Slope multi-mode monitoring and early warning system and method based on digital twinning

Abstract

The invention discloses a slope multi-modal monitoring and early warning system and method based on digital twinning, and relates to the technical field of slope monitoring, and the system comprises a data acquisition module, a data processing module, a digital twinning modeling module, a deformation prediction module, an early warning visualization module and a solar power supply module. According to the invention, slope multi-source monitoring data is obtained through the data acquisition module, and after the slope multi-source monitoring data is preprocessed by the data processing module, a slope digital twinborn model is constructed and calibrated by the digital twinborn modeling module; deformation prediction is realized by combining a space-time prediction model which contains a space correlation and time sequence analysis module and is optimized by multi-scale neighborhood aggregation, a DTW loss function and transfer learning, and a future-state digital twin is generated through an early warning visualization module and risks are visualized. And the solar power supply module is matched to ensure the operation of a scene without an external power supply and the flexible deployment of the assembly connection module, so that a multi-mode, high-precision and sustainable slope monitoring and early warning scheme is formed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of slope monitoring, in particular to a slope multi-modal monitoring and early warning system and method based on digital twinning. BACKGROUND

[0002] With the rapid advancement of infrastructure construction in China, the demand for slope safety monitoring in the fields of expressways, mines and water conservancy projects is increasingly prominent. As a key structure in engineering construction, the stability of the slope is directly related to the safety of the project and the surrounding environment. In actual working conditions, the slope is easily deformed or even unstable due to the influence of multiple factors such as geological structure, rainfall and human engineering activities. The traditional monitoring system is difficult to fully cope with complex working conditions due to technical limitations. In the past, slope monitoring mainly relied on single type sensors or manual inspection. The former, such as GNSS displacement sensors, can capture local displacement data, but it is difficult to integrate multi-source heterogeneous information such as rainfall and geological structure, resulting in a significant data isolation phenomenon. The latter has problems such as low efficiency, limited coverage and inability to provide real-time early warning. With the development of sensing technology and three-dimensional modeling technology, some systems attempt to introduce multi-sensor collaborative monitoring and three-dimensional model display, but still face challenges such as insufficient spatio-temporal registration accuracy of multi-source data and lagging deformation trend prediction. For example, the traditional threshold alarm mechanism can only trigger after deformation occurs, making it difficult to capture potential risks in advance. Two-dimensional visualization methods also cannot intuitively present the overall situation evolution of the three-dimensional space of the slope.

[0003] In recent years, the application of digital twinning technology in the engineering field has provided a new approach to slope monitoring, but existing solutions still have technical shortcomings in dynamic matching of models and physical entities, deep fusion of multi-modal data and future state extrapolation. Especially in the face of slope monitoring under complex geological conditions, how to achieve spatio-temporal unified expression of multi-source data, build accurate prediction models to provide early warnings, and support decision-making through efficient visualization technology has become a technical problem that needs to be solved.

[0004] Therefore, the present application proposes a slope multi-modal monitoring and early warning system and method based on digital twinning, which solves the problems of data fusion difficulty, early warning lag and lack of three-dimensional situation awareness in traditional systems by integrating multi-source heterogeneous data, building spatio-temporal hybrid prediction models and dynamic digital twinning reconstruction technology, providing an intelligent solution for slope safety management SUMMARY

[0005] To solve the above technical problems, a slope multi-modal monitoring and early warning system and method based on digital twinning are provided, which solves the above problems.

[0006] To achieve the above purposes, the technical scheme adopted by the present application is as follows:

[0007] A slope multi-modal monitoring and early warning system based on digital twinning, comprising:

[0008] a data acquisition module, a data processing module, a digital twinning modeling module, a deformation prediction module, an early warning visualization module, and a solar power supply module;

[0009] The data acquisition module is used to collect multi-source monitoring data of the slope, including displacement data obtained through displacement monitoring devices, environmental data obtained through environmental sensing devices, and slope surface image data obtained through aerial photography equipment, and to transmit the preliminary integrated multi-source data to the data processing module;

[0010] The data processing module is electrically connected with the data acquisition module and is used to preprocess the multi-source monitoring data, including data cleaning, standardization, and spatio-temporal alignment, to generate a structured monitoring data set, and to perform three-dimensional processing on the image data to generate a digital ground model;

[0011] The digital twinning modeling module is electrically connected with the data processing module and is used to construct a three-dimensional geometric model of the slope based on the digital ground model, to dynamically calibrate the three-dimensional geometric model through a spatial matching algorithm based on the measured data of the displacement monitoring devices, and to form an initial slope digital twinning model;

[0012] The deformation prediction module is electrically connected with the data processing module and the digital twinning modeling module, is used to convert the structured monitoring data set into a spatio-temporal data sequence, input the pre-trained spatio-temporal prediction model, and output deformation prediction data of key areas of the slope in the future period, and the spatio-temporal prediction model includes a spatial correlation module and a time series analysis module;

[0013] The early warning visualization module is electrically connected with the deformation prediction module and the digital twinning modeling module, is used to generate a predicted displacement field on the three-dimensional geometric model based on the deformation prediction data, to perform deformation reconstruction on the model to form a future state digital twinning body, to dynamically display the slope topography evolution and risk area distribution through three-dimensional visualization technology, and to transmit the results to a monitoring terminal;

[0014] The solar power supply module is electrically connected with the data acquisition module, the data processing module, and the early warning visualization module, is used to convert light energy into electrical energy, provides continuous and stable green energy for the entire system, and ensures normal operation of the system in the absence of external power supply.

[0015] Preferably, the data acquisition module specifically includes:

[0016] a displacement monitoring unit: used to deploy displacement monitoring devices, to collect displacement data of key monitoring points of the slope in real time, including horizontal displacement and vertical displacement data;

[0017] An environmental sensing unit is configured to deploy environmental sensing devices to collect environmental data of the slope region, including rainfall, soil moisture, temperature, and wind speed data.

[0018] An image acquisition unit is configured to acquire image data of the slope surface by using aerial photography equipment, covering the overall topography and local cracks and collapse information of the slope.

[0019] A data preliminary integration unit is configured to associate the displacement data, environmental data, and image data according to the collection time stamp, remove abnormal data, form preliminary integrated multi-source monitoring data, and transmit the data to a data processing module.

[0020] Preferably, the data processing module specifically includes:

[0021] A data preprocessing unit is configured to clean, standardize, and spatiotemporally align the multi-source monitoring data to generate a structured monitoring data set.

[0022] An image three-dimensional processing unit is configured to perform three-dimensional reconstruction processing on the image data of the slope surface, generate a digital terrain model through feature point matching and dense reconstruction algorithms, and synchronously extract geometric parameters such as slope surface texture, slope, and slope direction, and transmit the data to a digital twin modeling module.

[0023] Preferably, the spatiotemporal prediction model specifically includes:

[0024] A spatial correlation module is configured to use a graph neural network architecture, take the slope monitoring points as graph nodes, represent the spatial correlation between the monitoring points through a topological connection weight matrix, extract the spatial propagation features of the slope deformation, including the correlation between local deformation and overall displacement.

