Xinjiang cold and arid slope reinforcement evaluation method based on digital twinning

Abstract

The application discloses a Xinjiang cold and arid slope reinforcement evaluation method based on digital twinning, and belongs to the field of digital evaluation of cold and arid region slope reinforcement state, and aims at solving the problem that the slope reinforcement components, drainage facilities, slope surface peeling and deep slip risk in the Xinjiang cold and arid region are difficult to continuously grasp. The application realizes the technical effects of continuous evaluation of the cold and arid region slope reinforcement state and output of maintenance priority by establishing a three-dimensional digital twinning model, collecting multi-source continuous data, constructing a heterogeneous space-time graph and performing time series coding, graph attention propagation and physical stability calibration.

Description

Technical Field

[0001] This invention relates to the field of digital assessment of slope reinforcement status in cold and arid regions, and particularly to a digital twin-based assessment method for slope reinforcement in cold and arid regions of Xinjiang. Background Technology

[0002] With the continuous construction of slope reinforcement projects for highways, mines, and water conservancy in cold and arid regions, reinforcement measures such as anchor bolts, anti-slide piles, grid beams, and interception and drainage structures have been widely used for slope stability control. Existing projects typically monitor the condition of components through completion acceptance, regular inspections, single-point monitoring, and phased testing, and combine this with on-site experience to determine subsequent maintenance and reinforcement.

[0003] However, slopes in the cold and arid regions of Xinjiang are subject to the combined effects of freeze-thaw cycles, wind erosion, salinization, and seasonal fluctuations in drainage capacity. A spatiotemporal coupling relationship exists between the slope's mechanical properties, surface protection status, and deep displacement. Existing methods struggle to continuously correlate data such as anchor axial force, anti-slide pile displacement, lattice beam cracks, drainage flow, temperature, humidity, and salinity environment, slope surface point clouds, and deep inclinometer readings. This can easily lead to localized reinforcement failure, surface spalling, and delayed identification of deep slippage trends.

[0004] Therefore, there is a need for an assessment method for reinforcing cold and arid slopes in Xinjiang that can address the shortcomings of existing technologies. Summary of the Invention

[0005] One objective of this invention is to propose a digital twin-based assessment method for the reinforcement of cold and arid slopes in Xinjiang. Addressing the limitations of existing technologies in continuously monitoring the service status of reinforcement components, seasonal failures of drainage facilities, and the coupled risks of slope surface spalling and deep slippage on cold and arid slopes, this invention proposes a technical solution that involves establishing a three-dimensional digital twin model, collecting and aligning multi-source continuous data, extracting characteristics of cold and arid degradation and drainage failure, constructing a heterogeneous spatiotemporal map, and using temporal coding, graph attention propagation, and physical stability calibration to output status risks and maintenance priorities. This invention effectively improves the timeliness of local failure identification, coupled slippage risk assessment, and maintenance decision-making.

[0006] This invention provides a digital twin-based assessment method for reinforcing arid and cold slopes in Xinjiang, comprising: S1, establishing a three-dimensional digital twin model of the arid and cold slope, dividing the slope into slope units, and registering the spatial locations, component attributes, and connection relationships of anchor bolts, anti-slide piles, lattice beams, drainage structures, and monitoring points in the three-dimensional digital twin model to obtain a set of twin foundation objects; S2, collecting multi-source continuous data corresponding to the set of twin foundation objects according to a unified timestamp, and performing coordinate registration and time alignment on the multi-source continuous data to obtain a spatiotemporally aligned dataset; S3, extracting freeze-thaw degradation features, wind erosion features, salt erosion features, and drainage failure features from the spatiotemporally aligned dataset, and... S4. Generate a slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector from the aforementioned features; S5. Construct a heterogeneous spatiotemporal graph based on the aforementioned twin-based object set and the aforementioned features, and update the weight matrices of the force transmission edge, seepage drainage edge, and deep slope surface coupling edge according to the aforementioned slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector; S6. Input the updated heterogeneous spatiotemporal graph into the evaluation model, and have the evaluation model perform temporal encoding, graph attention propagation based on the aforementioned weight matrix, and physical stability calibration, and output the component service status, slope surface spalling risk, deep slippage trend, drainage failure category, and comprehensive maintenance priority.

[0007] Optionally, S1 includes:

[0008] A three-dimensional coordinate system was established based on slope design data, topographic survey data, and geological exploration data.

