Slope reinforcing system and method for municipal building

Abstract

The invention relates to the technical field of slope reinforcement, and particularly discloses a slope reinforcement system and method for building municipal administration, slope rock mass strain time sequence data is collected through a distributed optical fiber sensing system, slope deformation monitoring data is obtained through a spaceborne interferometric synthetic aperture radar, and a three-dimensional deformation monitoring data set is constructed; an LSTM-rock constitutive model fusing a DFOS strain field and an InSAR deformation field is established, strain time sequence data and deformation data are input into the model for three-dimensional stress field inversion, and shear modulus degradation rate parameters, creep coefficient parameters and displacement increment prediction data are output; shear modulus degradation rate parameters, creep coefficient parameters and displacement increment prediction data are processed, regulation and control strategy codes of anchor rod tensioning force gradients and grouting pressure intervals are generated and uploaded to a control terminal, the precision and timeliness of slope stability analysis are remarkably improved, accurate intervention can be conducted in time according to potential risks of a slope, and the stability of the slope is improved. The probability of geological disasters such as landslide is reduced, and life and property safety of people is protected.

Description

Technical Field

[0001] This invention belongs to the field of slope reinforcement technology, and relates to a slope reinforcement system and method for building and municipal applications. Background Technology

[0002] Slope reinforcement is of paramount importance and necessity in civil engineering and geological engineering, especially in mountainous and hilly areas and urban construction, where slope stability directly affects personal safety and the integrity of infrastructure. With the acceleration of urbanization, more and more buildings and infrastructure are being constructed near slopes. Slope instability can lead to severe geological disasters such as landslides and collapses, causing huge economic losses and casualties. Therefore, timely and effective slope reinforcement can not only improve slope stability and ensure the safety of the surrounding environment, but also extend the service life of infrastructure and reduce long-term maintenance costs. Furthermore, a scientifically sound slope reinforcement plan can improve land use efficiency and provide a favorable foundation for urban development.

[0003] However, current technologies still have many technical defects and drawbacks in practical implementation, especially in the output adjustment of the anchor tensioning hydraulic station and the pressure control of the grouting machine. For example, the output adjustment of the anchor tensioning hydraulic station relies on preset parameters, but in actual construction, geological conditions are complex and varied, and a single parameter setting is difficult to adapt to the mechanical properties of different soil and rock layers, resulting in uneven distribution of anchor prestress, which may lead to local stress concentration or anchorage failure. In addition, the precision control of the hydraulic system has limitations. If the equipment is not calibrated in time or the sensor response is lagging, the tension force may exceed the design range, thereby affecting the long-term stability of the anchor. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides a slope reinforcement system and method for building and municipal applications to solve the above-mentioned technical problems.

[0005] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for slope reinforcement in building and municipal applications, the method comprising: Step S1: Collect time-series data of slope rock mass strain through a distributed optical fiber sensing system, and simultaneously acquire slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset. Step S2: Establish an LSTM-soil constitutive model that integrates the DFOS strain field and the InSAR deformation field. Input the strain time series data and deformation data into the model to perform three-dimensional stress field inversion and output the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Step S3: Process the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data to generate the control strategy code for the anchor tension gradient and grouting pressure range and upload it to the control terminal.

[0006] For example, the operation logic of step S1 is as follows: Step S11: Collect time-series data of strain on the surface of the slope rock mass through a distributed optical fiber sensing system, deploy optical fiber measuring points at 0.5-meter intervals, generate a strain data matrix with spatial location labels, perform temperature drift compensation calibration on the matrix, and output strain calibration data. Step S12: Synchronously acquire slope surface deformation monitoring data from spaceborne interferometric synthetic aperture radar, extract vertical deformation rate and horizontal displacement components, and generate a set of deformation monitoring values; Step S13: Perform time window synchronization processing on the strain calibration data and deformation monitoring value set, and use a sliding time window comparison algorithm to align the acquisition timestamps of DFOS and InSAR to eliminate time dimension deviation and generate a spatiotemporal synchronization data packet. Step S14: Map the DFOS strain data in the spatiotemporal synchronization data packet to the three-dimensional slope mesh model, synchronously overlay InSAR deformation data to the corresponding mesh vertices, perform spatial resolution error compensation, and generate spatially continuous three-dimensional fused data. Step S15: Extract strain-deformation coupling parameters for each grid point based on the 3D fusion data, calculate the transverse shear stress, longitudinal compressive strain and the angle between the deformation vector according to the elevation gradient, and integrate them into a 3D deformation monitoring dataset containing multiple mechanical indicators.

[0007] For example, step S15 includes the following steps: Step S151: Perform grid point decoupling processing on the generated three-dimensional fusion data, extract the DFOS strain parameters and InSAR deformation vector parameters corresponding to each grid point, and generate a set of grid parameters containing strain amplitude, deformation rate and spatial coordinates. Step S152: Based on the slope elevation gradient distribution data, the grid parameter set is divided into layers, the lateral strain difference rate between adjacent grid points in each elevation layer is calculated, and the lateral shear stress components are deduced by combining the reference value of rock mass elastic modulus to generate layered lateral shear stress distribution data. Step S153: Based on the DFOS strain parameters in the grid parameter set, calculate the cumulative strain change along the gravity direction in each layer according to elevation, and combine the Poisson's ratio of the rock mass obtained from geological exploration to correct the longitudinal compressive strain intensity, and output the corrected longitudinal compressive strain data of each layer. Step S154: Perform three-dimensional directional decomposition on the InSAR deformation vector parameters, calculate the cosine of the angle between the deformation vector of the grid points in the same elevation layer and the main slip direction of the slope, and combine it with the slope rock layer dip angle data for weighted processing to generate a set of deformation vector angle values. Step S155: Align and fuse the layered transverse shear stress distribution data, longitudinal compressive strain correction data, and deformation vector angle data according to the elevation layer in multiple dimensions, remove data with conflicting dimensions, and generate a standardized three-dimensional deformation monitoring dataset.

