Dam safety monitoring system and method based on digital twinning

Abstract

The invention discloses a dam safety monitoring system and method based on digital twinning, and belongs to the technical field of data processing, and the method comprises the steps: collecting dam surface acoustic emission signals and internal infrared thermal image data in real time through a multi-source sensor, and generating a multi-dimensional damage feature matrix; constructing a disaster node topology network in combination with density peak clustering and dynamic weight; on the basis of gravitational field intensity calculation and improved Monte Carlo simulation, a damage chain reaction path is accurately predicted; generating a dynamic risk thermodynamic diagram through space-time folding mapping and a bidirectional diffraction equation; and extracting hidden danger indexes by adopting multi-axis domain characteristic distillation, dynamically comparing a safety envelope function, and generating a flood discharge scheduling and reinforcing scheme on line. According to the method, cross-modal fusion, a gravitational field dynamic topology network and a time-space folding mapping technology are adopted, deep fusion of accurate simulation of a damage energy conduction path, carbonization-deformation bidirectional coupling prediction and intelligent risk hidden danger disposal decision can be realized, and the accuracy and initiative of dam safety early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a dam safety monitoring system and method based on digital twins. Background Art

[0002] Traditional dam safety monitoring technologies mostly rely on the discrete measurement of a single physical quantity, such as local parameter monitoring such as displacement sensors and piezometers, which have problems such as difficulty in fusing multi-source heterogeneous data and insufficient revelation of potential damage evolution mechanisms. Existing methods make it difficult to achieve cross-scale dynamic coupling analysis of dam structural damage, resulting in limited prediction accuracy of hidden danger chain reaction paths. Although digital twin technology has been applied to dam monitoring, the existing digital model focuses on geometric information mapping and data visualization, and fails to deeply integrate algorithmic innovations such as cross-modal convolution kernels, gravitational field dynamic topology, and space-time folding mapping, resulting in the inability of the twin to truly reflect the disaster process under the coupling of multiple physical fields. Therefore, it is urgent to build an integrated monitoring system that can connect the evolution of microscopic damage mechanisms and dynamic assessment of macroscopic risks to improve the accuracy and initiative of dam safety warnings. Summary of the invention

[0003] To solve the above problems, the present invention provides a dam safety monitoring system and method based on digital twins, which adopts cross-modal fusion, gravitational field dynamic topological network and space-time folding mapping technology, and can realize the deep integration of accurate simulation of damage energy conduction path, carbonization-deformation bidirectional coupling prediction and intelligent disposal decision-making of risk hazards, thereby improving the accuracy and initiative of dam safety early warning.

[0004] The above objectives can be achieved through the following solutions: A dam safety monitoring method based on digital twins comprises: obtaining a distribution map of acoustic emission signals on the dam surface and an internal infrared thermal image matrix, and generating a multidimensional damage feature matrix including a strain energy density field through cross-modal convolution kernel fusion calculation; establishing a disaster node topological network based on the spatial coordinates of the multidimensional damage feature matrix; inputting the node weights in the topological network into a gravitational field strength calculation model, simulating the energy conduction path, and outputting a set of chain reaction probability values ​​between adjacent nodes in the path; performing a spatiotemporal folding mapping operation on the multidimensional damage feature matrix according to the chain reaction probability value set, and establishing a bidirectional derivative relationship between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation. The carbonization-deformation coupling prediction curve is generated by using the projection equation. Based on the carbonization-deformation coupling prediction curve, the damage energy exchange rate at each monitoring point is calculated, and the energy exchange rate is multiplied by the chain reaction probability value at the corresponding position by nonlinear Gram matrix multiplication to generate a standardized risk coupling coefficient heat map. Multi-axis domain feature distillation is performed on the risk coupling coefficient heat map to extract hidden danger quantitative indicators including the sudden change gradient of gravitational field intensity, the spectrum entropy of the seepage phase, and the convergence degree of approaching energy, and dynamically compare them with the safety envelope function in the knowledge base. When any hidden danger quantitative indicator exceeds the preset envelope boundary, a disposal plan is generated including a microscopic repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector map.

[0005] Optionally, the cross-modal convolution kernel fusion calculation includes: performing wavelet packet decomposition on the acoustic emission signal distribution map to generate a multi-band energy distribution tensor; performing edge enhancement based on a morphological gradient operator on the internal infrared thermal imaging matrix, and outputting a spatial gradient matrix of the thermal conduction abnormality area; inputting the multi-band energy distribution tensor and the spatial gradient matrix into a dual-channel residual network for cross-modal feature fusion, optimizing the fusion weights through the strain energy density field loss function, and generating the multi-dimensional damage feature matrix.

[0006] Optionally, the establishment of the disaster node-based topological network includes: based on the strain energy density field distribution of the multi-dimensional damage characteristic matrix, using a density peak clustering algorithm to screen the potential disaster node set; performing Delaunay hierarchical decomposition on the node set, constructing a three-dimensional space grid model and dividing it into multiple grid units; calculating the energy gradient change rate between adjacent grid units to generate a dynamic weight matrix; and using the dynamic weight matrix to output a weighted disaster node-based topological network.

[0007] Optionally, the simulated energy conduction path includes: inputting the dynamic weight matrix of the topological network into the gravitational field model to calculate the potential field matrix between nodes; using the potential field matrix and the node set, adopting an improved Markov chain Monte Carlo algorithm to simulate the energy conduction path, outputting a set of chain reaction probability values ​​between adjacent nodes, and storing it in a dynamic risk probability database.

[0008] Optionally, the space-time folding mapping operation includes: dividing the multi-dimensional damage feature matrix into frames according to time series, combining the space-time distribution of the chain reaction probability value set, and constructing a four-dimensional space-time hypercube; performing discrete cosine transform on the carbonization depth axis of the four-dimensional space-time hypercube to generate a frequency domain projection subspace; performing second-order difference calculation on the macroscopic displacement accumulation axis of the four-dimensional space-time hypercube to extract frequency domain-time domain features; solving the coupling coefficient tensor of the bidirectional diffraction equation, generating a relationship curve between the carbonization depth change rate and the macroscopic displacement, and obtaining a carbonization-deformation coupling prediction curve.

[0009] Optionally, the calculation of the damage energy exchange rate includes: extracting the instantaneous slope sequence of each monitoring point along the carbonization-deformation coupling prediction curve; aligning the instantaneous slope sequence with the chain reaction probability value set according to the topological network node coding; fusing the aligned coding data with the ambient temperature and humidity time series data based on the Lagrange interpolation algorithm to generate a dynamic exchange rate correction factor; eliminating noise interference through principal component analysis and outputting the damage energy exchange rate.