[0025] A time series analysis module is configured to use a structure combining a long short-term memory network and an attention mechanism to capture the time series features of the historical deformation trend in the spatiotemporal data sequence, filter the evolution law of the key time steps through a gating unit, and highlight the deformation features of important time periods.

[0026] A model optimization unit is configured to introduce a multi-scale neighborhood aggregation mechanism, a dynamic time warping loss function, and a transfer learning strategy to optimize the model. The multi-scale neighborhood aggregation mechanism aggregates the neighborhood features of the monitoring points through different size sliding windows, the DTW loss function optimizes the long-term deformation prediction accuracy, and the transfer learning strategy shortens the model convergence time.

[0027] Preferably, the model optimization unit specifically includes:

[0028] A multi-scale neighborhood aggregation subunit is configured to aggregate the neighborhood features of the monitoring points through 3x3 and 5x5 sliding windows, retain the multi-scale features of the local subtle deformation and overall macro displacement of the slope, and improve the completeness of the feature extraction of the mixed geological slope.

[0029] The loss function optimization subunit combines the dynamic time warping loss function with the traditional mean square error to calculate the time offset similarity of the predicted sequence and the historical true sequence, optimize the long-term deformation rate prediction accuracy, and solve the time series phase offset problem.

[0030] The transfer learning subunit pre-trains the model parameters with the historical monitoring data of similar slopes, and then fine-tunes the model weights with the real-time monitoring data of the target slope, thereby shortening the model convergence time and improving the initial prediction accuracy in the new scene.

[0031] Preferably, the early warning visualization module specifically comprises:

[0032] The digital twin reconstruction unit is used to generate a predicted displacement field on a three-dimensional geometric model based on deformation prediction data through a spatial interpolation algorithm, and to perform deformation reconstruction on the model grid according to the displacement field, thereby forming a digital twin that reflects the future 1-7 day slope state.

[0033] The risk assessment unit is used to construct a risk assessment table mapping deformation parameters and risk levels, match the deformation rate, displacement amount and duration of key areas in the predicted displacement field with the threshold values of the table, assess the risk level, including low risk, medium risk and high risk, and label the early warning response suggestions.

[0034] The three-dimensional visualization unit is used to dynamically display the slope morphology evolution process through three-dimensional rendering technology, mark the risk area distribution using color coding, superimpose the monitoring data curves synchronously, and transmit the results to the monitoring terminal to support mouse click query of specific monitoring point data.

[0035] Preferably, the solar power supply module specifically comprises:

[0036] The solar energy conversion unit is used to convert light energy into electrical energy through a solar panel made of monocrystalline silicon material, which has wind sand resistance and low temperature resistance;

[0037] The electrical energy transmission unit is used to transmit the converted electrical energy to each module of the system through a connecting rod, which has an internal insulation wire and corrosion prevention function.

[0038] The energy storage backup unit is used to store excess electrical energy in a lithium battery pack to power the system when there is insufficient light.

[0039] Preferably, it further comprises an assembly connection module, which comprises:

[0040] The assembly connection bracket is welded from upper and lower horizontal round corner steel plates and a middle vertical steel plate, and is provided with standardized connection hole positions for fixing the data acquisition module and the data processing module to realize line regulation.

[0041] Rotatable support rod: bottom with bolt hole, top flange plate matches with bracket lower plate circular slot, side surface is provided with annular steel hoop installation structure, support bolt fixation and trunk binding two deployment modes, can adjust equipment angle and height according to side slope topography.

[0042] A slope multi-modal monitoring and early warning method based on digital twinning includes:

[0043] S1: collecting slope multi-source monitoring data;

[0044] Deploy GNSS displacement sensors, rain stations and optional environmental sensors at key positions of the target slope, wherein the GNSS displacement sensors are used to collect real-time three-dimensional displacement time series data of key points of the slope. Let the coordinates of the key points of the slope in the spatial rectangular coordinate system be (X, Y, Z), and the coordinates collected at time t be (X t ,Y t ,Z t ), the displacement in three-dimensional space can be calculated by the following formula:

[0045]

[0046] Where, △X t , △Y t , △Z t are the displacement amounts in the X, Y, and Z axis directions at time t, X t-1 , Y t-1 , Z t-1 are the coordinate values at time t-1. The rain station collects real-time rainfall and other environmental parameters, denoted as rainfall Pt at time t. Meanwhile, according to actual needs, optional environmental sensors such as temperature and humidity, wind speed and direction can be deployed to collect corresponding environmental data;

[0047] Use a multi-angle tilt photography device-equipped unmanned aerial vehicle to periodically take pictures of the slope surface according to a preset flight path, obtain high-resolution multi-view images, and ensure comprehensive and complete coverage of the slope surface by following a specific flight path planning algorithm. The flight height of the unmanned aerial vehicle during the shooting process is H, and the shooting angle parameters are θ (tilt angle) and φ (azimuth angle). By reasonably setting these parameters, slope images at different angles can be obtained.

[0048] Convert the obtained multi-view image data into slope surface three-dimensional point cloud, digital orthographic image map (DOM), and digital surface model (DSM) through three-dimensional reconstruction technology. During three-dimensional reconstruction, based on the multi-view stereo matching (MVS) principle, a feature point matching algorithm is used to search for homonymous points. Let two images I1 and I2 be obtained from different angles. For the feature point p1 (x2, y2) in image I1, calculate the three-dimensional point coordinates (X, Y, Z) according to the triangulation principle:

[0049]

[0050] Where p1 and p2 are the projection matrices corresponding to images I1 and I2, respectively, and f1 and f2 are the focal lengths of the two images. By calculating a large number of corresponding points, a three-dimensional point cloud model of the slope surface is constructed.

[0051] Based on the constructed 3D point cloud model, a digital orthophoto map (DOM) and a digital surface model (DSM) are further generated. When generating the DOM, an orthorectification algorithm is used to correct the oblique image into an orthophoto image. The principle is based on the following formula:

[0052]

[0053] Where K is the camera intrinsic parameter matrix, R is the rotation matrix, T is the translation vector, (x ortho (X,Y,Z) are the coordinates in the orthophoto, ((X,Y,Z) are the coordinates in the 3D point cloud model, and the generation of is based on the elevation information of each point in the 3D point cloud model to directly construct the terrain surface model.

[0054] Finally, the three-dimensional displacement time series data of key slope points collected by GNSS displacement sensors, environmental parameters collected by rain gauges and environmental sensors, and the three-dimensional point cloud, DOM and DSM data of the slope surface obtained by three-dimensional reconstruction are combined to form a multimodal monitoring dataset, which is used for subsequent slope condition analysis and monitoring and early warning.