[0008] For example, step S2 includes the following steps: Step S21: Construct a three-channel gated recurrent neural network architecture, including strain feature extraction branch, deformation feature extraction branch and parameter inversion branch; Step S22: Input the DFOS strain time series data into the strain feature extraction branch and perform time dimension convolution processing to generate strain spatiotemporal feature vectors; Step S23: Input the InSAR deformation data into the deformation feature extraction branch and perform spatial dimension pooling processing to generate a deformation spatial distribution matrix; Step S24: Weighted concatenation of the strain spatiotemporal feature vector and the deformation spatial distribution matrix to form a fused feature tensor; Step S25: Perform bidirectional recursive calculation on the fused feature tensor through the parameter inversion branch, and output the original parameter set containing the shear modulus degradation rate, creep coefficient and displacement increment.

[0009] For example, steps S22 to S24 specifically include: Step S221: Perform sliding window noise reduction on the DFOS strain time series data, with a window length of 1 hour and a step size of 15 minutes; Step S222: The noise-reduced data is divided into data segments of equal duration using the overlapping segmentation method, with each data segment containing N sampling points; Step S223: Perform a one-dimensional convolution operation in the time dimension, with the convolution kernel size set to 5 time steps, to generate a primary temporal feature map; Step S224: Perform max pooling on the primary time series feature map to extract the most representative strain feature peaks; Step S225: Standardize the pooled feature sequence to generate a 32-dimensional strain spatiotemporal feature vector; Step S231: Divide the InSAR deformation data into 10m×10m grid cells according to the slope engineering coordinates; Step S232: Calculate the mean and standard deviation of deformation data within each grid cell to form a spatial distribution statistical matrix; Step S233: Use the spatial pyramid pooling method to process the statistical matrix and retain the first 80% of the high-frequency deformation features; Step S234: Perform region clustering on the filtered features and merge adjacent grids with similar deformation patterns; Step S235: Generate a 16×16 spatial distribution matrix that reflects the overall deformation trend of the slope; Step S241: Assign a weight value of 0.6 to the strain spatiotemporal eigenvector and a weight value of 0.4 to the deformation spatial distribution matrix; Step S242: Concatenate the weighted feature vectors and the distribution matrix along the channel dimension to form a 48-dimensional fused feature tensor; Step S243: Verify whether the numerical range of the fused feature tensor is within the preset [0,1] interval; Step S244: When a feature value exceeds the threshold, activate the adaptive scaling mechanism to adjust the feature dimensions; Step S245: Output the validated normalized fusion feature tensor.

[0010] For example, step S25 includes: Step S251: Perform recursive calculation of the forward time step on the fused feature tensor to generate the forward propagation state vector; Step S252: Perform recursive calculation of the fused feature tensor in reverse time step to generate the backpropagation state vector; Step S253: Weighted concatenation of the forward propagation state vector and the backward propagation state vector to form a bidirectional state tensor; Step S254: Process the bidirectional state tensor through a fully connected layer and output initial values ​​containing three parameters: initial value of shear modulus degradation rate, initial value of creep coefficient, and initial value of displacement increment. Step S255: Verify the physical feasibility of the initial parameter values ​​based on the constitutive constraints of the soil and rock mass, and automatically reduce the gain coefficient when the shear modulus degradation rate exceeds the allowable threshold. Step S256: Perform numerical optimization on the verified initial parameter values ​​to generate the final shear modulus degradation rate parameter, creep coefficient parameter, and displacement increment prediction data.

[0011] For example, step S3 includes the following steps: Step S31: The shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data are fused and cleaned to remove sensor noise points and generate a three-dimensional spatiotemporally aligned slope mechanical state data package. Step S32: Construct a residual reinforcement learning model, input the shear modulus degradation rate time-series change rate and creep coefficient transient fluctuation value from the mechanical state data packet, predict the current slope safety index, and use the displacement increment prediction value as a residual correction term to dynamically calibrate the prediction result; Step S33: Based on the percentage of real-time displacement increment to the preset warning value, calculate the formula: displacement increment / warning value × 100%, dynamically allocate the weights of shear modulus, creep coefficient, and displacement increment, and generate an adaptive weight parameter group. Step S34: Perform mixed integer programming operation on the safety index output by the residual reinforcement learning model and the adaptive weight parameter group to generate a control strategy code that includes the anchor tension gradient and grouting pressure range; Step S35: Verify the feasibility of the three-dimensional control strategy coding. If the anchor tension and grouting pressure of the same grid unit conflict, the back propagation mechanism is triggered to optimize the MILP constraint conditions until conflict-free integrated control parameters are generated.

[0012] A second aspect of the present invention provides a slope reinforcement system for building and municipal applications, the system comprising: Monitoring Data Set Acquisition Module: Collects time-series data of slope rock mass strain through a distributed fiber optic sensing system, and acquires slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset; 3D stress field inversion module: Establish an LSTM-soil constitutive model that integrates DFOS strain field and InSAR deformation field, input strain time series data and deformation data into the model to perform 3D stress field inversion, and output shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Grouting pressure regulation module: Processes shear modulus degradation rate parameters, creep coefficient parameters, and displacement increment prediction data to generate control strategy codes for anchor tension gradient and grouting pressure range, and uploads them to the control terminal.