[0010] Optionally, generating a standardized risk coupling coefficient heat map includes: mapping the damage energy exchange rate and the chain reaction probability value to the grid vertices of the topological network; constructing a three-dimensional covariance Gram matrix, and strengthening the nonlinear coupling effect through a hyperbolic tangent activation function; using the t-SNE algorithm in manifold learning to reduce the dimension to a two-dimensional plane; dynamically superimposing the risk coefficient distribution according to the time label of the space-time hypercube, and outputting a standardized heat map.

[0011] Optionally, the multi-axis domain feature distillation includes: calculating the directional differential operator along the thermal gradient axis of the standardized risk coupling coefficient heat map to extract the geometric feature vector of the region where the gravitational field intensity mutation occurs; performing fast Fourier transform on the thermal frames of continuous time periods to calculate the power spectrum entropy as the percolation phase spectrum entropy; fitting the Lyapunov exponent in combination with the energy convergence trajectory of the space-time hypercube to generate an approximate energy convergence degree; and outputting a set of hidden danger quantification indicators including mutation gradient, spectrum entropy and convergence degree.

[0012] Optionally, the dynamic comparison includes: inputting the hidden danger quantitative indicator set into the fuzzy Petri net reasoning engine in the knowledge base; matching the critical interval of the gravitational field strength elastic modulus and the seepage phase safety domain in the safety envelope function; when any indicator breaks through the envelope boundary, triggering the Monte Carlo tree search protocol of the three-dimensional decision engine; based on the optimal convergence solution of the energy conduction path and the coupling prediction curve, outputting the reinforcement strength vector diagram and the flood discharge scheduling sequence.

[0013] Based on the same inventive concept, the present invention also provides a dam safety monitoring system based on digital twins, the system comprising: a feature acquisition module, used to obtain the distribution map of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and generate a multidimensional damage feature matrix containing a strain energy density field through cross-modal convolution kernel fusion calculation; a disaster network construction module, used to establish a disaster node topological network based on the spatial coordinates of the multidimensional damage feature matrix; a reaction probability calculation module, used to input the node weights in the topological network into the gravitational field strength calculation model, simulate the energy conduction path, and output a set of chain reaction probability values ​​between adjacent nodes in the path; a carbonization-deformation coupling prediction curve generation module, used to perform space-time folding mapping operations on the multidimensional damage feature matrix according to the chain reaction probability value set, and establish a daily change rate of microscopic carbonization depth. A bidirectional diffraction equation between the sequence and the macroscopic displacement accumulation is used to generate a carbonization-deformation coupling prediction curve; a risk coupling coefficient heat map generation module is used to calculate the damage energy exchange rate at each monitoring point based on the carbonization-deformation coupling prediction curve, and perform nonlinear Gram matrix multiplication of the energy exchange rate with the chain reaction probability value at the corresponding position to generate a standardized risk coupling coefficient heat map; a hidden danger quantitative index extraction module is used to implement multi-axis domain feature distillation on the risk coupling coefficient heat map, extract hidden danger quantitative indicators including gravitational field intensity mutation gradient, seepage phase spectrum entropy, and approach energy convergence degree, and dynamically compare them with the safety envelope function in the knowledge base; a solution generation module is used to generate a disposal solution including a microscopic repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector map when any hidden danger quantitative indicator breaks through the preset envelope boundary.

[0014] Compared with the prior art, the present invention has the following advantages: 1. The present invention obtains the distribution spectrum of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and uses cross-modal convolution kernel fusion calculation to generate a multi-dimensional damage feature matrix containing the strain energy density field; this method effectively integrates multi-source heterogeneous data, overcomes the limitations of traditional single physical quantity monitoring, and provides more comprehensive and accurate damage information; 2. Based on the spatial coordinates of the multi-dimensional damage feature matrix, the present invention establishes a disaster node topological network and simulates the energy conduction path through the gravitational field strength calculation model; this method can accurately simulate the propagation process of damage inside the dam, providing a scientific basis for predicting potential disasters; 3. The present invention establishes a bidirectional diffraction equation between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation through time-space folding mapping operation, and generates a carbonization-deformation coupling prediction curve; this bidirectional coupling analysis can more accurately reveal the dam damage evolution mechanism and improve the prediction accuracy; 4. When any quantitative indicator of hidden danger exceeds the preset envelope boundary, the present invention can automatically generate a treatment plan including a microscopic repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector diagram; this intelligent treatment decision-making mechanism can take effective measures in time when a safety hazard occurs in the dam to avoid disasters.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of a dam safety monitoring method based on digital twins according to an embodiment of the present invention.

[0018] Figure 2 Schematic diagram of the carbonization-deformation coupling prediction curve of the embodiment of the present invention.

[0019] Figure 3 It is a structural schematic diagram of a dam safety monitoring system based on digital twins in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Reference Figure 1 An embodiment of the present invention proposes a dam safety monitoring method based on digital twins, which adopts cross-modal fusion, gravitational field dynamic topological network and space-time folding mapping technology, and can achieve deep integration of accurate simulation of damage energy conduction path, carbonization-deformation bidirectional coupling prediction and intelligent disposal decision-making of risk hazards, thereby improving the accuracy and initiative of dam safety early warning.

[0022] The method of this embodiment specifically includes: Obtain the distribution spectrum of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and generate a multi-dimensional damage feature matrix containing the strain energy density field through cross-modal convolution kernel fusion calculation; Based on the spatial coordinates of the multi-dimensional damage feature matrix, a disaster node topological network is established; Input the node weights in the topological network into the gravitational field strength calculation model, simulate the energy conduction path, and output a set of chain reaction probability values ​​between adjacent nodes in the path; According to the chain reaction probability value set, a time-space folding mapping operation is performed on the multi-dimensional damage characteristic matrix, a bidirectional diffraction equation between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation is established, and a carbonization-deformation coupling prediction curve is generated; Based on the carbonization-deformation coupling prediction curve, the damage energy exchange rate at each monitoring point is calculated, and the energy exchange rate is multiplied by the chain reaction probability value at the corresponding position by nonlinear Gram matrix to generate a standardized risk coupling coefficient heat map; Perform multi-axis domain feature distillation on the risk coupling coefficient heat map to extract hidden danger quantitative indicators including gravitational field intensity mutation gradient, percolation phase spectrum entropy, and convergence degree of approaching energy, and dynamically compare them with the safety envelope function in the knowledge base; When any quantitative indicator of hidden danger exceeds the preset envelope boundary, a disposal plan is generated including a micro-repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector diagram.

[0023] Specifically, this method establishes a digital twin through cross-modal fusion, dynamic topological modeling and multi-dimensional space-time analysis to achieve the connection between damage evolution mechanism and macro-risk. The principle is to map multi-source data such as acoustic emission and infrared thermal imaging into a unified damage feature matrix, construct a weighted topological network to simulate the energy conduction path, use space-time mapping and bidirectional diffraction equations to couple microscopic carbonization and macroscopic deformation parameters, and distill key hidden danger indicators through risk heat maps. The technical effect is reflected in improving the simulation accuracy of damage propagation paths, enhancing the spatiotemporal correlation of hidden danger prediction, realizing bidirectional coupling analysis of carbonization and deformation, and being able to actively generate targeted disposal plans, improving the early warning capability and response efficiency of dam safety monitoring.

