Geotechnical infrastructure risk early warning method and system based on multi-scale damage model

Abstract

This application relates to a method and system for risk early warning of geotechnical infrastructure based on a multi-scale damage model. The method includes: acquiring environmental monitoring data of a target geotechnical area based on a preset monitoring scale; transforming the environmental monitoring data into a multi-scale damage model through multi-scale coupling analysis modeling; extracting spatiotemporal features of damage from the multi-scale damage model using a pre-trained convolutional neural network, and determining risk patterns by matching with a preset risk pattern library; inputting the spatiotemporal features of damage and risk patterns into a pre-trained risk diffusion probability prediction model to obtain risk propagation prediction results; constructing a multi-dimensional risk decision boundary using a nonlinear classifier based on the risk patterns and risk propagation prediction results; and generating graded early warning instructions through preset dynamic risk decision rules. This method can achieve accurate and dynamic early warning of risks to geotechnical infrastructure, preventing safety accidents caused by multi-scale damage evolution.

Description

Technical Field

[0001] This invention belongs to the field of geotechnical infrastructure safety management, and in particular relates to a method and system for risk early warning of geotechnical infrastructure based on a multi-scale damage model. Background Technology

[0002] With the development of technologies in the field of geotechnical engineering and infrastructure safety management, damage monitoring and risk early warning technologies for geotechnical infrastructure have emerged. These technologies can achieve preliminary risk identification and early warning by monitoring the state parameters of the soil and rock mass, becoming an important means to ensure the safe operation of geotechnical infrastructure. In traditional technologies, damage judgment and risk warning are mostly made by acquiring macroscopic monitoring data of the soil and rock mass and combining it with empirical formulas or simple models. However, the state of soil and rock particles and samples at the microscopic and mesoscopic scales is not monitored and analyzed, and the risk level is judged only by manually set fixed thresholds.

[0003] Current early warning methods have the following limitations: single-scale data cannot fully reflect the multi-scale damage evolution of soil and rock masses; simple analysis models cannot capture the spatiotemporal characteristics of damage and the risk propagation pattern; and fixed threshold judgment methods lack dynamic adaptability, resulting in inaccurate risk identification, delayed early warning, or misjudgment, making it difficult to meet the high-precision and intelligent requirements of risk early warning for soil and rock infrastructure. Summary of the Invention

[0004] Therefore, it is necessary to provide a risk early warning method and system for geotechnical infrastructure based on a multi-scale damage model that can solve the above problems.

[0005] Firstly, this application provides a risk early warning method for geotechnical infrastructure based on a multi-scale damage model, including:

[0006] Based on a preset monitoring scale, environmental monitoring data of the target soil and rock area is obtained;

[0007] Based on environmental monitoring data, a multi-scale coupled analysis model is performed to obtain a multi-scale damage model.

[0008] The multi-scale damage model is input into a pre-trained convolutional neural network to extract spatiotemporal damage features. Based on these features, and combined with a pre-defined risk pattern library, risk patterns are matched and determined.

[0009] The spatiotemporal characteristics of damage and risk patterns are input into a pre-trained risk diffusion probability prediction model to make predictions and obtain risk propagation prediction results.

[0010] Based on risk pattern and risk propagation prediction results, a nonlinear classifier is used to construct a multidimensional risk decision boundary;

[0011] Based on multidimensional risk decision boundaries, hierarchical early warning instructions are generated using preset dynamic risk decision rules.

[0012] In one embodiment, the environmental monitoring data includes particle-scale monitoring data, sample-scale monitoring data, and macro-scale monitoring data. The particle-scale monitoring data includes particle geometric parameters and particle state parameters. The sample-scale monitoring data includes pore distribution parameters and water content parameters. The macro-scale monitoring data includes topographic parameters and geological structure parameters.

[0013] Based on environmental monitoring data, multi-scale coupled analysis and modeling are performed to obtain a multi-scale damage model, including:

[0014] Based on particle geometric parameters and particle state parameters, particle mechanical parameters are inverted using mechanical theory. Combining particle geometric parameters, particle state parameters, and particle mechanical parameters, a micro-contact network of soil and rock particles is constructed using the discrete element method.

[0015] Based on pore distribution parameters and water content parameters, the material constitutive matrix is ​​generated by meshing using the finite element method.

[0016] By performing geometric transformation and alignment of the micro-contact network of soil particles and the material constitutive matrix, a cross-scale correlation parameter matrix is ​​obtained.

[0017] Based on topographic and geological parameters, a three-dimensional model is created to obtain a macroscopic geotechnical model.

[0018] Based on the macroscopic geotechnical geometric model and the cross-scale correlation parameter matrix, combined with the preset damage index system, multi-physics field coupling is carried out to obtain a multi-scale damage model.

[0019] In one embodiment, a multi-scale damage model is input into a pre-trained convolutional neural network to extract spatiotemporal damage features. Based on these features and a pre-defined risk pattern library, risk patterns are matched and determined, including:

[0020] A multi-frequency spatiotemporal feature matrix is ​​obtained by mapping and transforming the model parameters of the multi-scale damage model. This multi-frequency spatiotemporal feature matrix includes a high-frequency spatiotemporal feature matrix, a mid-frequency spatiotemporal feature matrix, and a low-frequency spatiotemporal feature matrix. The high-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the particle-scale model parameters in the multi-scale damage model. The mid-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the sample-scale model parameters in the multi-scale damage model. The low-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the macroscopic-scale model parameters in the multi-scale damage model.

[0021] The multi-band spatiotemporal feature matrix is ​​input into a pre-trained convolutional neural network for feature extraction, resulting in multi-scale spatiotemporal features as damage spatiotemporal features. The feature extraction operation is performed using four sets of pre-set dilation rate dilated convolutional kernels. The first set of dilated convolutional kernels is used to extract local deformation features of model parameters at the particle scale, the second set of dilated convolutional kernels is used to extract mechanical coupling features of model parameters at the sample scale, the third set of dilated convolutional kernels is used to extract geological structural features of model parameters at the macroscopic scale, and the fourth set of dilated convolutional kernels is used to integrate local deformation features, mechanical coupling features, and geological structural features to obtain multi-scale spatiotemporal features.

[0022] Based on the spatiotemporal characteristics of damage, a feature channel allocation mechanism based on the damage index system is adopted to generate a spatiotemporal feature map of damage.

[0023] Calculate the feature similarity between the spatiotemporal feature map of damage and the baseline features of each risk pattern in the preset risk pattern library;

[0024] Define feature matching rules based on feature similarity, and determine risk patterns based on feature similarity and feature matching rules.

[0025] In one embodiment, the formula for calculating the feature similarity between the damage spatiotemporal feature map and the baseline features of each risk pattern in the preset risk pattern library is as follows:

[0026]

[0027] in, For feature similarity, represents the dynamic weighting coefficient for the k-th scale, where k=1 is the particle scale, k=2 is the sample scale, and k=3 is the macroscopic scale. Let be the normalized feature tensor at the k-th scale. This serves as the baseline feature for the l-th risk model in the risk model library. Let k be the Gaussian kernel standard deviation at the k-th scale. This represents the cross-scale correlation strengthening coefficient. The feature covariance matrix is ​​a cross-scale feature matrix. For matrix transpose, This is a weight matrix generated based on the cross-scale correlation parameter matrix.

