Rock-soil body deformation identification method and device based on neural network, and electronic equipment

Abstract

The invention provides a rock-soil body deformation identification method and device based on a neural network and electronic equipment, relates to the technical field of artificial intelligence, and realizes dual guarantee of physical consistency and data authenticity by pre-constructing a multi-source rock-soil fusion tensor. And corresponding multi-source rock-soil fusion tensor matching is carried out on environment factor data corresponding to the monitoring data of the rock-soil body, so that feature recognition with physical and mechanical consistency and field data authenticity is realized. Moreover, by extracting branch features, strain sensitive features and multi-scale spatio-temporal features of data, a complete feature system covering mechanical essence, real-time response and spatio-temporal evolution is formed, and the deformation essence of the rock-soil body is comprehensively captured. And the features are classified cooperatively based on a self-attention mechanism, so that a deformation mode can be accurately determined in a deformation recognition scene adapted to different geological conditions, and rock-soil body deformation analysis with physical consistency and data driving advantages is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus and electronic device for identifying soil and rock deformation based on neural networks. Background Technology

[0002] In the field of geotechnical engineering, engineering activities such as slope stability, tunnel excavation, foundation bearing capacity, and underground space development are highly dependent on the deformation and failure patterns of soil and rock masses. Traditional research methods mainly rely on numerical simulation and field monitoring data for analysis, but both have significant limitations. Numerical simulation can describe the overall deformation process of soil and rock masses from a theoretical and mechanical perspective, but its parameters are often idealized and cannot fully reflect the actual situation under complex environmental conditions. While monitoring data can realistically reflect the field conditions, it often suffers from problems such as locality, noise interference, and uneven sampling. How to organically integrate the two to form a soil and rock deformation analysis that combines physical consistency with the advantages of data-driven approaches is a key scientific and engineering problem that urgently needs to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a method, device, and electronic device for identifying soil and rock deformation based on neural networks, which can realize soil and rock deformation analysis that combines the advantages of physical consistency and data-driven approach.

[0004] In a first aspect, embodiments of the present invention provide a method for identifying deformation of soil and rock masses based on neural networks. The method includes: monitoring the target soil and rock mass to obtain monitoring data of the target soil and rock mass, as well as environmental factor data of the environment in which the monitoring data is located; determining the multi-source soil and rock fusion tensor corresponding to the monitoring data from a pre-constructed set of multi-source soil and rock fusion tensors based on the environmental factor data; constructing the multi-source soil and rock fusion tensor set based on the monitoring data, simulation data, and environmental factor data of the preset soil and rock mass; extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil and rock fusion tensor; the branch features include elastic features and plastic features; performing collaborative classification of the branch features, strain-sensitive features, and multi-scale spatiotemporal features based on a pre-constructed self-attention mechanism to determine the deformation mode corresponding to the target soil and rock mass; and performing deformation analysis on the target soil and rock mass based on the deformation mode.

[0005] In conjunction with the first aspect, the present invention provides a first implementation of the first aspect, wherein the method further includes: acquiring pre-collected numerical simulation tensors, monitoring tensors, and corresponding environmental factor tensors for a preset soil and rock mass; linearly combining the numerical simulation tensors, monitoring tensors, and environmental factor tensors to generate an interaction tensor for the preset soil and rock mass; unifying the dimensions of the numerical simulation tensors, monitoring tensors, and environmental factor tensors, and fusing the dimension-unified numerical simulation tensors, monitoring tensors, and environmental factor tensors with the interaction tensors to generate a multi-source soil and rock fusion tensor set for the preset soil and rock mass.

[0006] In conjunction with the first aspect, this embodiment of the invention provides a second implementation of the first aspect, wherein the step of extracting the branch features of the multi-source soil-rock fusion tensor includes: performing a first convolution processing on the multi-source soil-rock fusion tensor using a preset neural network model to extract the elastic feature tensor corresponding to the multi-source soil-rock fusion tensor; performing a second convolution processing on the monitoring data using the preset neural network model to determine the gating weights corresponding to the multi-source soil-rock fusion tensor; performing range compression on the elastic feature tensor; and performing gating selection on the range-compressed elastic feature tensor based on the gating weights to determine the plastic feature tensor of the multi-source soil-rock fusion tensor; and performing feature concatenation between the elastic feature tensor and the plastic feature tensor to determine the branch features of the multi-source soil-rock fusion tensor. In conjunction with the first aspect, this invention provides a third implementation of the first aspect, wherein the step of extracting the strain-sensitive features of the multi-source soil-rock fusion tensor includes: extracting the strain components of the multi-source soil-rock fusion tensor, determining the gradients of the strain components in the spatial direction and time dimension, and generating a strain gradient vector; identifying the importance of the positional features of the strain gradient vector, and determining the strain-sensitive weights corresponding to the positional features; and using the strain-sensitive weights to perform feature weighting on the multi-source soil-rock fusion tensor to determine the strain-sensitive features of the multi-source soil-rock fusion tensor. In conjunction with the first aspect, this invention provides a fourth implementation of the first aspect, wherein the step of extracting the multi-scale spatiotemporal features of the multi-source soil-rock fusion tensor includes: performing a dilated convolution operation on the multi-source soil-rock fusion tensor using multiple parallel pre-set dilated convolution branches with different expansion rates to capture the spatiotemporal features of the multi-source soil-rock fusion tensor at different scales; and summing the spatiotemporal features to obtain the multi-scale spatiotemporal features of the multi-source soil-rock fusion tensor.

[0007] In conjunction with the first aspect, this invention provides a fifth implementation of the first aspect, wherein the step of determining the deformation mode corresponding to the target soil and rock mass by co-classifying branch features, strain-sensitive features, and multi-scale spatiotemporal features based on a pre-constructed self-attention mechanism includes: converting branch features, strain-sensitive features, and multi-scale spatiotemporal features into global feature vectors; mapping the global feature vectors to a query space to generate a query representation matrix; performing dot product processing on the query representation matrix using a scaling dot product attention mechanism to determine the mode association graph corresponding to the global feature vectors; the mode association graph includes the deformation mode similarity between samples; performing matrix multiplication processing on the mode association graph and the global feature vectors to generate a fusion output; and determining the deformation mode corresponding to the target soil and rock mass based on the fusion output.

[0008] In conjunction with the first aspect, this invention provides a sixth implementation of the first aspect, wherein the method further includes: extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor through a preset neural network model; wherein the preset neural network model is trained based on a preset physical information hybrid loss function; the method further includes: training the preset neural network model using a preset training sample set to determine an initial prediction output; calculating the cross-entropy loss of the preset neural network model based on the initial prediction output, and generating a dynamic stiffness matrix based on the initial prediction output; the dynamic stiffness matrix is ​​determined based on the material constitutive relation corresponding to the training samples and the finite element discretization method of the preset neural network model; constructing the physical constraint residual loss of the preset neural network model based on the stiffness matrix; and weightedly fusing the physical constraint residual loss and the cross-entropy loss to generate a physical information hybrid loss function. In conjunction with the first aspect, the present invention provides a seventh implementation of the first aspect, wherein the above method further includes: obtaining pre-constructed multi-source soil-rock fusion tensor training samples; applying physical constraints to the multi-source soil-rock fusion tensor training samples based on the linear physical response of the soil-rock bodies corresponding to the multi-source soil-rock fusion tensor training samples to generate a physical constraint tensor; and constructing a training sample set based on the physical constraint tensor.

[0009] Secondly, embodiments of the present invention provide a soil and rock deformation identification device based on a neural network. The device includes: a data monitoring module for monitoring a target soil and rock mass, obtaining monitoring data of the target soil and rock mass, and environmental factor data of the environment in which the monitoring data is located; a data processing module for determining the multi-source soil and rock fusion tensor corresponding to the monitoring data from a pre-constructed multi-source soil and rock fusion tensor set based on the environmental factor data; the multi-source soil and rock fusion tensor set is constructed based on the monitoring data, simulation data, and environmental factor data of the preset soil and rock mass; a feature extraction module for extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil and rock fusion tensor; the branch features include elastic features and plastic features; an execution module for collaboratively classifying the branch features, strain-sensitive features, and multi-scale spatiotemporal features based on a pre-constructed self-attention mechanism to determine the deformation mode corresponding to the target soil and rock mass; and an output module for performing deformation analysis on the target soil and rock mass based on the deformation mode.

[0010] Thirdly, embodiments of the present invention provide an electronic device including a processor and a memory, the memory storing machine-executable instructions executable by the processor, the processor executing the machine-executable instructions to implement the methods of any of the above embodiments.