[0055] S2: Preprocess multi-source monitoring data and construct an initial digital twin model;

[0056] The CloudCompare software was used to process 3D point cloud data. Outliers were removed using statistical filtering. The average k-neighbor distance of a point pi in the point cloud was calculated. and standard deviation σ i The calculation is as follows:

[0057]

[0058] like (where n is the threshold coefficient), then the outlier is removed. The point cloud is simplified using the voxel mesh method, dividing the point cloud into a grid with side length v. The coordinates of the points within the grid are represented by prep. Calculate (V is the set of grid points, m is the number of points) by calculating the covariance matrix. And eigenvalue decomposition, taking the eigenvector corresponding to the smallest eigenvalue as the point p i Normal vectors are used to extract feature points by setting thresholds and constructing a 3D point cloud model of the digital substrate.

[0059] SIFT algorithm of OpenCV is used to detect feature points from the point cloud model, and the difference of Gaussian image D o,s (x,y)=L o,s+1 (x,y)-L o,s (x,y) to determine the position of the feature points and extract 128-dimensional descriptors;

[0060] Based on the GNSS measured coordinates, the point cloud is registered by using the ICP algorithm, assuming that the source point cloud P and the target point cloud Q, the objective function is minimized by iteration After SVD solution, the rotation matrix R and the translation vector T are obtained;

[0061] According to R and T, the rigid transformation p i '=Rp i +T is carried out on the three-dimensional point cloud model; if there is local deformation, TPS transformation is used, and the control point grid correction is realized to achieve accurate matching between the digital base plate and the real slope;

[0062] S3: predict the development trend of the slope deformation;

[0063] The multi-modal sensing data stream is preprocessed by cleaning, interpolation and standardization. In the data cleaning process, abnormal data points are removed by setting a reasonable threshold range. For the three-dimensional displacement data collected by the GNSS displacement sensor, if the displacement amount △X t , △Y t , △Z t at a certain time satisfies:

[0064]

[0065] Among them, are the threshold values of the displacement amount in X, Y and Z axis directions, and the data points exceeding the range are regarded as abnormal points and are removed.

[0066] For missing data, linear interpolation method is used for filling. Let the data at the i and j time points on the time series be x i and y j (i<j), and there are n missing data points in between, and the difference result x k of the k missing point (i<k<j) is:

[0067]

[0068] Data standardization adopts normalization method to map the data to the interval [0,1]. For a certain dimension of data x, the normalized value x' is calculated as:

[0069]

[0070] Where xmin and xmax are the minimum and maximum values ​​of the data in this dimension, respectively. After preprocessing, a multidimensional spatiotemporal input sequence X = [x1, x2, ... x...] is formed. T ], where T is the time step;

[0071] Preprocessed data is input into a pre-trained GCN-LSTM hybrid model. In the GCN layer, a graph structure G = (V, E) is constructed based on the spatial topological relationships of the monitoring points, where V is the set of monitoring points and E is the set of edges connecting the monitoring points. Let A be the adjacency matrix of the graph, and D be a diagonal matrix whose diagonal elements... Normalize the input feature matrix X to obtain The calculation process of the GCN layer can be represented as follows:

[0072] Among them, H (l) Let H be the feature matrix of the l-th hidden layer. (0) =X;W (l) Let σ be the weight matrix of the l-th layer; σ is the activation function. The normalized adjacency matrix is ​​used to extract the spatial dependency between displacement and environmental data through multi-layer GCN computation. LSTM layers are used to capture the dynamic evolution patterns in the time dimension. LSTM units contain forget gates f. t Input gate i t Output gate o t and cell state C t The forget gate determines how much information to retain from the previous cell state, the input gate determines how much of the current input information will be added to the cell state, and the output gate controls how much of the cell state will be output as the current hidden state. The cell state transmits information over time, is adjusted by the forget and input gates, updated with the current input information, and then passed through the output gate to obtain the current hidden state. At time t, the calculations for each part are as follows:

[0073] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0074] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0075]

[0076] o t =σ(W o ·[h t-1 ,xt ]+b o )

[0077] h t =o t ⊙tanh(C t )

[0078] where W f , W i , W C , W o are weight matrices, b f , b i , b C , b o are bias vectors, σ is a sigmoid activation function, tanhh is a hyperbolic tangent activation function, ⊙ represents element multiplication, h t-1 is the hidden state at the previous time.

[0079] The model output and the predicted GCN-LSTM hybrid model finally outputs the displacement change prediction sequence of the key monitoring points of the slope in the future preset time period after multi-layer calculation, where N is the time step of prediction. The prediction sequence reflects the displacement change of each monitoring point in the future time period, and the potential deformation trend of the slope is judged by analyzing the change trend of the prediction sequence;

[0080] S4: generate a future state digital twin and assess risks;

[0081] Based on the prediction results of step S3, when generating the global predicted displacement field using the Kriging interpolation algorithm, it is assumed that there are n known displacement prediction values {y1, y2, …, y n} of the monitoring points, and the corresponding spatial position coordinates are {x1, x2, …, x n}. For the displacement value y0 of the to-be-predicted position x0, the Kriging interpolation formula can be expressed as:

[0082] where λ i is the weight coefficient, which is obtained by solving the following linear equation system: A·X=B

[0083] In the formula:

[0084]

[0085] C(x1,x1) is a covariance function for describing the spatial correlation between monitoring points, σ 2 is the white noise variance, and μ is the Lagrange multiplier.

[0086] If the deformation field simulation algorithm is used, the slope area is divided into m units based on the finite element method, and the displacement field U satisfies the following balance equation:

[0087] where σ(U) is the stress tensor, a function of displacement field U, and f is the external force vector. By solving this equation, combined with boundary conditions, the predicted displacement field of the whole domain is obtained.

[0088] When fusing the predicted displacement field into the 3D point cloud model through geometric transformation, an affine transformation matrix T is adopted:

[0089]

[0090] For a point p = [x, y, z, 1] in the 3D point cloud model, T the transformed point p' is: p' = T·p

[0091] where a ij represents rotation, scaling, and other transformation parameters, t x , t y , t z represents the translation parameter, and the corresponding transformation parameter value is calculated by the predicted displacement field to realize the geometric update of the point cloud model.

[0092] In terms of attribute update, for the elevation attribute in the digital surface model (DSM), let the original elevation value be h, and according to the displacement amount Δh in the vertical direction of the predicted displacement field, the updated elevation value h' is: h' = h + Δh

[0093] Through the above geometric transformation and attribute update, a digital twin is reconstructed to reflect the topography and state of the slope at a specific future time point;

[0094] S5: Realize the visualization and transmission of early warning information;

[0095] When constructing a 3D visualization scene, shader programming of the OpenGL graphics library is adopted. In the vertex shader, the 3D point coordinates v = [x, y, z, 1] of the digital twin are subjected to model-view-projection transformation: T

[0096] v' = P·V·M·v

[0097] where M is the model matrix, used to process the local transformation of the model; V is the view matrix, defining the observation angle; P is the projection matrix, projecting the 3D scene to a 2D plane.