[0028] In one embodiment, the spatiotemporal characteristics of the damage and the risk pattern are input into a pre-trained risk propagation probability prediction model for prediction, resulting in a risk propagation prediction outcome, including:

[0029] Based on multi-scale spatiotemporal feature maps and risk patterns, feature concatenation is performed to generate composite risk feature vectors;

[0030] The composite risk feature vector is input into the pre-trained risk diffusion probability prediction model, which includes a diffusion coefficient matrix and a feature space transformation mechanism. The diffusion coefficient matrix and the feature space transformation mechanism are trained from historical risk diffusion data.

[0031] Based on the diffusion coefficient matrix, a spatiotemporal correlation weighting calculation is performed on the composite risk feature vector to generate a weighted risk feature tensor.

[0032] A feature space transformation mechanism is used to perform topological mapping on the weighted risk feature tensor to obtain the risk propagation path network, which is then used as the risk propagation prediction result.

[0033] In one embodiment, based on risk patterns and risk propagation prediction results, a multidimensional risk decision boundary is constructed using a nonlinear classifier, including:

[0034] Based on the risk pattern, extract the damage risk threshold parameter;

[0035] Based on the risk propagation prediction results, the risk pattern topology is extracted;

[0036] Based on the damage risk threshold parameter and the risk pattern topology, a multi-dimensional risk correlation matrix is ​​constructed using the local linear embedding algorithm.

[0037] Based on a multi-dimensional risk correlation matrix and a pre-configured risk attenuation coefficient matrix, a risk decision hyperplane is constructed using a support vector machine.

[0038] Extract the intersection features of the risk decision hyperplane and the multidimensional risk correlation matrix, and based on the intersection features, use principal component analysis to extract the dominant risk dimension;

[0039] Based on the dominant risk dimension and combined with the risk pattern topology, the risk decision hyperplane is optimized using Lagrange multipliers to obtain a multidimensional risk decision boundary.

[0040] In one embodiment, the dynamic risk decision-making rules include a decision tree classifier and a hierarchical early warning plan library;

[0041] Based on multidimensional risk decision boundaries, a tiered early warning system is generated using pre-defined dynamic risk decision rules, including:

[0042] Extract the temporal variation characteristics of the multidimensional risk decision boundary;

[0043] Based on the temporal variation characteristics, a decision tree classifier is used to classify risk levels and obtain graded early warning labels;

[0044] Based on the tiered early warning labels, a rule engine is used to match tiered early warning instructions from the tiered early warning plan library.

[0045] Secondly, this application also provides a geotechnical infrastructure risk early warning system based on a multi-scale damage model, including:

[0046] The monitoring data acquisition module is used to acquire environmental monitoring data of the target soil and rock area based on a preset monitoring scale.

[0047] The multi-scale model building module is used to perform multi-scale coupled analysis and modeling based on environmental monitoring data to obtain a multi-scale damage model.

[0048] The risk pattern matching module is used to input the multi-scale damage model into the pre-trained convolutional neural network, extract the spatiotemporal features of the damage, and determine the risk pattern based on the spatiotemporal features of the damage and the preset risk pattern library.

[0049] The risk propagation prediction module is used to input the spatiotemporal characteristics of damage and risk patterns into a pre-trained risk diffusion probability prediction model to make predictions and obtain risk propagation prediction results.

[0050] The decision boundary construction module is used to construct a multidimensional risk decision boundary based on risk patterns and risk propagation prediction results using a nonlinear classifier.

[0051] The graded early warning generation module is used to generate graded early warning instructions based on multi-dimensional risk decision boundaries and using preset dynamic risk decision rules.

[0052] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for risk warning of geotechnical infrastructure based on a multi-scale damage model.

[0053] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for risk warning of geotechnical infrastructure based on a multi-scale damage model.

[0054] The aforementioned method and system for risk early warning of geotechnical infrastructure based on a multi-scale damage model acquires environmental monitoring data based on a preset monitoring scale, achieving comprehensive collection of multi-dimensional data and overcoming the limitations of traditional single-scale monitoring. It constructs a multi-scale damage model through multi-scale coupling analysis, coupling the interaction between microscopic particles and macroscopic geology to eliminate prediction bias caused by scale separation. It utilizes a pre-trained convolutional neural network to extract spatiotemporal features of damage and match risk patterns, reducing reliance on human experience. It obtains risk propagation prediction results through a risk diffusion probability prediction model, dynamically assesses risk paths, and improves the timeliness of early warning. It employs a nonlinear classifier to construct a multi-dimensional risk decision boundary, enhancing classification accuracy and avoiding false alarms and missed alarms. Based on dynamic risk decision rules, it generates hierarchical early warning instructions, achieving precise response and improving the accuracy and reliability of risk early warning for geotechnical infrastructure. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of the geotechnical infrastructure risk early warning method based on a multi-scale damage model according to the present invention.

[0057] Figure 2 This is a structural diagram of the geotechnical infrastructure risk early warning system based on a multi-scale damage model according to the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] In one embodiment, such as Figure 1As shown, a risk early warning method for geotechnical infrastructure based on a multi-scale damage model is provided. This embodiment illustrates the application of this method to a risk early warning terminal. It is understood that this method can also be applied to a risk early warning server, and can also be applied to a risk early warning system including a risk early warning terminal and a risk early warning server, and can be implemented through the interaction between the risk early warning terminal and the risk early warning server. The implementation environment of this application includes: terminal computing devices (including: monitoring terminals, such as a sensor group composed of multiple sensor devices of different scales, used to collect and preprocess environmental monitoring data of the target geotechnical area in real time; response terminals, used to receive early warning commands and respond), and servers (used to perform computational tasks such as analysis modeling, feature extraction, and risk prediction). The application scenario includes: when geotechnical infrastructure (such as slopes or tunnels) has a need for damage evolution, the terminal collects and preprocesses monitoring data, and transmits it to the server through an information network; after receiving the data, the server calls the pre-trained model to construct a multi-scale damage model, match risk patterns, and predict propagation, generates decision results, and feeds them back to the terminal; the terminal responds according to the decision results, realizing real-time risk prevention and control through end-to-cloud collaboration.

[0060] In this embodiment, the method includes the following steps:

[0061] S01, based on a preset monitoring scale, acquire environmental monitoring data of the target soil and rock area.

[0062] Optionally, the risk warning terminal can comprehensively collect environmental monitoring data of the target soil and rock area through pre-set multi-dimensional monitoring scales (such as micro, meso, and macro scales) to overcome the limitations of traditional single-scale monitoring. The preset monitoring scales can be multiple spatial or temporal scales pre-configured according to the characteristics of the soil and rock infrastructure and the needs of risk warning. Environmental monitoring data can include the physicochemical parameters of the soil and rock mass, such as particle geometry, pore distribution, water content, topography, and geological structure. The risk warning terminal can acquire the physicochemical parameters of the soil and rock mass in real time through sensor networks, remote sensing technology, or field monitoring equipment, and integrate monitoring data at different scales into a unified environmental dataset through multi-source data fusion technology to support subsequent data analysis.