[0011] The embodiments of this invention bring the following beneficial effects: This invention provides a method, device, and electronic device for identifying soil and rock deformation based on neural networks. By pre-constructing a multi-source soil-rock fusion tensor, it achieves dual assurance of physical consistency and data authenticity. Furthermore, it performs corresponding multi-source soil-rock fusion tensor matching on environmental factor data corresponding to the monitoring data of the soil and rock mass, achieving feature identification with both physical and mechanical consistency and on-site data authenticity. Moreover, by extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features from the data, a complete feature system covering the mechanical essence, real-time response, and spatiotemporal evolution is formed, comprehensively capturing the essence of soil and rock deformation and achieving a comprehensive characterization of the deformation essence. Based on a self-attention mechanism, the above features are collaboratively classified, enabling accurate determination of deformation patterns in deformation identification scenarios adaptable to different geological conditions (cohesive soil, rock mass, sand and gravel), achieving soil and rock deformation analysis with both physical consistency and data-driven advantages.

[0012] Other features and advantages of the present invention will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above. To make the above-described objects, features, and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating a method for identifying soil and rock deformation based on a neural network, provided in an embodiment of the present invention; Figure 2 A flowchart of another method for identifying soil and rock deformation based on neural networks provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the recognition accuracy of different methods provided in an embodiment of the present invention; Figure 4 A schematic diagram illustrating the comparison of prediction errors provided in an embodiment of the present invention; Figure 5 A schematic diagram illustrating the prediction error distribution of different methods on various strain components, provided in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the effect of physical constraint strength on recognition performance, provided by an embodiment of the present invention. Figure 7 A schematic diagram illustrating the performance changes of different methods under gradually increasing noise levels, as provided in an embodiment of the present invention. Figure 8 A schematic diagram of the structure of a soil and rock deformation identification device based on a neural network provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0016] This invention provides a method, device, and electronic device for identifying soil and rock deformation based on neural networks, which can achieve soil and rock deformation analysis that combines physical consistency with the advantages of data-driven approaches.

[0017] To facilitate understanding, the method for identifying soil and rock deformation based on a neural network, provided in an embodiment of the present invention, will first be described, referring to... Figure 1 The method includes the following steps: Step S102: Data monitoring is performed on the target rock and soil mass to obtain monitoring data of the target rock and soil mass and environmental factor data of the environment in which the monitoring data is located.

[0018] By monitoring the target soil and rock mass to be deformed, the monitoring data can reflect the true state of the soil and rock mass, thereby determining its spatial displacement and degree of deformation. Furthermore, this embodiment of the invention also simultaneously determines the external influence data corresponding to the environment in which the soil and rock mass is located, i.e., environmental factor data, establishing a spatiotemporal match between environmental factor data and monitoring data, facilitating accurate classification of deformation patterns under the tracing of deformation factors. The monitoring data can be collected through on-site deployed sensor equipment, including strain gauge azimuth and burial depth (determined by installation records), temperature (measured by a temperature sensor), moisture content (measured by a humidity sensor), strain values ​​(x-direction normal strain, y-direction normal strain, z-direction normal strain, xy shear strain), and displacement (x-direction displacement, y-direction displacement, z-direction displacement). This data is acquired in real time by strain gauges and displacement sensors. The environmental factor data comes from external observation equipment, including rainfall (recorded by meteorological station data), groundwater level (acquired from water level monitoring well data), and vibration intensity (measured by a vibration sensor). Among them, the dimensions of each attribute of the above multi-source data can be eliminated by normalization, or the units can be uniformly removed after numerical collection according to the specified units to maintain the consistency of dimensions.

[0019] Step S104: Based on environmental factor data, determine the multi-source soil-rock fusion tensor corresponding to the monitoring data from the pre-constructed multi-source soil-rock fusion tensor set.

[0020] Traditional methods for identifying soil and rock deformation by establishing theoretical numerical simulation data or using field monitoring data often fail to accurately analyze deformation patterns due to variations in environmental conditions. This invention pre-establishes an organically integrated dataset (i.e., a multi-source soil and rock fusion tensor set) comprising theoretical simulation data, field monitoring data, and environmental impact data. This allows field monitoring data to be integrated with theoretical and historical field data under a unified environmental dimension, preserving the authenticity of the monitoring data while using simulation data to compensate for the localization of monitoring, thus overcoming the shortcomings of idealized or localized single-data sources.

[0021] Furthermore, using environmental factor data as a matching basis, fusion tensors highly consistent with the field environment (such as actual rainfall and groundwater depth) are selected from the tensor set. This allows for pattern recognition under the real deformation mechanism based on multi-source soil-rock fusion tensors that fully cover the mechanical essence, external driving forces, and field manifestations of deformation, improving the efficiency and accuracy of deformation cause tracing and pattern recognition. Numerical simulation data can be generated using numerical simulation software (such as finite element analysis tools), including attributes such as mesh size, element type, elastic modulus, Poisson's ratio, cohesion, internal friction angle, permeability coefficient, and load step size. These attributes are derived from material tests or literature values, and the software simulates the soil-rock response under different working conditions. The collected data can then be used to construct a database for model training. Each sample corresponds to an independent soil-rock case, containing N samples and time-series data with T time steps. Each sample can be assigned a deformation mode category label, including four categories: elastic deformation, elastoplastic deformation, creep, and shear failure. These categories correspond to the typical deformation behavior of soil and rock masses, ensuring that the training sample set can support the subsequent model's task of identifying deformation modes. The data labeling process can be based on expert knowledge or historical deformation records.

[0022] Step S106: Extract the branching features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor.

[0023] After determining the multi-source soil-rock fusion tensor corresponding to the current monitoring data, multi-dimensional feature extraction can be performed based on the current data to comprehensively capture deformation influencing factors from aspects such as mechanical properties (e.g., branching features, including elastic and plastic features), deformation correlation (e.g., strain-sensitive features), and spatiotemporal distribution (e.g., multi-scale spatiotemporal features). By screening key features with mechanical correlation and trend indication of soil-rock deformation from massive data, redundant data interference can be avoided. Among them, elastic features are used to characterize the reversible deformation capacity and elastic mechanical properties of soil-rock, while plastic features are used to characterize the permanent deformation capacity and damage accumulation characteristics of soil-rock, and to characterize the evolution process of soil-rock from reversible elastic deformation to irreversible plastic deformation. Based on soil-rock mechanics theory, two types of features can be extracted from the attribute dimension through tensor decomposition algorithms (e.g., PARAFAC decomposition). Strain-sensitive features are used to characterize the real-time degree, development trend, and instability risk of soil-rock deformation, focusing on strain parameters directly related to deformation. Shear deformation risk can be assessed by calculating the maximum shear strain at each spatiotemporal point in the tensor, and deformation rate can be reflected by the axial strain rate. Sensitive data points with strain values ​​exceeding the standard threshold can be screened to enhance the trend indicativeness of the features. Multi-scale spatiotemporal features are used to characterize the evolution rhythm of soil and rock deformation in different time dimensions and the distribution differences in different spatial dimensions. By conducting cross-scale analysis, the essence of deformation can be fully presented. Multi-resolution analysis methods can be used to decompose short-term (1-7 days), medium-term (1-3 months), and long-term (3-12 months) deformation features in the time dimension through wavelet transform. In the spatial dimension, through downsampling (taking one representative point every 5 monitoring points) and interpolation (Kriging interpolation to fill blank areas), spatial deformation features of local points (within 10m), regions (within 100m), and the whole (the entire engineering area) can be obtained, ultimately forming a multi-scale spatiotemporal feature matrix.

[0024] Step S108: Based on the pre-constructed self-attention mechanism, branch features, strain-sensitive features and multi-scale spatiotemporal features are collaboratively classified to determine the deformation mode corresponding to the target rock and soil mass.

[0025] Step S110: Based on the deformation mode, perform deformation analysis on the target rock and soil mass.

[0026] By employing a self-attention mechanism to achieve collaborative classification, the deformation patterns of target soil and rock masses can be determined. This approach enables the synergistic fusion of mechanical properties, deformation response, and spatiotemporal distribution characteristics based on the inherent connections among three types of features, avoiding the one-sidedness caused by single-feature analysis. For example, a higher weight can be assigned to the combination of a sudden increase in strain rate and concentrated local displacement at the toe of the slope. Furthermore, the core direction of the deformation pattern can be anchored by focusing on corresponding key information. For instance, if the classification result indicates creep deformation dominated by long-term time characteristics, priority should be given to analyzing the time-dependent properties of the soil and rock mass (such as viscosity coefficient) and long-term environmental effects (such as seasonal groundwater level changes). This achieves precise positioning from experience-based judgment to data-driven analysis, enabling accurate deformation analysis of soil and rock masses that combines physical consistency with the advantages of data-driven approaches.