[0024] Optionally, the cross-modal convolution kernel fusion calculation includes: Performing wavelet packet decomposition on the acoustic emission signal distribution spectrum to generate a multi-band energy distribution tensor; Specifically, wavelet packet decomposition is a signal processing method that decomposes the original signal into sub-signals covering different frequency bands by performing a finer frequency band division on the distribution spectrum of the acoustic emission signal, and generates a multi-band energy distribution tensor, which is a three-dimensional data structure used to record the distribution intensity of each frequency band energy at different spatial positions. Multi-band energy analysis can capture the low-frequency fracture waves and high-frequency microcrack signals generated during the material damage process, providing a basis for the feature identification of different damage modes.

[0025] Performing edge enhancement based on a morphological gradient operator on the internal infrared thermal image matrix, and outputting a spatial gradient matrix of the thermal conduction abnormality area; Specifically, edge enhancement based on morphological gradient operators is an image processing technology. The morphological gradient operator calculates the difference between expansion and corrosion in the local area of ​​the infrared thermal image matrix, strengthens the temperature gradient change characteristics of the abnormal heat conduction area, and outputs a spatial gradient matrix. The matrix value represents the degree of temperature change at each pixel point. High gradient values ​​correspond to potential heat conduction obstruction or material defect areas, which can effectively locate internal damage below the surface.

[0026] The multi-band energy distribution tensor and the spatial gradient matrix are input into a dual-channel residual network for cross-modal feature fusion, and the fusion weights are optimized through the strain energy density field loss function to generate the multi-dimensional damage feature matrix.

[0027] Specifically, the dual-channel residual network is a deep learning architecture with dual input branches that process the multi-band energy distribution tensor of acoustic emission and the spatial gradient matrix of infrared thermal image in parallel. The detailed information of the original mode is retained through residual connection in the feature fusion stage. The strain energy density field loss function is the optimization objective function during network training. By calculating the difference in energy distribution between the predicted strain energy density field and the actual physical model, the fusion weight of cross-modal features is dynamically adjusted. The multi-dimensional damage feature matrix finally generated integrates the frequency domain energy characteristics of the acoustic emission signal and the spatial temperature gradient characteristics of the infrared thermal image.

[0028] Among them, the input layer of the dual-channel residual network is the acoustic emission tensor (64×64×8) and the infrared gradient matrix (256×256), which are downsampled to the same dimension by the convolution kernel 3×3 plus the step size 2; the residual block contains 4 residual units per channel, and the number of channels is aligned by the 1×1 convolution kernel in the cross-modal fusion stage, and the fusion weight is optimized by back propagation of the strain energy density loss function; the training parameters are Dropout rate 0.3, Adam optimizer initial learning rate 0.001, batch size 32, and the training set is a cross-modal image database containing 200 dam damage cases.

[0029] Among them, the strain energy density field loss function is defined as: , In the formula, For the The predicted strain energy density value corresponding to each grid cell is: For the The actual measured value of strain energy density of each grid cell is is the total variation regularization coefficient, is the total variation regularization term, is the number of grid cells.

[0030] Specifically, the damage characterization capability is improved through fine processing of cross-modal data and deep feature fusion. The principle is to use wavelet packet decomposition to analyze the frequency domain characteristics of acoustic emission signals, combine morphological gradient operators to extract spatial anomaly characteristics of thermal imaging data, and fuse heterogeneous modal information through a dual-channel network. The technical effect is reflected in enhancing the accuracy of capturing weak signals in the early stage of material damage, suppressing environmental noise interference of single sensor data, and ensuring that the fused multi-dimensional damage feature matrix contains both frequency domain energy distribution and spatial thermodynamic anomaly information, providing a more accurate input basis for subsequent topological network construction and risk prediction.

[0031] Optionally, the establishment of the disaster node topology network includes: Based on the strain energy density field distribution of the multi-dimensional damage characteristic matrix, a density peak clustering algorithm is used to screen the potential disaster node set; Specifically, the density peak clustering algorithm is an unsupervised learning method based on the local density and relative distance of data points. By calculating the local density of each area in the strain energy density field and the minimum distance to the higher density point, the potential node set of disasters is screened out, that is, the relatively isolated damage core area with high strain energy density. The algorithm can automatically identify high energy aggregation points and exclude low-density noise interference, ensuring that the node set covers the real potential damage location.

[0032] Among them, in the density peak clustering algorithm, the neighborhood radius is set according to the geometric dimensions of the dam, and the neighborhood radius is 0.1 times the maximum span of the dam.

[0033] Perform Delaunay hierarchical decomposition on the node set to construct a three-dimensional space grid model and divide it into multiple grid units; Specifically, Delaunay hierarchical decomposition is a three-dimensional space meshing method that generates a non-overlapping tetrahedral unit network according to the triangulation criterion for the points where the disaster nodes are concentrated, and constructs a three-dimensional space mesh model with optimized structure. Each mesh unit uses the Delaunay condition to ensure that any point in the space is not located within the circumscribed sphere of other tetrahedrons. The formed mesh topology has geometric stability and can accurately describe the spatial distribution characteristics of the damaged nodes.

[0034] Calculate the energy gradient change rate between adjacent grid cells and generate a dynamic weight matrix; Specifically, the energy gradient change rate is a calculation index for the change rate of strain energy density between adjacent grid units. By comparing the average strain energy density difference in adjacent tetrahedral units and normalizing it according to the grid edge length, a dynamic weight matrix is ​​generated. This matrix quantifies the energy transfer intensity between nodes, and the weight value reflects the driving potential of damage diffusion from high-energy units to low-energy units, giving the topological network dynamic evolution characteristics.

[0035] The dynamic weight matrix is ​​used to output a disaster node topological network with weights.

[0036] Specifically, the weighted disaster node topological network is a network model that integrates the spatial distribution of nodes and the intensity of energy conduction. The dynamic weight matrix annotates the connection strength between nodes in the form of edge weights to achieve a simulatable expression of the damage propagation path. The network can characterize the interaction relationship between multiple damage areas in the dam and provide structured input data for subsequent gravitational field simulations.

[0037] Specifically, a dynamic weighted disaster topology network is constructed through density clustering and spatial subdivision. The principle is based on screening key damaged core nodes and dynamically quantifying the interaction intensity with energy gradients. The technical effect is reflected in improving the recognition accuracy of the spatial distribution of damaged nodes, enhancing the expression flexibility of complex structures through tetrahedral mesh subdivision, and accurately describing the unevenness of energy conduction paths using dynamic weight matrices, providing a network topology foundation with clear physical meaning for subsequent chain reaction probability calculations, thereby improving the authenticity and reliability of dam damage evolution simulation.