[0063] S02. Based on environmental monitoring data, multi-scale coupling analysis and modeling are performed to obtain a multi-scale damage model.

[0064] Optionally, scale coupling analysis modeling is a cross-scale integrated computation method that can be used to eliminate scale separation bias. The risk warning terminal can associate data at different scales through numerical simulation techniques (such as discrete element method, finite element method or other physical driving models).

[0065] For example, a risk warning terminal can first construct a particle contact network based on microscopic data, generate a material constitutive matrix by combining it with mesoscopic data, obtain a cross-scale parameter matrix through geometric transformation alignment, and integrate a macroscopic three-dimensional geological model. Based on a preset damage index system, it can couple multiple physical fields (such as mechanical and seepage fields) to generate a multi-scale damage model that characterizes the damage evolution law of soil and rock masses. In implementation, damage simulation of the soil and rock environment can be achieved through data standardization, model parameter mapping, and coupling algorithms (such as tensor decomposition or machine learning-assisted scale bridging).

[0066] S03. Input the multi-scale damage model into the pre-trained convolutional neural network to extract the spatiotemporal features of the damage, and based on the spatiotemporal features of the damage, combine with the preset risk pattern library to match and determine the risk pattern.

[0067] Alternatively, the pre-trained convolutional neural network can be a deep learning model pre-trained based on historical data, which automatically extracts features through multi-layer convolutional operations and is specifically configured to process multi-scale data.

[0068] Optionally, a pre-trained convolutional neural network can capture local deformation at the particle scale, mechanical coupling at the sample scale, and geological structural features at the macro scale using dilated convolutional kernels at different expansion rates, and integrate them into a unified spatiotemporal damage feature. This spatiotemporal damage feature can be a multi-dimensional tensor, dynamically reflecting the distribution pattern of damage in time and space. During feature extraction, a multi-band feature matrix can be generated through model parameter mapping (such as tensor decomposition), and then converted into a spatiotemporal damage feature map through a feature channel allocation mechanism. The preset risk pattern library can be a pre-established set of benchmark features containing various typical risk scenarios (such as landslides and subsidence). During matching, the risk warning terminal can calculate the similarity between the spatiotemporal damage feature map and the benchmark features in the library (using methods such as cosine similarity or Gaussian weighted similarity calculation formulas), and apply feature matching rules (such as threshold comparison or optimization search) to determine the appropriate risk pattern, reducing reliance on human experience and improving recognition accuracy.

[0069] S04. The spatiotemporal characteristics of the damage and the risk pattern are input into the pre-trained risk diffusion probability prediction model for prediction, and the risk propagation prediction results are obtained.

[0070] Optionally, the pre-trained risk diffusion probability prediction model can be a machine learning or probabilistic model pre-trained based on historical risk diffusion data. This model can be configured with a diffusion coefficient matrix and a feature space transformation mechanism to simulate the propagation behavior of risk in soil and rock masses. In implementation, the risk warning terminal can concatenate the spatiotemporal characteristics of damage with the risk pattern to generate a composite risk feature vector. It then performs spatiotemporal correlation weighting calculations using the diffusion coefficient matrix in the pre-trained model to generate a weighted risk feature tensor. A feature space transformation mechanism (such as topological mapping or graph neural networks) is then used to map the tensor into a risk propagation path network, outputting the risk propagation prediction result. This result can be used to quantify the risk diffusion probability and path, improving the timeliness of early warnings. Different probabilistic models (such as Bayesian networks) or real-time data streams can also be used to adjust the prediction parameters, enabling dynamic prediction of risk propagation in various soil and rock environments.

[0071] S05, based on risk patterns and risk propagation prediction results, uses a nonlinear classifier to construct a multidimensional risk decision boundary.

[0072] Optionally, the nonlinear classifier can be a machine learning model capable of handling complex nonlinear relationships. The nonlinear classifier can be constructed based on, but is not limited to, support vector machines, neural networks, or ensemble learning algorithms. It can be used for classification in a multi-dimensional feature space. The multidimensional risk decision boundary can refer to the hyperplane or decision surface constructed by the classifier. This boundary can be used to distinguish different risk levels across multiple risk dimensions (such as damage severity and propagation speed) and achieve dynamic threshold adjustment. In implementation, the risk warning terminal can extract damage risk threshold parameters (such as critical stress or deformation) based on risk patterns and combine them with the topological structure (such as path connectivity) in the risk propagation prediction results. It can then construct a multidimensional risk correlation matrix through feature extraction (such as local linear embedding or principal component analysis) to capture the nonlinear interactions between risk factors. The risk warning terminal can apply a nonlinear classifier (such as a support vector machine configured with a kernel function) to generate an initial risk decision hyperplane. Furthermore, it can refine the initial risk decision hyperplane using optimization techniques (such as the Lagrange multiplier method) to eliminate overfitting and obtain an adaptive multidimensional risk decision boundary, thereby improving classification accuracy and robustness.

[0073] S06, based on multi-dimensional risk decision boundaries, generates tiered early warning instructions using preset dynamic risk decision rules.

[0074] Optionally, the preset dynamic risk decision-making rules can be a pre-configured intelligent rule set, which may include a decision tree classifier, a hierarchical early warning plan library, and a rule engine. These preset dynamic risk decision-making rules are used to adaptively adjust early warning strategies based on real-time data. The hierarchical early warning instructions can be response instructions generated based on risk levels (such as different levels of alarms or action plans), and can be used for risk prevention and control. In implementation, the risk early warning terminal can extract the temporal change characteristics of multi-dimensional risk decision boundaries (such as boundary offset or stability indicators), classify risk levels (such as low, medium, and high risk) using a decision tree classifier, generate hierarchical early warning labels, and use the rule engine to match corresponding early warning instructions from the plan library. The risk early warning terminal can also employ different classification algorithms (such as random forests) or real-time rule optimization (such as reinforcement learning) to adapt to various geotechnical environments and computing resource conditions, thereby achieving risk early warning for geotechnical infrastructure.

[0075] In one embodiment, the environmental monitoring data may include particle-scale monitoring data, sample-scale monitoring data and macro-scale monitoring data. The particle-scale monitoring data may include particle geometric parameters and particle state parameters. The sample-scale monitoring data may include pore distribution parameters and water content parameters. The macro-scale monitoring data may include topographic parameters and geological structure parameters.

[0076] Based on environmental monitoring data, multi-scale coupled analysis and modeling are performed to obtain a multi-scale damage model, which may include:

[0077] S11, based on particle geometric parameters and particle state parameters, uses mechanical theory to invert particle mechanical parameters, and combines particle geometric parameters, particle state parameters and particle mechanical parameters to construct a micro-contact network of soil and rock particles using the discrete element method.

[0078] For example, the risk warning terminal can invert particle mechanical parameters (such as particle shape and size) and particle state parameters (such as position and velocity) based on particle-scale geometric parameters (such as particle shape and size) and particle state parameters (such as position and velocity) using mechanical theories (such as Hooke's law or Mohr-Coulomb criterion). By constructing a micro-contact network of soil and rock particles through the discrete element method, the risk warning terminal can lay the foundation for micro-damage analysis by simulating the interaction force chains and motion behavior between particles.

