Geological disaster early warning device

Abstract

The invention relates to the technical field of geological disaster early warning, and discloses a geological disaster early warning device, which comprises an acoustic electromagnetic fusion sensor array for collecting multi-physical field data related to the internal structure change of a rock-soil body; the data pre-processing and feature extraction module is used for performing pre-processing and feature extraction on the original multi-physics field data so as to identify the change features of the internal structure of the rock-soil body; the multi-physics field collaborative inversion module reconstructs the internal structure distribution of the rock-soil body by adopting a multi-physics field collaborative inversion algorithm; the structure change cause analysis and risk assessment module is used for analyzing causes of internal structure change of the rock-soil body and assessing potential geological disaster risks; the grading early warning information generation module is used for generating targeted grading early warning information; according to the invention, high-precision monitoring of the internal structure change of the rock-soil body is realized through an acoustic electromagnetic fusion sensing technology, the monitoring precision is improved, and possible landslide or collapse events can be predicted in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster early warning technology, and more specifically, to a geological disaster early warning device. Background Art

[0002] Geological disasters are geological phenomena caused by geological processes that result in loss of life and property or damage to the environment, such as landslides, debris flows, and ground collapses. Because geological disasters are sudden and destructive, establishing an effective geological disaster early warning system is crucial for disaster prevention and mitigation.

[0003] Currently, geological disaster early warning technologies mainly use methods such as surface displacement monitoring, groundwater level monitoring, or single physical field monitoring. These methods have obvious limitations:

[0004] First, single physical field monitoring can only obtain single parameter information, which is difficult to fully reflect the complex change process of the underground medium, especially when distinguishing different causes (such as changes caused by water infiltration and changes caused by stress); second, surface monitoring methods can only obtain surface information, which is difficult to timely and accurately reflect the early changes in the internal structure of the rock and soil, and these internal structure changes are often important precursors to disasters; in addition, the existing multi-physical field data processing methods have high computational complexity and a long inversion process, which is difficult to meet the real-time requirements of geological disaster early warning. Existing technologies are also difficult to establish a reliable quantitative correlation between the characteristics of internal structural changes in rock and soil and the occurrence of potential disasters. Early warnings often rely on empirical judgment and have limited accuracy; finally, due to the inability to accurately identify the causes of disasters, existing early warning systems can often only provide general early warning information, and it is difficult to provide targeted early warning suggestions for potential disasters of different types and causes.

[0005] Therefore, there is an urgent need for a geological disaster early warning technology that can accurately monitor the internal structural changes of rock and soil, scientifically analyze the causes of structural changes, and provide targeted early warning information. Summary of the Invention

[0006] The present invention provides a geological disaster early warning device to solve the technical problems existing in related technologies, such as the limitation of single physical field monitoring, the lack of quantitative correlation between internal structure changes and disaster occurrence, and the lack of pertinence of early warning information.

[0007] The present invention provides a geological disaster early warning device, comprising:

[0008] Acoustic and electromagnetic fusion sensor arrays are used to collect multi-physics field data related to changes in the internal structure of rock and soil;

[0009] The data preprocessing and feature extraction module is used to preprocess and extract features from the original multi-physics field data collected by the acoustic and electromagnetic fusion sensor array to identify the internal structural change characteristics of the rock and soil mass;

[0010] The multi-physics collaborative inversion module is used to reconstruct the internal structural distribution of the rock and soil mass using a multi-physics collaborative inversion algorithm based on the internal structural change characteristics of the rock and soil mass identified by the data preprocessing and feature extraction modules;

[0011] The structural change cause analysis and risk assessment module is used to analyze the causes of internal structural changes in rock and soil based on the internal structural distribution results reconstructed by the multi-physics field collaborative inversion module, and to assess the potential geological disaster risks;

[0012] The graded warning information generation module is used to generate targeted graded warning information based on the potential geological disaster risk results assessed by the structural change cause analysis and risk assessment module.

[0013] Furthermore, the collection of multi-physics field data includes:

[0014] Deploy acoustic and electromagnetic fusion sensor arrays based on the topographic and geological conditions of the monitoring area;

[0015] Using acoustic electromagnetic fusion sensor array to simultaneously collect acoustic parameters and electromagnetic parameters;

[0016] The collected raw data are preliminarily verified and transmitted to the data processing center.

[0017] Furthermore, the preprocessing and feature extraction of the original multi-physics field data collected by the acoustic electromagnetic fusion sensor array includes:

[0018] Apply adaptive filtering algorithm to pre-process the raw data to eliminate environmental noise and system errors;

[0019] Perform spectrum analysis on the preprocessed acoustic data to extract acoustic features;

[0020] Analyze the preprocessed electromagnetic data and extract electromagnetic features;

[0021] The cross-correlations between acoustic and electromagnetic parameters are calculated, and the relationship expressions between physical parameters are established.

[0022] Furthermore, the multi-physics field collaborative inversion algorithm includes:

[0023] Apply the eigenmode decomposition algorithm to reduce the dimensionality of the extracted high-dimensional feature data;

[0024] Construct a Bayesian multi-parameter inversion model;

[0025] The Markov chain Monte Carlo method is used for numerical solution;

[0026] Acoustic data and electromagnetic data are used simultaneously to collaboratively constrain the inversion results.

[0027] Furthermore, the Markov Chain Monte Carlo method adopts an improved Metropolis Hastings algorithm and improves sampling efficiency through adaptive proposal distribution.

[0028] Furthermore, the structural change cause analysis and risk assessment module includes:

[0029] Using multi-physics inversion results, we can distinguish between structural changes caused by hydraulic and stress effects.

[0030] Apply the transfer entropy method to conduct causal analysis and determine the correlation between internal structural changes and potential disasters;

[0031] Build a time-varying attribute graph model to represent underground structures as a time-varying attribute graph, and apply graph convolutional neural networks to predict future states;

[0032] Based on the analysis of the causes of structural changes and evolution prediction results, a comprehensive assessment of the potential disaster risk level is conducted.

[0033] Furthermore, the time-varying attribute graph model includes:

[0034] Spatial graph convolution layer, which captures the spatial relationship between nodes;

[0035] Temporal attention layer, which captures the temporal evolution of node attributes;

[0036] The spatiotemporal fusion layer fuses spatial and temporal information to generate predictions of the future states of nodes.

[0037] Furthermore, the hierarchical warning information generation module includes:

[0038] Determine the warning level based on the risk value and risk level obtained from the risk assessment;

[0039] Generate warning information content based on the warning level and structural change cause analysis results;

[0040] Generate targeted prevention and control suggestions based on structural changes of different causes;

[0041] Push early warning information to relevant management departments and technical personnel.