[0027] This invention pre-constructs a multi-source soil-rock fusion tensor to ensure both physical consistency and data authenticity. It then performs multi-source soil-rock fusion tensor matching on environmental factor data corresponding to monitoring data of the soil-rock mass, achieving feature identification with both physical and mechanical consistency and on-site data authenticity. Furthermore, by extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features from the data, a complete feature system covering the mechanical essence, real-time response, and spatiotemporal evolution is formed, comprehensively capturing the deformation essence of the soil-rock mass and achieving a holistic characterization of deformation. Based on a self-attention mechanism, the aforementioned features are collaboratively classified, enabling accurate determination of deformation patterns in deformation identification scenarios adaptable to different geological conditions (cohesive soil, rock mass, sand and gravel), achieving soil-rock mass deformation analysis that combines physical consistency and data-driven advantages.

[0028] Furthermore, with the development of deep learning, more and more research attempts to apply neural networks to soil and rock deformation prediction and disaster identification. However, most existing methods are limited to utilizing a single data source, typically employing simple feature concatenation to process multi-source information, leading to the neglect of potential physical coupling relationships between cross-source data. This invention combines a projection matrix for unified dimensional mapping, adaptively capturing the nonlinear correlation between cross-source attributes to solve the feature mismatch problem caused by traditional simple concatenation methods. In addition, the elastic and plastic deformations of soil and rock are highly nonlinear and time-varying, making it difficult for existing methods to effectively distinguish between the two types of features, resulting in feature coupling and information loss. Especially in the early stages of failure, weak local high-strain signals are often masked, making timely identification difficult. This invention uses a dual-branch decoupled encoder combined with a gating mechanism to effectively separate elastic and plastic features. Simultaneously, this invention introduces a strain-sensitive weight matrix, automatically amplifying the importance of local high-strain regions based on strain gradients, thereby highlighting key deformation signals and solving the problems of feature confusion and weak signal capture.

[0029] In summary, based on the above embodiments, this invention also provides another method for identifying soil and rock deformation based on neural networks. Figure 2A flowchart of an embodiment of the present invention is shown, with reference to Figure 2 The method includes the following steps: Step S202: Data monitoring is performed on the target rock and soil mass to obtain monitoring data of the target rock and soil mass and environmental factor data of the environment in which the monitoring data is located.

[0030] Step S204: Based on environmental factor data, determine the multi-source soil-rock fusion tensor corresponding to the monitoring data from the pre-constructed multi-source soil-rock fusion tensor set.

[0031] Based on the above embodiments, geotechnical data exhibits multi-scale and heterogeneous characteristics. Numerical simulation data contains high-dimensional field variables, while monitoring data is sparse in time and space. Conventional processing methods typically employ simple splicing, which easily leads to feature mismatch and makes it difficult to establish effective correlations across data sources, thus affecting the accuracy of deformation identification. This invention constructs numerical simulation tensors, monitoring tensors, and environmental factor tensors, and utilizes learnable projection matrices and interaction tensors for data fusion, thereby mapping multi-source data to a unified dimension, enhancing the correlation across data sources, and solving the feature mismatch problem. The specific steps are as follows: 1) Obtain the numerical simulation tensor, monitoring tensor, and corresponding environmental factor tensor pre-collected for the preset soil and rock mass.

[0032] Numerical simulation data, monitoring data, and environmental factors are organized into three-dimensional tensors, including: constructing the numerical simulation tensor. , dimension , The number of attributes is simulated numerically. The numerical simulation attributes include: mesh size, element type, elastic modulus, Poisson's ratio, cohesion, internal friction angle, permeability coefficient, and load step size; a monitoring tensor is constructed. , dimension , To monitor the number of attributes, The monitored attributes include: strain gauge azimuth, burial depth, temperature, moisture content, strain value, displacement, settlement rate, and tilt angle change. The strain values ​​include normal strain in the x-direction, normal strain in the y-direction, normal strain in the z-direction, and xy shear strain. The displacements include displacement in the x-direction, y-direction, and z-direction. An environmental factor tensor is constructed. , dimension , The number of environmental factor attributes. Environmental factors: rainfall, groundwater level, vibration intensity; For the sample size, For time step.

[0033] In practical implementation, the data sources, attributes, collection methods, significance, and units / dimensions of multi-source data are introduced as follows:

[0034]

[0035] 2) Linearly combine the numerical simulation tensor, monitoring tensor, and environmental factor tensor to generate the interaction tensor of the preset soil and rock mass.

[0036] By linearly combining all attribute vectors of the numerical simulation tensor, monitoring tensor, and environmental factor tensor at each sample and time step through fusion weight vectors, and then applying a hyperbolic tangent activation function for nonlinear transformation, an interaction tensor is generated. The elements of the interaction tensor represent the cross-source interaction strength, thereby enhancing the correlation between different data sources, as expressed below:

[0037] In the formula, The interaction tensor represents the feature value of the j-th attribute of the i-th sample at time t, which is used to characterize the cross-source interaction strength; Let J be the fusion weight vector of the j-th attribute, with dimension J. , is a trainable parameter used for linear combination of multi-source features; This represents a vector concatenation operation that combines three components into a single vector. This is the hyperbolic tangent activation function, used to implement nonlinear transformations and limit the output range; Let represent the vector of all attributes at time t for the i-th sample, i.e. Represents numerical simulation tensor The vector of all attributes of the i-th sample at time t. Represents the monitoring tensor The vector of all attributes of the i-th sample at time t. Represents environmental factor tensors The vector of all attributes of the i-th sample at time t; i is the first dimension index of the tensor, which also represents the row index; j is the second dimension index of the tensor, which also represents the column index; t is the third dimension index of the tensor, which also represents the time step index.

[0038] It should be noted that the interaction tensor Each element By dynamically quantifying the correlation strength between different data source attributes through nonlinear activation, such as the coupling relationship between elastic modulus and displacement sensors, the implicit influence of environmental factors on material parameters can be adaptively corrected. For example, when the environmental factor tensor... When groundwater levels change abruptly, the interaction tensor automatically enhances the correlation weight between the permeability coefficient and the strain value, whereas conventional methods require manual setting of correction coefficients, which cannot achieve the same effect.

[0039] 3) Unify the dimensions of the numerical simulation tensor, monitoring tensor, and environmental factor tensor, and then fuse the unified numerical simulation tensor, monitoring tensor, and environmental factor tensor with the interaction tensor to generate a multi-source soil-rock fusion tensor set for the preset soil-rock body.

[0040] In practical implementation, a projection matrix can be used to map the numerical simulation tensor, monitoring tensor, and environmental factor tensor along the attribute dimension to a unified dimension. Then, tensor-matrix multiplication is used to add the mapped tensor to the interaction tensor to obtain a fused tensor, capturing the nonlinear interactions between data, represented as:

[0041] In the formula, This represents a fusion tensor used to integrate information from multiple sources. The modulo product of a tensor and a matrix (Mode-2-Product), also known as the 2-modulo product, is a tensor-matrix multiplication performed along the attribute dimension. The projection matrix of the numerical simulation data has dimensions of . Mapping simulated data to a unified dimension provides trainable parameters. To unify attribute dimensions, It is the sum of the number of numerical simulation attributes and the number of monitored attributes; The projection matrix of the monitoring data has dimensions of . Mapping monitoring data to a unified dimension provides trainable parameters. This is the projection matrix of the environmental factor data, with dimensions of . Mapping environmental factor data to a unified dimension provides trainable parameters. For the interaction tensor, the dimension is It is used to enhance cross-source correlations and capture nonlinear interactions between data. It is an interaction tensor In the The sample index, the first The index of the attribute and the first The element value at each time step index, i.e., the interaction tensor. It is made by all The tensor formed by this. It should be noted that the projection matrix... , , It can bridge the dimensional gap problem by mapping attribute dimensions to a unified space while preserving the interpretability of the original physical quantities. For example, it can monitor the "xy shear strain" in a tensor through a projection matrix. After mapping, it still maintains a mechanical connection with the "cohesion" of the numerical simulation in the same dimension, while conventional feature extraction would destroy this physical correspondence.