[0038] Optionally, the simulated energy conduction path includes: Inputting the dynamic weight matrix of the topological network into the gravitational field model to calculate the potential field matrix between nodes; Specifically, the gravitational field model is a mechanical simulation method based on the classical Newton's law of gravity and topological weights. The connection strength between nodes in the dynamic weight matrix is ​​mapped to the gravitational interaction relationship. The potential field matrix is ​​a mathematical representation matrix of the gravitational potential energy between nodes. It is obtained by calculating the spatial relationship between the node mass attributes and the topological weights. Its element values ​​reflect the interaction potential energy intensity of adjacent nodes in the energy conduction process, and are used to quantitatively describe the mechanical correlation between the damaged areas inside the dam.

[0039] Among them, the node quality is the integral value of the strain energy density of the corresponding grid unit, the gravitational constant Adjust according to the material Poisson's ratio, ,in is the elastic modulus of the material, is Poisson's ratio; weight matrix Step by time according to , where , is the aging coefficient, which is set according to the concrete grade. For example, for C30 concrete, =0.05; the iteration termination condition is that the path energy standard deviation is less than 0.1 times the maximum potential energy peak or the maximum step size condition is met.

[0040] The potential energy field matrix and the node set are used to simulate the energy conduction path using an improved Markov chain Monte Carlo algorithm, and a chain reaction probability value set between adjacent nodes is output and stored in a dynamic risk probability database.

[0041] Specifically, the improved Markov Chain Monte Carlo algorithm is a random path simulation method that combines topological weights and potential energy fields. It simulates the energy conduction process between nodes by iteratively generating state transition sequences. The improvements include the introduction of potential energy gradient as a transfer probability correction factor and a dynamic weight attenuation mechanism. The chain reaction probability value set is a multidimensional data set that records the energy conduction probability between adjacent nodes. The dynamic risk probability database is a structured storage system that stores the evolution of risk probability at different time steps, and supports real-time query and update.

[0042] Among them, the improvements in the improved Markov chain Monte Carlo algorithm include: using the node potential energy difference Modify the transition probability for the constraint condition and set the annealing coefficient Dynamic adjustment rules; set the iteration termination condition to the potential energy convergence of 100 consecutive simulations ≤0.01% or the maximum step length reaches , The value is related to the fatigue properties of the material. .

[0043] The transition probability formula is corrected to: , In the formula, For slave nodes Transfer to Node The probability value of For Node To Node The dynamic weight of is the annealing coefficient, For Node To Node The potential energy difference, The meaning of node All neighboring nodes The transfer probability is normalized to ensure that the probability sum is 1. Its physical meaning is to ensure that the energy is at the node The conduction path selection is completed within the neighborhood of .

[0044] Specifically, the mechanical interaction relationship between nodes is quantified through the gravitational field model, and the random path of energy conduction is simulated based on the improved Monte Carlo method. The principle is to use the potential energy calculation driven by physics to reveal the mechanical mechanism of damage propagation, and combine the probability simulation to capture the uncertainty of the conduction path. The technical effect is reflected in the dynamic weight and potential energy gradient synergistically driving the path simulation process, improving the physical rationality and computational efficiency of damage chain reaction prediction. The generated chain reaction probability value set can accurately reflect the spatiotemporal correlation of risk propagation between multiple nodes, and provide highly reliable dynamic risk data support for subsequent risk heat map generation and safety warning.

[0045] Optionally, the spatiotemporal folding mapping operation includes: Divide the multi-dimensional damage feature matrix into frames according to time series, and construct a four-dimensional space-time hypercube by combining the space-time distribution of the chain reaction probability value set; Specifically, the space-time folding mapping operation is a method for analyzing the space-time correlation of multidimensional data. It refers to the method of analyzing cross-scale correlation by projecting the time and space dimensions into the same hypercube coordinate system and using topological equivalent transformations. It is mathematically realized as the Hilbert space embedding of a four-dimensional tensor. The multi-dimensional damage feature matrix is ​​divided into frames according to the time series, and the spatial distribution structure of the chain reaction probability value is integrated to construct a four-dimensional space-time hypercube. The four-dimensional space-time hypercube is a high-dimensional data structure formed with three-dimensional space coordinates and time axis as the basis, which can simultaneously record the spatial evolution trajectory and probability coupling relationship of damage features in different time slices.

[0046] Performing discrete cosine transformation on the carbonization depth axis of the four-dimensional space-time hypercube to generate a frequency domain projection subspace; Specifically, discrete cosine transform is a signal frequency domain processing technology. Discrete cosine transform of the carbonization depth axis means converting the time-varying sequence of carbonization depth into a frequency domain projection subspace. By extracting the characteristics of different frequency components, the periodic evolution law of the carbonization rate is revealed. The frequency domain projection subspace is the main frequency energy distribution matrix of the carbonization process, which can distinguish the spectral differences between normal carbonization and abnormal accelerated carbonization.

[0047] Performing second-order difference calculation on the macroscopic displacement cumulant axis of the four-dimensional space-time hypercube to extract frequency-domain-time-domain features; Specifically, the second-order difference calculation is a method for extracting dynamic features of time series. The second-order difference calculation of the macro-displacement cumulative axis refers to quantifying the acceleration characteristics of the displacement change through the quadratic difference of adjacent time points. The frequency domain-time domain feature is a multidimensional feature set that integrates the frequency domain principal component and the time domain acceleration, which characterizes the inertial trend and sudden change fluctuation characteristics of the macro-displacement change.

[0048] The coupling coefficient tensor of the bidirectional diffraction equation is solved to generate the relationship curve between the carbonization depth change rate and the macroscopic displacement, and the carbonization-deformation coupling prediction curve is obtained.

[0049] Specifically, the two-way diffraction equation is a partial differential equation that couples carbonization and displacement variables. Solving the coupling coefficient tensor requires a set of matrix equations constructed by jointly constructing the frequency domain projection subspace and the second-order difference characteristics. The resulting carbonization-deformation coupling prediction curve is a dynamic correlation function of the daily change rate of carbonization depth and the accumulated displacement, which reflects the mutual influence mechanism between the two through the two-way action term.

[0050] in, , , where is the daily change rate of carbonization depth, is the accumulated macroscopic displacement, is the carbonization diffusion coefficient tensor, which is associated with the strain energy density field, is the coupling coefficient, which is calibrated by finite element simulation. The solution of the coupling coefficient is firstly normalized by orthogonalization of the time domain characteristic matrix and the frequency domain projection matrix, and then the least squares method is used to fit the displacement-carbonization coupling residual function.

[0051] Specifically, a bidirectional coupling model of carbonization and deformation is established through space-time folding mapping and multi-domain feature fusion. The principle is to decompose the four-dimensional space-time data into frequency domain-time domain features and solve the bidirectional action equation in parallel. The technical effect is reflected in breaking through the limitations of traditional single variable analysis, significantly enhancing the dynamic correlation modeling ability of microscopic carbonization process and macroscopic deformation, and accurately predicting the damage co-evolution trend through quantitative bidirectional coupling mechanism, providing multi-dimensional physical correlation basis for the full life cycle assessment of dam structure safety.