[0079] S12, based on pore distribution parameters and water content parameters, uses the finite element method to generate the material constitutive matrix through mesh generation.

[0080] Optionally, the risk warning terminal can use the pore distribution parameters (such as pore size and connectivity) and water content parameters at the sample scale to perform mesh generation using the finite element method, thereby generating a material constitutive matrix. This material constitutive matrix can characterize the stress-strain relationship of the soil and rock mass, and mechanical consistency can be ensured through numerical integration methods.

[0081] S13, the micro-contact network of soil particles and the material constitutive matrix are geometrically transformed and aligned to obtain the cross-scale correlation parameter matrix.

[0082] Optionally, the risk warning terminal can perform geometric transformations to align the micro-contact network and the material constitutive matrix, such as by eliminating scale differences through coordinate mapping or interpolation algorithms, to obtain a cross-scale correlation parameter matrix. This cross-scale correlation parameter matrix can integrate micro- and meso-scale data to reveal the transmission law of damage.

[0083] S14, based on topographic parameters and geological structure parameters, performs three-dimensional modeling to obtain a macroscopic geotechnical model.

[0084] Optionally, the risk warning terminal can generate a macroscopic geotechnical geometric model based on macroscopic topographic parameters (such as slope and elevation) and geological structural parameters (such as fault distribution) using 3D modeling software (such as GIS or CAD tools). This macroscopic geotechnical geometric model can reflect the spatial characteristics of the geological structure.

[0085] S15, based on the macroscopic geotechnical geometric model and the cross-scale correlation parameter matrix, combined with the preset damage index system, performs multi-physics field coupling to obtain a multi-scale damage model.

[0086] Optionally, the risk warning terminal can combine a preset damage index system (such as strain threshold or damage index) and integrate the macroscopic geotechnical geometric model with the cross-scale correlation parameter matrix through a multi-physics coupling algorithm (such as finite element-discrete element coupling) to perform synchronous simulation of mechanical fields, seepage fields, etc., to obtain a multi-scale damage model. This multi-scale damage model can dynamically characterize the damage evolution of geotechnical mass from particles to macroscopic scales, providing data support for subsequent risk warning.

[0087] In one embodiment, a multi-scale damage model is input into a pre-trained convolutional neural network to extract spatiotemporal damage features. Based on these features and a pre-defined risk pattern library, risk patterns are matched and determined, including:

[0088] S21, the model parameters of the multi-scale damage model are mapped and transformed to obtain a multi-frequency spatiotemporal feature matrix; wherein, the multi-frequency spatiotemporal feature matrix includes a high-frequency spatiotemporal feature matrix, a mid-frequency spatiotemporal feature matrix, and a low-frequency spatiotemporal feature matrix; the high-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the particle-scale model parameters in the multi-scale damage model; the mid-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the sample-scale model parameters in the multi-scale damage model; the low-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the macroscopic-scale model parameters in the multi-scale damage model.

[0089] Specifically, the risk warning terminal can use tensor decomposition algorithms (such as CP decomposition or Tucker decomposition) to map and transform model parameters of multi-scale damage models.

[0090] For example, the risk warning terminal can transform particle-scale model parameters into a high-frequency spatiotemporal feature matrix, which can capture rapid changes in microscopic damage; the risk warning terminal can transform sample-scale model parameters into a mid-frequency spatiotemporal feature matrix, which can reflect the microscopic mechanical response; the risk warning terminal can transform macroscopic-scale model parameters into a low-frequency spatiotemporal feature matrix, which can characterize the slow-changing characteristics of geological structures; the risk warning terminal model maps and transforms multi-scale damage models to generate multi-frequency spatiotemporal feature matrices, which can lay the foundation for multi-scale analysis.

[0091] S22, the multi-band spatiotemporal feature matrix is ​​input into a pre-trained convolutional neural network for feature extraction, resulting in multi-scale spatiotemporal features as damage spatiotemporal features. The feature extraction operation is performed using four sets of pre-set dilation rate dilated convolutional kernels. The first set of dilated convolutional kernels is used to extract local deformation features of model parameters at the particle scale, the second set of dilated convolutional kernels is used to extract mechanical coupling features of model parameters at the sample scale, the third set of dilated convolutional kernels is used to extract geological structural features of model parameters at the macroscopic scale, and the fourth set of dilated convolutional kernels is used to integrate local deformation features, mechanical coupling features, and geological structural features to obtain multi-scale spatiotemporal features.

[0092] Optionally, the risk warning terminal can input the multi-band spatiotemporal feature matrix into a pre-trained convolutional neural network, which can use four sets of dilated convolutional kernels with preset dilation rates for feature extraction.

[0093] For example, the first set of void convolution kernels (with an expansion rate of 1) can extract local deformation features at the particle scale, such as microcracks or particle displacement; the second set of void convolution kernels (with an expansion rate of 2) can extract mechanical coupling features at the sample scale, such as stress distribution or strain concentration; the third set of void convolution kernels (with an expansion rate of 4) can extract geological structural features at the macroscopic scale, such as rock and soil faults or rock and soil joints; the fourth set of void convolution kernels (with an expansion rate of 8) can integrate the above local deformation features, mechanical coupling features and geological structural features through convolution layers and activation functions to output multi-scale spatiotemporal features (i.e., damage spatiotemporal features), which can dynamically characterize the evolution of damage in time and space.

[0094] S23, based on the spatiotemporal characteristics of damage, adopts a feature channel allocation mechanism based on the damage index system to generate a spatiotemporal feature map of damage.

[0095] Optionally, the risk warning terminal can adopt a feature channel allocation mechanism based on the damage index system to allocate the spatiotemporal features of damage to different channels according to the scale dimension, and generate a spatiotemporal feature map of damage. This spatiotemporal feature map of damage can be in the form of a multidimensional tensor, and the spatiotemporal feature map of damage can intuitively display the hot spots of damage.

[0096] S24, calculate the feature similarity between the spatiotemporal feature map of damage and the baseline features of each risk model in the preset risk model library.

[0097] Optionally, the risk warning terminal can use similarity calculation methods (such as Euclidean distance method, cosine similarity method, etc.) to calculate the feature similarity between the damage spatiotemporal feature map and the baseline features of each risk model in the preset risk model library.

[0098] S25, set feature matching rules based on feature similarity, and determine risk patterns based on feature similarity and feature matching rules.

[0099] Optionally, the risk warning terminal can set feature matching rules based on feature similarity, such as setting a similarity threshold or using an optimized search algorithm. When the feature similarity reaches its maximum value or exceeds a critical value, the risk warning terminal can match and determine the corresponding risk pattern (such as a landslide or subsidence pattern) from the risk pattern library according to the feature matching rules, providing support for subsequent warnings. The risk warning terminal can improve the accuracy and timeliness of risk identification through multi-scale feature fusion and intelligent matching.