[0042] Furthermore, there are four warning levels, namely green warning, yellow warning, orange warning and red warning;

[0043] The warning information is expressed in different ways according to the recipients.

[0044] The present invention provides a computer storage medium, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned geological disaster early warning device.

[0045] The beneficial effects of the present invention are: high-precision monitoring of internal structural changes of rock and soil is achieved through acoustic and electromagnetic fusion sensing technology, which improves monitoring accuracy and can predict possible landslides or collapse events in advance, solving the technical problem that traditional methods are difficult to timely reflect early changes in the internal structure of rock and soil;

[0046] The use of multi-physics field collaborative analysis methods accurately distinguishes structural changes caused by hydraulic and stress effects, improving the pertinence of early warnings, reducing false alarm rates, and solving the technical problem of traditional single-physics field monitoring's difficulty in distinguishing structural changes caused by different factors.

[0047] Through eigenmode decomposition and parallel computing technology, the computational efficiency of multi-physics field inversion is improved, enabling the system to achieve real-time monitoring and analysis even with limited computing resources, solving the technical problem of balancing computational complexity and real-time performance.

[0048] Based on transfer entropy causal relationship analysis and graph convolutional neural network prediction, a quantitative correlation model between internal structural changes and potential disasters was constructed, which improved the accuracy of early warnings and solved the technical problem of the lack of quantitative correlation between internal structural changes and disaster occurrence.

[0049] Based on the analysis results of the causes of structural changes, targeted prevention and control suggestions can be generated, making disaster prevention and mitigation measures more efficient and solving the technical problem of lack of targeted early warning information. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a module diagram of a geological disaster early warning device in the present invention;

[0051] Figure 2 is a bar chart comparing the performance of different early warning methods in the present invention;

[0052] Figure 3 It is a radar chart for multi-dimensional performance evaluation in the present invention;

[0053] Figure 4 It is a scatter plot of the correlation between structural changes and disaster risks in the present invention;

[0054] Figure 5 It is a line graph showing the evolution of the performance of the early warning system over time in the present invention. DETAILED DESCRIPTION

[0055] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0056] At least one embodiment of the present invention discloses a geological disaster early warning device, such as Figure 1 As shown, including:

[0057] Acoustic and electromagnetic fusion sensor arrays are used to collect multi-physics field data related to changes in the internal structure of rock and soil;

[0058] According to one embodiment of the present application, this step uses an acoustic-electromagnetic fusion sensor array to collect multi-physics field data related to changes in the internal structure of the rock and soil mass, and specifically includes the following sub-steps:

[0059] Step 1.1, deployment of acoustic electromagnetic fusion sensor array;

[0060] The optimal layout of the acoustic-electromagnetic fusion sensor is determined based on the topographic and geological conditions of the monitoring area. In areas prone to geological disasters, the sensor layout density is appropriately increased to ensure the monitoring accuracy of key areas. The acoustic-electromagnetic fusion sensor array consists of acoustic sensing units and electromagnetic sensing units:

[0061] Acoustic sensing unit: includes an acoustic wave transmitter and a receiver, which are used to transmit acoustic wave signals of a specific frequency and receive reflected waves and transmitted waves after passing through the rock and soil;

[0062] Electromagnetic sensing unit: includes electromagnetic wave transmitter and receiver, used to measure the changes of electromagnetic parameters of rock and soil.

[0063] The acoustic sensing unit and the electromagnetic sensing unit use time synchronization technology to ensure that the collection process of the two types of data has an accurate time correspondence.

[0064] Step 1.2, synchronous acquisition of multi-physics field data;

[0065] Use acoustic and electromagnetic fusion sensor arrays to simultaneously collect acoustic and electromagnetic parameters:

[0066] Acoustic parameters: including sound wave velocity (v p and v so , representing the longitudinal and shear wave velocities respectively), reflection coefficient (R ref ), attenuation coefficient (α att )wait;

[0067] Electromagnetic parameters: including conductivity (σ cond ), dielectric constant (ε), etc.

[0068] During data collection, the system adaptively adjusts the sampling frequency based on the dynamic characteristics of the monitored target. For areas or parameters that change rapidly, the sampling frequency is increased; for areas or parameters that change slowly, the sampling frequency is appropriately reduced to optimize system resource utilization efficiency.

[0069] Step 1.3, preliminary verification and transmission of collected data;

[0070] The collected raw data undergoes preliminary verification to remove significant outliers and noise. Verified data is then transmitted to the data processing center via a wired or wireless network. Data compression algorithms are used to reduce bandwidth requirements, while data verification mechanisms are employed to ensure data integrity during transmission.

[0071] Through the above steps, this application realizes high-precision, real-time and synchronous collection of multi-physical field parameters of rock and soil in the monitoring area, providing a reliable original data basis for subsequent data processing and analysis.

[0072] The data preprocessing and feature extraction module is used to preprocess and extract features from the original multi-physics field data collected by the acoustic and electromagnetic fusion sensor array to identify the internal structural change characteristics of the rock and soil mass;

[0073] This step preprocesses and extracts features from the collected raw multi-physics field data to identify the internal structural changes of the rock and soil mass. It specifically includes the following sub-steps:

[0074] Step 2.1, data preprocessing and noise elimination;

[0075] According to one embodiment of the present application, an adaptive filtering algorithm is applied to preprocess the raw data to eliminate environmental noise and system errors. For acoustic data, a bandpass filter is used to remove low-frequency environmental vibrations and high-frequency electronic noise; for electromagnetic data, median filtering and wavelet denoising techniques are used to eliminate interference. The adaptive filtering algorithm dynamically adjusts the filtering parameters based on the noise characteristics. The filtering process can be expressed as:

[0076]

[0077] Where y(n1) is the filtered data, x(n1) is the original data, and h k (n1) is the adaptive filter coefficient before updating, N filt is the filter order, k represents the filter tap index, n1 represents the discrete time index, and ∑ represents the summation operator symbol.

[0078] Filter coefficient h k(n1) is continuously updated according to the error signal e(n1) through the least mean square error (LMS) algorithm:

[0079] h k (n1+1)=h k (n1)+μ step ·e(n1)·x(n1-k);

[0080] where μ step is the step size parameter, which controls the coefficient of the filter convergence speed; e(n1) is the error between the expected output and the actual output; h k (n1+1) and h k (n1) represents the adaptive filter coefficients before and after updating respectively; x(n1-k) represents the input signal delayed by k sampling points, and n1 represents the discrete time index.