[0052] Optionally, the calculation of the damage energy exchange rate includes: Extracting the instantaneous slope sequence of each monitoring point along the carbonization-deformation coupling prediction curve; Specifically, the instantaneous slope sequence is a set of real-time change rates at each monitoring point of the carbonization-deformation coupling prediction curve, which is obtained by calculating the difference between adjacent time points on the curve, reflecting the dynamic relationship between the local deformation energy transfer rate and the carbonization process. This sequence can quantify the damage energy efficiency conversion rate at different locations at a specific moment, providing transient dynamic characteristics for energy exchange rate calculation.

[0053] Aligning the instantaneous slope sequence with the chain reaction probability value set according to topological network node coding; Specifically, the topological network node coding alignment is an operation that matches the instantaneous slope sequence with the chain reaction probability value according to the spatial coordinate and topological structure mapping relationship. It is necessary to achieve accurate correspondence between the time and space dimensions through the node unique identifier to ensure the homology of the deformation energy rate and the risk probability data, and establish the spatiotemporal consistency at the data level for subsequent fusion analysis.

[0054] The aligned encoded data is fused with the ambient temperature and humidity time series data based on the Lagrange interpolation algorithm to generate a dynamic exchange rate correction factor; Specifically, the Lagrange interpolation algorithm is a function interpolation method based on polynomial fitting. It generates a dynamic exchange rate correction factor by interpolating and fusing the aligned data with the ambient temperature and humidity time series data at the same time node. This factor quantifies the modulation effect of temperature and humidity changes on energy conduction efficiency and weakens the interference of environmental noise on the calculation of energy exchange rates.

[0055] The noise interference is eliminated by principal component analysis and the damage energy exchange rate is output.

[0056] Specifically, principal component analysis is a dimensionality reduction and denoising method that extracts the main components of the data through orthogonal transformation, decomposes the covariance matrix of the fused multi-source data, retains a few principal components that characterize the essential characteristics of the energy exchange rate, eliminates sensor noise and redundant coupling characteristics, outputs a damage energy exchange rate index with a high signal-to-noise ratio, and accurately characterizes the energy conduction efficiency per unit time of each monitoring point.

[0057] Specifically, the calculation accuracy of the damage energy exchange rate is improved through the spatiotemporal alignment of multi-source data and the correction of environmental parameters. The principle is to combine the dynamic characteristics of deformation slope and the risk probability information of topological network, integrate the influence of environmental factors through interpolation algorithm, and use principal component analysis to extract core features. The technical effect is reflected in enhancing the quantitative characterization ability of the damage energy conduction process, reducing the misleading of external environmental disturbances and data noise to the assessment model, ensuring that the generated energy exchange rate data can reliably support the calculation of subsequent risk coupling heat maps, and improving the robustness and accuracy of the dynamic assessment of dam safety risks.

[0058] Optionally, generating a standardized risk coupling coefficient heat map includes: Mapping the damage energy exchange rate and the chain reaction probability value to the grid vertices of the topological network; Specifically, the topological grid vertex is a set of connection points of the three-dimensional space model in the disaster node topological network. The grid vertex mapped to the topological network means that the damage energy exchange rate and the chain reaction probability value are assigned to the grid node as dynamic attributes, so that each vertex carries energy conduction efficiency and risk diffusion probability data at the same time. This operation establishes a strong correlation between data and spatial position, providing structured input for subsequent matrix operations.

[0059] Construct a three-dimensional covariance Gram matrix and enhance the nonlinear coupling effect through the hyperbolic tangent activation function; Specifically, the three-dimensional covariance Gram matrix is ​​a high-dimensional expansion matrix of the multivariate covariance relationship, which is generated by calculating the interaction covariance between the damage energy exchange rate, the chain reaction probability value and the space-time coordinates; the hyperbolic tangent activation function is a nonlinear function with an S-shaped curve, which compresses the input to the [-1,1] interval and enhances the nonlinear association, strengthens the complex coupling effect between the energy exchange rate and the probability value, and significantly improves the dynamic interaction feature expression ability between network nodes.

[0060] Use the t-SNE algorithm in manifold learning to reduce the dimension to a two-dimensional plane; Specifically, the t-SNE algorithm in manifold learning is a data dimensionality reduction method based on probability distribution. It projects the three-dimensional covariance matrix data onto a two-dimensional plane by retaining the probability distribution characteristics of local similarities in high-dimensional space. In the process of dimensionality reduction, the algorithm effectively distinguishes the clustering pattern and abnormal characteristics of energy conduction, and generates a risk distribution base image that can be intuitively interpreted.

[0061] The risk coefficient distribution is dynamically superimposed according to the time labels of the space-time hypercube, and a standardized heat map is output.

[0062] Specifically, the time label of the space-time hypercube is the index identifier that records the monitoring time in the four-dimensional data structure. The dynamic superposition of risk coefficient distribution refers to the weighted fusion of thermal layers under different time slices. The output of the standardized thermal map eliminates the dimensional difference through normalization processing, dynamically displays the gradient distribution of the global risk coupling intensity evolving over time, and realizes the spatial and temporal continuity expression of risk visualization.

[0063] Specifically, the risk heat map is generated through multi-dimensional matrix construction and nonlinear enhanced dimensionality reduction. The principle is to integrate the spatial covariance relationship between energy exchange rate and probability value, and reflect the risk evolution process through dynamic time superposition. The technical effect is reflected in breaking through the limitations of traditional static heat maps, enhancing the quantitative characterization ability of nonlinear coupling relationships, intuitively displaying the spatiotemporal aggregation characteristics and diffusion trends of damage propagation, providing high-resolution visualization tools for hidden danger identification, and supporting dynamic monitoring and decision-making intervention of risk evolution.

[0064] Optionally, the multi-axis domain feature distillation includes: Calculating a directional differential operator along the thermal gradient axis of the standardized risk coupling coefficient thermal map to extract geometric characteristic vectors of the region where the gravitational field intensity changes suddenly; Specifically, the directional differential operator is a mathematical tool for calculating the axial local variation intensity of the thermal gradient. Calculating the directional differential operator along the thermal gradient axis means extracting the geometric characteristic vector of the region where the gravitational field intensity changes suddenly by performing partial derivative operations on the thermal map along the spatial coordinate axis. This vector contains geometric properties such as the gradient amplitude and directional curvature of the sudden change region, and can quantitatively describe the degree of distortion and spatial distribution of the stress field inside the dam.