[0013] As described above, the slope reinforcement system and method for building and municipal applications provided by the present invention have at least the following beneficial effects: This invention presents a slope monitoring and control method based on the fusion of a distributed fiber optic sensing system (DFOS) and spaceborne interferometric synthetic aperture radar (InSAR). It achieves multi-scale and all-round coverage from internal rock mass strain to surface slope deformation, making up for the shortcomings of single monitoring methods in terms of spatial resolution, monitoring range, or data continuity. It provides a more comprehensive and accurate data foundation for subsequent analysis, significantly improves the accuracy and timeliness of slope stability analysis, and enables timely and precise intervention for potential slope risks, reducing the probability of geological disasters such as landslides and protecting people's lives and property. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0016] Figure 2 This is a schematic diagram showing the connections of the various modules in the system of the present invention. Detailed Implementation

[0017] The following description, in conjunction with the implementation of this invention, is merely an example and illustration of the concept of this invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in these claims, all of which should fall within the protection scope of this invention.

[0018] Example 1 Please see Figure 1 As shown, a method for slope reinforcement in building and municipal applications includes the following steps: Step S1: Collect time-series data of slope rock mass strain through a distributed optical fiber sensing system, and simultaneously acquire slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset. The operation logic for step S1 is as follows: Step S11: Collect time-series data of strain on the surface of the slope rock mass through a distributed optical fiber sensing system, deploy optical fiber measuring points at 0.5-meter intervals, generate a strain data matrix with spatial location labels, perform temperature drift compensation calibration on the matrix, and output strain calibration data. Step S12: Synchronously acquire slope surface deformation monitoring data from spaceborne interferometric synthetic aperture radar, extract vertical deformation rate and horizontal displacement components, and generate a set of deformation monitoring values; Step S13: Perform time window synchronization processing on the strain calibration data and deformation monitoring value set, and use a sliding time window comparison algorithm to align the acquisition timestamps of DFOS and InSAR to eliminate time dimension deviation and generate a spatiotemporal synchronization data packet. Step S14: Map the DFOS strain data in the spatiotemporal synchronization data packet to the three-dimensional slope mesh model, synchronously overlay InSAR deformation data to the corresponding mesh vertices, perform spatial resolution error compensation, and generate spatially continuous three-dimensional fused data. Step S15: Extract strain-deformation coupling parameters for each grid point based on the 3D fusion data, calculate the transverse shear stress, longitudinal compressive strain and the angle between the deformation vector according to the elevation gradient, and integrate them into a 3D deformation monitoring dataset containing multiple mechanical indicators.

[0019] In this embodiment of the invention, distributed optical fiber measuring points are first deployed on the slope surface at a grid spacing of 0.5 meters. Each measuring point collects strain data in real time using optical frequency domain reflection technology, while simultaneously integrating a temperature sensor to measure ambient temperature fluctuations. A temperature-strain decoupling algorithm is employed to remove temperature interference components from the original strain data: specifically, a temperature drift compensation model is established, using the temperature difference gradient between adjacent measuring points as input, and dynamically correcting the strain value based on the thermal expansion coefficient of the optical fiber material to generate a spatially continuous strain calibration data matrix. In step S12, the original phase data from the spaceborne InSAR undergoes multi-look processing and baseline correction. A phase unwrapping algorithm is used to extract the three-dimensional deformation field of the slope surface, and atmospheric delay errors are removed through statistical filtering. Finally, a set of deformation monitoring values ​​for the vertical deformation rate and horizontal displacement components is output. Step S13, time window synchronization processing, employs dynamic time warping technology. First, millisecond-level high-frequency data from DFOS and daily-level low-frequency data from InSAR are resampled in the time domain. Then, using the InSAR data time nodes as a reference, cubic spline interpolation is used to compress the DFOS data along the time axis, generating a spatiotemporal synchronization data packet with a unified time label. In the spatial mapping stage of step S14, DFOS strain data is projected onto the grid vertices of the 3D slope BIM model according to geographic coordinates. Simultaneously, the InSAR deformation vector is decomposed into the XYZ displacement of the grid nodes. Addressing the difference between the centimeter-level spatial resolution of DFOS and the meter-level resolution of InSAR, bilinear interpolation is used to refine the coarse-resolution areas, and Kleekin spatial interpolation is used to eliminate local data abrupt changes, ultimately generating 3D fused data with consistent resolution. In the parameter extraction process of step S15, the DFOS microstrain data are first converted into stress components based on the reference value of the rock mass elastic modulus. The stress difference rate of each layer of the transverse adjacent grid is statistically analyzed according to the elevation to infer the transverse shear stress distribution. The longitudinal strain data are accumulated by summing the cumulative compression in the vertical direction within each elevation layer to characterize the longitudinal compressive strain intensity. At the same time, based on the three-dimensional spatial decomposition results of the InSAR deformation vector, the cosine value of the spatial angle between the displacement direction of each grid point and the main slip surface of the slope is calculated. The three-dimensional deformation monitoring dataset is constructed by combining the above multi-dimensional mechanical indicators.

[0020] Step S15 includes the following steps: Step S151: Perform grid point decoupling processing on the generated three-dimensional fusion data, extract the DFOS strain parameters and InSAR deformation vector parameters corresponding to each grid point, and generate a set of grid parameters containing strain amplitude, deformation rate and spatial coordinates. Step S152: Based on the slope elevation gradient distribution data, the grid parameter set is divided into layers, the lateral strain difference rate between adjacent grid points in each elevation layer is calculated, and the lateral shear stress components are deduced by combining the reference value of rock mass elastic modulus to generate layered lateral shear stress distribution data. Step S153: Based on the DFOS strain parameters in the grid parameter set, calculate the cumulative strain change along the gravity direction in each layer according to elevation, and combine the Poisson's ratio of the rock mass obtained from geological exploration to correct the longitudinal compressive strain intensity, and output the corrected longitudinal compressive strain data of each layer. Step S154: Perform three-dimensional directional decomposition on the InSAR deformation vector parameters, calculate the cosine of the angle between the deformation vector of the grid points in the same elevation layer and the main slip direction of the slope, and combine it with the slope rock layer dip angle data for weighted processing to generate a set of deformation vector angle values. Step S155: Align and fuse the layered transverse shear stress distribution data, longitudinal compressive strain correction data, and deformation vector angle data according to the elevation layer in multiple dimensions, remove data with conflicting dimensions, and generate a standardized three-dimensional deformation monitoring dataset.