[0081] Step 2.2, acoustic parameter feature extraction;

[0082] According to a specific embodiment of the present application, spectrum analysis is performed on the pre-processed acoustic data to extract acoustic features. First, the propagation characteristics of the sound wave in the rock and soil are calculated, including:

[0083] Change in acoustic wave velocity: The acoustic wave velocity is calculated by measuring the propagation time of the acoustic wave and the known sensor spacing. The relationship between the acoustic wave velocity and the rock mass parameters can be expressed as:

[0084]

[0085] where v p is the longitudinal wave velocity (the rate of longitudinal wave propagation in the rock mass), K is the bulk modulus (indicating the material's ability to resist volume change), G1 is the shear modulus (indicating the material's ability to resist shear deformation), and ρ is the density.

[0086] Sound wave attenuation characteristics: Calculate the energy attenuation of sound waves during propagation, attenuation coefficient α att Related to structural anomalies such as cracks and cavities inside the rock mass:

[0087]

[0088] Where A(x) is the amplitude at a distance x from the sound source (the intensity of the sound wave at a propagation distance x), A0 is the initial amplitude (the intensity of the sound wave at the sound source), α att is the attenuation coefficient (indicating the rate at which sound wave energy decays with distance), and e is the base of the natural logarithm.

[0089] Spectral characteristics: Perform a short-time Fourier transform (STFT) on the sound wave signal to analyze the changes in different frequency components. A window of length L is taken around time m, and then a Fourier transform is performed within the window:

[0090]

[0091] Where X(m, ω) is the STFT coefficient of the signal x(n2) at time m and frequency ω (indicating the signal strength of a specific frequency at a specific time point), x(n2) is the original time domain signal, and w win (n2) is the window function (used to intercept a specific time period of the signal), is the complex exponential function (the core of the Fourier transform), j is the imaginary unit, n2 is the time index, ω is the angular frequency, ∑ represents the summation operator, and L / 2 represents half the window length.

[0092] Through the above analysis, the system identifies structural abnormalities such as cracks and cavities inside the rock and soil body and outputs the acoustic parameter feature matrix.

[0093] Step 2.3, electromagnetic parameter feature extraction;

[0094] In another embodiment of the present application, the pre-processed electromagnetic data is analyzed to extract electromagnetic features. This mainly includes:

[0095] Conductivity changes: Conductivity is closely related to the moisture content and ion concentration within the rock and soil. By measuring conductivity changes, structural changes caused by water infiltration can be identified:

[0096]

[0097] where σ cond is the electrical conductivity of rock and soil (a parameter indicating the electrical conductivity of rock and soil), σ w is the pore water conductivity (the electrical conductivity of water in the pores), φ is the porosity (the ratio of the pore volume to the total volume of the rock mass), S w is the water saturation (the proportion of pores filled with water), m emp1 and n emp2 represent the pore structure cementation factor and saturation index, respectively.

[0098] Dielectric constant change: The dielectric constant is related to the mineral composition, porosity and water content of the rock and soil. By measuring the change in the dielectric constant, the water infiltration situation can be further confirmed:

[0099] ε r =(1-φ)ε matrix +φS w ε w+φ(1-S w )ε air ;

[0100] where ε r is the relative dielectric constant of the rock and soil (a parameter indicating the charge storage capacity of the rock and soil), ε matrix , ε w and ε air are the relative dielectric constants of rock matrix, water and air respectively, φ is the porosity, S w is the water saturation, (1-φ) represents the rock matrix volume fraction, φS w represents the water volume fraction, φ(1-S w ) represents the air volume fraction.

[0101] Through the above analysis, the system identifies characteristics such as water infiltration and mineral composition changes inside the rock and soil, and outputs the electromagnetic parameter characteristic matrix.

[0102] Step 2.4, physical parameter correlation analysis;

[0103] In addition, this application calculates the cross-correlation between acoustic and electromagnetic parameters to establish a relationship expression between physical parameters. By analyzing the correlation between different physical parameters, the reliability of parameter changes can be further verified and constraints can be provided for subsequent collaborative inversion. The cross-correlation coefficient is calculated as follows:

[0104]

[0105] where R xy (k) is the cross-correlation coefficient between the sequences x(n2) and y(n2) (a measure of the similarity between the two signals), N filt is the data length (number of sample points), k delay is the time delay (the time offset between the two signals), x(n2) and y(n2) represent the acoustic parameter sequence and electromagnetic parameter sequence respectively, n2 is the time index, and ∑ represents the summation operator symbol.

[0106] Through this step, the system extracts characteristic information of internal structural changes of rock and soil from the original data of multi-physics fields, and provides processed characteristic data for subsequent collaborative inversion calculations.

[0107] The multi-physics collaborative inversion module is used to reconstruct the internal structural distribution of the rock and soil mass using a multi-physics collaborative inversion algorithm based on the internal structural change characteristics of the rock and soil mass identified by the data preprocessing and feature extraction modules;

[0108] This step uses a multi-physics collaborative inversion algorithm to reconstruct the internal structure distribution of the rock and soil based on the results of feature extraction. It specifically includes the following sub-steps:

[0109] Step 3.1, eigenmode decomposition and dimensionality reduction;

[0110] In the embodiment of the present application, in order to improve computational efficiency, the extracted high-dimensional feature data is first subjected to dimensionality reduction processing. The original feature matrix is ​​decomposed into several eigenmodes using the eigenmode decomposition algorithm, retaining the main information and reducing computational complexity. The mathematical expression of eigenmode decomposition is:

[0111] A≈U·SIG·V T ;

[0112] Where A is the original feature matrix (containing all extracted feature data), U and V are the left singular vector and right singular vector matrices respectively (representing the orthogonal basis of the feature space), SIG is the singular value diagonal matrix (indicating the importance of each eigenmode), and T represents the matrix transpose operation.

[0113] By retaining larger singular values ​​(for example, retaining singular values ​​with energy accounting for 95%), effective dimensionality reduction can be achieved.

[0114] The feature matrix after dimensionality reduction is expressed as:

[0115]

[0116] Among them A r is the feature matrix after dimensionality reduction (retaining the low-dimensional representation of the main information), U r , SIG r and V r are the truncated left singular vector, singular value and right singular vector matrices respectively, and T represents the matrix transpose operation.