[0098] The fragment shader calculates the color value according to the material attribute and the lighting model. The final color I of the pixel point is calculated as follows using the Phong lighting model:

[0099]

[0100] where I a is the ambient light intensity, k​a is the ambient light reflection coefficient; I d,i , I s,i is the diffuse reflection, specular reflection intensity of the i th point light source, k d,i , k s,i is the reflection coefficient; n is the surface normal vector, l i is the vector pointing to the i th light source, v is the vector pointing to the viewpoint, r i is the reflection vector, s is the highlight index, n l is the number of point light sources.

[0101] Through real-time rendering, the digital twin is visualized and displayed on the terminal interface;

[0102] Set multi-level warning thresholds, and the displacement prediction value of the key monitoring point of the slope The warning rules are as follows:

[0103]

[0104] Where, β low , β mid , β high are low, medium and high warning thresholds.

[0105] When the warning is triggered, the warning information is pushed through SMS, email and other ways, and the warning area is highlighted in the visual interface in the form of flashing and color changing.

[0106] Compared with the prior art, the beneficial effects of the present application are:

[0107] 1. The present application provides a slope monitoring and warning method based on digital twin, which constructs a monitoring data set through multi-source data acquisition and pixel-level fusion, establishes a slope deformation database containing multi-working-condition characteristics, solves the problem of insufficient model generalization ability caused by single data in traditional methods, and realizes accurate characterization of the slope state under complex geological conditions.

[0108] 2. The present application introduces the coupling architecture of graph neural network and long short-term memory network in the spatio-temporal prediction model, breaks through the limitation of traditional models in capturing the spatio-temporal coupling characteristics of slope deformation, and significantly improves the recognition ability of multi-scale deformation characteristics of landslide precursors, providing more reliable prediction basis for disaster warning.

[0109] 3. The spatio-temporal prediction model optimized by the present application realizes the collaborative analysis of local deformation and overall displacement of the slope through multi-scale neighborhood aggregation, dynamic time warping loss function and transfer learning strategy, effectively solves the problem of insufficient prediction accuracy under complex geological conditions, shortens the deployment cycle of the model in new scenes, and improves the response efficiency of the warning.

[0110] 4. The application constructs a slope risk assessment table based on deformation parameters and historical cases, realizes the automatic conversion from deformation prediction to risk level division through a dynamic updating mechanism and three-dimensional visualization technology, breaks through the subjectivity and lag of traditional manual assessment, and can realize real-time identification and early warning response of the slope risk area.

[0111] 5. The slope monitoring and early warning device provided by the application adopts solar power supply, multi-source data acquisition and edge computing integrated design, overcomes the limitations of complex wiring and high maintenance cost of traditional monitoring systems, realizes long-term unattended monitoring in the wild, and improves the on-site data processing efficiency through lightweight algorithm deployment, and is suitable for flexible deployment under different terrain conditions. BRIEF DESCRIPTION OF DRAWINGS

[0112] Figure 1 is a system framework diagram of the application;

[0113] Figure 2 is a method step flowchart of the application;

[0114] Figure 3 is a multi-source monitoring data acquisition and preprocessing flowchart of step S1;

[0115] Figure 4 is a digital bottom plate construction and dynamic calibration flowchart of step S2;

[0116] Figure 5 is a slope deformation trend prediction model flowchart of step S3;

[0117] Figure 6 is a future state digital twin reconstruction flowchart of step S4;

[0118] Figure 7 is an early warning information visualization flowchart. DETAILED DESCRIPTION

[0119] The following description is used to disclose the application so that those skilled in the art can implement the application. The preferred embodiments in the following description are only as examples, and those skilled in the art can think of other obvious modifications.

[0120] Referring to Figure 1 The figure shows a slope multi-modal monitoring and early warning system based on digital twinning, which comprises:

[0121] a data acquisition module, a data processing module, a digital twin modeling module, a deformation prediction module, an early warning visualization module, and a solar power supply module;

[0122] The data acquisition module is used for collecting slope multi-source monitoring data, including displacement data obtained through a displacement monitoring device, environmental data obtained through an environmental sensing device, and slope surface image data obtained through a flight equipment, and transmitting the preliminary integrated multi-source data to the data processing module;

[0123] The data processing module is electrically connected with the data acquisition module, and is used for pre-processing the multi-source monitoring data, including data cleaning, standardization and spatio-temporal alignment, generating a structured monitoring data set, and performing three-dimensional processing on the image data to generate a digital ground surface model;

[0124] The digital twin modeling module is electrically connected with the data processing module, and is used for constructing a three-dimensional geometric model of the slope based on the digital ground surface model, dynamically calibrating the three-dimensional geometric model through a spatial matching algorithm based on the measured data of the displacement monitoring device, and forming an initial slope digital twin model;

[0125] The deformation prediction module is electrically connected with the data processing module and the digital twin modeling module, and is used for converting the structured monitoring data set into a spatio-temporal data sequence, inputting the spatio-temporal prediction model pre-trained, and outputting deformation prediction data of a key area of the slope in a future period, wherein the spatio-temporal prediction model comprises a spatial correlation module and a time series analysis module;

[0126] The early warning visualization module is electrically connected with the deformation prediction module and the digital twin modeling module, and is used for generating a predicted displacement field on the three-dimensional geometric model based on the deformation prediction data, performing deformation reconstruction on the model to form a future state digital twin, dynamically displaying the slope topography evolution and the risk area distribution through three-dimensional visualization technology, and transmitting the result to a monitoring terminal;

[0127] The solar power supply module is electrically connected with the data acquisition module, the data processing module and the early warning visualization module, and is used for converting light energy into electrical energy to provide continuous and stable green energy for the whole system, and ensuring normal operation of the system in a scenario without external power supply.

[0128] The data acquisition module specifically comprises:

[0129] The displacement monitoring unit is used for deploying a displacement monitoring device, and collecting displacement data of key monitoring points of the slope in real time, including horizontal displacement data and vertical displacement data;

[0130] The environmental sensing unit is used for deploying an environmental sensing device, and collecting environmental data of the slope area, including rainfall, soil moisture, temperature and wind speed data;

[0131] The image acquisition unit is used for obtaining slope surface image data through a flight equipment, covering overall topography, local cracks and collapse information of the slope;

[0132] Data preliminary integration unit: used for associating displacement data, environmental data, image data according to collection time stamps, removing abnormal data, forming preliminary integrated multi-source monitoring data, and transmitting to a data processing module.

[0133] The data processing module specifically includes:

[0134] Data preprocessing unit: used for cleaning, standardizing and spatio-temporal aligning multi-source monitoring data, and generating a structured monitoring data set;

[0135] Image three-dimensional processing unit: used for three-dimensional reconstruction processing of slope surface image data, generating a digital ground surface model through feature point matching and dense reconstruction algorithm, synchronously extracting geometric parameters such as slope surface texture, slope and slope direction, and transmitting to a digital twin modeling module.