[0021] In step S151, the three-dimensional fused data is first analyzed point by point using a spatial grid decoupling algorithm: for each grid vertex, the micro-strain amplitude data collected by the DFOS system and the three-dimensional spatial components (eastward, northward, and vertical displacement rates) of the InSAR deformation vector are separated, and their spatial coordinate labels are associated to generate a multi-dimensional grid parameter set. In the layered transverse shear stress calculation process of step S152, the grid parameter set is spatially cut based on the pre-imported elevation layer data. Within each elevation layer, the strain difference rate calculation method of adjacent grid points (defined as the ratio of the absolute value of the strain difference between two points to the distance between the points) is used. Combined with the reference value of the rock mass elastic modulus (dynamically loaded according to the geological exploration report), linear extrapolation is performed. The strain difference rate is multiplied by the elastic modulus to convert it into shear stress components, and finally, a transverse shear stress distribution cloud map bound to the elevation layer is generated. In step S153, during the longitudinal compressive strain correction, the cumulative vertical (Z-axis) strain of all grid points within each elevation layer is accumulated to characterize the overall compressive deformation trend of that layer. Simultaneously, a longitudinal strain-lateral expansion correlation model is established based on the Poisson's ratio parameter of the rock mass, and a scaling algorithm is used to anisotropically correct the original compression, eliminating the interference of lateral deformation on the compressive strain intensity. In step S154, during the deformation vector direction analysis, the three-dimensional components of the InSAR deformation vector are first mapped to the main slip direction of the slope through spatial geometric projection, and the cosine value of their included angle is calculated. Then, a rock layer dip angle influence factor is introduced; when the rock layer dip angle is greater than 30 degrees, a weighting coefficient of 1.2 is applied to the cosine value to enhance the characteristic response of the bedding slip direction, ultimately generating a set of vector included angle values ​​with dip angle weighting labels. In the data standardization stage of step S155, the three types of heterogeneous data, namely shear stress, compressive strain and angle, are Z-score standardized by a dynamic dimension converter, so that the values ​​of different dimensions are uniformly mapped to the interval [-1,1], and then the data are aligned and fused according to the elevation layer to generate a three-dimensional deformation monitoring dataset.

[0022] Step S2: Establish an LSTM-soil constitutive model that integrates the DFOS strain field and the InSAR deformation field. Input the strain time series data and deformation data into the model to perform three-dimensional stress field inversion and output the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Step S2 includes the following steps: Step S21: Construct a three-channel gated recurrent neural network architecture, including strain feature extraction branch, deformation feature extraction branch and parameter inversion branch; Step S22: Input the DFOS strain time series data into the strain feature extraction branch and perform time dimension convolution processing to generate strain spatiotemporal feature vectors; Step S23: Input the InSAR deformation data into the deformation feature extraction branch and perform spatial dimension pooling processing to generate a deformation spatial distribution matrix; Step S24: Weighted concatenation of the strain spatiotemporal feature vector and the deformation spatial distribution matrix to form a fused feature tensor; Step S25: Perform bidirectional recursive calculation on the fused feature tensor through the parameter inversion branch, and output the original parameter set containing the shear modulus degradation rate, creep coefficient and displacement increment.

[0023] In the neural network architecture construction process of step S21, a three-channel gated recurrent unit model was designed: the strain feature branch uses a group of one-dimensional dilated convolutional kernels to extract multi-scale temporal features from DFOS strain time series data, with each convolutional layer followed by a gated linear unit activation function to preserve the abrupt changes in strain evolution; the deformation feature branch processes InSAR deformation field data through a spatial pyramid pooling structure, compressing the spatial dimension and extracting the dynamic distribution pattern of terrain displacement at different scales; the parameter inversion branch deploys a bidirectional GRU recursive layer, whose forward propagation captures the time-dependent features of mechanical parameters, and backpropagation strengthens the constraint effect of historical states on current parameters. The three branches achieve information interaction through a cross-channel attention mechanism. In the strain feature extraction process of step S22, a sliding window convolution operation is performed on the millisecond-level DFOS strain time series data, with a window length of 1000 sample points and a stride of 200 points. Short-term micro-strain fluctuation features and long-period creep trend features are extracted through two layers of cascaded convolution, and finally the feature vectors of different time scales are concatenated into a 256-dimensional spatiotemporal feature vector. In step S23, the deformation data processing stage, a quadtree adaptive pooling algorithm is used for the spatial grid data of the InSAR deformation field: high-resolution features are preserved in high-gradient deformation regions, while feature dimensionality reduction and compression are performed in low-gradient regions to form a deformation distribution feature matrix containing different spatial precisions. In step S24, feature fusion employs a dynamic weighted stitching strategy, automatically assigning fusion weights based on the coefficients of variation of the feature vectors and matrices. Strain feature weights are 0.6-0.8, and deformation feature weights are 0.2-0.4. Axial expansion integrates the two types of heterogeneous data into a four-dimensional feature tensor, with dimensions of: time step × spatial grid × number of channels × feature depth. In step S25, during parameter inversion, a bidirectional GRU network recursively parses the fused feature tensor along the time dimension. The forward layer calculates the nonlinear decay path of the shear modulus degradation rate, and the backward layer traces the hysteresis response law of the creep coefficient. Finally, a fully connected decoder synchronously outputs the spatiotemporal distribution sets of three types of parameters: shear modulus degradation rate, viscoelastic creep coefficient, and predicted displacement increment.