[0117] Step 3.2, Bayesian multi-parameter inversion model construction;

[0118] According to one embodiment of the present application, a Bayesian multi-parameter inversion model is constructed based on the feature data after dimensionality reduction. First, a forward model between the observed data and the underground rock and soil parameters is established:

[0119] d=G2(m)+ε err ;

[0120] Where d is the observation data vector (actually measured physical field data), m is the geotechnical parameter vector to be solved (underground structure parameters that need to be inverted), G2(·) is the forward operator (a function that maps underground parameters to observation data), and ε err is the observation error (noise and error in the measurement process), G2(m) represents the theoretical observation data calculated by the forward operator G2(·) using the rock and soil parameter vector m to be solved.

[0121] The role of the forward model is to describe how changes in underground structural parameters affect the observed multi-physics data and is the basis of the inversion problem;

[0122] Through the forward model, the actual observation data can be compared with the theoretical prediction data, and the most likely distribution of underground structure parameters can be inferred.

[0123] Bayesian inversion is based on Bayes’ theorem, which transforms the inversion problem into the estimation of the posterior probability distribution:

[0124] p(m|d)∝p(d|m)p(m);

[0125] Where p(m|d) is the posterior probability distribution of the parameter under the given observation data conditions (the goal of inversion), p(d|m) is the likelihood function (the degree of match between the observation data and the model parameters), p(m) is the prior probability distribution of the parameter (the parameter distribution based on prior knowledge), and ∝ indicates a positive proportional relationship.

[0126] The likelihood function is usually assumed to be Gaussian:

[0127]

[0128] Where p(m|d) is the likelihood function, which is the degree of matching between the observed data and the model parameters; C d is the data covariance matrix (indicating the uncertainty of the observed data), exp is the exponential function, (d-G2(m)) represents the residual between the observed data vector and the theoretical observed data, T represents the matrix transpose, is the inverse matrix of the covariance matrix, and ∝ indicates a positive proportional relationship.

[0129] In a specific embodiment of the present application, the parameter prior probability distribution is constructed based on geological prior knowledge. For the physical parameters of the rock mass, a constrained distribution based on geological knowledge can be used; for the geological structure boundary, a Markov random field prior based on local smoothness can be used. For example, for the rock mass elastic modulus E and Poisson's ratio ν, the following prior distribution can be set:

[0130]

[0131] where p(E) and p(v) represent the prior probability distributions of the elastic modulus E and Poisson's ratio v, respectively; E0 and ν0 are the prior mean values ​​based on geological knowledge (i.e., the expected values ​​or common values ​​of the parameters); σ E and σ ν is the corresponding standard deviation (indicating the quantification of parameter uncertainty based on prior knowledge); E>0 and 0<v<0.5 are the effective range constraints of the physical parameters, ensuring that the parameter values ​​are consistent with the physical meaning and actual geological conditions; ∝ indicates a direct proportional relationship, and exp is an exponential function.

[0132] Through the above-mentioned prior distribution, professional knowledge in the geological field can be effectively integrated into the inversion process, improving the rationality and physical interpretability of the inversion results.

[0133] For the spatial distribution of the internal structure of the rock mass, the total variation prior is used to promote the piecewise constant characteristics of the structure:

[0134]

[0135] Where p(m) is the prior probability of the parameter, is the regularization parameter (coefficient controlling the strength of the prior), is the spatial gradient of the parameter (indicating the spatial rate of change of the parameter), is the absolute value of the gradient (indicating the magnitude of the change), is the total variation integral (integral over the entire spatial domain), is the spatial differential element; it represents a direct proportional relationship.

[0136] Step 3.3, Markov chain Monte Carlo sampling calculation;

[0137] In another embodiment of the present application, since the posterior probability distribution cannot usually be solved directly, the Markov Chain Monte Carlo (MCMC) method is used for numerical solution. The present application adopts the improved Metropolis Hastings algorithm for MCMC sampling, and the sampling process is as follows:

[0138] Initialize parameter vector m0;

[0139] For iteration i1:

[0140] From the proposed distribution Extract candidate samples m * ;

[0141] Calculate the probability of acceptance:

[0142]

[0143] in is the acceptance probability, which determines whether to accept the candidate sample; min means taking the minimum value symbol; p(m * |d) is the candidate parameter m under the observation data condition * The posterior probability of Under the condition of observation data d, the current parameter The posterior probability of is the parameter vector of the i1-1th iteration; For the candidate sample m * Return to current sample Proposed distribution probability of ; From the current sample Generate candidate sample m * The proposed distribution probability is ; the superscript * represents the candidate sample, the subscript i1-1 represents the sample of the previous iteration, and d is the observed data.

[0144] By probability Accept candidate sample m * As otherwise

[0145] Repeat the iteration until the preset number of iterations is reached.

[0146] To improve sampling efficiency, this application uses an adaptive proposal distribution, dynamically adjusting the parameters of the proposal distribution based on the distribution characteristics of the accepted samples. In addition, parallel computing technology is used to run multiple Markov chains simultaneously to further improve computational efficiency.

[0147] Step 3.4, collaborative constraint of multi-physics data;

[0148] According to the embodiments of the present application, during the inversion process, acoustic data and electromagnetic data are used simultaneously to constrain the inversion results, solving the ill-posedness problem of single physical field inversion. The multi-physics field collaborative constraint is expressed as a combination of likelihood functions:

[0149] p′(d|m)=p(d acoustic |m)·p(d EM |m);

[0150] where p′(d|m) is the likelihood function (the degree of matching between the observed data and the model parameters) that uses both acoustic and electromagnetic data to constrain the inversion results, and p(d acoustic |m) and p(d EM |m) are the likelihood functions of acoustic data and electromagnetic data respectively (indicating the degree of matching between each type of data and the model parameters).

[0151] In addition, this application uses the correlation between different physical field data to further constrain the parameter space and improve the stability and reliability of the inversion. Through this collaborative constraint mechanism, the ambiguity of single physical field inversion is effectively eliminated and the accuracy of the inversion results is improved.

[0152] In some embodiments, the multi-physics field collaborative constraint can also adopt an alternating iterative strategy. Specifically, first perform inversion based on acoustic data to obtain a preliminary result m acoustic ; Then m acoustic As the initial value, the inversion is performed based on the electromagnetic data to obtain the updated result m EM ; Then m EMAs the initial value, the inversion is performed based on the acoustic data. Through this alternating iterative method, the inversion results will gradually converge to a solution that satisfies the constraints of the two physical field data.