[0136] The spatio-temporal prediction model specifically includes:

[0137] Spatial correlation module: adopting a graph neural network architecture, taking slope monitoring points as graph nodes, representing spatial correlation between monitoring points through a topological connection weight matrix, and extracting spatial propagation features of slope deformation, including correlation between local deformation and overall displacement;

[0138] Time series analysis module: adopting a structure combining a long short-term memory network and an attention mechanism, capturing time series features of historical deformation trends in spatio-temporal data sequences, filtering evolution laws of key time steps through a gate unit, and highlighting deformation features in important time periods;

[0139] Model optimization unit: used for introducing a multi-scale neighborhood aggregation mechanism, a dynamic time warping loss function and a transfer learning strategy to optimize the model, the multi-scale neighborhood aggregation mechanism aggregates neighborhood features of monitoring points through different size sliding windows, the DTW loss function optimizes long-term deformation prediction accuracy, and the transfer learning strategy shortens model convergence time.

[0140] The model optimization unit specifically includes:

[0141] Multi-scale neighborhood aggregation subunit: aggregating features of monitoring point neighborhoods through 3x3 and 5x5 sliding windows, retaining multi-scale features of local subtle deformation and overall macro displacement of slopes, and improving feature extraction integrity of mixed geological slopes;

[0142] Loss function optimization subunit: combining a dynamic time warping loss function with a traditional mean square error, calculating time offset similarity between a predicted sequence and a historical real sequence, optimizing long-term deformation rate prediction accuracy, and solving time series phase offset problems;

[0143] The migration learning subunit: pre-train the model parameters with the historical monitoring data of the same type of slope, and then fine-tune the model weight with the real-time monitoring data of the target slope, so as to shorten the model convergence time and improve the initial prediction accuracy in the new scene.

[0144] The early warning visualization module specifically includes:

[0145] The digital twin reconstruction unit: based on the deformation prediction data, a prediction displacement field is generated on the three-dimensional geometric model through a spatial interpolation algorithm, and the model grid is deformed and reconstructed according to the displacement field, forming a digital twin that reflects the slope state in the next 1-7 days;

[0146] The risk assessment unit: for constructing a risk assessment table mapping deformation parameters and risk levels, matching the deformation rate, displacement and duration of the key area in the prediction displacement field with the threshold values in the table, assessing the risk level, including low risk, medium risk and high risk, and marking the early warning response suggestions;

[0147] The three-dimensional visualization unit: for dynamically displaying the slope morphology evolution process through three-dimensional rendering technology, using color coding to mark the risk area distribution, synchronously superimposing the monitoring data curve, transmitting the results to the monitoring terminal, and supporting mouse click to query specific monitoring point data.

[0148] The solar power supply module specifically includes:

[0149] The solar energy conversion unit: for converting light energy into electrical energy through a solar panel made of monocrystalline silicon, which has wind sand resistance and low temperature resistance;

[0150] The electrical energy transmission unit: for transmitting the converted electrical energy to each module of the system through a connecting rod with built-in insulated wires, which has corrosion prevention function;

[0151] The energy storage backup unit: equipped with a lithium battery pack to store excess electrical energy and provide power to the system when there is insufficient light.

[0152] It also includes an assembly connection module, which includes:

[0153] The assembly connection bracket: welded by upper and lower transverse round corner steel plates and a middle vertical steel plate, provided with standardized connection holes, used to fix the data acquisition module and the data processing module, and realize line regulation;

[0154] The rotatable support rod: with bolt holes at the bottom, a flange plate at the top matching the circular slot of the lower plate of the bracket, and a ring-shaped steel hoop installation structure on the side, supporting two deployment methods of bolt fixation and trunk binding, and can adjust the angle and height of the equipment according to the slope terrain.

[0155] Refer to Figure 2As shown, a slope multi-modal monitoring and early warning method based on digital twinning includes:

[0156] S1: Collecting multi-source monitoring data of the slope;

[0157] Deploy displacement monitoring devices and environmental sensing devices in the slope area, and obtain image data of the slope surface through aerial photography equipment; generate a digital ground model through three-dimensional processing of the image data, and fuse the displacement data and environmental data to form preliminary monitoring data, which is transmitted to a data processing module;

[0158] S2: Preprocessing multi-source monitoring data and constructing an initial digital twinning model;

[0159] The data processing module cleans, standardizes and time-space aligns the preliminary monitoring data to generate a structured monitoring data set; the digital twinning modeling module constructs a three-dimensional geometric model of the slope based on the digital ground model, and calibrates the model position through a spatial matching algorithm combined with displacement measurement data to form an initial digital twinning model;

[0160] S3: Predicting the deformation development trend of the slope;

[0161] Convert the structured monitoring data set into a time-space data sequence, input it into a pre-trained time-space prediction model, extract the topological features of the monitoring points through a spatial correlation module, capture the dynamic evolution law through a time series analysis module, and output the deformation prediction data of the key areas of the slope in the future period;

[0162] S4: Generating a future state digital twinning body and evaluating the risk;

[0163] Based on the deformation prediction data, generate a predicted displacement field on the three-dimensional geometric model through spatial calculation methods, perform deformation reconstruction on the model to form a future state digital twinning body, match the deformation parameters of the key areas to evaluate the risk level according to a deformation parameter and risk level reference table, and label the early warning response suggestion;

[0164] S5: Realize early warning information visualization and transmission;

[0165] The early warning visualization module dynamically displays the slope topography evolution and risk area distribution through three-dimensional visualization technology, synchronously transmits the reconstructed digital twinning model, risk level, and monitoring data curve to the monitoring terminal, and supports the operation and maintenance personnel to view and make decisions in real time;

[0166] Embodiment 1: Practical application of the slope monitoring and early warning method based on digital twinning

[0167] Step S1: Multi-source monitoring data acquisition and feature labeling

[0168] First, deploy monitoring equipment in the target slope area, including GNSS receivers (to obtain overall three-dimensional displacement of the slope), tilt sensors (to monitor local slope surface deformation), rainfall sensors (to record precipitation data), and soil moisture sensors (to collect soil humidity); second, use a UAV carrying a high-resolution camera to periodically take aerial photographs of the slope to obtain visual information such as slope texture and crack development. Import the sensor data and aerial images into the data processing system. First, denoise the sensor data (such as Kalman filter to remove GNSS signal noise), normalize the data (to unify the data dimension), and stitch and correct the aerial images (to generate a slope orthographic image and a three-dimensional point cloud model); then use a labeling tool (such as the ArcGIS labeling module or PointLabeler) to spatially correlate and label key geological features (such as fault lines, joint surfaces, and potential sliding zones) in the three-dimensional point cloud and the sensor monitoring points, and establish a multi-dimensional sample set containing "geometric features - environmental parameters - displacement data".