[0024] Steps S22 to S24 specifically include: Step S221: Perform sliding window noise reduction on the DFOS strain time series data, with a window length of 1 hour and a step size of 15 minutes; Step S222: The noise-reduced data is divided into data segments of equal duration using the overlapping segmentation method, with each data segment containing N sampling points; Step S223: Perform a one-dimensional convolution operation in the time dimension, with the convolution kernel size set to 5 time steps, to generate a primary temporal feature map; Step S224: Perform max pooling on the primary time series feature map to extract the most representative strain feature peaks; Step S225: Standardize the pooled feature sequence to generate a 32-dimensional strain spatiotemporal feature vector; Step S231: Divide the InSAR deformation data into 10m×10m grid cells according to the slope engineering coordinates; Step S232: Calculate the mean and standard deviation of deformation data within each grid cell to form a spatial distribution statistical matrix; Step S233: Use the spatial pyramid pooling method to process the statistical matrix and retain the first 80% of the high-frequency deformation features; Step S234: Perform region clustering on the filtered features and merge adjacent grids with similar deformation patterns; Step S235: Generate a 16×16 spatial distribution matrix that reflects the overall deformation trend of the slope; Step S241: Assign a weight value of 0.6 to the strain spatiotemporal eigenvector and a weight value of 0.4 to the deformation spatial distribution matrix; Step S242: Concatenate the weighted feature vectors and the distribution matrix along the channel dimension to form a 48-dimensional fused feature tensor; Step S243: Verify whether the numerical range of the fused feature tensor is within the preset [0,1] interval; Step S244: When a feature value exceeds the threshold, activate the adaptive scaling mechanism to adjust the feature dimensions; Step S245: Output the validated normalized fusion feature tensor.

[0025] In the strain feature extraction step S22, the raw strain time-series data acquired by DFOS is first subjected to sliding window denoising, using a window length of 1 hour and a step size of 15 minutes. A moving average algorithm is used to eliminate high-frequency noise interference. The denoised data is then divided into segments of equal duration, each containing N consecutive sampling points, forming a standardized time-series input unit. During one-dimensional convolution operations in the time dimension, a convolution kernel group with a width of 5 time steps is used. The hyperbolic tangent activation function is used to extract local strain fluctuation pattern features, generating a primary feature map containing spatiotemporal correlations. Max pooling is then performed on the primary feature map to select peak strain feature points within each data segment, retaining the feature information with the largest amplitude. Next, the pooling result is normalized using the Z-score normalization method, and then dimensionality-reduced into a 32-dimensional strain spatiotemporal feature vector using a linear projection algorithm, ensuring the horizontal comparability of data from different measurement points.

[0026] In step S23, the deformation data processing involves dividing the InSAR-acquired deformation monitoring data into spatial grid cells with a precision of 10 meters according to the slope engineering coordinate system. Statistical analysis is performed on the deformation monitoring values ​​within each grid, calculating the average value to characterize the overall deformation trend and the standard deviation to reflect the local deformation dispersion, forming a spatial distribution matrix containing statistical features. A spatial pyramid pooling method is used to perform multi-level feature filtering on the statistical matrix, retaining the top 80% of high-frequency deformation area data based on feature contribution. A Gaussian filtering algorithm is then used to smooth noise interference in edge areas. Further density-based clustering analysis is performed on the selected features, merging grids with similar deformation trends and spatially adjacent areas into homogeneous regions, ultimately generating a 16×16 precision spatial distribution feature matrix characterizing the overall deformation situation of the slope.

[0027] In step S24, the feature fusion stage assigns a weight coefficient of 0.6 to the spatiotemporal feature vector and a weight coefficient of 0.4 to the spatial distribution matrix of deformation. The weight allocation is based on the variance contribution ratio of the two data sources in the historical validation set. The weighted feature vector and matrix are concatenated along the channel dimension through a channel expansion operation, forming a fusion tensor containing 48 feature dimensions. Subsequently, the fusion tensor undergoes data range verification. If feature values ​​exceed the preset [0,1] interval, an adaptive scaling mechanism is triggered: feature values ​​exceeding the upper limit are reduced in magnitude using a logarithmic compression algorithm, while feature values ​​below the lower limit are linearly interpolated to a reasonable range, ultimately outputting a standardized fusion feature tensor. This process employs a distributed streaming computing architecture for real-time processing, ensuring the integrity of the physical meaning of the fused features while keeping data processing latency within 200 milliseconds, meeting the real-time requirements of slope monitoring.

[0028] Step S25 includes: Step S251: Perform recursive calculation of the forward time step on the fused feature tensor to generate the forward propagation state vector; Step S252: Perform recursive calculation of the fused feature tensor in reverse time step to generate the backpropagation state vector; Step S253: Weighted concatenation of the forward propagation state vector and the backward propagation state vector to form a bidirectional state tensor; Step S254: Process the bidirectional state tensor through a fully connected layer and output initial values ​​containing three parameters: initial value of shear modulus degradation rate, initial value of creep coefficient, and initial value of displacement increment. Step S255: Verify the physical feasibility of the initial parameter values ​​based on the constitutive constraints of the soil and rock mass, and automatically reduce the gain coefficient when the shear modulus degradation rate exceeds the allowable threshold. Step S256: Perform numerical optimization on the verified initial parameter values ​​to generate the final shear modulus degradation rate parameter, creep coefficient parameter, and displacement increment prediction data.