[0065] Perform fast Fourier transform on the thermal frames of continuous time periods and calculate the power spectrum entropy as the percolation phase spectrum entropy; Specifically, the fast Fourier transform is a classic algorithm for converting time domain signals into frequency domain. The fast Fourier transform of thermal frames in continuous time periods refers to converting the local thermal distribution data of the time series into spectral components. The power spectrum entropy is an entropy value indicator obtained by calculating the distribution disorder of the spectrum energy. As the seepage phase spectrum entropy, it directly reflects the complexity and randomness of the seepage process in the frequency domain. High entropy values ​​indicate that the seepage path is disordered or there is abnormal seepage behavior.

[0066] The time-frequency transformation parameters are FFT window length 2048 points, Hanning window, overlap rate 50%; the frequency band is divided into 16 sub-bands according to the critical frequency of concrete permeability, and the critical frequency is 0.1-10Hz; the power spectrum entropy , where For the The sub-band energy ratio is normalized to between 0 and 1; the safety domain is determined when the power spectrum entropy >0.85 and lasting for >1 hour, an abnormal seepage alarm is triggered.

[0067] Fitting the Lyapunov exponent in combination with the energy convergence trajectory of the space-time hypercube to generate an approximate energy convergence degree; The output includes a set of hidden danger quantitative indicators including mutation gradient, spectral entropy and convergence degree.

[0068] Specifically, the energy convergence trajectory is the historical evolution path of energy density in the space-time hypercube. The Lyapunov index is fitted with the energy convergence trajectory to calculate the divergence characteristic index by statistically analyzing the energy decay or accumulation rate over time. The generated approximate energy convergence degree is used to evaluate the stability of the energy conduction process in the dam damage area. The Lyapunov index is a dynamic index that quantifies the sensitivity of the system to initial conditions. Negative values ​​indicate that the energy tends to converge stably, and positive values ​​indicate a potential unstable state.

[0069] Among them, based on the energy convergence trajectory data of the space-time hypercube, the Wolf algorithm is used to fit the maximum Lyapunov exponent, the data sampling interval Δt=1h, and the embedding dimension d=3.

[0070] Specifically, the quantitative indicators of hidden dangers are extracted by distilling the multi-axis domain features of the heat map. The principle is to analyze the risk characteristics from three dimensions: geometric mutation, frequency domain chaos and dynamic stability. The technical effect is reflected in the comprehensiveness and physical interpretability of hidden danger indicators by integrating geometric, spectral and dynamic methods, locating stress concentration areas by quantifying mutation gradients, evaluating the degree of seepage anomalies by using spectral entropy, and predicting system stability by combining energy convergence, providing a multi-dimensional evaluation basis for the refined classification and precise intervention of dam safety risks.

[0071] Optionally, the dynamic comparison includes: Inputting the hidden danger quantitative indicator set into the fuzzy Petri net reasoning engine in the knowledge base; Match the critical interval of elastic modulus of gravitational field strength and the safe region of seepage phase in the safety envelope function; Specifically, the fuzzy Petri net reasoning engine is an intelligent decision-making tool that combines fuzzy logic and Petri net dynamic deduction rules. Inputting the hidden danger quantitative index set into the engine means defining the degree of deviation of the fuzzy membership function quantitative index from the safety threshold, triggering the fuzzy reasoning rule chain preset in the knowledge base, matching the critical interval of the gravitational field strength elastic modulus and the seepage phase safety domain in the safety envelope function. It is necessary to compare the membership relationship between the measured index and the envelope boundary to confirm whether each index exceeds the safety threshold of the material mechanical properties or seepage dynamics. Among them, the critical interval of the gravitational field strength elastic modulus is the allowable stress fluctuation range set according to the material performance experiment, and the seepage phase safety domain is the permitted operating domain of the seepage velocity and flow angle.

[0072] Among them, the critical interval of the elastic modulus of the safety envelope function is calibrated by the yield strength threshold of the dam material tensile-compression test, and the elastic modulus fluctuation range is generated as the safety domain boundary by combining the Monte Carlo sampling method, and the fluctuation range is ±10% of the reference value. For example, 100 groups of dam body concrete core samples are collected, and the elastic modulus is tested according to the ASTM C469 standard to calculate the dynamic fluctuation threshold. The dynamic fluctuation threshold is ±3 times the standard deviation of the reference value, and the confidence interval of ±10% is taken after excluding abnormal values; through finite element simulation verification: when the local elastic modulus value exceeds the dynamic fluctuation threshold, the displacement increment error is greater than 15%.

[0073] When any indicator breaks through the envelope boundary, the Monte Carlo tree search protocol of the three-dimensional decision engine is triggered; Based on the optimal convergence solution of the energy conduction path and the coupling prediction curve, a reinforcement strength vector diagram and a flood discharge scheduling sequence are output.

[0074] Specifically, the Monte Carlo tree search protocol of the three-dimensional decision engine is a random search algorithm based on a tree structure. After the protocol is triggered, the optimal response strategy is calculated by simulating countless potential decision paths. The optimal convergence solution is a comprehensive optimal solution that satisfies the low dissipation of the energy conduction path and the stability of the carbonization-deformation coupling prediction curve. The reinforcement strength vector diagram is a three-dimensional spatial distribution diagram that characterizes the strength and direction of reinforcement materials in different regions. The flood discharge scheduling sequence is an optimized operation instruction set that controls the reservoir flood discharge flow and time interval to ensure that the water pressure load is reduced before the risk is eliminated.

[0075] Specifically, fuzzy reasoning and Monte Carlo tree search are used to realize the dynamic generation of hidden danger warning and disposal solutions. The principle is to use fuzzy Petri nets to capture the fuzzy state migration of indicators that cross the boundary, and combine the tree search algorithm to quickly plan the optimal response strategy. The technical effect is reflected in breaking through the single judgment mode of traditional threshold alarms, realizing intelligent reasoning and hierarchical response of multi-indicator associated risks, and the generated reinforcement vector map and flood discharge sequence accurately match the current actual damage state and environmental load of the dam, significantly improving the scientific nature and implementation efficiency of safety disposal measures. Embodiment 1:

[0076] In order to verify the feasibility and effectiveness of the present invention in dam safety monitoring, a large concrete gravity dam is taken as an example for implementation. The dam is 185 meters high and has a total storage capacity of 12 billion cubic meters. It is equipped with an acoustic emission sensor array and an infrared thermal imaging system. The acoustic emission sensor array has 64 channels and the infrared thermal imaging system has a resolution of 0.05°C. A digital monitoring network covering the surface and internal structure of the dam is constructed.

[0077] During the flood season in 2023, the system collected acoustic emission signals on the dam surface and internal infrared thermal imaging data in real time. The sampling rate of the acoustic emission signal was 1MHz, and the internal infrared thermal imaging data was a full-dam scan every 30 minutes. At 10:23 on July 15, a sudden abnormality in the power spectrum of the acoustic emission signal was detected in the C12 area of ​​the dam section, and the characteristic frequency jumped from 25kHz to 82kHz. At the same time, the infrared thermal image showed that there was a low-temperature gradient abnormal area with a diameter of 3.2m in the area, ΔT=1.8℃ / m.