[0042] Furthermore, the embodiments of the present invention also compare the recognition accuracy of different methods on various deformation modes to verify the recognition performance advantages of this technology on different deformation modes of soil and rock (elastic deformation, elastoplastic deformation, creep, shear failure). Figure 3 This diagram illustrates a comparison of recognition accuracy for different methods. By comparing this invention with three conventional techniques (multi-source splicing method, single-branch convolutional neural network, and physical constraint network), it highlights the comprehensive superiority of this technology in multi-pattern recognition. The vertical axis represents recognition accuracy (%), with higher values ​​indicating better performance. The experiment used the same dataset (containing samples of four deformation patterns) and compared the four methods under the same training conditions. In the experiment, this technology fully implemented multi-source tensor fusion, physical constraint embedding, and dual-branch decoupled encoding. The multi-source splicing method simply spliced ​​numerical simulation / monitoring / environmental data and then input it into a single-branch network. The single-branch convolutional neural network used only a basic convolutional network trained with monitoring data. The physical constraint network only embedded strain coordination constraints but did not achieve multi-source fusion. Experimental results show that this technology (blue bars) significantly outperforms other methods in recognizing all four deformation patterns. Conventional methods performed weakest in creep and shear failure recognition (lowest bar height), as these patterns require collaborative analysis of multi-source data. The tensor fusion mechanism of this technology effectively solves the feature mismatch problem. The physical constraint network (purple pillars) performs reasonably well in elastic deformation because physical constraints are effective for simple modes, but its effectiveness drops sharply in nonlinear deformation (such as creep), highlighting the necessity of the dual mechanism of "physical constraint + multi-source fusion" in this technology.

[0043] Step S206: Extract the branching features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor.

[0044] 1) Extracting Branch Features: Deformation features in soil and rock masses exhibit strong coupling characteristics. For example, elastic deformation and plastic flow often occur simultaneously and influence each other. Conventional convolutional neural networks, using a single-branch structure, cannot effectively separate these coupled features, easily leading to feature representation confusion and reduced classification accuracy. This invention uses a dual-branch structure to extract two types of features separately (e.g., elastic and plastic features), combined with a gating mechanism to achieve feature separation, decoupling coupled features and improving the clarity of feature representation.

[0045] The specific steps are as follows: Step S1: The first convolution of the multi-source soil-rock fusion tensor is performed using a preset neural network model to extract the elastic feature tensor corresponding to the multi-source soil-rock fusion tensor.

[0046] A 3D convolution operation is performed on the physical constraint tensor using an elastic convolution kernel weight tensor and a bias vector. Then, a linear rectified activation function is applied to extract the elastic feature tensor, which is used to characterize the recoverable elastic deformation mode, as follows:

[0047] In the formula, It represents the elastic characteristic tensor, which characterizes the recoverable elastic deformation mode; It is a linear rectified activation function used to achieve nonlinear transformation while preserving sparsity; The elastic convolution kernel weight tensor has a dimension of . Specifically designed to capture recoverable elastic deformation features, these are trainable parameters; This represents a 3D convolution operation used to extract spatiotemporal features; The bias vector of the elastic branch is used to adjust the baseline value of the output feature and is a trainable parameter. The size of the elastic convolution kernel, such as, Controlling the size of the receptive field; The number of channels for elastic feature output controls feature richness; for example, setting... .

[0048] Step S2: Perform a second convolution process on the monitoring data using a preset neural network model to determine the gating weights corresponding to the multi-source soil-rock fusion tensor.

[0049] Step S3: Perform range compression on the elastic feature tensor, and based on the gating weight, perform gating selection on the range-compressed elastic feature tensor to determine the plastic feature tensor of the multi-source soil-rock fusion tensor.

[0050] A 3D convolution operation is performed on the physical constraint tensor using a plastic convolution kernel weight tensor and a bias vector. Then, a sigmoid activation function is applied to output the gating weights. Simultaneously, the elastic feature tensor is compressed using a hyperbolic tangent activation function. Finally, gating selection is achieved through a Hadamard product to extract the plastic feature tensor, which is used to characterize the permanent plastic deformation mode, represented as follows:

[0051] In the formula, It represents the plastic characteristic tensor, characterizing the permanent plastic deformation mode; This is the Sigmoid activation function, used to output the gate weights; The plastic convolution kernel weight tensor has dimension 1. Specifically designed to capture features of permanent plastic deformation, it is a trainable parameter; The bias vector of the plastic branch is used to adjust the baseline value of the output feature and is a trainable parameter. This represents the Hadamard product, used to implement gating selection of features; It is a hyperbolic tangent activation function used to compress the range of elastic eigenvalues ​​to adapt to gating; The size of the plastic convolution kernel, such as, Controlling the size of the receptive field; The number of channels output for plastic features controls feature richness; for example, setting... .

[0052] Step S4: Perform feature splicing on the elastic feature tensor and the plastic feature tensor to determine the branch characteristics of the multi-source soil-rock fusion tensor.

[0053] The elastic and plastic feature tensors are concatenated along the channel dimension, represented as follows: This leads to the formation of decoupled coding features. The decoupled encoding features simultaneously contain both elastic and plastic feature information; among them, To decouple the encoded features, the number of channels is It characterizes the output features of the dual-branch decoupled encoder, while also containing elastic and plastic feature information; This represents a vector concatenation operation that combines two components into a single vector. It should be noted that while conventional dual-branch networks extract features in simple parallel processes, the strong coupling of the elasto-plastic deformation of soil and rock masses leads to feature confusion. This invention addresses this by... The range of compressive elastic eigenvalues ​​is up to Then, a Hadamard product is performed with the plastic gating weights to achieve a subtraction operation in the feature space. When elastic deformation dominates, such as in the early stages of loading, the gating value approaches 1, fully preserving the plastic characteristics. When plastic flow intensifies, such as in the yielding stage, the gating value automatically decays in the shear band region, suppressing the interference of elastic characteristics on the plastic branches. Based on this, it can be found that creep and elastic recovery can be accurately separated. Under cyclic loading, the gating mechanism follows... The oscillating synchronous switching of the term uses a plastic channel to simulate the Bauschinger effect. It should also be noted that the elastic branch uses the convolution kernel weight tensor. and bias vector It is specifically designed to capture recoverable elastic deformation; the plastic branch uses convolution kernel weight tensors. and bias vector This architecture is specifically designed to capture irreversible plastic deformation. These parameters are updated independently during training via backpropagation and gradient descent, allowing the network to learn to respond to different types of deformation patterns. Furthermore, different activation functions guide the network to learn different types of responses. The elastic branch uses a linear rectified activation function, suitable for capturing non-negative, sparse, and recoverable elastic responses, consistent with the physical characteristics of elastic deformation. The plastic branch uses the Sigmoid function as gate weights, which are then multiplied by the elastic features compressed by the hyperbolic tangent activation function using the Hadamard product. The Sigmoid function's output is a gate value between 0 and 1, used to suppress or enhance the plastic deformation-related parts of the elastic features, achieving selective extraction of plastic deformation. Simultaneously, it makes the plastic branch dependent on the output of the elastic branch, thus achieving elastoplastic decoupling in the feature space.

[0054] 2) Addressing Strain-Sensitive Features: During the deformation of soil and rock masses, the contribution of strain at different spatial locations and time points to the overall deformation varies significantly. Conventional average pooling or max pooling methods cannot distinguish these differences, leading to the dilution of strain information in key areas and the loss of important features, thus affecting classification accuracy. This invention adaptively weights features by calculating the strain gradient tensor and constructing a strain-sensitive weight matrix, highlighting the feature importance of high strain gradient regions, thereby enhancing the feature representation of key areas. The specific steps are as follows: Step S1: Extract the strain components of the multi-source soil-rock fusion tensor, determine the gradient of the strain components in the spatial direction and time dimension, and generate the strain gradient vector.

[0055] Strain components are extracted from the input data, and their gradients in the spatial and temporal dimensions are calculated to form a strain gradient vector, which characterizes the intensity and direction of strain change, and is expressed as:

[0056] In the formula, Indicates the first The strain gradient vector at position t at time t is used to characterize the intensity and direction of strain change; It represents the spatial gradient of strain in the x-direction. The gradient is calculated using discrete difference methods, such as central difference, in both spatial and temporal dimensions, reflecting the rate of change of strain in the horizontal direction. It represents the spatial gradient of strain in the y-direction. The gradient is calculated using discrete difference methods, such as central difference, in both spatial and temporal dimensions, reflecting the rate of change of strain in the vertical direction. It represents the gradient of strain in the time dimension. The gradient calculation uses discrete difference methods, such as central difference, to perform the calculation in both spatial and temporal dimensions, reflecting the rate of change of strain over time. This represents a vector concatenation operation that combines three components into a single vector. It should be noted that... Item and The term characterizes the spatial gradient and can identify potential slip surfaces. The term characterizes the time gradient and is associated with rate-sensitive properties, such as clay creep.

[0057] Step S2: Identify the importance of the positional features of the strain gradient vector and determine the strain sensitivity weights corresponding to the positional features.

[0058] Specifically, based on the L2 norm of the strain gradient vector, the strain sensitivity weights at each location and time can be calculated using exponential and softmax functions to quantify the importance of the feature at that location. Regions with larger strain gradients receive higher weights, as expressed below:

[0059] In the formula, Indicates the first The strain-sensitive weight at position t is used to quantify the importance of the feature at that position; For sensitivity factors, such as, It is used to adjust the sharpness of the weight distribution and control the degree of attention concentration; Represents the L2 norm; This represents an exponential function, used to amplify gradient differences; Indicates the first Position No. The strain gradient vector at time t; Indicates a spatial location index; Indicates difference from Spatial location index; This is to distinguish it from the time step index t.