[0169] For example, the data collection of the present application takes into account the complexity of slope deformation. First, considering that slope instability is often caused by the combined action of "overall displacement - local deformation - environmental triggers", multi-scale displacement monitoring is achieved through GNSS and tilt sensors (GNSS reflects the macro trend, and tilt sensors capture micro deformation), avoiding one-sided judgment of slope stability by a single sensor (such as ignoring the instability caused by local crack expansion using only GNSS). Second, the visual information provided by the UAV images (such as crack length and width changes) can be integrated with the sensor data to construct a more realistic digital twin model of the slope. During the labeling stage, the spatial position of potential sliding zones (such as the loose rock-soil mass region on the hanging wall of a fault) is deliberately marked, and its historical displacement data and environmental parameters (such as the impact of soil moisture content mutation after rainfall on this area) are associated, so that the subsequent deep learning model not only learns the displacement pattern, but also identifies the coupling effect of geological structure and environmental factors (for example: fault zone + high water content + heavy rainfall → high instability risk). In addition, when denoising the data, an adaptive filtering algorithm (such as wavelet transform-based denoising) is used to address the problem of GNSS susceptibility to multipath effect interference, ensuring the authenticity of the displacement data. If raw GNSS data is used directly, noise may cause the model to misjudge the slope "pseudo-displacement", reducing the accuracy of the warning. Through the above steps, the sample set established not only contains the spatial geometric features of the slope, but also integrates dynamic monitoring data and environmental triggers, providing a "spatio-temporal - multi-source - coupled" training basis for subsequent deformation prediction models.

[0170] Step S2: Digital substrate construction and dynamic calibration

[0171] Operation process: the three-dimensional point cloud data of the slope obtained in step S1 is imported into a professional processing software, and is sequentially subjected to filtering and denoising, point cloud simplification and feature extraction to construct a three-dimensional point cloud digital baseboard reflecting the true geometric shape of the slope; the feature point detection algorithm is used to identify the feature point set corresponding to the GNSS monitoring point from the point cloud model, and the spatial transformation parameters are calculated based on the GNSS measured coordinates through the point cloud registration algorithm; the three-dimensional point cloud model is subjected to rigid transformation or local deformation correction to realize the accurate matching of the digital baseboard and the real slope in space, and a dynamically updated geometric reference is formed.

[0172] Examples: the present application focuses on "multi-source data spatial consistency" and "slope deformation adaptability" when constructing the digital baseboard. Firstly, in view of the problem that the traditional static model is easily affected by noise interference or coordinate deviation, multi-level filtering and feature extraction are used to effectively remove the interference of non-slope targets, improve the purity of point cloud data, and ensure that the digital baseboard truly reflects the geological features; secondly, the point cloud registration algorithm is used to solve the problem of sensor coordinate system, and the model coordinate system is accurately converted into the GNSS coordinate system through feature point matching, which significantly improves the spatial positioning accuracy. In view of the possible local deformation of the slope, the present method supports non-rigid correction - through marking key areas such as potential sliding zones, the area is adaptively adjusted based on registration, so that the digital baseboard can not only present the overall shape of the slope, but also adapt to local deformation. In addition, the dynamic calibration process is linked with the GNSS real-time monitoring data, and when the displacement of the monitoring point exceeds the preset threshold, the system automatically triggers local correction of the model, ensuring that the digital baseboard is updated with the real shape evolution of the slope, providing a "dynamic-accurate-adaptive" geometric reference for subsequent analysis, and solving the defect that the traditional static model cannot reflect the subtle changes of the slope in real time. Through the above steps, the digital baseboard has high-precision initial geometric features and dynamic calibration capability, laying a reliable foundation for multi-source monitoring data fusion and digital twin model construction.

[0173] Step S3: prediction of slope deformation development trend

[0174] Operation process: first, the monitoring data set constructed in step S1 is preprocessed, including denoising (such as using moving average filtering or wavelet transform to remove abnormal fluctuations), normalization (unifying the dimensions of multi-source data) and time series alignment, forming a spatio-temporal data sequence containing spatial position, displacement parameter and environmental factor. The data sequence is input into the pre-trained spatio-temporal prediction model, which extracts the topological features (such as the correlation weight of the displacement of adjacent points) between monitoring points through the spatial correlation module (such as based on graph neural network architecture), and captures the dynamic evolution law (such as the time sequence coupling relationship between rainfall period and deformation rate) through the time series analysis module (such as the combination of long short-term memory network and attention mechanism). Finally, the deformation prediction data of the key areas (such as slope shoulder, fault zone and potential sliding surface) of the slope in the future period are output by the model, including displacement, deformation rate and deformation trend direction.

[0175] Example: The present application predicts the breakthrough of the traditional model of single dimension analysis limitation, and realizes multi-scale feature capture through the coupling modeling of "spatial correlation + time evolution". First, the spatial correlation module abstracts the monitoring points as graph nodes, and quantifies the displacement conduction mechanism in the slope (such as the influence degree of the abnormal displacement of a monitoring point on the surrounding area) through the topological connection weight matrix, solving the problem that the traditional method ignores the "cooperative effect of local deformation on overall stability" (for example: the sudden displacement of the slope foot may cause chain deformation of the slope top through the stress conduction of the stratum, and this module can identify such spatial propagation characteristics). Second, the time series analysis module selects key time steps (such as abnormal periods of continuous multi-day deformation rate increase) through the gating unit, and dynamically allocates the weight of historical data by combining the attention mechanism, effectively capturing the superposition effect of long-term deformation trend and short-term sudden signal (such as the gradual slope instability precursor caused by continuous rainfall in the rainy season). Compared with the traditional method which only relies on displacement threshold or single time series analysis, the present model significantly improves the prediction ability of the "slow deformation-sudden instability" mode under complex geological conditions (for example: a certain slope suddenly landslides after a few months of small displacement, the model identifies the spatio-temporal coupling characteristics of displacement increase rate and water content anomaly in advance, and the warning time is greatly ahead of the traditional method). In addition, the model supports targeted prediction of key areas of the slope—by marking geological sensitive areas such as fault zones and weak interlayers, it outputs the exclusive deformation parameters of the area, provides accurate basis for engineering reinforcement and emergency disposal, and solves the deficiency of the traditional model that "global prediction ignores local risk". Through the above steps, the spatio-temporal prediction model realizes the upgrade from "data-driven" to "mechanism cognition", and provides more reliable technical support for slope instability risk assessment.

[0176] Step S4: Future state digital twin generation

[0177] Operation flow: Based on the slope deformation prediction data output in step S3, the predicted displacement field is generated on the three-dimensional geometric model constructed in step S2 through spatial calculation method—mapping the predicted displacement amount and deformation direction of each monitoring point to the model grid node to form a continuous distribution of displacement vector set; according to the displacement field, the three-dimensional model is deformed and reconstructed, the overall model is executed with rigid transformation to reflect the macro displacement trend, and the local high-risk area (such as fault zone, weak interlayer) is implemented with non-rigid deformation processing (such as differential displacement adjustment of grid nodes), and finally the digital twin model reflecting the future period of slope morphological evolution is generated, which intuitively presents the displacement distribution and deformation mode of the key area.