[0029] During parameter inversion in step S25, the fused feature tensor is first calculated bidirectionally recursively: the forward time-step recursion uses gated loop units to analyze the feature evolution trend along the time axis, and the state vector at each time node is composed of the gated weighted result of the current feature input and the previous state; the reverse time-step recursion traces the temporal correlation of features in reverse, and amplifies the influence of key historical states by suppressing redundant features. The state vectors generated by forward and backward propagation are dynamically weighted and fused, with forward weights of 0.55-0.65 and reverse weights of 0.35-0.45, forming a bidirectional state tensor that combines historical and future correlation features. The fully connected layer performs dimensionality reduction mapping on the bidirectional state tensor, and uses a piecewise linear activation function to synchronously output the initial values ​​of shear modulus degradation rate, viscoelastic creep coefficient, and displacement increment, where the activation function threshold is dynamically set according to the physical properties of the geotechnical materials. During the physical feasibility verification phase of the initial parameters, the shear modulus degradation rate was compared in real time with long-term rock mass rheological test data. When the degradation rate exceeded the critical threshold of 0.8% per hour, a gain attenuation mechanism was triggered, and the gain coefficient was linearly reduced to 30%-50% of its original value to suppress parameter drift. Simultaneously, the dynamic correlation between the creep coefficient and displacement increment was verified to ensure that their changing trends conformed to the slope mechanical response law. The parameter optimization process employed a constrained gradient descent algorithm, using the stress-strain relationship defined by the constitutive equation as the boundary condition. Through multiple rounds of iterative fine-tuning, the errors in shear modulus degradation rate, creep coefficient, and displacement increment were all kept below the set thresholds.

[0030] Step S3: Process the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data to generate the control strategy code for the anchor tension gradient and grouting pressure range and upload it to the control terminal.

[0031] In slope reinforcement projects, dynamically adjusting the anchor tension gradient and grouting pressure range is the core strategy for achieving precise and intelligent control. Its necessity stems from the time-varying nature of rock and soil and the complexity of engineering risks. Under long-term geological activity and external loads, the mechanical properties of rock slopes continuously change, manifesting as shear modulus degradation, creep coefficient increase, and displacement increment accumulation. Using statically preset reinforcement parameters may lead to reinforcement failure due to their inability to adapt to the dynamic deterioration of the rock mass. Real-time monitoring of the shear modulus degradation rate parameter can quantitatively assess the degree of damage to the internal structure of the rock mass, reflecting the stress relaxation effect in the anchoring area; the creep coefficient parameter reveals the changing trend of the rock mass's rheological properties, directly affecting the long-term collaborative deformation capacity of the grouting material and the rock mass interface; and displacement increment prediction data can predict the development trend of potential sliding surfaces in advance through spatiotemporal evolution patterns. The integrated analysis of these three parameters can construct a comprehensive state profile of the rock mass, encompassing damage, rheology, and deformation.

[0032] The essence of generating dynamic control strategies through data processing is to transform the complex response of rock mechanics into digitally executable instructions for engineering. For example, when the shear modulus degradation rate exceeds a threshold, it indicates a decrease in the bearing capacity of the rock mass in the anchorage zone, requiring a gradual increase in the anchor tension to compensate for stress loss. If the creep coefficient continues to increase and the displacement increment approaches the warning value, the grouting pressure range needs to be dynamically expanded to fill rock mass fissures and suppress the rheological rate through high-pressure grouting. Traditional reinforcement systems are prone to anchor prestress relaxation and grout cracking, while dynamic control based on real-time mechanical parameters can effectively maintain the long-term stability of the reinforcement system and significantly improve the risk management capabilities of infrastructure throughout its entire life cycle.

[0033] Step S3 includes the following steps: Step S31: The shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data are fused and cleaned to remove sensor noise points and generate a three-dimensional spatiotemporally aligned slope mechanical state data package. Step S32: Construct a residual reinforcement learning model, input the shear modulus degradation rate time-series change rate and creep coefficient transient fluctuation value from the mechanical state data packet, predict the current slope safety index, and use the displacement increment prediction value as a residual correction term to dynamically calibrate the prediction result; Step S33: Based on the percentage of real-time displacement increment to the preset warning value, calculate the formula: displacement increment / warning value × 100%, dynamically allocate the weights of shear modulus, creep coefficient, and displacement increment, and generate an adaptive weight parameter group. Step S34: Perform mixed integer programming operation on the safety index output by the residual reinforcement learning model and the adaptive weight parameter group to generate a control strategy code that includes the anchor tension gradient and grouting pressure range; Step S35: Verify the feasibility of the three-dimensional control strategy coding. If the anchor tension and grouting pressure of the same grid unit conflict, the back propagation mechanism is triggered to optimize the MILP constraint conditions until conflict-free integrated control parameters are generated.

[0034] During the execution of step S3, three-dimensional spatiotemporal alignment is first performed on the shear modulus degradation rate parameter, creep coefficient parameter, and displacement increment prediction data: Data from heterogeneous sensors is matched and verified based on geographic coordinate labels, and data points with signal amplitude jumps exceeding three standard deviations are removed as noise interference. Missing values ​​are then filled using the Kriging interpolation algorithm, forming a three-dimensional data package containing timestamps, spatial locations, and mechanical state parameters. Subsequently, a residual reinforcement learning model is constructed. Its input layer receives the hourly rate of change of the shear modulus degradation rate and the minute-level fluctuations of the creep coefficient. Temporal features are extracted through a bidirectional long short-term memory network to predict the current slope safety index. Simultaneously, the displacement increment prediction value is dynamically fed back to the model's hidden layer as a residual term, and a gradient descent algorithm is used to update the network weights to reduce prediction bias. The percentage of displacement increment relative to a preset warning value is calculated in real time. When this percentage exceeds 60%, the parameter weight allocation strategy is automatically adjusted: the shear modulus degradation rate weight gradually decreases from the initial 60% to 40%, while the displacement increment weight increases from 20% to 40%.