[0060] Step S3: Use strain-sensitive weights to perform feature weighting on the multi-source soil-rock fusion tensor to determine the strain-sensitive features of the multi-source soil-rock fusion tensor.

[0061] The strain-sensitive weight tensor is element-wise multiplied with the feature tensor (such as the multi-source soil-rock fusion tensor mentioned above, or alternatively, the branched features mentioned above) to achieve adaptive weighting of the features, resulting in a weighted feature tensor that highlights the characteristics of high strain gradient regions. The following formula is for branched features. (That is, the decoupling coding feature) is explained as follows:

[0062] In the formula, This represents the weighted feature tensor, highlighting the characteristics of the high strain gradient region; For the strain-sensitive weight tensor, the dimension is , It is a strain-sensitive weight tensor In the Position No. The element value at time t, i.e., the strain-sensitive weight tensor. It is made by all The tensor that makes up the tensor; The spatial height of the weighted feature tensor; The spatial width of the weighted feature tensor; It is the total time step, which is determined based on the length of the time series of the input data, such as the number of time points in monitoring data or numerical simulation data; This represents element-wise multiplication of tensors. It should be noted that, through... The quantitative strain change rate corresponds physically to the localization intensity of deformation, and is an exponential function. Items and Sensitivity Factors The combination of these factors produces a superlinear amplification effect. When the strain gradient at a certain point exceeds a critical value, such as... Its weight rises sharply to become dominant, which can solve the problem of capturing weak deformation signals in the early stage of creep and achieve automatic focusing on the accelerated creep point.

[0063] 3) Addressing Multi-Scale Spatiotemporal Characteristics: Soil and rock deformation exhibits significant multi-scale spatiotemporal characteristics, such as the coexistence of instantaneous settlement and long-term creep, and the interaction of local shearing and global slip. Conventional single-scale convolutional neural networks cannot simultaneously capture these features at different scales, resulting in insufficient ability to identify complex deformation patterns. This invention constructs a multi-scale convolutional structure and performs multi-scale convolution operations, using hollow convolutional kernels with different expansion rates to extract multi-scale spatiotemporal features in parallel, and then fuses these features to comprehensively characterize the complex deformation behavior of soil and rock masses. The specific steps are as follows: Step S1 involves performing a dilated convolution operation on the multi-source soil-rock fusion tensor using multiple pre-set dilated convolution branches in parallel with different expansion rates, thereby capturing the spatiotemporal characteristics of the multi-source soil-rock fusion tensor at different scales.

[0064] This can be achieved by employing three parallel dilated convolution branches, each using a different dilation rate to perform dilated convolution operations, in order to capture spatiotemporal features at different scales, as shown below. .in, Indicates the first The output of the dilated convolution operation performed by each dilated convolution branch; For the first The dilation rate of each dilated convolution branch, taking a value of... These correspond to three receptive field scales: small, medium, and large. The kernel size is, for example, The three branches maintain the same size but have different receptive fields; The branch index of the dilated convolution has a value of This can be achieved by processing the aforementioned multi-source soil-rock fusion tensor, and further feature extraction can be performed on the aforementioned strain-sensitive characteristics. The above formula is based on strain-sensitive characteristics. (That is, the aforementioned weighted feature tensor) Explanation. Specifically, the receptive field is expanded through dilated convolution. Corresponding to the grain scale, it captures the rearrangement of clay particles. Match joint spacing and monitor the opening of rock mass structural surfaces. Covers engineering scales, such as tunnel convergence deformation.

[0065] Step S2 involves summing the spatiotemporal features to obtain the multi-scale spatiotemporal features of the multi-source soil-rock fusion tensor.

[0066] The outputs of each dilated convolution branch are summed to obtain a multi-scale fused feature tensor, which integrates multi-scale information to comprehensively characterize the complex deformation behavior of soil and rock masses, and is represented as:

[0067] In the formula, This represents a multi-scale fusion feature tensor that integrates multi-scale information, with dimensions of . ; The number of channels output by multi-scale convolution. It should be noted that the "interval sampling" characteristic of dilated convolution can match the non-homogeneity of soil and rock media; in soil-rock mixtures, The convolution kernel automatically skips rocky regions and focuses on the soil matrix, significantly improving the feature signal-to-noise ratio compared to conventional uniform convolution. A uniform kernel size ensures a balance of parameters across different branches, preventing large-scale branches from dominating training. Further, proceed according to step S108. Among the deformation types of soil and rock masses, there are physical correlations, such as elastic deformation, elastoplastic deformation, creep, and shear failure. Conventional single-label classification methods treat each category as independent, ignoring the cooperative and exclusive relationships between patterns, leading to a decline in classification performance for complex composite deformation patterns. This invention improves the ability to identify composite deformation patterns by using feature global pooling and query mapping, pattern association graph construction, and cooperative classification prediction, leveraging a self-attention mechanism to model the correlations between deformation patterns and performing cooperative classification. See steps S208-S212 below for details.

[0068] Step S208: Convert the branch features, strain-sensitive features, and multi-scale spatiotemporal features into global feature vectors, map the global feature vectors to the query space, and generate a query representation matrix.

[0069] In one implementation, feature transformation can be performed on the three types of features (i.e., branch features, strain-sensitive features, and multi-scale spatiotemporal features) separately. Alternatively, feature transformation can be performed after further determining the strain-sensitive features among the branch features and the multi-scale spatiotemporal features. In a specific implementation, high-dimensional spatiotemporal features are compressed into sample-level global feature vectors by performing global average pooling on the corresponding data. Then, the global feature vectors are multiplied by a trainable query projection matrix to map to the query space, resulting in a query representation matrix.

[0070] Multi-scale fusion feature tensor To explain, global average pooling calculates the average value of all spatial locations and time steps for each channel, resulting in a global pooled feature vector. , dimension Furthermore, the global pooling feature vectors are... With dimension Query projection matrix Multiply to obtain the query representation matrix. , dimension This allows for the mapping of global features of each sample to the query space; where, To query the projection matrix, which is a trainable parameter with dimension 1. This maps features to the query space. To query the dimensions of the space, such as, .

[0071] Step S210: The query representation matrix is ​​processed by a scaling dot product attention mechanism to determine the pattern association graph corresponding to the global feature vector; the pattern association graph includes the similarity of deformed patterns between samples.

[0072] Combining the above steps, the similarity matrix between samples is calculated using the query representation matrix. Normalized attention weights are obtained through a scaled dot product attention mechanism and a softmax function. A pattern association graph is then constructed to express the association strength between different samples, as follows: In the formula, Query representation matrix Transpose of; This is a pattern association graph with dimension 1. The first pattern association graph Line 1 Column elements Indicates the first The sample and the first The strength of the association between individual samples; The row index of the pattern association graph is the sample index; The column index for the schema association graph is the sample index; The normalized exponential function ensures that the sum of all elements in each row is 1, forming a probability distribution. It should be noted that the matrix is ​​represented by a query function. Mapping samples to the physical semantic space, their dot product The similarity of deformation patterns between samples is calculated and divided by . The scaling operation of items can balance the problem of similarity numerical drift in high-dimensional space, making Stability reflects the physical nature of things, such as the similar instability modes of two slopes.

[0073] Step S212: Perform matrix multiplication on the pattern association diagram and global feature vector to generate a fusion output, and determine the deformation mode corresponding to the target soil and rock mass based on the fusion output.

[0074] In practical implementation, the pattern association graph is multiplied by the global pooled features to achieve information propagation and fusion based on pattern associations. Furthermore, the fused features can be input into a multilayer perceptron classifier, represented as... Output classification matrix .in, It is a multilayer perceptron classifier; The output matrix is ​​for classification, with dimension 1. Each row represents the predicted probability of a sample belonging to each deformation mode; The number of categories for the deformation pattern is 4. In one implementation, the multilayer perceptron classifier employs a two-layer perceptron structure, with an input layer dimension of... The hidden layer has a dimension of 64, and the output layer has a dimension of [missing information]. It corresponds to four deformation modes: elastic, elastoplastic, creep, and shear failure. The activation function is the ReLU activation function.

[0075] It should be noted that, The project can achieve knowledge transfer. When a new sample lacks data on a certain type of deformation, the association graph automatically refers to the characteristics of similar samples, such as predicting the long-term deformation of an unknown sample through a known creep sample.

[0076] Step S214: Based on the deformation mode, perform deformation analysis on the target rock and soil mass.