[0100] In one embodiment, S31, the formula for calculating the feature similarity between the damage spatiotemporal feature map and the baseline features of each risk pattern in the preset risk pattern library is as follows:

[0101]

[0102] in, For feature similarity, represents the dynamic weighting coefficient for the k-th scale, where k=1 is the particle scale, k=2 is the sample scale, and k=3 is the macroscopic scale. Let be the normalized feature tensor at the k-th scale. This serves as the baseline feature for the l-th risk model in the risk model library. Let k be the Gaussian kernel standard deviation at the k-th scale. This represents the cross-scale correlation strengthening coefficient. The feature covariance matrix is ​​a cross-scale feature matrix. For matrix transpose, This is a weight matrix generated based on the cross-scale correlation parameter matrix.

[0103] For example, the purpose of the feature similarity calculation formula is to quantify the similarity between the spatiotemporal feature map of damage and the benchmark features in the risk pattern library, in order to determine the most suitable risk pattern and provide a data-driven basis for subsequent early warning decisions. The risk early warning terminal can use this calculation formula to combine intra-scale similarity measurement and cross-scale correlation enhancement to achieve a comprehensive evaluation of multi-dimensional features: In this calculation formula, the summation term... Calculating normalized characteristic tensors at three scales: particle, sample, and macroscopic, using Gaussian kernel functions. Compared with the benchmark features The Euclidean distance similarity is calculated using dynamic weight coefficients. Adjustments are made to the contributions at each scale to reflect the differences in damage from micro to macro; additional terms in the calculation formula. Including cross-scale feature covariance matrix and weight matrix The trace operation, this additional term, strengthens the correlation between scales and avoids bias caused by scale separation. The relationship between the whole and the variables in this calculation formula is expressed as follows: [The formula's...] Values ​​are determined by variables , , Parameters are adjusted, among which It can be used to dynamically adjust weights based on real-time data. It can be used to control the sensitivity of the Gaussian kernel. It can be used to enhance cross-scale consistency. and As input features, it can be used to connect the implementation with the damage indicator system. The calculation formula relies on the generation mechanism of the damage spatiotemporal feature map (such as feature channel allocation) and serves as the input to the risk pattern matching rule, which can be used to drive subsequent risk propagation prediction and decision boundary construction. The risk warning terminal can use the calculation formula for mathematical optimization to enhance the robustness and scalability of the warning and solve the problem of single similarity calculation in traditional methods.

[0104] In one embodiment, the spatiotemporal characteristics of the damage and the risk pattern are input into a pre-trained risk propagation probability prediction model for prediction, resulting in a risk propagation prediction outcome, including:

[0105] S41, based on multi-scale spatiotemporal feature maps and risk patterns, performs feature splicing to generate a composite risk feature vector.

[0106] Specifically, the risk warning terminal can perform feature stitching operations based on multi-scale spatiotemporal feature maps and matched risk patterns (such as landslide or subsidence patterns).

[0107] For example, a risk warning terminal can integrate multi-dimensional data (such as time series and spatial distribution) of feature maps with the category encoding of risk patterns through tensor connection or fusion algorithms (such as concatenation or weighted superposition) to generate a composite risk feature vector. This composite risk feature vector can be used to capture the correlation between damage and risk patterns.

[0108] S42, input the composite risk feature vector into the pre-trained risk diffusion probability prediction model. The risk diffusion probability prediction model includes a diffusion coefficient matrix and a feature space transformation mechanism, which are trained from historical risk diffusion data.

[0109] Optionally, the risk warning terminal can input the composite risk feature vector into the pre-trained risk diffusion probability prediction model. The pre-trained risk diffusion probability prediction model can be pre-trained based on historical risk diffusion data (such as cases of soil and rock failure). The pre-trained risk diffusion probability prediction model can include a diffusion coefficient matrix (obtained through machine learning algorithms such as gradient descent optimization, which can be used to quantify the propagation weight of risk in time and space) and a feature space transformation mechanism (such as a nonlinear mapping function or a graph neural network layer). The diffusion coefficient matrix and the feature space transformation mechanism can ensure that the pre-trained risk diffusion probability prediction model can adapt to different soil and rock environments.

[0110] S43, based on the diffusion coefficient matrix, performs spatiotemporal correlation weighting calculation on the composite risk feature vector to generate a weighted risk feature tensor.

[0111] Optionally, the risk warning terminal can perform spatiotemporal correlation weighted calculation on the composite risk feature vector based on the diffusion coefficient matrix.

[0112] For example, the risk warning terminal can use matrix multiplication or convolution operations to apply weights to the spatiotemporal dimensions of the composite risk feature vector, such as adjusting the influencing factors according to distance decay or geological continuity, to generate a weighted risk feature tensor. This weighted risk feature tensor can enhance the risk warning terminal's ability to pay attention to high risks.

[0113] S44 employs a feature space transformation mechanism to perform topological mapping on the weighted risk feature tensor, obtaining a risk propagation path network, and uses the risk propagation path network as the risk propagation prediction result.

[0114] Optionally, the risk warning terminal can use a feature space transformation mechanism to perform topological structure mapping on the weighted risk feature tensor.

[0115] For example, a risk warning terminal can use graph theory algorithms (such as shortest path analysis or community detection) to convert tensors into a network structure, obtaining a risk propagation path network. Nodes in the risk propagation path network represent risk points, and edges represent propagation probabilities. This risk propagation path network can serve as a prediction result, dynamically displaying the direction and probability of risk spread.

[0116] In one embodiment, based on risk patterns and risk propagation prediction results, a multidimensional risk decision boundary is constructed using a nonlinear classifier, including:

[0117] S51, based on risk patterns, extracts damage risk threshold parameters.

[0118] For example, the risk warning terminal can extract damage risk threshold parameters based on the matched risk patterns (such as landslide or settlement patterns), such as obtaining critical stress or strain values ​​by querying a preset damage index system. The risk threshold parameters can be used as a benchmark for risk classification.

[0119] S52, based on the risk propagation prediction results, extract the risk pattern topology.

[0120] Optionally, the risk warning terminal can extract the risk pattern topology based on the risk propagation prediction results (i.e., the risk propagation path network).

[0121] For example, the risk warning terminal can use graph theory algorithms (such as node degree analysis or path connectivity calculation) to quantify the node and edge weights of the risk propagation path network in the risk propagation prediction results, so as to capture the spatiotemporal characteristics of risk diffusion.

[0122] S53, based on damage risk threshold parameters and risk pattern topology, uses a local linear embedding algorithm to construct a multi-dimensional risk correlation matrix.

[0123] Optionally, the risk warning terminal can construct a multi-dimensional risk correlation matrix based on damage risk threshold parameters and risk pattern topology using a local linear embedding algorithm.

[0124] For example, a risk warning terminal can reduce the dimensionality of high-dimensional features (such as thresholds and topological indicators) through neighborhood-preserving mapping to generate a low-dimensional manifold representation. The low-dimensional manifold representation can be used to capture the nonlinear interactions between risk factors, thereby forming a multi-dimensional risk correlation matrix, which can serve as a classification basis.

[0125] S54, based on a multi-dimensional risk correlation matrix and combined with a pre-configured risk decay coefficient matrix, uses a support vector machine to construct a risk decision hyperplane.

[0126] Optionally, the risk warning terminal can construct a risk decision hyperplane using a support vector machine based on a multi-dimensional risk correlation matrix and a pre-configured risk decay coefficient matrix (a weight matrix trained with historical data to adjust the decay effect of risk over time or space).