[0153] Optionally, the multiphysics synergy constraint can also be combined with the structural similarity constraint, expressed as:

[0154] p″(d|m)=p(d acoustic |m)·p(d EM |m)·p struct (m);

[0155] where p″(d|m) is the likelihood function (the degree of match between observed data and model parameters) combined with the structural similarity constraint, and p(d acoustic |m) and p(d EM |m) are the likelihood functions of acoustic data and electromagnetic data respectively (indicating the degree of matching between each type of data and model parameters), p struct (m) is the structural similarity constraint (which forces different physical parameters to be consistent on the structural boundary), which forces the reconstruction results of acoustic parameters and electromagnetic parameters to be consistent on the structural boundary:

[0156]

[0157] in is a parameter that controls the strength of the structural similarity constraint (adjusts the weight of the structural constraint), and are the gradient amplitudes of acoustic parameters and electromagnetic parameters (indicating the severity of parameter changes), represents the difference in the magnitude of the gradients of the two parameters (the degree of inconsistency in the structural boundaries), is the integral over the entire spatial domain, and dx2 is the spatial differential element.

[0158] Therefore, through the above-mentioned multi-physics field collaborative inversion calculation, this application achieves high-precision reconstruction of the internal structural parameters of the rock and soil mass, including the distribution of the physical and mechanical parameters of the rock and soil mass and the location of internal structural anomalies, providing reliable basic data for subsequent structural change cause analysis and risk assessment.

[0159] The structural change cause analysis and risk assessment module is used to analyze the causes of internal structural changes in rock and soil based on the internal structural distribution results reconstructed by the multi-physics field collaborative inversion module, and to assess the potential geological disaster risks;

[0160] Based on the inversion results, this step analyzes the causes of changes in the internal structure of the rock mass and assesses potential disaster risks. It specifically includes the following sub-steps:

[0161] Step 4.1, classification of causes of structural change;

[0162] According to one embodiment of the present application, the causes of structural changes in rock and soil are analyzed using multi-physics field inversion results. The structural changes caused by hydraulic action and stress action are mainly distinguished by the following methods:

[0163] Hydraulic action identification: This is primarily based on changes in electromagnetic parameters (conductivity and dielectric constant). Structural changes caused by water infiltration typically manifest as increases in conductivity and dielectric constant, accompanied by decreases in acoustic velocity and increased attenuation. Structural changes caused by hydraulic action often have distinct spatial distribution patterns, such as expansion along the path of rainwater infiltration or along the direction of groundwater flow.

[0164] Stress effects are primarily identified through changes in acoustic parameters. Structural changes caused by stress are typically manifested as directional shifts in acoustic velocity (increased anisotropy) and changes in reflection characteristics. Furthermore, structural changes caused by stress are often concentrated in structural weaknesses or areas of stress concentration.

[0165] By establishing a discrimination model for hydraulic and stress effects, the system can automatically classify the causes of changes in the internal structure of rock and soil, improving the pertinence of early warning. The discrimination model is based on the following criterion function:

[0166] f water (m) = w 1,1 Δσ cond +W 2,1 Δε r +w 3,1 Δv p +w 4,1 Δα att ;

[0167] where f water (m) and f stress (m) represents the discriminant function of structural change caused by hydraulic action and the discriminant function of structural change caused by stress, Δσ cond , Δε r , Δv p , Δα att 、 and ΔR ref represent the changes in the conductivity, dielectric constant, acoustic velocity, attenuation coefficient, acoustic velocity anisotropy, and reflection coefficient of the rock and soil (the changes in parameters relative to the reference state), respectively. 1，1 、w 2，1 、w 3，1 、w 4，1 are the weight coefficients of rock and soil conductivity, dielectric constant, acoustic wave velocity and attenuation coefficient in the discriminant function of structural change caused by hydraulic action; w 5，1 、w 6，1、w 7，1 、w 8，1 are the weight coefficients of acoustic velocity anisotropy, reflection coefficient, conductivity, and dielectric constant in the discriminant function of structural change caused by stress.

[0168] When f water (m)>f stress If the load is less than 0.05, it is judged that the structural change is caused by hydraulic action; otherwise, it is judged that the structural change is caused by stress action.

[0169] Step 4.2, transfer entropy causal relationship analysis;

[0170] To determine the correlation between internal structural changes and potential disasters, the transfer entropy method is used to conduct causal relationship analysis. Transfer entropy is a measure in information theory that quantifies the degree of influence of one random process on another random process. It is defined as:

[0171]

[0172] Among them, TE X→Y is the transfer entropy from process X to process Y (indicating the causal influence of X on Y), H(·) is the Shannon entropy (uncertainty measure in information theory), represents the state of process Y at time t2, represents the state history of process Y from t2-k1 to t2-1 (the state sequence of the past k1 time steps), represents the state history of process X from t2-l1 to t2-1 (the state sequence of the past l1 time steps), represents the conditional entropy of the current state of Y given the past state of Y, It represents the conditional entropy of the current state of Y given both the past states of Y and X.

[0173] Therefore, by calculating the transfer entropy between different structural change characteristics and historical disaster events, we can quantify the contribution of various structural changes to potential disasters. The greater the transfer entropy, the stronger the causal relationship between the structural change characteristic and the disaster, and the higher the weight should be given in risk assessment.

[0174] Step 4.3: Modeling and evolution prediction of time-varying attribute graphs;

[0175] In one embodiment of the present application, in order to predict the future change trend of the internal structure of the rock and soil mass, a time-varying attribute graph model is constructed, and the underground structure is represented as a time-varying attribute graph:

[0176]

[0177] in The graph structure at time t3 (indicating the underground structure state at a specific time point), is the node set at time t3 (representing discrete points in space), is the edge set at time t3 (representing the connection relationship between nodes), It is the attribute set at time t3 (including the physical parameters of each node and edge).

[0178] According to the implementation of the present application, a series of time series diagrams are constructed based on historical monitoring data:

[0179]

[0180] in They represent the underground structure diagrams at time t3-n3, t3-n3+1, and t3 respectively, where n3 is the number of historical time steps;

[0181] Then apply graph convolutional neural network to predict future states:

[0182]

[0183] in Respectively represent the future time t3+1 and the future time t3 + 2 underground structure diagram;

[0184] The basic operations of graph convolutional neural networks are:

[0185]

[0186] in and are the node feature matrices of the l2th layer and the l2+1th layer respectively, To add a self-connected adjacency matrix (the original adjacency matrix plus the identity matrix so that the node contains its own information), for degree matrix (diagonal matrix, each diagonal element represents the degree of the corresponding node), is the weight matrix of the l2th layer (learnable parameters of the neural network), σ act is a non-linear activation function.