[0078] The acoustic emission signal was calculated by cross-modal convolution kernel fusion and decomposed into 8 frequency bands through wavelet packets to generate a 64×64×8 energy distribution tensor. After the infrared thermal image was processed by morphological gradient, the gradient peak was detected to be 4.8℃ / m². The final damage feature matrix showed that the strain energy density of the C12 area reached 5.7kJ / m³, with an average value of <1.2kJ / m³, which was determined to be a Class III damage area.

[0079] Based on the density peak clustering algorithm, 35 core nodes were selected to form the disaster network. Delaunay hierarchical subdivision generated 412 tetrahedral units, and the energy gradient between adjacent units was calculated. The energy gradient of the unit pair (C12-24, C12-25) was 4.7 kJ / m 4 , the energy gradient of the unit pair (C12-25, C12-26) is 8.3 kJ / m 4 , critical jump. The node weights are calculated by the gravitational field model, the gravitational constant G is 0.12MN·m / t², the concrete parameter is C30, and after 20,000 Monte Carlo simulations, the chain reaction probability values ​​of the C12 node group are output. The chain reaction probability of the associated nodes (C12-25, C13-03) is 0.78, the chain reaction probability of the associated nodes (C12-25, C11-17) is 0.65, and the chain reaction probability of the associated nodes (C12-25, spillway base F09) is 0.43.

[0080] The hypercube dimension generated by the space-time folding mapping has a space-time axis of 30 days, 72 time slices, and a coordinate accuracy of 0.1m; the deformation monitoring point AX05 shows a cumulative displacement of 12.7mm; the solution of the two-way diffraction equation shows that there is a strong coupling between the daily growth rate of carbonization depth and the displacement acceleration stage, such as Figure 2 As shown, the prediction curve shows that if the carbonization rate reaches 2.5 μm / d, the displacement will exceed the threshold after 28 days, and the fitting error is ±0.11 mm.

[0081] Through Gram matrix calculation, a standardized risk heat map is generated, which shows that the risk coefficient of the C12 core area is 0.93, the risk coefficient of the spillway base F09 is 0.68, and the risk coefficient of the left bank drainage corridor L24 is 0.55. The multi-axis domain feature distillation results are that the gravitational field gradient mutation is 0.43kN / m²·m, the threshold is greater than 0.33kN / m²·m; the seepage entropy value is 0.89, the safety domain is less than or equal to 0.8, the Lyapunov index is 0.12, and the steady state should be less than 0.

[0082] The comparison of the safety envelope function shows that the seepage entropy value is 0.89, which breaks the threshold and triggers the three-dimensional decision engine. The Monte Carlo tree search outputs the micro-repair plan: the coordinate set is the position of {C12-25(34.72M,121.85M,EL.152)}, and the epoxy mortar injection volume is 2.4m³; flood discharge scheduling optimization: the scheduling sequence is [0-2h: 500m³ / s to 2-6h:800m³ / s], and the gate opening G05 is controlled to be 62% and G07 is controlled to be 58%; reinforcement vector plan: the density of the anchor group is 3.2 roots / m² in the C12 area, the vector angle is 45 degrees, and the density in the F09 area is 2.1 roots / m², and the vector angle is 32 degrees.

[0083] After 48 hours of implementing the treatment plan, the strain energy density of the C12 area was re-measured and dropped to 1.8kJ / m³, a decrease of 68%; the seepage entropy dropped back to 0.7, within the qualified threshold; the displacement rate dropped to 0.13mm / d, a decrease of 59%. This embodiment verifies the dynamic monitoring capability of the system in a complex environment. Through multi-source data fusion and intelligent decision-making models, closed-loop management from early detection of damage to precise treatment is achieved, providing innovative solutions for the safe operation of the dam.

[0084] Based on the same inventive concept, Figure 3 As shown, the present invention also provides a dam safety monitoring system based on digital twins, the system comprising: The feature acquisition module is used to obtain the distribution map of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and generate a multi-dimensional damage feature matrix including the strain energy density field through cross-modal convolution kernel fusion calculation; A disaster network construction module is used to establish a disaster node topological network based on the spatial coordinates of the multi-dimensional damage feature matrix; A reaction probability calculation module, used to input the node weights in the topological network into the gravitational field strength calculation model, simulate the energy conduction path, and output a set of chain reaction probability values ​​between adjacent nodes in the path; A carbonization-deformation coupling prediction curve generation module is used to perform a spatiotemporal folding mapping operation on the multi-dimensional damage feature matrix according to the chain reaction probability value set, establish a bidirectional diffraction equation between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation, and generate a carbonization-deformation coupling prediction curve; A risk coupling coefficient heat map generation module is used to calculate the damage energy exchange rate at each monitoring point based on the carbonization-deformation coupling prediction curve, perform nonlinear Gram matrix multiplication on the energy exchange rate and the chain reaction probability value at the corresponding position, and generate a standardized risk coupling coefficient heat map; A hidden danger quantitative index extraction module is used to implement multi-axis domain feature distillation on the risk coupling coefficient heat map, extract hidden danger quantitative indicators including gravitational field intensity mutation gradient, percolation phase spectrum entropy, and approximate energy convergence, and dynamically compare them with the safety envelope function in the knowledge base; The solution generation module is used to generate a treatment plan including a micro-repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector diagram when any quantitative indicator of hidden danger exceeds the preset envelope boundary.

[0085] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean direct connection of the lines, and the indirect connection mode can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above is only an exemplary embodiment of the present invention and cannot be used to limit the scope of the present invention.

[0086] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. After considering the disclosure of the specification and practice, those skilled in the art will easily think of other embodiments of the present invention. This application is intended to cover any variation, use or adaptation of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the art that are not described in the present invention.

Claims

1. A dam safety monitoring method based on digital twins, characterized in that: The method comprises: Obtain the distribution spectrum of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and generate a multi-dimensional damage feature matrix containing the strain energy density field through cross-modal convolution kernel fusion calculation; Based on the spatial coordinates of the multi-dimensional damage feature matrix, a disaster node topological network is established; Input the node weights in the topological network into the gravitational field strength calculation model, simulate the energy conduction path, and output a set of chain reaction probability values ​​between adjacent nodes in the path; According to the chain reaction probability value set, a time-space folding mapping operation is performed on the multi-dimensional damage characteristic matrix, a bidirectional diffraction equation between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation is established, and a carbonization-deformation coupling prediction curve is generated; Based on the carbonization-deformation coupling prediction curve, the damage energy exchange rate at each monitoring point is calculated, and the energy exchange rate is multiplied by the chain reaction probability value at the corresponding position by nonlinear Gram matrix to generate a standardized risk coupling coefficient heat map; Perform multi-axis domain feature distillation on the risk coupling coefficient heat map to extract hidden danger quantitative indicators including gravitational field intensity mutation gradient, percolation phase spectrum entropy, and convergence degree of approaching energy, and dynamically compare them with the safety envelope function in the knowledge base; When any quantitative indicator of hidden danger exceeds the preset envelope boundary, a disposal plan is generated including a micro-repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector diagram.