[0178] Example: When generating the digital twin, the present application focuses on "predictive data space mapping" and "geological structure adaptability", solving the problem that traditional models cannot dynamically deduce slope deformation. First, the discrete monitoring point prediction data is converted into a continuous displacement field through spatial calculation method, avoiding the isolation of single point data (such as extending the displacement of sparse monitoring points to the entire slope grid through interpolation algorithm, ensuring the spatial continuity of deformation trend); second, the deformation reconstruction process combines the geological characteristics of the slope - for homogeneous rock area, rigid transformation is used to maintain the consistency of the overall displacement, and for complex structure area such as fault and joint, non-rigid deformation is used (such as allowing the relative displacement of grid nodes on both sides of the fault surface), so that the digital twin can accurately simulate the coupled deformation mode of "overall slip - local shear" (traditional methods often assume that the slope is a homogeneous rigid body, ignoring the differential response of geological interface, this method significantly improves the authenticity of the model through non-rigid adjustment). In addition, the digital twin supports multi-dimensional visual analysis - through color coding of displacement in different areas (such as red for high deformation area and blue for stable area), superimposed display of historical state, current measurement and future prediction three-dimensional model, intuitive identification of deformation evolution path and potential danger area (for example: a slope prediction model shows that the displacement of the slope shoulder and the slope foot increases synchronously, combined with the geological structure, it can be predicted that the middle soft interlayer may become a sliding surface). The digital twin constructed in this step is not only a visualization tool, but also a core carrier of risk assessment, which provides spatial distribution basis for subsequent risk level division, solves the "data fragmentation, lack of spatial correlation" of traditional early warning system, and provides dynamic and three-dimensional digital support for forward-looking research and decision of slope disaster and engineering decision.

[0179] Step S5: Early warning information visualization

[0180] Operation flow: The digital twin model generated in step S4 is transmitted to the monitoring center terminal, and the three-dimensional visualization engine is used to dynamically render the slope topography evolution process - the space-time distribution of the predicted displacement field is displayed through grid deformation animation, and the color mapping scheme (such as cold tone representing stability and warm tone representing high risk) is used to intuitively present the deformation degree of each area. The system automatically identifies the displacement abnormal area and superimposes the risk level label (such as using different shaped icons to distinguish low, medium and high risk) according to the preset risk assessment rules, and generates comprehensive early warning information by associating environmental data (such as rainfall intensity, water content), which is provided for monitoring personnel to check and make decisions in real time through the interactive interface.

[0181] Exemplary: The early warning visualization of the application focuses on "multi-dimensional information fusion" and "risk intuitive expression", solving the problem of "data scattering and early warning ambiguity" of traditional systems. First, based on the deep learning algorithm of YOLOv8-Seg segmentation architecture, an image contour preprocessing module is introduced into the network. Through the Sobel algorithm, the edge gradient information of the target is extracted to enhance the visual saliency of geological features such as cracks and faults (to avoid feature omission caused by noise), and combined with the SE attention mechanism, the enhanced features are weighted. Assign, suppress irrelevant noise interference, and improve the segmentation accuracy of the risk area. In the visualization process, the digital twin model dynamically synchronizes the measured data and the prediction results, supports multi-temporal comparison (such as the superimposed display of historical 3D models and current prediction models), and clearly presents the evolution path of the slope deformation by adjusting the transparency. In addition, the early warning information integrates multi-factor analysis results such as displacement trend, geological structure, and environmental incentives (such as the simultaneous triggering of "high displacement acceleration + continuous heavy rainfall + abnormal water content" three early warnings in a fault zone), avoiding single index misjudgment risk. Compared with the traditional interface that only displays displacement values, this method converts complex deformation data into intuitive and easy-to-understand risk maps through three-dimensional dynamic models, color coding, icon labeling and other multi-modal visualization means, significantly improving the response efficiency and decision-making accuracy of monitoring personnel, and providing an efficient human-computer interaction interface for real-time early warning of slope disasters.

[0182] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection required by the present application is defined by the appended claims and their equivalents.

Claims

1. A slope multi-modal monitoring and early warning system based on digital twinning, characterized in that, The application relates to a slope deformation prediction and early warning system based on digital twin technology. The system comprises a data acquisition module, a data processing module, a digital twin modeling module, a deformation prediction module, an early warning visualization module and a solar power supply module. The data acquisition module is used for collecting slope multi-source monitoring data, including displacement data obtained through a displacement monitoring device, environmental data obtained through an environmental sensing device and slope surface image data obtained through aerial photography equipment, and the multi-source data is preliminarily integrated and then transmitted to the data processing module. The data processing module is electrically connected with the data acquisition module and is used for pre-processing the multi-source monitoring data, including data cleaning, standardization and time-space alignment, generating a structured monitoring data set, and performing three-dimensional processing on the image data to generate a digital ground model. The digital twin modeling module is electrically connected with the data processing module and is used for constructing a three-dimensional geometric model of the slope based on the digital ground model, dynamically calibrating the three-dimensional geometric model through a space matching algorithm based on the measured data of the displacement monitoring device, and forming an initial slope digital twin model. The deformation prediction module is electrically connected with the data processing module and the digital twin modeling module, is used for converting the structured monitoring data set into a time-space data sequence, inputting the time-space prediction model, outputting deformation prediction data of a key area of the slope in a future period, and the time-space prediction model comprises a space correlation module and a time series analysis module. The early warning visualization module is electrically connected with the deformation prediction module and the digital twin modeling module, is used for generating a predicted displacement field on the three-dimensional geometric model based on the deformation prediction data, performing deformation reconstruction on the model to form a future state digital twin body, dynamically displaying the slope topography evolution and the risk area distribution through three-dimensional visualization technology, and transmitting the results to a monitoring terminal. The solar power supply module is electrically connected with the data acquisition module, the data processing module and the early warning visualization module, is used for converting light energy into electric energy, providing continuous and stable green energy for the whole system, and ensuring normal operation of the system in a scene without external power supply.

2. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, The data acquisition module specifically comprises a displacement monitoring unit, an environmental sensing unit and an image acquisition unit. The displacement monitoring unit is used for deploying displacement monitoring devices, collecting displacement data of key monitoring points of the slope in real time, and collecting horizontal displacement and vertical displacement data. The environmental sensing unit is used for deploying environmental sensing devices, collecting environmental data of the slope area, and collecting rainfall, soil humidity, temperature and wind speed data. The image acquisition unit is used for acquiring slope surface image data through aerial photography equipment, and covers overall topography, local cracks and collapse information of the slope. The data preliminary integration unit is used for associating the displacement data, the environmental data and the image data according to collection time stamps, removing abnormal data, forming preliminarily integrated multi-source monitoring data, and transmitting the data to the data processing module.

3. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, The data processing module specifically comprises a data preprocessing unit and an image three-dimensional processing unit. The data preprocessing unit is used for cleaning, standardizing and time-space aligning the multi-source monitoring data, and generating a structured monitoring data set. The image three-dimensional processing unit is used for performing three-dimensional reconstruction processing on the slope surface image data, generating a digital ground model through feature point matching and dense reconstruction algorithms, synchronously extracting geometric parameters such as slope surface texture, slope and slope direction, and transmitting the parameters to the digital twin modeling module.

4. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, The spatio-temporal prediction model specifically comprises: a spatial correlation module: adopting a graph neural network architecture, taking the slope monitoring points as graph nodes, representing the spatial correlation between the monitoring points through a topological connection weight matrix, extracting the spatial propagation characteristics of the slope deformation, including the correlation between local deformation and overall displacement; a time series analysis module: adopting a structure combining a long short-term memory network and an attention mechanism, capturing the time series characteristics of the historical deformation trend in the spatio-temporal data sequence, filtering the evolution law of the key time steps through a gating unit, and highlighting the deformation characteristics of the important time periods; a model optimization unit: used for introducing a multi-scale neighborhood aggregation mechanism, a dynamic time warping loss function and a transfer learning strategy to optimize the model, the multi-scale neighborhood aggregation mechanism aggregates the neighborhood features of the monitoring points through different size sliding windows, the DTW loss function optimizes the long-term deformation prediction accuracy, and the transfer learning strategy shortens the model convergence time.

5. The slope multi-modal monitoring and early warning system based on digital twinning of claim 4, wherein, The model optimization unit specifically comprises: a multi-scale neighborhood aggregation subunit: aggregating the neighborhood features of the monitoring points through 3*3 and 5*5 sliding windows, retaining the multi-scale characteristics of the local subtle deformation and overall macro displacement of the slope, and improving the integrity of the feature extraction of the mixed geological slope; a loss function optimization subunit: combining the dynamic time warping loss function with the traditional mean square error, calculating the time offset similarity between the predicted sequence and the historical true sequence, optimizing the long-term deformation rate prediction accuracy, and solving the time series phase offset problem; a transfer learning subunit: pre-training the model parameters based on the historical monitoring data of the same type of slope, and then fine-tuning the model weight through real-time monitoring data of the target slope, thereby shortening the model convergence time and improving the initial prediction accuracy in the new scene.

6. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, The early warning visualization module specifically comprises: a digital twin reconstruction unit: used for generating a predicted displacement field on a three-dimensional geometric model based on the deformation prediction data through a spatial interpolation algorithm, deforming the model grid according to the displacement field, and forming a digital twin reflecting the slope state in the next 1-7 days; a risk assessment unit: used for constructing a risk assessment table mapping deformation parameters and risk levels, matching the deformation rate, displacement amount and duration of the key areas in the predicted displacement field with the threshold values of the table, assessing the risk level, including low risk, medium risk and high risk, and labeling the early warning response suggestions; a three-dimensional visualization unit: used for dynamically displaying the slope topography evolution process through three-dimensional rendering technology, labeling the risk area distribution with color coding, synchronously superimposing the monitoring data curve, transmitting the results to the monitoring terminal, and supporting mouse click query of specific monitoring point data.

7. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, The solar power supply module specifically comprises: a solar conversion unit: used for converting light energy into electrical energy through a solar panel, the solar panel is made of monocrystalline silicon material and has wind sand resistance and low temperature resistance; an electrical energy transmission unit: used for transmitting the converted electrical energy to each module of the system through a connecting rod, the connecting rod is internally provided with an insulating wire and has a corrosion prevention function; an energy storage backup unit: used for being provided with a lithium battery pack to store excess electrical energy and supply power to the system when the light is insufficient.

8. The slope multi-modal monitoring and early warning system based on digital twinning of claim 1, wherein, Further comprising an assembly connection module, the assembly connection module comprises: Assembly connection support: welded by upper and lower transverse round corner steel plates and middle vertical steel plate, provided with standardized connection hole positions, used for fixing data acquisition module, data processing module equipment, realizing line regulation; Rotatable support rod: bottom with bolt hole, top flange plate matched with support lower plate circular hollow groove, side surface provided with annular steel hoop installation structure, supporting bolt fixation and trunk binding two deployment modes, can adjust equipment angle and height according to slope terrain.

9. A slope multi-modal monitoring and early warning method based on digital twinning, applicable to the slope multi-modal monitoring and early warning system based on digital twinning in any one of claims 1-8, characterized in that, Comprise: S1: collecting slope multi-source monitoring data; Deploy displacement monitoring devices and environmental sensing devices in the slope area, obtain slope surface image data through aerial photography equipment; generate a digital ground model through three-dimensional processing of the image data, and fuse the image data with displacement data and environmental data to form preliminary monitoring data, which is transmitted to a data processing module; S2: preprocessing multi-source monitoring data and constructing an initial digital twin model; The data processing module cleans, standardizes and time-space aligns the preliminary monitoring data to generate a structured monitoring data set; the digital twin modeling module constructs a three-dimensional geometric model of the slope based on the digital ground model, calibrates the model position through a spatial matching algorithm combined with displacement measurement data, and forms an initial digital twin model; S3: predicting the development trend of slope deformation; The structured monitoring data set is converted into a time-space data sequence, which is input into a pre-trained time-space prediction model. The spatial correlation module extracts the topological features of the monitoring points, the time series analysis module captures the dynamic evolution law, and the deformation prediction data of the key areas of the slope in the future period is output; S4: generating a future state digital twin and evaluating risks; Based on the deformation prediction data, a predicted displacement field is generated on the three-dimensional geometric model through spatial calculation methods to reconstruct the model and form a future state digital twin. Combined with the deformation parameters and the risk level reference table, the key area deformation parameters are matched to evaluate the risk level and label the warning response suggestions; S5: realizing warning information visualization and transmission; The warning visualization module dynamically displays the slope topography evolution and risk area distribution through three-dimensional visualization technology, synchronously transmits the reconstructed digital twin model, risk level and monitoring data curve to the monitoring terminal, and supports real-time viewing and decision-making by operation and maintenance personnel.

10. The slope multi-modal monitoring and early warning method based on digital twinning of claim 9, wherein, The S3 specifically comprises: S301: constructing a time-space prediction model; Build a time-space prediction model containing a spatial correlation module and a time series analysis module, introduce a multi-scale neighborhood aggregation mechanism, a DTW loss function and a transfer learning strategy to optimize the model; S302: dividing data sets and training models; Divide the historical monitoring data into training set and test set according to the ratio of 8:2, use the training set to train the time-space prediction model, optimize the model parameters through back propagation algorithm, and verify the model performance by using the test set; S303: input data and predict deformation; Convert the structured monitoring data set generated by S2 into a time-space data sequence, input the trained time-space prediction model, set the prediction time length, and output the prediction data of the deformation rate and cumulative displacement of the key areas of the slope; S304: verify the prediction result; The prediction data is compared with subsequent measured data, a prediction error is calculated, and if the error exceeds a preset threshold, the model parameters are fine-tuned by a transfer learning strategy using new measured data to improve subsequent prediction accuracy.

Images