[0127] For example, a risk warning terminal can use radial basis function kernels to process nonlinear data and generate an initial hyperplane to distinguish different risk levels.

[0128] S55. Extract the intersection features of the risk decision hyperplane and the multi-dimensional risk correlation matrix, and based on the intersection features, use principal component analysis to extract the dominant risk dimension.

[0129] Optionally, the risk warning terminal can extract the intersection features of the risk decision hyperplane and the multi-dimensional risk correlation matrix (i.e., the risk warning terminal can find the region where the hyperplane and the correlation matrix overlap through matrix operations); based on the intersection features, the risk warning terminal can use principal component analysis to identify the risk dimension whose contribution to classification exceeds the contribution threshold, and this risk dimension can be used as the dominant risk dimension.

[0130] S56, based on the dominant risk dimension and combined with the risk pattern topology, uses Lagrange multipliers to optimize the risk decision hyperplane, thus obtaining a multidimensional risk decision boundary.

[0131] Optionally, based on the dominant risk dimension, the risk warning terminal can combine the risk pattern topology and use Lagrange multipliers to optimize the risk decision hyperplane.

[0132] For example, the risk warning terminal can adjust the parameters of the risk decision hyperplane by transforming the classification problem into a constrained optimization problem to eliminate overfitting, thereby obtaining a multidimensional risk decision boundary that can distinguish different risk levels. The multidimensional risk decision boundary can be used to respond to dynamic changes in the geotechnical environment.

[0133] In one embodiment, the dynamic risk decision-making rules include a decision tree classifier and a hierarchical early warning plan library;

[0134] Based on multidimensional risk decision boundaries, a tiered early warning system is generated using pre-defined dynamic risk decision rules, including:

[0135] S61, extract the temporal variation characteristics of the multidimensional risk decision boundary.

[0136] Specifically, the risk warning terminal can extract the temporal change characteristics of multidimensional risk decision boundaries.

[0137] For example, a risk warning terminal can capture the temporal variation characteristics of risk by analyzing the evolution of the multidimensional risk decision boundary over time (including using sliding window technology or time series analysis algorithms to calculate the boundary offset, stability index or rate of change). The temporal variation characteristics can reflect the real-time progress of soil and rock damage and provide input for risk classification.

[0138] S62, based on the time-series change characteristics, uses a decision tree classifier to classify risk levels and obtain graded early warning labels.

[0139] Optionally, based on the time-series change characteristics, the risk warning terminal can use a decision tree classifier to classify risk levels.

[0140] For example, the decision tree classifier can be configured based on historical training data. The risk warning terminal can use the decision tree classifier to construct a tree structure through feature selection (such as information gain or Gini impurity), map time-series features (such as boundary fluctuation frequency) to discrete risk levels (such as low, medium, and high risk), and output graded warning labels.

[0141] S63, based on the hierarchical warning labels, uses a rule engine to match hierarchical warning instructions from the hierarchical warning plan library.

[0142] Optionally, based on the tiered warning labels, the risk warning terminal can use a rule engine to match tiered warning instructions from the tiered warning plan library.

[0143] For example, the risk warning terminal can use a rule engine to match the response schemes (based on expert knowledge predefined) of the tag and the hierarchical warning plan library in real time according to logical rules (such as if-then statements); for example, when the tag indicates high risk, it can trigger corresponding instructions (such as evacuation alarms or engineering reinforcement); the risk warning terminal can achieve dynamic adaptation of warnings by integrating time series analysis, machine learning classification and rule reasoning, thereby improving the accuracy of geotechnical infrastructure safety management. This design solves the lag problem of traditional fixed threshold methods.

[0144] The aforementioned risk early warning method for geotechnical infrastructure based on a multi-scale damage model acquires environmental monitoring data of the target geotechnical area based on a preset monitoring scale. This data includes particle geometric and state parameters at the particle scale, pore distribution and water content parameters at the sample scale, and topographic and geological structural parameters at the macro scale. This achieves comprehensive multi-dimensional data acquisition from micro to macro, overcoming the limitations of traditional single-scale monitoring. Through multi-scale coupled analysis and modeling, the method utilizes the discrete element method to invert particle mechanical parameters and constructs a microscopic contact network of geotechnical particles. It then combines this with the finite element method to generate a material constitutive matrix, aligns it through geometric transformation to obtain a cross-scale correlation parameter matrix, and couples it with the macroscopic geotechnical geometric model to form a multi-physics field, thus eliminating prediction bias caused by scale separation and improving the model's dynamic representation ability of damage evolution. The multi-scale damage model is input into a pre-trained convolutional neural network. Tensor decomposition yields a multi-frequency spatiotemporal feature matrix, and four sets of cavitary convolutional kernels are used to extract local deformation, mechanical coupling, and geological structural features, generating a spatiotemporal damage feature map. The feature similarity with the baseline features of the risk pattern library is calculated, reducing reliance on manual experience and enabling automatic identification of risk patterns. The spatiotemporal characteristics of damage and risk patterns are input into a pre-trained risk diffusion probability prediction model. A composite risk feature vector is generated through feature concatenation, and spatiotemporal weighting and topological mapping are performed using a diffusion coefficient matrix to output a risk propagation path network. This dynamically assesses the risk diffusion probability and path, improving the timeliness and foresight of early warnings. Based on risk patterns and propagation prediction results, a nonlinear classifier such as a support vector machine is used to construct a multi-dimensional risk correlation matrix through local linear embedding. Principal component analysis and Lagrange multiplier optimization of the decision hyperplane are combined to form an adaptive multi-dimensional risk decision boundary, enhancing risk classification accuracy. Based on the temporal variation characteristics of the multi-dimensional risk decision boundary, a decision tree classifier is used to classify risk levels. A rule engine is used to match instructions generated from a hierarchical early warning plan library, enabling precise and dynamic response to geotechnical infrastructure risks. This improves the accuracy and reliability of early warnings, effectively preventing safety accidents caused by multi-scale damage evolution, and achieving intelligent management across the entire chain from data collection to early warning response.

[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0146] Based on the same inventive concept, this application also provides a multi-scale damage model-based geotechnical infrastructure risk early warning system for implementing the aforementioned multi-scale damage model-based geotechnical infrastructure risk early warning method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the multi-scale damage model-based geotechnical infrastructure risk early warning system provided below can be found in the limitations of the multi-scale damage model-based geotechnical infrastructure risk early warning method described above, and will not be repeated here.

[0147] In one exemplary embodiment, such as Figure 2 As shown, a geotechnical infrastructure risk early warning system based on a multi-scale damage model is provided, including:

[0148] The monitoring data acquisition module 101 can be used to acquire environmental monitoring data of the target soil and rock area based on a preset monitoring scale;

[0149] The multi-scale model building module 102 can be used to perform multi-scale coupled analysis modeling based on environmental monitoring data to obtain a multi-scale damage model.

[0150] The risk pattern matching module 103 can be used to input a multi-scale damage model into a pre-trained convolutional neural network, extract damage spatiotemporal features, and determine risk patterns based on damage spatiotemporal features and a preset risk pattern library.