[0187] In the specific implementation of this application, the time-varying attribute graph model adopts a multi-layer structure, including:

[0188] Spatial graph convolution layer: Captures the spatial relationship between nodes, where each node represents a spatial location within the monitoring area. Node features include the physical parameters of that location (such as acoustic wave velocity, electrical conductivity, etc.). The weight of the spatial edge is determined based on the physical distance and geological similarity between nodes:

[0189]

[0190] in is the weight of the edge between node i2 and node j2 (indicating the connection strength between the two nodes), is the spatial distance between node i2 and node j2 (Euclidean distance in physical space), and are the node feature vectors of node i2 and node j2 respectively, and are the scaling parameters of spatial distance and eigenvector difference, respectively. exp(·) is the exponential function, and ||·|| represents the vector norm.

[0191] Temporal Attention Layer: Captures the temporal evolution of node attributes and learns the importance weights of data at different time points through the self-attention mechanism:

[0192]

[0193] in For node i3 at time t4 The attention weight (representing the historical time point importance to the current forecast), Indicates that node i3 has a historical moment at time t4 The unnormalized attention score of Indicates that node i3 has a historical moment at time t4 The unnormalized attention score of is the hidden state of node i3 at time t4, Indicates that node i3 is at the historical moment The hidden state, and are the learnable parameter matrices for transforming query vectors and key vectors, respectively, and a score (·) is the attention score function (a function that calculates the similarity between two vectors), represents the sum of the past T1 time steps, and exp(·) is an exponential function.

[0194] Spatiotemporal fusion layer: Fusion of spatial and temporal information to generate predictions of future node states:

[0195]

[0196] in is the predicted hidden state of node i4 at time t5+1 (the predicted feature representation at the next moment), is the neighbor set of node i4 (all nodes directly connected to node i4), is the spatial transformation matrix (parameter matrix for processing spatial information), is the time transformation matrix (parameter matrix for processing time information), is the output transformation matrix (generates the parameter matrix of the final prediction), σ act (·) is a nonlinear activation function (introducing nonlinear transformation capabilities), ∑ represents the summation operator, is the weight of the edge between node i4 and node j3, For node i4 at time t5 The attention weight, is the hidden state of node j3 at time t5, [·||·] represents the concatenation of spatial aggregation information and temporal aggregation information, Indicates that node i4 is at the historical moment The hidden state.

[0197] In typical applications, such as landslide early warning in mountainous areas, the model can identify the formation and development of potential sliding surfaces by analyzing the spatiotemporal patterns of acoustic and electromagnetic parameters. The system updates model predictions in real time, and as monitoring data accumulates, prediction accuracy continues to improve.

[0198] Graph convolutional neural networks capture the spatial relationships and temporal evolution characteristics between nodes through multi-layer graph convolution operations, and predict the future state of the internal structure of geotechnical bodies.

[0199] Step 4.4, risk level assessment and early warning;

[0200] Based on the analysis of the causes of structural changes and the results of evolution prediction, a comprehensive assessment of the potential disaster risk level is conducted. The risk assessment model considers the following factors:

[0201] The severity of structural changes: including crack size, density and distribution;

[0202] The rate of development of structural change: Analyze the acceleration of structural change through time series data;

[0203] Type of structural change cause: hydraulic action and stress action have different risk characteristics;

[0204] Historical case similarity: the degree of similarity with historical disaster cases.

[0205] In addition, the risk level calculation formula is:

[0206] R risk =S sev ·V rate ·C cause ·H hist ;

[0207] where R risk is the comprehensive risk value (indicating the comprehensive risk level of potential disasters), S sevis the severity of structural change (quantifying the spatial extent and intensity of structural change), V rate is the development rate (the time derivative of structural change, indicating the speed of change), C cause is the risk coefficient corresponding to the cause type (risk weight of different cause types), H hist is the historical case similarity (the degree of matching with historical disaster cases).

[0208] According to the calculated risk value R risk The system divides risk levels into four levels: low risk (green warning), medium risk (yellow warning), high risk (orange warning) and extremely high risk (red warning).

[0209] Therefore, through this step, the system realizes the precise analysis of the causes of changes in the internal structure of rock and soil and the scientific assessment of potential disaster risks, providing a decision-making basis for the subsequent generation of graded warning information.

[0210] A graded warning information generation module is used to generate targeted graded warning information based on the potential geological disaster risk results assessed by the structural change cause analysis and risk assessment module;

[0211] This step generates targeted graded warning information based on the risk assessment results, and specifically includes the following sub-steps:

[0212] Step 5.1, determine the warning level;

[0213] According to the embodiment of the present application, the warning level is determined based on the risk value and risk level obtained by risk assessment. The warning level is divided into four levels:

[0214] Green warning (Level 1): The risk value is low, the geological structure changes are minor, and the possibility of disasters in the short term is low;

[0215] Yellow Alert (Level 2): ​​The risk value is moderate, the geological structure changes are obvious, and enhanced monitoring is required;

[0216] Orange Alert (Level 3): The risk value is high, the geological structure changes are serious and accelerating, and there is a high possibility of disasters;

[0217] Red alert (Level 4): The risk value is extremely high, the geological structure changes are extremely serious and developing rapidly, and there is a high possibility of disasters occurring in the short term.

[0218] The warning level is determined by considering the absolute size and relative change trend of the risk value. When the risk value exceeds the preset threshold or the risk growth rate exceeds the preset rate, the corresponding level of warning is triggered.

[0219] Step 5.2, generating warning information content;

[0220] In another embodiment of the present application, based on the warning level and the structural change cause analysis results, the warning information content is generated. The warning information includes the following main contents:

[0221] Warning level and area: clearly state the warning level and the scope of impact;

[0222] Description of structural changes: describe the location, extent, degree and type of structural changes within the rock mass;

[0223] Cause analysis: explain the main causes of structural changes (hydraulic action or stress action);

[0224] Disaster type prediction: predict the type of disaster that may occur (such as landslide, collapse, etc.);

[0225] Development trend: describe the development trend and possible time window of structural change;

[0226] Prevention and control recommendations: Provide corresponding prevention and control recommendations based on different causes and disaster types.

[0227] The content of the warning information is adjusted according to the different recipients (such as professional and technical personnel or the general public) to ensure that the information is accurate and easy to understand.

[0228] In some embodiments, for professional technicians, the early warning information may include detailed monitoring data charts, three-dimensional visualization models of structural changes, and quantitative risk assessment indicators; while for the general public, it provides concise and easy-to-understand warning level descriptions, intuitive maps of possible impact areas, and clear response recommendations.

