the slope creep deformation time series corresponding to the i-th numerical sample.
11. The construction method according to claim 10, characterized in that: The weight function ω(t) associated with the time step t is implemented by the following formula: Among them, a and b are weight coefficients, and γ is the nonlinear rate of weight growth.
12. The construction method according to claim 9, characterized in that: The combined loss is determined by combining the predicted loss of the intermediate implicit feature and the predicted loss of the slope creep deformation time series, which is implemented by the following formula: in, represents the prediction loss of the slope creep deformation time series corresponding to the i-th numerical sample, represents the predicted slope creep deformation time series corresponding to the i-th numerical sample, Y i represents the first slope creep deformation time series corresponding to the i-th numerical sample, represents the prediction loss of the intermediate implicit feature corresponding to the i-th numerical sample, Represents the predicted intermediate implicit feature corresponding to the i-th numerical sample, Z i represents the first intermediate implicit feature corresponding to the i-th numerical sample, represents the combined loss function corresponding to the i-th numerical sample, α represents the hyperparameter, 0≤α≤1.
13. A non-transitory computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor to implement the method for constructing a slope creep deformation proxy model as claimed in any one of claims 1 to 12.
14. A device for constructing a slope creep deformation proxy model, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is read and executed by the processor, the method for constructing a slope creep deformation proxy model as described in any one of claims 1 to 12 can be implemented.
Citation Information
Patent Citations
Slope deformation prediction method
CN111144636A
Method for dynamically evaluating slope safety
CN114330168A
Bank slope deformation early warning method based on Bayesian updating and water-induced degradation model
CN116861762A
Concrete dam deformation prediction method based on IPSO-Transform model
CN117540265A
Cited By
Slope multi-physics field fusion early warning decision-making system based on digital twinning
CN120995574A
Slope multi-physical field fusion early warning decision system based on digital twinning
CN120995574B
Slope displacement prediction method and system based on physical information neural network and gating circulation unit
CN121072360A