[0151] The risk propagation prediction module 104 can be used to input the spatiotemporal characteristics of damage and risk patterns into a pre-trained risk diffusion probability prediction model to make predictions and obtain risk propagation prediction results.

[0152] The decision boundary construction module 105 can be used to construct a multidimensional risk decision boundary based on risk patterns and risk propagation prediction results using a nonlinear classifier.

[0153] The graded early warning generation module 106 can be used to generate graded early warning instructions based on multi-dimensional risk decision boundaries and using preset dynamic risk decision rules.

[0154] In one embodiment, in the multi-scale model construction module 102, the environmental monitoring data includes particle-scale monitoring data, sample-scale monitoring data and macro-scale monitoring data. The particle-scale monitoring data includes particle geometric parameters and particle state parameters. The sample-scale monitoring data includes pore distribution parameters and water content parameters. The macro-scale monitoring data includes topographic parameters and geological structure parameters.

[0155] The multi-scale model building module 102 can also be used for:

[0156] Based on particle geometric parameters and particle state parameters, particle mechanical parameters are inverted using mechanical theory. Combining particle geometric parameters, particle state parameters, and particle mechanical parameters, a micro-contact network of soil and rock particles is constructed using the discrete element method.

[0157] Based on pore distribution parameters and water content parameters, the material constitutive matrix is ​​generated by meshing using the finite element method.

[0158] By performing geometric transformation and alignment of the micro-contact network of soil particles and the material constitutive matrix, a cross-scale correlation parameter matrix is ​​obtained.

[0159] Based on topographic and geological parameters, a three-dimensional model is created to obtain a macroscopic geotechnical model.

[0160] Based on the macroscopic geotechnical geometric model and the cross-scale correlation parameter matrix, combined with the preset damage index system, multi-physics field coupling is carried out to obtain a multi-scale damage model.

[0161] In one embodiment, the risk pattern matching module 103 can also be used for:

[0162] A multi-frequency spatiotemporal feature matrix is ​​obtained by mapping and transforming the model parameters of the multi-scale damage model. This multi-frequency spatiotemporal feature matrix includes a high-frequency spatiotemporal feature matrix, a mid-frequency spatiotemporal feature matrix, and a low-frequency spatiotemporal feature matrix. The high-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the particle-scale model parameters in the multi-scale damage model. The mid-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the sample-scale model parameters in the multi-scale damage model. The low-frequency spatiotemporal feature matrix is ​​obtained by tensor decomposition of the macroscopic-scale model parameters in the multi-scale damage model.

[0163] The multi-band spatiotemporal feature matrix is ​​input into a pre-trained convolutional neural network for feature extraction, resulting in multi-scale spatiotemporal features as damage spatiotemporal features. The feature extraction operation is performed using four sets of pre-set dilation rate dilated convolutional kernels. The first set of dilated convolutional kernels is used to extract local deformation features of model parameters at the particle scale, the second set of dilated convolutional kernels is used to extract mechanical coupling features of model parameters at the sample scale, the third set of dilated convolutional kernels is used to extract geological structural features of model parameters at the macroscopic scale, and the fourth set of dilated convolutional kernels is used to integrate local deformation features, mechanical coupling features, and geological structural features to obtain multi-scale spatiotemporal features.

[0164] Based on the spatiotemporal characteristics of damage, a feature channel allocation mechanism based on the damage index system is adopted to generate a spatiotemporal feature map of damage.

[0165] Calculate the feature similarity between the spatiotemporal feature map of damage and the baseline features of each risk pattern in the preset risk pattern library;

[0166] Define feature matching rules based on feature similarity, and determine risk patterns based on feature similarity and feature matching rules.

[0167] In one embodiment, the risk propagation prediction module 104 can also be used for:

[0168] Based on multi-scale spatiotemporal feature maps and risk patterns, feature concatenation is performed to generate composite risk feature vectors;

[0169] The composite risk feature vector is input into the pre-trained risk diffusion probability prediction model, which includes a diffusion coefficient matrix and a feature space transformation mechanism. The diffusion coefficient matrix and the feature space transformation mechanism are trained from historical risk diffusion data.

[0170] Based on the diffusion coefficient matrix, a spatiotemporal correlation weighting calculation is performed on the composite risk feature vector to generate a weighted risk feature tensor.

[0171] A feature space transformation mechanism is used to perform topological mapping on the weighted risk feature tensor to obtain the risk propagation path network, which is then used as the risk propagation prediction result.

[0172] In one embodiment, the decision boundary construction module 105 can also be used to:

[0173] Based on the risk pattern, extract the damage risk threshold parameter;

[0174] Based on the risk propagation prediction results, the risk pattern topology is extracted;

[0175] Based on the damage risk threshold parameter and the risk pattern topology, a multi-dimensional risk correlation matrix is ​​constructed using the local linear embedding algorithm.

[0176] Based on a multi-dimensional risk correlation matrix and a pre-configured risk attenuation coefficient matrix, a risk decision hyperplane is constructed using a support vector machine.

[0177] Extract the intersection features of the risk decision hyperplane and the multidimensional risk correlation matrix, and based on the intersection features, use principal component analysis to extract the dominant risk dimension;

[0178] Based on the dominant risk dimension and combined with the risk pattern topology, the risk decision hyperplane is optimized using Lagrange multipliers to obtain a multidimensional risk decision boundary.

[0179] In one embodiment, the dynamic risk decision-making rules in the hierarchical early warning generation module 106 include a decision tree classifier and a hierarchical early warning plan library;

[0180] The graded early warning generation module 106 can also be used for:

[0181] Extract the temporal variation characteristics of the multidimensional risk decision boundary;

[0182] Based on the temporal variation characteristics, a decision tree classifier is used to classify risk levels and obtain graded early warning labels;

[0183] Based on the tiered early warning labels, a rule engine is used to match tiered early warning instructions from the tiered early warning plan library.

[0184] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a geotechnical infrastructure risk early warning method based on a multi-scale damage model as described above.

[0185] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0186] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0187] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for risk warning of geotechnical infrastructure based on multi-scale damage model, characterized in that, The method comprises: obtaining environmental monitoring data of a target rock-soil region based on a preset monitoring scale; performing multi-scale coupling analysis modeling based on the environmental monitoring data to obtain a multi-scale damage model; inputting the multi-scale damage model into a pre-trained convolutional neural network to extract damage spatiotemporal features, and matching and determining a risk mode based on the damage spatiotemporal features and a preset risk mode library; inputting the damage spatiotemporal features and the risk mode into a pre-trained risk diffusion probability prediction model for prediction to obtain a risk propagation prediction result; constructing a multi-dimensional risk decision boundary using a nonlinear classifier based on the risk mode and the risk propagation prediction result; generating a hierarchical early warning instruction using a preset dynamic risk decision rule based on the multi-dimensional risk decision boundary.