2. A dam safety monitoring method based on digital twins according to claim 1, characterized in that: The cross-modal convolution kernel fusion calculation includes: Performing wavelet packet decomposition on the acoustic emission signal distribution spectrum to generate a multi-band energy distribution tensor; Performing edge enhancement based on a morphological gradient operator on the internal infrared thermal image matrix, and outputting a spatial gradient matrix of the thermal conduction abnormality area; The multi-band energy distribution tensor and the spatial gradient matrix are input into a dual-channel residual network for cross-modal feature fusion, and the fusion weights are optimized through the strain energy density field loss function to generate the multi-dimensional damage feature matrix.

3. A dam safety monitoring method based on digital twins according to claim 2, characterized in that: The establishment of the disaster node topology network includes: Based on the strain energy density field distribution of the multi-dimensional damage characteristic matrix, a density peak clustering algorithm is used to screen the potential disaster node set; Perform Delaunay hierarchical decomposition on the node set to construct a three-dimensional space grid model and divide it into multiple grid units; Calculate the energy gradient change rate between adjacent grid cells and generate a dynamic weight matrix; The dynamic weight matrix is ​​used to output a disaster node topological network with weights.

4. A dam safety monitoring method based on digital twins according to claim 3, characterized in that: The simulated energy conduction path includes: Inputting the dynamic weight matrix of the topological network into the gravitational field model to calculate the potential field matrix between nodes; The potential energy field matrix and the node set are used to simulate the energy conduction path using an improved Markov chain Monte Carlo algorithm, and a chain reaction probability value set between adjacent nodes is output and stored in a dynamic risk probability database.

5. A dam safety monitoring method based on digital twins according to claim 4, characterized in that: The space-time folding mapping operation includes: Divide the multi-dimensional damage feature matrix into frames according to time series, and construct a four-dimensional space-time hypercube by combining the space-time distribution of the chain reaction probability value set; Performing discrete cosine transformation on the carbonization depth axis of the four-dimensional space-time hypercube to generate a frequency domain projection subspace; Performing second-order difference calculation on the macroscopic displacement cumulant axis of the four-dimensional space-time hypercube to extract frequency-domain-time-domain features; The coupling coefficient tensor of the bidirectional diffraction equation is solved to generate the relationship curve between the carbonization depth change rate and the macroscopic displacement, and the carbonization-deformation coupling prediction curve is obtained.

6. A dam safety monitoring method based on digital twins according to claim 5, characterized in that: The calculation of the damage energy exchange rate includes: Extracting the instantaneous slope sequence of each monitoring point along the carbonization-deformation coupling prediction curve; Aligning the instantaneous slope sequence with the chain reaction probability value set according to topological network node coding; The aligned encoded data is fused with the ambient temperature and humidity time series data based on the Lagrange interpolation algorithm to generate a dynamic exchange rate correction factor; The noise interference is eliminated by principal component analysis and the damage energy exchange rate is output.

7. A dam safety monitoring method based on digital twins according to claim 6, characterized in that: The generating of the standardized risk coupling coefficient heat map comprises: Mapping the damage energy exchange rate and the chain reaction probability value to the grid vertices of the topological network; Construct a three-dimensional covariance Gram matrix and enhance the nonlinear coupling effect through the hyperbolic tangent activation function; Use the t-SNE algorithm in manifold learning to reduce the dimension to a two-dimensional plane; The risk coefficient distribution is dynamically superimposed according to the time labels of the space-time hypercube, and a standardized heat map is output.

8. A dam safety monitoring method based on digital twins according to claim 7, characterized in that: The multi-axis domain feature distillation comprises: Calculating a directional differential operator along the thermal gradient axis of the standardized risk coupling coefficient thermal map to extract geometric characteristic vectors of the region where the gravitational field intensity changes suddenly; Perform fast Fourier transform on the thermal frames of continuous time periods and calculate the power spectrum entropy as the percolation phase spectrum entropy; Fitting the Lyapunov exponent in combination with the energy convergence trajectory of the space-time hypercube to generate an approximate energy convergence degree; The output includes a set of hidden danger quantitative indicators including mutation gradient, spectral entropy and convergence degree.

9. A dam safety monitoring method based on digital twins according to claim 8, characterized in that: The dynamic comparison includes: Inputting the hidden danger quantitative indicator set into the fuzzy Petri net reasoning engine in the knowledge base; Match the critical interval of elastic modulus of gravitational field strength and the safe region of seepage phase in the safety envelope function; When any indicator breaks through the envelope boundary, the Monte Carlo tree search protocol of the three-dimensional decision engine is triggered; Based on the optimal convergence solution of the energy conduction path and the coupling prediction curve, a reinforcement strength vector diagram and a flood discharge scheduling sequence are output.

10. A dam safety monitoring system based on digital twins, applied to the dam safety monitoring method based on digital twins according to any one of claims 1 to 9, characterized in that: The system comprises: The feature acquisition module is used to obtain the distribution map of acoustic emission signals on the dam surface and the internal infrared thermal image matrix, and generate a multi-dimensional damage feature matrix including the strain energy density field through cross-modal convolution kernel fusion calculation; A disaster network construction module is used to establish a disaster node topological network based on the spatial coordinates of the multi-dimensional damage feature matrix; A reaction probability calculation module, used to input the node weights in the topological network into the gravitational field strength calculation model, simulate the energy conduction path, and output a set of chain reaction probability values ​​between adjacent nodes in the path; A carbonization-deformation coupling prediction curve generation module is used to perform a spatiotemporal folding mapping operation on the multi-dimensional damage feature matrix according to the chain reaction probability value set, establish a bidirectional diffraction equation between the microscopic carbonization depth daily change rate sequence and the macroscopic displacement accumulation, and generate a carbonization-deformation coupling prediction curve; A risk coupling coefficient heat map generation module is used to calculate the damage energy exchange rate at each monitoring point based on the carbonization-deformation coupling prediction curve, perform nonlinear Gram matrix multiplication on the energy exchange rate and the chain reaction probability value at the corresponding position, and generate a standardized risk coupling coefficient heat map; A hidden danger quantitative index extraction module is used to implement multi-axis domain feature distillation on the risk coupling coefficient heat map, extract hidden danger quantitative indicators including gravitational field intensity mutation gradient, percolation phase spectrum entropy, and approximate energy convergence, and dynamically compare them with the safety envelope function in the knowledge base; The solution generation module is used to generate a treatment plan including a micro-repair coordinate set, a flood discharge path rescheduling sequence, and a reinforcement strength vector diagram when any quantitative indicator of hidden danger exceeds the preset envelope boundary.

Images