[0009] The slope unit is generated using the soil-rock stratification boundary, slope grid zoning, and monitoring profile as segmentation conditions;

[0010] Register the length, inclination angle, axial force design value, and anchorage section location of the anchor rod as anchor rod node attributes; register the pile location, pile length, cross-sectional dimensions, and pile top displacement design limit of the anti-slide pile as anti-slide pile node attributes; register the beam grid location, cross-sectional dimensions, and crack monitoring location of the lattice beam as lattice beam node attributes; and register the ditch direction, cross-sectional dimensions, and water collection unit of the drainage structure as drainage structure node attributes.

[0011] The connection relationships are generated according to spatial adjacency, component connection, and drainage connectivity.

[0012] Optionally, S2 includes:

[0013] The multi-source continuous data includes anchor axial force, anchor displacement, anti-slide pile top displacement, anti-slide pile body strain, lattice beam crack width, drainage structure flow rate, drainage structure water level, slope temperature, slope moisture content, pore water pressure, salt content, electrical conductivity, wind speed, snowfall, slope surface point cloud elevation, and deep inclinometer displacement.

[0014] The coordinate registration includes mapping the elevation of the slope surface point cloud, the deep inclinometer displacement, and the coordinates of each monitoring point to the three-dimensional coordinate system;

[0015] The time alignment includes resampling each data channel according to a preset sampling period, and forming the spatiotemporal aligned dataset with the mean, peak, valley and rate of change within the same time window.

[0016] Furthermore, after forming the spatiotemporal aligned dataset, the data credibility is calculated for each data channel. The data credibility is obtained by normalizing and weighting the online duration ratio, the deviation from the median value of adjacent monitoring points, and the deviation from the predicted value of the physical model.

[0017] When the data credibility is less than a preset credibility threshold, the node features of the corresponding data channel are replaced with the fusion value of the weighted interpolation of adjacent nodes and the prediction value of the physical model, and the edge weights related to the corresponding data channel are reduced.

[0018] The replaced node features and adjusted edge weights are input into the evaluation model.

[0019] Optionally, S3 includes:

[0020] The freeze-thaw degradation characteristics are generated by the number of times the slope temperature crosses the freezing point, the duration of freezing, the duration of thawing, and the peak moisture content.

[0021] The wind erosion characteristics are generated by the elevation difference of point clouds on the slope surface, the cumulative wind speed, and the proportion of exposed area on the slope surface.

[0022] The salt corrosion characteristics are generated by changes in salt content, electrical conductivity, and pore water pressure.

[0023] The drainage failure characteristics are generated by the flow attenuation rate of the intercepting and drainage structure, the water retention time, and the siltation ratio of the ditch.

[0024] The freeze-thaw degradation characteristics, the salt erosion characteristics, and the deep displacement increment are input into the first weight table obtained by fitting historical calibration samples to obtain the slope mechanical degradation coefficient.

[0025] The wind erosion characteristics and the change in cracks in the lattice beams are input into the second weight table obtained by fitting historical calibration samples to obtain the slope protection reduction coefficient.

[0026] The drainage failure characteristics and the slope moisture content increment are input into the third weight table obtained by fitting historical calibration samples to obtain the drainage failure coefficient.

[0027] The physical residual vector is composed of the difference between the monitored displacement and the displacement calculated by physical stability, the difference between the monitored drainage flow and the hydraulically calculated flow, and the difference between the monitored internal forces of the components and the calculated internal forces of the structure.

[0028] Optionally, S4 includes:

[0029] The heterogeneous spatiotemporal diagram includes slope unit nodes, anchor nodes, anti-slide pile nodes, lattice beam nodes, drainage structure nodes, and monitoring point nodes;

[0030] The force transmission side connects the slope unit node with the anchor node, the anti-slide pile node and the lattice beam node; the seepage drainage side connects the slope unit node with the interception and drainage structure node; and the slope surface deep coupling side connects the slope surface monitoring point node, the slope unit node and the deep displacement monitoring point node.

[0031] The initial edge weights are determined by the spatial distance between nodes, the design connection stiffness, and the hydraulic connectivity direction;

[0032] The slope mechanical deterioration coefficient, the slope surface protection reduction coefficient, and the drainage failure coefficient are mapped to a closed interval of 0 to 1, and the force transmission edge, the seepage drainage edge, and the deep coupling edge of the slope surface are corrected respectively by combining the normalized components of the physical residual vector to obtain the updated weight matrix.