[0077] Furthermore, the embodiments of the present invention also compare the displacement prediction errors at different time steps to verify the stability of this technology in the time-series prediction of soil and rock displacement. Figure 4 A schematic diagram comparing prediction errors according to an embodiment of the present invention is shown. Figure 4The vertical axis represents the displacement prediction error (mm), with lower values ​​indicating higher accuracy. The experiment compares the errors of the two methods over 100 hours to demonstrate the improvement in long-term prediction accuracy through physical constraint embedding. A typical slope case was selected, and its z-direction displacement (meters) was monitored over 100 hours. The absolute errors (mm) of the predicted and actual displacements of the two methods were compared, with a time step of 5 hours (horizontal axis: hours). Experimental results show that the proposed technique (blue line) maintains low error fluctuation throughout (the curve is close to the horizontal axis), while the error of the multi-source stitching method (orange line) increases significantly and fluctuates drastically over time. During the 60-80 hour period (potential instability stage), the error of the proposed technique remains stable, while the comparative method exhibits a spike error, proving that the physical stiffness matrix effectively suppresses prediction bias under instability conditions. Furthermore, the area under the curve of the proposed technique (blue filled area) is much smaller than that of the comparative method, reflecting that it reduces the cumulative error effect through strain coordination constraints.

[0078] Furthermore, this embodiment of the invention extracts branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor through a preset neural network model, achieving a fusion of data-driven and model-driven capabilities. The data-driven capability, jointly realized by multi-source tensor fusion, dual-branch decoupled encoding, multi-scale convolution, and self-attention mechanism, ensures the model's learning ability and fitting of complex soil-rock monitoring data patterns. The physical-driven capability, jointly realized by strain coordination constraints, physical stiffness matrix, and physical information hybrid loss, ensures the physical consistency and reliability of the prediction results. The combination of data-driven and model-driven capabilities avoids the "black box" problem of purely data-driven approaches and breaks through the idealized assumptions of purely physical modeling, enabling robust and reliable identification of soil-rock deformation in real engineering environments. Traditional model loss function designs are mostly based on classification or regression errors, without incorporating prior knowledge of soil-rock instability mechanisms. When soil-rock enters the instability stage, mechanical properties such as stiffness degradation and yield expansion are crucial to system stability, but existing methods often fail to reflect this sensitivity to critical states in the loss function, thus weakening the ability to predict catastrophic critical points. This invention proposes a hybrid form of classification loss and physical constraint residual loss, trained based on a pre-set physical information hybrid loss function. In this embodiment, a stiffness matrix is ​​dynamically generated based on the predicted deformation pattern, achieving instability reinforcement constraints under abrupt stiffness changes, thereby improving the model's ability to identify and warn of critical failure points. In the task of classifying and identifying deformation in soil and rock masses, the conventional cross-entropy loss function only focuses on the consistency between the predicted and true labels, ignoring the physical laws that the deformation process of soil and rock masses must follow, such as equilibrium equations and constitutive relations. This may lead to the model predicting physically unreliable deformation patterns, reducing its reliability and generalization ability in real-world scenarios. This invention calculates cross-entropy classification loss and constructs physical constraint residual loss, then forms a hybrid loss function through weighted fusion, ensuring that the model's prediction results conform to physical laws. The specific steps are as follows: Step S10: Train the preset neural network model using the preset training sample set to determine the initial prediction output.

[0079] Calculate the cross-entropy loss between the probability distribution of the predicted deformation patterns from the network and the true labels. This is used to ensure the basic accuracy of the model's classification prediction, and is expressed as:

[0080] In the formula, Cross-entropy represents the classification loss, used to measure the difference between the predicted probability and the true label; For the first The one-hot encoded value of the true label of a sample in the c-th category is a scalar of 0 or 1; The model predicts the first The probability value of a sample belonging to the c-th category of deformation pattern is a scalar between 0 and 1. It is a classification output matrix The element in the i-th row and c-th column represents the element predicted by the model. The probability that a sample belongs to the c-th category; It is a logarithmic function, with the default base being the natural constant.

[0081] Step S20: Calculate the cross-entropy loss of the preset neural network model based on the initial prediction output, and generate a dynamic stiffness matrix according to the initial prediction output. The dynamic stiffness matrix is ​​determined based on the material constitutive relation corresponding to the training samples and the finite element discretization method of the preset neural network model.

[0082] Step S30: Construct the physical constraint residual loss of the preset neural network model based on the stiffness matrix.

[0083] 2) Constructing physical constraint residual loss The stiffness matrix is ​​dynamically generated based on the deformation mode categories predicted by the model. Then, the L2 norm squared of the residual between the predicted displacement vector and the load vector is calculated as the physical constraint residual loss to ensure that the prediction conforms to physical laws. This is expressed as:

[0084] In the formula, This represents the physical constraint residual loss, used to ensure that the prediction conforms to physical laws; According to the first The stiffness matrix generated by the predicted deformation mode categories of each sample has dimensions of [dimensional value missing]. Its construction depends on the material constitutive relation and the finite element discretization method; For the first The displacement vector corresponding to each sample is derived from the monitoring tensor; For the first The load vector corresponding to each sample is derived from the numerical simulation tensor. The dimension of the displacement vector depends on the number of degrees of freedom of the finite element model. Assume the model has... There are nodes, each node has One degree of freedom, such as in a 3D model ,but In one implementation, the stiffness matrix Dynamically generated based on the deformation pattern categories predicted by the model; specifically, from the classification output matrix. Get the first Predicted probability distribution of each sample These correspond to four categories: elastic, elastoplastic, creep, and shear failure; the category with the highest probability is selected as the first category. Prediction patterns for individual samples According to the first Prediction patterns for individual samples Select material parameters and calculate the equivalent elastic modulus. Specifically:

[0085] Based on equivalent elastic modulus Compared to Poisson Generate the physical stiffness matrix, represented as: ;in, The elastic modulus, derived from the numerical simulation tensor, characterizes the material's ability to resist elastic deformation. Poisson's ratio, derived from the numerical simulation tensor, characterizes the ratio of lateral deformation to longitudinal deformation of a material; For the first Kronecker symbol, when Its value is 1 when it is active and 0 otherwise, used to indicate the position of matrix elements, such as... , ; For the second Kronecker symbol, for a fixed index ,because If it is always equal to itself, then ,like, hour , hour ; For the third Kronecker symbol, for a fixed index ,because If it is always equal to itself, then ,like, hour , hour It should be noted that conventional physical losses typically use fixed equations, but soil and rock failure involves abrupt changes in stiffness; this invention is based on a predictive model. Dynamically generate stiffness matrix When predicting shear failure, Make the stiffness matrix Exoticization, forcing The number of terms increases dramatically in unstable states. The instability reinforcement mechanism can improve the model's sensitivity to catastrophic inflection points, which is something that constant weight loss cannot achieve.

[0086] Step S40: Weighted fusion of physical constraint residual loss and cross-entropy loss to generate a physical information hybrid loss function.

[0087] The cross-entropy classification loss and the physical constraint residual loss are linearly superimposed according to a preset physical constraint strength coefficient to form the final hybrid loss function, which is used to guide model training, and is expressed as:

[0088] In the formula, This represents the hybrid loss function, used to optimize model parameters; This refers to the physical constraint strength coefficient, used to adjust the relative importance of physical constraint terms in the total loss, such as... In summary, the model iterative training and parameter update process is based on a hybrid loss function to optimize and converge the model parameters. Specifically, the model is trained iteratively using the training dataset. In each iteration, batch sample data is input, and the fusion tensor, physical constraint tensor, decoupled encoded features, weighted feature tensor, and multi-scale fusion feature tensor are calculated through forward propagation, ultimately outputting the classification prediction result. Then, the total loss value is calculated according to the hybrid loss function, and the gradient of the loss with respect to each trainable parameter is calculated through backpropagation. Simultaneously, the Adam optimizer is used to update these parameters, adjusting the step size to minimize the loss function. The parameter update process continues until the model converges. The criteria for stopping model iteration include: training is terminated when the hybrid loss function no longer decreases significantly over multiple consecutive iterations (e.g., 10 iterations) on the validation set (the rate of change is below a preset threshold, e.g., 0.001), or when the preset maximum number of iterations (e.g., 1000 iterations) is reached, to prevent overfitting and ensure generalization performance.

[0089] Furthermore, the deformation of soil and rock mass must satisfy the strain compatibility equation, but noise and errors exist in actual monitoring data, leading to physical inconsistencies. Traditional neural networks often only pursue prediction accuracy, lacking consideration of strain compatibility, constitutive relations, and mechanical constraints, which may cause prediction results to violate mechanical laws and affect the reliability of the model. This embodiment of the invention also generates a physical stiffness matrix and constructs strain compatibility constraint terms. The strain compatibility equation and physical stiffness matrix are embedded as physical constraints into the training samples (such as the multi-source soil-rock fusion tensor mentioned above). The prediction results are corrected through strain compatibility constraints and curl operators, and the fusion tensor is physically consistent to ensure that the strain satisfies the compatibility equation. This ensures that the trained model strictly conforms to the laws of continuum mechanics while maintaining data-driven learning capabilities, improving the physical consistency and reliability of the prediction. In specific implementation, the following steps are included: Step S1: Obtain pre-constructed multi-source soil-rock fusion tensor training samples.