[0035] After integrating the safety index and adaptive weight parameter set, a mixed-integer programming (MILP) approach is used to solve for the optimal control strategy. The upper limit of anchor bolt tension is set to 70% of the material yield strength, and the grouting pressure range is ±30% of the base value as hard constraints. The optimization objectives are to maximize the safety index and minimize control energy consumption. The anchor bolt tension gradient and grouting pressure range for each grid cell are output simultaneously. Finally, the physical feasibility of the generated control strategy is verified. If the anchor bolt tension demand within the same grid exceeds the upper limit of the grouting pressure threshold, a backpropagation mechanism is triggered to adjust the safety margin coefficient in the MILP constraints, and the calculation is iterated again until the rock mass strength-pressure balance equation is satisfied. The verified three-dimensional control strategy code is uploaded to the control terminal in real time to drive the anchor bolt hydraulic station and grouting machine to perform dynamic coordinated control.

[0036] Example 2 Please see Figure 2 As shown, a slope reinforcement system for building and municipal applications includes a monitoring dataset acquisition module, a three-dimensional stress field inversion module, and a grouting pressure adjustment module. The modules described above are connected via wired and / or wireless means to enable data transmission between them. Monitoring Data Set Acquisition Module: Collects time-series data of slope rock mass strain through a distributed fiber optic sensing system, and acquires slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset; 3D stress field inversion module: Establish an LSTM-soil constitutive model that integrates DFOS strain field and InSAR deformation field, input strain time series data and deformation data into the model to perform 3D stress field inversion, and output shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Grouting pressure regulation module: Processes shear modulus degradation rate parameters, creep coefficient parameters, and displacement increment prediction data to generate control strategy codes for anchor tension gradient and grouting pressure range, and uploads them to the control terminal.

[0037] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0038] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0039] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0040] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for slope reinforcement in building and municipal applications, characterized in that, Includes the following steps: Step S1: Collect time-series data of slope rock mass strain through a distributed optical fiber sensing system, and simultaneously acquire slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset. Step S2: Establish an LSTM-soil constitutive model that integrates the DFOS strain field and the InSAR deformation field. Input the strain time series data and deformation data into the model to perform three-dimensional stress field inversion and output the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Step S3: Process the shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data to generate the control strategy code for the anchor tension gradient and grouting pressure range and upload it to the control terminal.

2. The method for slope reinforcement in building and municipal engineering according to claim 1, characterized in that, The operation logic for step S1 is as follows: Step S11: Collect time-series data of strain on the surface of the slope rock mass through a distributed optical fiber sensing system, deploy optical fiber measuring points at 0.5-meter intervals, generate a strain data matrix with spatial location labels, perform temperature drift compensation calibration on the matrix, and output strain calibration data. Step S12: Synchronously acquire slope surface deformation monitoring data from spaceborne interferometric synthetic aperture radar, extract vertical deformation rate and horizontal displacement components, and generate a set of deformation monitoring values; Step S13: Perform time window synchronization processing on the strain calibration data and deformation monitoring value set, and use a sliding time window comparison algorithm to align the acquisition timestamps of DFOS and InSAR to eliminate time dimension deviation and generate a spatiotemporal synchronization data packet. Step S14: Map the DFOS strain data in the spatiotemporal synchronization data packet to the three-dimensional slope mesh model, synchronously overlay InSAR deformation data to the corresponding mesh vertices, perform spatial resolution error compensation, and generate spatially continuous three-dimensional fused data. Step S15: Extract strain-deformation coupling parameters for each grid point based on the 3D fusion data, calculate the transverse shear stress, longitudinal compressive strain and the angle between the deformation vector according to the elevation gradient, and integrate them into a 3D deformation monitoring dataset containing multiple mechanical indicators.

3. The method for slope reinforcement in building and municipal engineering according to claim 2, characterized in that, Step S15 includes the following steps: Step S151: Perform grid point decoupling processing on the generated three-dimensional fusion data, extract the DFOS strain parameters and InSAR deformation vector parameters corresponding to each grid point, and generate a set of grid parameters containing strain amplitude, deformation rate and spatial coordinates. Step S152: Based on the slope elevation gradient distribution data, the grid parameter set is divided into layers, the lateral strain difference rate between adjacent grid points in each elevation layer is calculated, and the lateral shear stress components are deduced by combining the reference value of rock mass elastic modulus to generate layered lateral shear stress distribution data. Step S153: Based on the DFOS strain parameters in the grid parameter set, calculate the cumulative strain change along the gravity direction in each layer according to elevation, and combine the Poisson's ratio of the rock mass obtained from geological exploration to correct the longitudinal compressive strain intensity, and output the corrected longitudinal compressive strain data of each layer. Step S154: Perform three-dimensional directional decomposition on the InSAR deformation vector parameters, calculate the cosine of the angle between the deformation vector of the grid points in the same elevation layer and the main slip direction of the slope, and combine it with the slope rock layer dip angle data for weighted processing to generate a set of deformation vector angle values. Step S155: Align and fuse the layered transverse shear stress distribution data, longitudinal compressive strain correction data, and deformation vector angle data according to the elevation layer in multiple dimensions, remove data with conflicting dimensions, and generate a standardized three-dimensional deformation monitoring dataset.