2. The method of claim 1, wherein, The environmental monitoring data comprises particle scale monitoring data, sample scale monitoring data and macro scale monitoring data, the particle scale monitoring data comprises particle geometric parameters and particle state parameters, the sample scale monitoring data comprises pore distribution parameters and water content parameters, and the macro scale monitoring data comprises topographic and geomorphic parameters and geological structure parameters. The multi-scale coupling analysis modeling based on the environmental monitoring data comprises: inverting particle mechanical parameters using a mechanical theory based on the particle geometric parameters and the particle state parameters, and constructing a rock-soil particle micro contact network using a discrete element method based on the particle geometric parameters, the particle state parameters and the particle mechanical parameters; generating a material constitutive matrix through mesh partitioning using a finite element method based on the pore distribution parameters and the water content parameters; aligning the rock-soil particle micro contact network and the material constitutive matrix through geometric transformation to obtain a cross-scale correlation parameter matrix; performing three-dimensional modeling based on the topographic and geomorphic parameters and the geological structure parameters to obtain a macro rock-soil geometric model; performing multi-physical field coupling based on the macro rock-soil geometric model and the cross-scale correlation parameter matrix in combination with a preset damage index system to obtain the multi-scale damage model.

3. The method of claim 2, wherein, The inputting of the multi-scale damage model into a pre-trained convolutional neural network to extract damage spatiotemporal features, and the matching and determining of a risk mode based on the damage spatiotemporal features and a preset risk mode library, comprise: performing model parameter mapping transformation on the multi-scale damage model to obtain a multi-band spatiotemporal feature matrix; wherein the multi-band spatiotemporal feature matrix comprises a high-frequency spatiotemporal feature matrix, a medium-frequency spatiotemporal feature matrix and a low-frequency spatiotemporal feature matrix; the high-frequency spatiotemporal feature matrix is obtained by tensor decomposition of model parameters of a particle scale in the multi-scale damage model; the medium-frequency spatiotemporal feature matrix is obtained by tensor decomposition of model parameters of a sample scale in the multi-scale damage model; and the low-frequency spatiotemporal feature matrix is obtained by tensor decomposition of model parameters of a macro scale in the multi-scale damage model. inputting the multi-frequency band spatio-temporal feature matrix into the pre-trained convolutional neural network to perform feature extraction, to obtain multi-scale spatio-temporal features as the damage spatio-temporal features; wherein the feature extraction operation is completed by using four groups of preset dilation rate hole convolution kernels, the first group of hole convolution kernels are used to extract local deformation features of the model parameters at the particle scale, the second group of hole convolution kernels are used to extract mechanical coupling features of the model parameters at the sample scale, the third group of hole convolution kernels are used to extract geological structure features of the model parameters at the macroscopic scale, and the fourth group of hole convolution kernels are used to integrate the local deformation features, the mechanical coupling features and the geological structure features to obtain the multi-scale spatio-temporal features; based on the damage spatio-temporal features, a feature channel allocation mechanism based on the damage index system is used to generate a damage spatio-temporal feature map; calculating the feature similarity between the damage spatio-temporal feature map and each risk mode reference feature in the preset risk mode library; setting a feature matching rule based on the feature similarity, and matching and determining the risk mode based on the feature similarity and the feature matching rule.

4. The method of claim 3, wherein, The formula for calculating the feature similarity between the damage spatio-temporal feature map and each risk mode reference feature in the preset risk mode library is: wherein, is the feature similarity, is the dynamic weight coefficient of the k-th scale, k=1 for the particle scale, k=2 for the sample scale, and k=3 for the macro scale, is the normalized feature tensor of the k-th scale, is the reference feature of the l-th risk pattern in the risk pattern library, is the Gaussian kernel standard deviation of the k-th scale, is the cross-scale correlation reinforcement coefficient, is the cross-scale feature covariance matrix, is the matrix transpose, is the weight matrix generated based on the cross-scale correlation parameter matrix.

5. The method of claim 3, wherein, inputting the damage spatio-temporal features and the risk mode into a pre-trained risk diffusion probability prediction model to perform prediction, to obtain a risk propagation prediction result, including: based on the multi-scale spatio-temporal feature map and the risk mode, performing feature splicing to generate a composite risk feature vector; inputting the composite risk feature vector into the pre-trained risk diffusion probability prediction model, the risk diffusion probability prediction model including a diffusion coefficient matrix and a feature space transformation mechanism, the diffusion coefficient matrix and the feature space transformation mechanism being trained from historical risk diffusion data; based on the diffusion coefficient matrix, performing spatio-temporal correlation weighted calculation on the composite risk feature vector to generate a weighted risk feature tensor; using the feature space transformation mechanism to perform topological structure mapping on the weighted risk feature tensor to obtain a risk propagation path network, and taking the risk propagation path network as the risk propagation prediction result.

6. The method of claim 5, wherein, based on the risk mode and the risk propagation prediction result, a multi-dimensional risk decision boundary is constructed using a nonlinear classifier, including: based on the risk mode, extracting damage risk threshold parameters; based on the risk propagation prediction result, extracting risk mode topological structures; based on the damage risk threshold parameters and the risk mode topological structures, a multi-dimensional risk correlation matrix is constructed using a local linear embedding algorithm; based on the multi-dimensional risk correlation matrix, a risk decision hyperplane is constructed using a support vector machine in combination with a preconfigured risk attenuation coefficient matrix; extracting intersection features of the risk decision hyperplane and the multi-dimensional risk correlation matrix, and based on the intersection features, extracting dominant risk dimensions using principal component analysis; Based on the dominant risk dimension, in combination with the risk pattern topology, the risk decision hyperplane is optimized by using a Lagrange multiplier, and a multi-dimensional risk decision boundary is obtained.

7. The method of claim 1, wherein, The dynamic risk decision rule includes a decision tree classifier and a hierarchical early warning plan library. The dynamic risk decision rule includes a decision tree classifier and a hierarchical early warning plan library. The dynamic risk decision rule includes a decision tree classifier and a hierarchical early warning plan library. The time sequence change characteristics of the multi-dimensional risk decision boundary are extracted. Based on the time sequence change characteristics, the risk level is divided by using the decision tree classifier, and a hierarchical early warning label is obtained.

8. A geotechnical infrastructure risk warning system based on a multi-scale damage model, characterized in that, Based on the hierarchical early warning label, the rule engine is used to match the hierarchical early warning plan library to obtain the hierarchical early warning instruction. The system includes: A monitoring data acquisition module is configured to acquire environmental monitoring data of a target rock-soil region based on a preset monitoring scale. A multi-scale model construction module is configured to perform multi-scale coupling analysis modeling based on the environmental monitoring data to obtain a multi-scale damage model. A risk pattern matching module is configured to input the multi-scale damage model into a pre-trained convolutional neural network to extract damage spatiotemporal characteristics, and based on the damage spatiotemporal characteristics, in combination with a preset risk pattern library, a risk pattern is matched and determined. A risk propagation prediction module is configured to input the damage spatiotemporal characteristics and the risk pattern into a pre-trained risk diffusion probability prediction model for prediction to obtain a risk propagation prediction result. A decision boundary construction module is configured to construct a multi-dimensional risk decision boundary by using a nonlinear classifier based on the risk pattern and the risk propagation prediction result. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. A hierarchical early warning generation module is configured to generate a hierarchical early warning instruction based on the multi-dimensional risk decision boundary by using a preset dynamic risk decision rule.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images