[0090] Step S2: Based on the linear physical response of the soil and rock bodies corresponding to the multi-source soil and rock fusion tensor training samples, physical constraints are applied to the multi-source soil and rock fusion tensor training samples to generate a physical constraint tensor.

[0091] In one implementation, a physical stiffness matrix can be dynamically generated based on the elastic modulus and Poisson's ratio. The elements of the physical stiffness matrix are calculated from the elastic modulus, Poisson's ratio, and Kronecker notation, and are used to construct the constitutive relation of the material, characterizing the linear physical response of the material, and are expressed as:

[0092] In the formula, The first element of the physical stiffness matrix represents the... Line 1 Column elements are used to construct the constitutive relations of materials; Represents the physical stiffness matrix, with dimensions of . It is used to characterize the linear physical response of a material and reflects the constraints of material parameters on deformation; The elastic modulus, derived from the numerical simulation tensor, characterizes the material's ability to resist elastic deformation. Poisson's ratio, derived from the numerical simulation tensor, characterizes the ratio of lateral deformation to longitudinal deformation of a material; For the first Kronecker symbol, when Its value is 1 when it is active and 0 otherwise, used to indicate the position of matrix elements, such as... , ; For the second Kronecker symbol, for a fixed index ,because If it is always equal to itself, then ,like, hour , hour ; For the third Kronecker symbol, for a fixed index ,because If it is always equal to itself, then ,like, hour , hour It should be noted that, and The double Kronecker notation not only simplifies the matrix expression of the three-dimensional Hooke's law, but more importantly, it... The terms can distinguish between the constraint strength of normal stress and shear stress, such as the strengthening constraint of Poisson's ratio on transverse deformation.

[0093] Furthermore, strain tensors can be extracted from the multi-source soil-rock fusion tensor training samples and reorganized into matrix form. Then, the vectorized strain tensors are linearly transformed using the physical stiffness matrix, and the strain compatibility constraint function is calculated by combining the strain compatibility penalty term. The strain compatibility penalty term calculates the Frobenius norm based on the curl and transpose of the strain field. The strain compatibility constraint function performs physical consistency correction on the strain to ensure that the strain satisfies the compatibility equation, expressed as:

[0094] In the formula, This represents the strain compatibility constraint function, used to perform physical consistency correction on the input strain; To fusion tensor The strain tensor extracted has a dimension of It characterizes the local deformation state of rock and soil masses; This indicates a vectorization operation, which expands the matrix into column vectors first, making matrix operations easier; For example, the constraint strength coefficient, It is used to adjust the importance of the strain compatibility penalty term and control the strictness of physical constraints; The Frobenius norm of the matrix is ​​used to measure the magnitude of strain compatibility deviations. This represents the curl operator, which is approximated by the central difference in the discrete case and is used to evaluate the compatibility of the strain field. This indicates the transpose operation.

[0095] In practical implementation, the strain tensor From fusion tensor The attribute dimension index corresponds to the strain component, including , , , and reorganized into Matrix, where, For the normal strain in the x-direction, For the normal strain in the y-direction, For the normal strain in the z-direction, Let x be the shear strain. It should be noted that the strain compatibility equation... It is for strain tensor The constraints are such that the strain tensor is a second-order symmetric tensor in continuum mechanics, expressed as... To compute curl and transpose operations, the strain components dispersed across the attribute dimensions need to be extracted and reassembled into a matrix form. The deformation of the soil and rock mass must satisfy the strain compatibility equation. By embedding the strain compatibility equation as a physical constraint into the network and calculating the strain compatibility deviation as a penalty term, the predicted strain is made to conform to physical laws, thereby improving the physical consistency and reliability of the prediction. It should also be noted that conventional physical constraints only use the stiffness matrix to correct static strain, while the curl operator... It can capture the dynamic propagation characteristics of strain compatibility. When local shear zones such as landslide surfaces form, the curl operator will amplify. The spatial discontinuity causes the penalty term to increase sharply in the unstable region, which weakens the constraints in the early creep stage when the strain gradient is small, and strengthens the constraints at the critical point of failure when the strain gradient changes abruptly.

[0096] Step S3: Construct a training sample set based on the physical constraint tensor.

[0097] In practical implementation, the output of the strain constraint function is projected back to the original data space through a linear mapping matrix, and combined with the fusion tensor to obtain the physical constraint tensor, thereby improving the physical consistency of the data, expressed as:

[0098] In the formula, For physical constraint tensors, it represents tensors after applying physical constraints (which can be used as training samples for model training). It is a linear mapping matrix with dimension . , used to The dimensional transformation of the term is related to the numerical simulation tensor. Consistency refers to trainable parameters. It should be noted that the linear mapping matrix... During training, it learns to ignore non-mechanically relevant feature dimensions and uses a linear mapping matrix. 9-dimensional The terms are projected back into a high-dimensional feature space, which can then automatically suppress spurious strain coordination caused by environmental noise, such as sensor drift caused by temperature changes.

[0099] Furthermore, embodiments of the present invention also use box plots to analyze the distribution of prediction errors of different methods on each strain component. Figure 5 A corresponding schematic diagram is shown for reference. Figure 5The horizontal axis represents four strain components: normal strain in the x-direction, normal strain in the y-direction, normal strain in the z-direction, and xy shear strain. The vertical axis represents the prediction error, using absolute error values; smaller values ​​indicate more accurate predictions. This evaluates the accuracy of each method in predicting specific physical quantities. The median line of the box plot shows that the method of this invention has the lowest prediction error across all strain components, and the error distribution is the most concentrated, indicating more stable and reliable prediction results. This advantage is particularly evident in the shear strain component, demonstrating the special handling of shear deformation by the strain field physical constraint embedding mechanism. Other methods have larger error distribution ranges and higher median positions, indicating poorer prediction stability. Experiments demonstrate that the method of this invention, through physical constraints and feature decoupling, not only improves classification accuracy but also provides more accurate and reliable predictions of specific physical quantities.

[0100] Furthermore, this embodiment of the invention also analyzes the impact of physical constraint strength on recognition performance, explores the optimization range of the physical constraint strength coefficient, and analyzes its balancing effect on recognition accuracy and physical consistency. Figure 6 A corresponding schematic diagram is shown. In practical implementation, other parameters are fixed through experiments, the physical constraint strength coefficient is adjusted, and two key indicators are measured: recognition accuracy (left vertical axis, unit: %) and physical residual (right vertical axis, dimensionless), reflecting the degree to which the prediction results violate physical laws. Experimental results show that when A value of 0 indicates no physical constraints, resulting in the highest physical residual (peak of the red curve), proving that ignoring physical laws leads to prediction distortion. At a value of 0.6 (gray dashed line), the recognition accuracy (blue curve) reaches its peak, while the physical residual (red curve) drops below 2, verifying the rationality of the parameter settings.

[0101] Furthermore, the embodiments of the present invention also analyze the performance changes of different methods under gradually increasing noise levels, and evaluate the robustness of each method against noise interference in real-world environments. Figure 7 A corresponding schematic diagram is shown. (Refer to...) Figure 7 The horizontal axis represents the noise level, gradually increasing from 0 to 0.5, representing the intensity of noise introduced into the data. The vertical axis represents the recognition accuracy, ranging from 0.4 to 1. Experimental results show that the performance of all methods decreases with increasing noise levels, but the method of this invention exhibits the smallest decrease, maintaining a relatively high accuracy even at high noise levels. Experiments demonstrate that the method of this invention, through strain-sensitive feature weighting and multi-scale feature extraction, can effectively suppress noise interference and extract robust feature representations. In contrast, traditional numerical simulation methods are most sensitive to noise and exhibit the most significant performance degradation, illustrating the limitations of a single data source in real-world environments.

[0102] Furthermore, this embodiment of the invention also provides a soil and rock deformation recognition device based on a neural network. Figure 8 A schematic diagram of an embodiment of the present invention is shown, with reference to... Figure 8 The device includes: a data monitoring module 100, used to monitor the target soil and rock mass, obtain monitoring data of the target soil and rock mass, and environmental factor data of the environment in which the monitoring data is located; a data processing module 200, used to determine the multi-source soil and rock fusion tensor corresponding to the monitoring data from a pre-constructed multi-source soil and rock fusion tensor set based on the environmental factor data; the multi-source soil and rock fusion tensor set is constructed based on the monitoring data, simulation data, and environmental factor data of the preset soil and rock mass; a feature extraction module 300, used to extract the branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil and rock fusion tensor; the branch features include elastic features and plastic features; an execution module 400, used to perform collaborative classification of the branch features, strain-sensitive features, and multi-scale spatiotemporal features based on a pre-constructed self-attention mechanism, and determine the deformation mode corresponding to the target soil and rock mass; and an output module 500, used to perform deformation analysis on the target soil and rock mass based on the deformation mode.