[0033] Optionally, S5 includes:

[0034] The evaluation model includes a temporal coding layer, a graph attention propagation layer, a physical stability calibration layer, and a multi-task output layer;

[0035] The temporal coding layer generates node temporal latent vectors based on the node feature sequence within the sliding time window;

[0036] The graph attention propagation layer concatenates the temporal latent vectors of adjacent nodes, the edge type embedding, and the weight matrix to calculate the edge attention coefficients, and normalizes the edge attention coefficients according to the edge type to obtain the node state vector.

[0037] The physical stability calibration layer generates a calibration vector based on the limit equilibrium safety factor, pile-anchor force utilization rate and drainage capacity ratio, and then merges the calibration vector with the node state vector to obtain the calibration state vector.

[0038] The multi-task output layer outputs the component service status, the slope surface spalling risk, the deep slippage trend, the drainage failure category, and the comprehensive maintenance priority from the calibration state vector;

[0039] Furthermore, the physical stability calibration layer also generates deformation attribute markers, which are determined as follows: within a continuous m sliding time window, the displacement fallback ratio of the deep displacement after the freeze-thaw temperature recovery is calculated, and the duration of the residual axial force of the anchor bolt, the residual displacement of the anti-slide pile, and the residual crack change of the lattice beam are calculated, where m is a preset positive integer;

[0040] When the displacement fallback ratio is not less than a preset recovery threshold and the duration is less than a preset duration threshold, the corresponding change is marked as seasonal reversible deformation.

[0041] When the displacement fallback ratio is less than a preset recovery threshold and the duration is not less than a preset duration threshold, the corresponding change will be marked as irreversible reinforcement damage.

[0042] The deformation attribute markers are used as the weight correction input for the deep coupling edge of the slope surface and the calculation input for the comprehensive maintenance priority;

[0043] Furthermore, the comprehensive maintenance priority is determined in the following way: the service status of the component is converted into a service status code, the slope surface spalling risk is converted into a spalling risk value, the deep slippage trend is converted into a slippage trend value, and the drainage failure category is converted into a drainage category code;

[0044] The service status code, the spalling risk value, the slippage trend value, the drainage category code, the norm of the physical residual vector, and the deformation attribute label are input into the priority weight table to obtain a comprehensive score.

[0045] When the comprehensive score is not less than the first score threshold, the first maintenance priority is output; when the comprehensive score is less than the first score threshold but not less than the second score threshold, the second maintenance priority is output; when the comprehensive score is less than the second score threshold, the third maintenance priority is output.

[0046] The first scoring threshold is greater than the second scoring threshold, and the first scoring threshold and the second scoring threshold are determined by the scoring percentile values ​​of historical maintenance samples;

[0047] Furthermore, the evaluation model is trained by acquiring the spatiotemporally aligned dataset of historical cold and drought slope engineering projects, component detection records, slope surface spalling records, deep slippage monitoring records, drainage maintenance records, and maintenance and treatment records.

[0048] Generate the heterogeneous spatiotemporal graph and the weight matrix for training using the same processing method as steps S3 and S4.

[0049] Training labels are generated using the component inspection records, the slope surface spalling records, the deep slippage monitoring records, the drainage maintenance records, and the maintenance and treatment records.

[0050] The evaluation model is obtained by using the weighted sum of component state classification loss, slope surface spalling regression loss, deep slippage trend regression loss, drainage category classification loss, maintenance priority classification loss, edge weight smoothing loss, and physical residual consistency loss as the training objective.

[0051] The beneficial effects of this invention are:

[0052] 1. By establishing a three-dimensional digital twin model and registering the spatial location, component attributes, and connection relationships of slope units, anchor bolts, anti-slide piles, grid beams, drainage structures, and monitoring points, it is possible to incorporate the dispersed component status, slope environment, slope surface morphology, and deep displacement into a unified object system, which facilitates continuous monitoring of the service status of cold and arid slope reinforcement projects.

[0053] 2. By extracting freeze-thaw degradation, wind erosion, salt erosion and drainage failure features from spatiotemporally aligned datasets, and generating slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient and physical residual vector, the effects of cold and drought environment can be correlated with the response of reinforced components, thereby improving the identification of slope surface spalling, drainage failure and deep slippage trends.

[0054] 3. By updating the weight matrices of the force transmission edge, seepage drainage edge, and deep slope coupling edge in the heterogeneous spatiotemporal diagram, and performing temporal encoding, graph attention propagation, and physical stability calibration in the evaluation model, it is possible to distinguish between seasonal reversible deformation and irreversible reinforcement damage, output comprehensive maintenance priorities, and reduce the risk of delayed maintenance and reinforcement decisions. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a flowchart of an assessment method for reinforcing cold and arid slopes in Xinjiang based on digital twins.