4. The method for slope reinforcement in building and municipal engineering according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Construct a three-channel gated recurrent neural network architecture, including strain feature extraction branch, deformation feature extraction branch and parameter inversion branch; Step S22: Input the DFOS strain time series data into the strain feature extraction branch and perform time dimension convolution processing to generate strain spatiotemporal feature vectors; Step S23: Input the InSAR deformation data into the deformation feature extraction branch and perform spatial dimension pooling processing to generate a deformation spatial distribution matrix; Step S24: Weighted concatenation of the strain spatiotemporal feature vector and the deformation spatial distribution matrix to form a fused feature tensor; Step S25: Perform bidirectional recursive calculation on the fused feature tensor through the parameter inversion branch, and output the original parameter set containing the shear modulus degradation rate, creep coefficient and displacement increment.

5. A method for reinforcing slopes in building and municipal applications according to claim 4, characterized in that, Steps S22 to S24 specifically include: Step S221: Perform sliding window noise reduction on the DFOS strain time series data, with a window length of 1 hour and a step size of 15 minutes; Step S222: The noise-reduced data is divided into data segments of equal duration using the overlapping segmentation method, with each data segment containing N sampling points; Step S223: Perform a one-dimensional convolution operation in the time dimension, with the convolution kernel size set to 5 time steps, to generate a primary temporal feature map; Step S224: Perform max pooling on the primary time series feature map to extract the most representative strain feature peaks; Step S225: Standardize the pooled feature sequence to generate a 32-dimensional strain spatiotemporal feature vector; Step S231: Divide the InSAR deformation data into 10m×10m grid cells according to the slope engineering coordinates; Step S232: Calculate the mean and standard deviation of deformation data within each grid cell to form a spatial distribution statistical matrix; Step S233: Use the spatial pyramid pooling method to process the statistical matrix and retain the first 80% of the high-frequency deformation features; Step S234: Perform region clustering on the filtered features and merge adjacent grids with similar deformation patterns; Step S235: Generate a 16×16 spatial distribution matrix that reflects the overall deformation trend of the slope; Step S241: Assign a weight value of 0.6 to the strain spatiotemporal eigenvector and a weight value of 0.4 to the deformation spatial distribution matrix; Step S242: Concatenate the weighted feature vectors and the distribution matrix along the channel dimension to form a 48-dimensional fused feature tensor; Step S243: Verify whether the numerical range of the fused feature tensor is within the preset [0,1] interval; Step S244: When a feature value exceeds the threshold, activate the adaptive scaling mechanism to adjust the feature dimensions; Step S245: Output the validated normalized fusion feature tensor.

6. A method for reinforcing slopes in building and municipal engineering according to claim 4, characterized in that, Step S25 includes: Step S251: Perform recursive calculation of the forward time step on the fused feature tensor to generate the forward propagation state vector; Step S252: Perform recursive calculation of the fused feature tensor in reverse time step to generate the backpropagation state vector; Step S253: Weighted concatenation of the forward propagation state vector and the backward propagation state vector to form a bidirectional state tensor; Step S254: Process the bidirectional state tensor through a fully connected layer and output initial values ​​containing three parameters: initial value of shear modulus degradation rate, initial value of creep coefficient, and initial value of displacement increment. Step S255: Verify the physical feasibility of the initial parameter values ​​based on the constitutive constraints of the soil and rock mass, and automatically reduce the gain coefficient when the shear modulus degradation rate exceeds the allowable threshold. Step S256: Perform numerical optimization on the verified initial parameter values ​​to generate the final shear modulus degradation rate parameter, creep coefficient parameter, and displacement increment prediction data.

7. A method for reinforcing slopes in building and municipal applications according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: The shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data are fused and cleaned to remove sensor noise points and generate a three-dimensional spatiotemporally aligned slope mechanical state data package. Step S32: Construct a residual reinforcement learning model, input the shear modulus degradation rate time-series change rate and creep coefficient transient fluctuation value from the mechanical state data packet, predict the current slope safety index, and use the displacement increment prediction value as a residual correction term to dynamically calibrate the prediction result; Step S33: Based on the percentage of real-time displacement increment to the preset warning value, dynamically allocate the weights of shear modulus, creep coefficient, and displacement increment to generate an adaptive weight parameter group. Step S34: Perform mixed integer programming operation on the safety index output by the residual reinforcement learning model and the adaptive weight parameter group to generate a control strategy code that includes the anchor tension gradient and grouting pressure range; Step S35: Verify the feasibility of the three-dimensional control strategy coding. If the anchor tension and grouting pressure of the same grid unit conflict, the back propagation mechanism is triggered to optimize the MILP constraint conditions until conflict-free integrated control parameters are generated.

8. A slope reinforcement system for building and municipal applications, characterized in that, It is implemented based on any one of claims 1-7, a method for slope reinforcement in building and municipal applications, comprising: Monitoring Data Set Acquisition Module: Collects time-series data of slope rock mass strain through a distributed fiber optic sensing system, and acquires slope deformation monitoring data through a spaceborne interferometric synthetic aperture radar to construct a three-dimensional deformation monitoring dataset; 3D stress field inversion module: Establish an LSTM-soil constitutive model that integrates DFOS strain field and InSAR deformation field, input strain time series data and deformation data into the model to perform 3D stress field inversion, and output shear modulus degradation rate parameter, creep coefficient parameter and displacement increment prediction data. Grouting pressure regulation module: Processes shear modulus degradation rate parameters, creep coefficient parameters, and displacement increment prediction data to generate control strategy codes for anchor tension gradient and grouting pressure range, and uploads them to the control terminal.

Images