[0103] The soil and rock deformation identification device based on neural network provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0104] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described... Figures 1 to 2 The steps of any of the methods shown. Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the above-described steps. Figures 1 to 2 The steps of any of the methods shown. Embodiments of the present invention also provide a structural schematic diagram of an electronic device, such as... Figure 9 The diagram shows the structure of the electronic device, which includes a processor 91 and a memory 90. The memory 90 stores computer-executable instructions that can be executed by the processor 91. The processor 91 executes the computer-executable instructions to implement the above-mentioned... Figures 1 to 2 Any of the methods shown. In Figure 9In the illustrated embodiment, the electronic device further includes a bus 92 and a communication interface 93, wherein the processor 91, the communication interface 93, and the memory 90 are connected via the bus 92. The memory 90 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk drive. Communication between this system network element and at least one other network element is achieved through at least one communication interface 93 (which can be wired or wireless), using the Internet, wide area network, local area network, metropolitan area network, etc. Bus 92 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, or an AMBA (Advanced Microcontroller Bus Architecture) bus. AMBA defines three types of buses: APB (Advanced Peripheral Bus), AHB (Advanced High-performance Bus), and AXI (Advanced deXtensible Interface). Bus 92 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9The diagram uses only a single double-headed arrow, but this does not imply a single bus or a single type of bus. Processor 91 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed through integrated logic circuits in the hardware of processor 91 or through software instructions. Processor 91 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor 91 reads information from the memory and, in conjunction with its hardware, completes the aforementioned tasks. Figures 1 to 2Any of the methods shown. The computer program product of the method, apparatus, and electronic device for identifying soil and rock deformation based on neural networks provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. Specific implementations can be found in the method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the preceding method embodiments, and will not be repeated here. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, and other media capable of storing program code. In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Finally, it should be noted that the above embodiments are merely specific implementations of this invention, used to illustrate the technical solutions of this invention, and not to limit it. The scope of protection of this invention is not limited thereto. Although this invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in this invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be covered within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for identifying deformation of soil and rock masses based on neural networks, characterized in that, The method includes: Data monitoring is performed on the target soil and rock mass to obtain monitoring data of the target soil and rock mass, as well as environmental factor data of the environment in which the monitoring data is located; Based on the environmental factor data, the multi-source soil and rock fusion tensor corresponding to the monitoring data is determined from a pre-constructed multi-source soil and rock fusion tensor set; the multi-source soil and rock fusion tensor set is constructed based on the monitoring data, simulation data and environmental factor data of the preset soil and rock body. Extract the branching features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor; the branching features include elastic features and plastic features; Based on a pre-built self-attention mechanism, the branch features, strain-sensitive features, and multi-scale spatiotemporal features are collaboratively classified to determine the deformation mode corresponding to the target rock and soil mass. Based on the deformation mode, deformation analysis is performed on the target rock and soil mass.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the numerical simulation tensor, monitoring tensor, and corresponding environmental factor tensor pre-collected for the preset soil and rock mass; The numerical simulation tensor, the monitoring tensor, and the environmental factor tensor are linearly combined to generate the interaction tensor of the preset soil and rock mass. The numerical simulation tensor, the monitoring tensor, and the environmental factor tensor are dimensionally unified, and the dimensionally unified numerical simulation tensor, the monitoring tensor, and the environmental factor tensor are fused with the interaction tensor to generate a multi-source soil-rock fusion tensor set for the preset soil-rock body.

3. The method according to claim 1, characterized in that, The steps for extracting the branching features of the multi-source soil-rock fusion tensor include: The elastic feature tensor corresponding to the multi-source soil-rock fusion tensor is extracted by performing the first convolution process on the multi-source soil-rock fusion tensor through a preset neural network model. Furthermore, the monitoring data is subjected to a second convolutional process through the preset neural network model to determine the gating weights corresponding to the multi-source soil-rock fusion tensor; The elastic feature tensor is range-compressed, and based on the gating weight, the range-compressed elastic feature tensor is gating-selected to determine the plastic feature tensor of the multi-source soil-rock fusion tensor. The elastic feature tensor and the plastic feature tensor are spliced ​​together to determine the branching features of the multi-source soil-rock fusion tensor.

4. The method according to claim 1, characterized in that, The steps for extracting the strain-sensitive features of the multi-source soil-rock fusion tensor include: Extract the strain components of the multi-source soil-rock fusion tensor, determine the gradients of the strain components in the spatial direction and time dimension, and generate a strain gradient vector; The importance of the positional features of the strain gradient vector is identified, and the strain sensitivity weights corresponding to the positional features are determined. The strain-sensitive weights are used to perform feature weighting on the multi-source soil-rock fusion tensor to determine the strain-sensitive features of the multi-source soil-rock fusion tensor.

5. The method according to claim 1, characterized in that, The steps for extracting the multi-scale spatiotemporal features of the multi-source soil-rock fusion tensor include: By using multiple pre-defined dilated convolution branches in parallel with different expansion rates, dilated convolution operations are performed on the multi-source soil-rock fusion tensor to capture the spatiotemporal characteristics of the multi-source soil-rock fusion tensor at different scales. The spatiotemporal features are summed to obtain the multi-scale spatiotemporal features of the multi-source soil-rock fusion tensor.

6. The method according to claim 1, characterized in that, The step of determining the deformation mode corresponding to the target soil and rock mass by co-classifying the branch features, strain-sensitive features, and multi-scale spatiotemporal features based on a pre-constructed self-attention mechanism includes: The branch features, strain-sensitive features, and multi-scale spatiotemporal features are converted into global feature vectors, and the global feature vectors are mapped to the query space to generate a query representation matrix. A scaling dot product attention mechanism is used to perform dot product processing on the query representation matrix to determine the pattern association graph corresponding to the global feature vector; the pattern association graph includes the deformed pattern similarity between samples. The pattern association graph and the global feature vector are subjected to matrix multiplication to generate a fusion output, and the deformation mode corresponding to the target soil and rock mass is determined based on the fusion output.

7. The method according to claim 1, characterized in that, The method further includes: The branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor are extracted by a preset neural network model; wherein, the preset neural network model is trained based on a preset physical information hybrid loss function; The method further includes: The preset neural network model is trained using a preset training sample set to determine the initial prediction output; The cross-entropy loss of the preset neural network model is calculated based on the initial prediction output, and a dynamic stiffness matrix is ​​generated according to the initial prediction output; the dynamic stiffness matrix is ​​determined based on the material constitutive relation corresponding to the training samples and the finite element discretization method of the preset neural network model. The physical constraint residual loss of the preset neural network model is constructed based on the stiffness matrix; The physical constraint residual loss and the cross-entropy loss are weighted and fused to generate a physical information hybrid loss function.

8. The method according to claim 7, characterized in that, The method further includes: Obtain pre-constructed multi-source geotechnical fusion tensor training samples; Based on the linear physical response of the soil and rock body corresponding to the multi-source soil and rock fusion tensor training sample, physical constraints are applied to the multi-source soil and rock fusion tensor training sample to generate a physical constraint tensor. A training sample set is constructed based on the physical constraint tensor.

9. A device for identifying soil and rock deformation based on a neural network, characterized in that, The device includes: The data monitoring module is used to monitor the target rock and soil mass, obtain the monitoring data of the target rock and soil mass, and the environmental factor data of the environment in which the monitoring data is located; The data processing module is used to determine the multi-source soil-rock fusion tensor corresponding to the monitoring data from a pre-constructed multi-source soil-rock fusion tensor set based on the environmental factor data; the multi-source soil-rock fusion tensor set is constructed based on the monitoring data, simulation data and environmental factor data of the preset soil-rock body; The feature extraction module is used to extract the branch features, strain-sensitive features, and multi-scale spatiotemporal features corresponding to the multi-source soil-rock fusion tensor; the branch features include elastic features and plastic features; The execution module is used to perform collaborative classification of the branch features, the strain-sensitive features and the multi-scale spatiotemporal features based on a pre-built self-attention mechanism to determine the deformation mode corresponding to the target rock and soil mass; The output module is used to perform deformation analysis on the target rock and soil mass based on the deformation mode.

10. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the neural network-based rock and soil deformation identification method according to any one of claims 1 to 8.

Images