[0229] Optionally, early warning information can be tailored to the characteristics of different disaster types. For example, for landslides, the focus is on the location of the potential sliding surface, the volume of the sliding body, and the possible direction of movement; for collapses, the focus is on the location of the surface subsidence center, the subsidence range, and the subsidence rate; and for rockfalls, the focus is on the location, volume, and possible collapse path of the unstable rock mass.

[0230] Step 5.3: Generate targeted prevention and control recommendations;

[0231] According to the implementation method of this application, the system automatically generates targeted prevention and control suggestions for structural changes caused by different factors:

[0232] For structural changes caused by hydraulic action:

[0233] Provide drainage measures, such as setting up drainage ditches and precipitation wells;

[0234] Provide suggestions for anti-seepage measures, such as anti-seepage membranes, grouting, etc.;

[0235] Provide water flow control recommendations, such as changing the path of surface runoff.

[0236] For structural changes caused by stress:

[0237] Provide suggestions for support and reinforcement measures, such as anchor rods, anti-slide piles, etc.;

[0238] Provide load adjustment suggestions, such as reducing overhead loads, adjusting slope angles, etc.

[0239] Provide stress relief suggestions such as pre-splitting blasting, unloading, etc.

[0240] In addition, the prevention and control recommendations are based on successful experiences in similar situations in the historical case library and are generated in combination with the specific conditions of the current monitoring area. They are highly targeted and operational.

[0241] Step 5.4: Push and feedback of warning information;

[0242] In another embodiment of the present application, the system pushes warning information to relevant management departments and technical personnel according to a preset push strategy. Simultaneously, a warning information feedback mechanism is established to receive user feedback and continuously optimize the warning information generation process. Through this closed-loop optimization, the system's warning effectiveness will be continuously improved.

[0243] Therefore, through this step, the system realizes the generation of geological disaster graded warning information based on multi-physical field monitoring data, providing a scientific basis and decision-making support for geological disaster prevention and control.

[0244] like Figures 2 to 5 As shown, there are bar charts comparing the performance of different early warning methods; radar charts for multi-dimensional performance evaluation; radar charts for model robustness comparison; scatter plots of the correlation between structural changes and disaster risks; and line charts of the evolution of early warning system performance over time.

[0245] A computer storage medium includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the above-mentioned geological disaster early warning device.

[0246] Here, the present invention provides an implementation example:

[0247] This example application scenario involves a mountain roadside slope, approximately 45 meters high and 38 degrees steep, composed primarily of heavily weathered granite and residual soil. This slope has historically experienced numerous small-scale landslides, often associated with heavy rainfall. To prevent landslides, the multi-physics collaborative inversion geological disaster early warning system of this application was deployed on this slope.

[0248] A total of 16 sets of acoustic and electromagnetic fusion sensors were deployed at the top, middle, and foot of the slope to form a monitoring network. The acoustic sensors operate in a frequency range of 50 Hz to 10 kHz, and the electromagnetic sensors operate in a frequency range of 1 kHz to 100 MHz. All sensors transmit data in real time via wireless transmission to a data processing center near the slope. The data processing center is equipped with industrial-grade computing servers and runs the multi-physics collaborative inversion algorithm and early warning information generation system proposed in this application.

[0249] In June 2023, the region experienced continuous heavy rainfall, with cumulative rainfall reaching 347 mm. The monitoring system continued to operate after the rainfall began, collecting multi-physics field data in real time and performing processing and analysis. The following is a detailed implementation of the monitoring process:

[0250] Multi-physics data acquisition: The system collects complete acoustic and electromagnetic data every hour. As rainfall continues, the system automatically increases the sampling frequency to every 15 minutes. Acoustic data collected includes parameters such as longitudinal wave velocity, shear wave velocity, and reflection coefficient; electromagnetic data includes parameters such as conductivity and dielectric constant.

[0251] Data Processing and Feature Extraction: The raw data was processed using an adaptive filtering algorithm to eliminate noise caused by rainfall and environmental vibration. The processed acoustic data showed that the longitudinal wave velocity at a depth of 10 to 15 meters in the middle of the slope gradually decreased from an initial 2840 m / s to 2480 m / s, a decrease of approximately 12.7%. Electromagnetic data showed that the electrical conductivity in the same area increased from 0.023 S / m to 0.086 S / m, an increase of approximately 274%.

[0252] Multi-physics collaborative inversion: The system uses a Bayesian multi-parameter inversion model, along with acoustic and electromagnetic data. The inversion results reveal the formation of an arc-shaped potential slip surface within the slope, ranging in depth from 8 to 16 meters and approximately 50 meters long. Pore water pressure in the slip surface area has increased, and shear strength has decreased by approximately 35%.

[0253] Structural Change Cause Analysis and Risk Assessment: The system determined through discriminant model analysis that the structural changes were primarily caused by water infiltration from rainfall, with hydraulic forces being the primary cause. Analyzing historical monitoring data using a time-varying attribute graph model, the system predicted that if rainfall continued, the sliding surface would be broken through within 36 hours, triggering a landslide. The calculated comprehensive risk value was 0.75, corresponding to an orange alert (Level 3).

[0254] Generation of graded warning information: The system generates targeted warning information, including: warning level (orange warning), impact area (slope and 100 meters below), expected time of occurrence (within 36 hours), and prevention and control recommendations (immediately close the highway, set up a temporary retaining wall at the foot of the slope, and excavate a drainage ditch at the top of the slope). Warning information is pushed to highway management departments and local emergency management departments.

[0255] Upon receiving the early warning, relevant authorities immediately took action, closing the highway, erecting a temporary retaining wall at the foot of the slope, and digging a diversion ditch at the top to divert runoff. After 28 hours of continuous rainfall, the monitoring system detected significant deformation within the slope. However, due to timely preventative measures, the sliding surface was not completely penetrated, resulting in only small, localized slippage on the slope surface, which avoided causing major damage.

[0256] Through this practical application, the two main technical effects of the technical solution of this application have been verified:

[0257] Timely warning: The system issued an accurate warning 28 hours before the landslide occurred, buying ample time for disaster prevention and mitigation. Compared to traditional monitoring methods (which typically issue warnings 6 to 12 hours before a landslide occurs), this system's warning time was 16 to 22 hours earlier, improving timeliness.

[0258] Early warning accuracy: The system accurately identified the primary cause of the landslide (water infiltration caused by rainfall) and the location of the potential sliding surface. Field verification and subsequent drilling confirmed that the actual sliding surface deviated from the system's predicted location by less than 1.5 meters, with depth prediction accuracy exceeding 90%. Compared to traditional single-field physical field monitoring, this system's inversion accuracy improved by approximately 42%.

[0259] In addition, through this practical application, the system has accumulated new monitoring data and cases, further enriched the historical case library, and provided a more reliable reference basis for future early warning work.

[0260] According to the embodiments of the present application, the geological disaster early warning method based on multi-physics field collaborative inversion has the following technical effects:

[0261] High-precision monitoring of the internal structure of rock and soil: Through acoustic and electromagnetic fusion sensing technology, high-precision monitoring of changes in the internal structure of rock and soil is achieved. Internal cracks with a diameter of more than 5 mm can be identified. The monitoring accuracy is 60% higher than that of traditional surface monitoring methods, and possible landslides or collapse events can be predicted 3 to 7 days in advance.

[0262] Accurate identification of the causes of structural changes: Using a multi-physics field collaborative analysis method, we can accurately distinguish between structural changes caused by hydraulic effects and stress effects, improving the relevance of early warnings by 45%, significantly reducing the false alarm rate, and enabling the targeted implementation of disaster prevention and mitigation measures.

[0263] Efficient computing processing capabilities: Through eigenmode decomposition and parallel computing technology, the computational efficiency of multi-physics field inversion is increased by 8 times, enabling the system to achieve real-time monitoring and analysis under complex geological conditions with limited computing resources.

[0264] Reliable disaster risk assessment: Based on transfer entropy causal relationship analysis and graph convolutional neural network prediction, a quantitative correlation model between internal structural changes and potential disasters was constructed. The early warning accuracy rate was increased from 70% of traditional methods to 92%, providing sufficient time and reliable basis for disaster prevention and evacuation.

[0265] Targeted prevention and control suggestions: Based on the different causes of geological disasters, the system can generate targeted prevention and control suggestions, making disaster prevention and mitigation measures more efficient and reducing the risk of casualties and property losses.

[0266] Strong system adaptability: This method is applicable to a variety of geological environments and disaster types, including mountain landslides, mining subsidence, reservoir bank stability monitoring, etc., and has wide application value.

[0267] like Figures 2 to 5 As shown in the figure, the performance comparison of different early warning methods; the results of multidimensional performance evaluation; the correlation between structural changes and disaster risks; and the evolution of early warning system performance over time are shown.

[0268] In summary, the geological disaster early warning method based on multi-physical field collaborative inversion provided by this application has improved the warning accuracy, timeliness, pertinence and reliability, and provides an innovative technical solution for geological disaster prevention and control.

[0269] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A geological disaster early warning device, characterized in that: include: Acoustic and electromagnetic fusion sensor arrays are used to collect multi-physics field data related to changes in the internal structure of rock and soil; The data preprocessing and feature extraction module is used to preprocess and extract features from the original multi-physics field data collected by the acoustic and electromagnetic fusion sensor array to identify the internal structural change characteristics of the rock and soil mass; The multi-physics collaborative inversion module is used to reconstruct the internal structural distribution of the rock and soil mass using a multi-physics collaborative inversion algorithm based on the internal structural change characteristics of the rock and soil mass identified by the data preprocessing and feature extraction modules; The structural change cause analysis and risk assessment module is used to analyze the causes of internal structural changes in rock and soil based on the internal structural distribution results reconstructed by the multi-physics field collaborative inversion module, and to assess the potential geological disaster risks; The graded warning information generation module is used to generate targeted graded warning information based on the potential geological disaster risk results assessed by the structural change cause analysis and risk assessment module.

2. A geological disaster early warning device according to claim 1, characterized in that: The acquisition of multi-physics field data includes: Deploy acoustic and electromagnetic fusion sensor arrays based on the topographic and geological conditions of the monitoring area; Using acoustic electromagnetic fusion sensor array to simultaneously collect acoustic parameters and electromagnetic parameters; The collected raw data are preliminarily verified and transmitted to the data processing center.

3. A geological disaster early warning device according to claim 1, characterized in that: The preprocessing and feature extraction of the original multi-physics field data collected by the acoustic electromagnetic fusion sensor array includes: Apply adaptive filtering algorithm to pre-process the raw data to eliminate environmental noise and system errors; Perform spectrum analysis on the preprocessed acoustic data to extract acoustic features; Analyze the preprocessed electromagnetic data and extract electromagnetic features; The cross-correlations between acoustic and electromagnetic parameters are calculated, and the relationship expressions between physical parameters are established.

4. A geological disaster early warning device according to claim 1, characterized in that: The multi-physics collaborative inversion algorithm includes: Apply the eigenmode decomposition algorithm to reduce the dimensionality of the extracted high-dimensional feature data; Construct a Bayesian multi-parameter inversion model; The Markov chain Monte Carlo method is used for numerical solution; Acoustic data and electromagnetic data are used simultaneously to collaboratively constrain the inversion results.

5. A geological disaster early warning device according to claim 4, characterized in that: The Markov Chain Monte Carlo method adopts an improved Metropolis Hastings algorithm and improves sampling efficiency through adaptive proposal distribution.

6. A geological disaster early warning device according to claim 1, characterized in that: The structural change cause analysis and risk assessment module includes: Using multi-physics inversion results, we can distinguish between structural changes caused by hydraulic and stress effects. Apply the transfer entropy method to conduct causal analysis and determine the correlation between internal structural changes and potential disasters; Build a time-varying attribute graph model to represent underground structures as a time-varying attribute graph, and apply graph convolutional neural networks to predict future states; Based on the analysis of the causes of structural changes and evolution prediction results, a comprehensive assessment of the potential disaster risk level is conducted.

7. A geological disaster early warning device according to claim 6, characterized in that: The time-varying attribute graph model includes: Spatial graph convolution layer, which captures the spatial relationship between nodes; Temporal attention layer, which captures the temporal evolution of node attributes; The spatiotemporal fusion layer fuses spatial and temporal information to generate predictions of the future states of nodes.

8. A geological disaster early warning device according to claim 1, characterized in that: The hierarchical warning information generation module includes: Determine the warning level based on the risk value and risk level obtained from the risk assessment; Generate warning information content based on the warning level and structural change cause analysis results; Generate targeted prevention and control suggestions based on structural changes of different causes; Push early warning information to relevant management departments and technical personnel.

9. A geological disaster early warning device according to claim 8, characterized in that: There are four levels of warning, namely green warning, yellow warning, orange warning and red warning; The warning information is expressed in different ways according to the recipients.

10. A computer storage medium, characterized in that It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement a geological disaster early warning device according to any one of claims 1 to 9.

Images