[0057] Figure 2 This is a flowchart of step S5 of the present invention for evaluating the model. Detailed Implementation

[0058] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0059] refer to Figures 1-2A digital twin-based assessment method for reinforcing arid and cold slopes in Xinjiang includes: S1. Establishing a three-dimensional digital twin model of the arid and cold slope, dividing the slope into slope units, and registering the spatial locations, component attributes, and connection relationships of anchor bolts, anti-slide piles, lattice beams, drainage structures, and monitoring points in the three-dimensional digital twin model to obtain a set of twin foundation objects; S2. Collecting multi-source continuous data corresponding to the set of twin foundation objects according to a unified timestamp, and performing coordinate registration and time alignment on the multi-source continuous data to obtain a spatiotemporally aligned dataset; S3. Extracting freeze-thaw degradation characteristics, wind erosion characteristics, salt erosion characteristics, and drainage failure characteristics from the spatiotemporally aligned dataset, and then... The following steps are taken: S4. Based on the twin-based object set and the features, a heterogeneous spatiotemporal graph is constructed, and the weight matrices of the force transmission edge, seepage drainage edge, and deep coupling edge of the slope surface are updated according to the slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector; S5. The updated heterogeneous spatiotemporal graph is input into the evaluation model, which performs time-series encoding, graph attention propagation based on the weight matrix, and physical stability calibration, and outputs the component service status, slope surface spalling risk, deep slippage trend, drainage failure category, and comprehensive maintenance priority.

[0060] In this specific embodiment, S1 includes:

[0061] Based on the slope toe line, slope crest line, anchor bolt layout diagram, anti-slide pile layout diagram, lattice beam layout diagram, and drainage structure layout diagram in the slope design data, combined with the slope surface point cloud, control point coordinates, and elevation data in the topographic survey data, and the rock and soil stratification boundaries, groundwater level, and monitoring profile locations in the geological survey data, a three-dimensional coordinate system is established for the cold and arid slope. The slope toe control point is used as the origin, and the slope direction is used as the coordinate axis. The positive direction of the axis is defined by the horizontal normal pointing from the inside of the slope to the outside. The positive direction of the axis, with the direction of increasing elevation as the reference. In the positive direction of the axis, the design coordinates, measurement coordinates, and exploration coordinates are uniformly converted into spatial coordinates in a three-dimensional coordinate system. The coordinate transformation relationship is as follows: ,in, Indicates the first The coordinate vectors of an object point in the three-dimensional coordinate system. Indicates the first The object points along the The coordinate values ​​along the axis. Indicates the first The object points along the The coordinate values ​​along the axis. Indicates the first The object points along the The coordinate values ​​along the axis. Indicates the first The coordinate vectors of each object point in the original data coordinate system. This represents the coordinate vector of the control point at the toe of the slope in the original data coordinate system. Indicated by the axis, shaft and A three-dimensional rotation matrix composed of the cosines of the axial directions. The object point number is represented by T, and the vector transpose is represented by T. After the coordinate unification is completed, the slope is divided into entities using the soil and rock layer boundary, the slope grid partition and the monitoring profile as common dividing conditions. When the soil and rock layer boundary intersects with the slope grid partition or the monitoring profile, the intersection line is used as the new unit boundary, so that each slope unit has a unique spatial range, soil and rock layer type, grid partition and monitoring profile number, and the slope unit is registered as a slope unit node.

[0062] Node objects are established for anchor bolts, anti-slide piles, lattice beams, drainage structures, and monitoring points. Anchor bolt node attributes include anchor bolt number, anchor bolt start coordinates, anchor bolt end coordinates, length, inclination angle, axial force design value, and anchorage section location. Anti-slide pile node attributes include anti-slide pile number, pile location coordinates, pile length, cross-sectional dimensions, pile top displacement design limit, and corresponding control slope unit. Lattice beam node attributes include lattice beam number, beam grid location, cross-sectional dimensions, slope unit covered by beam grid, and crack monitoring location. Drainage structure node attributes include drainage structure number, ditch start and end coordinates, ditch direction, cross-sectional dimensions, catchment unit, drainage direction, and design drainage capacity. Monitoring point node attributes include monitoring point number, monitoring type, installation coordinates, monitoring object, and sampling channel.

[0063] In the three-dimensional digital twin model, connection relationships are further generated. Spatial adjacency relationships are determined by whether two slope units share a boundary surface. Component connection relationships are determined by whether the anchor bolt anchorage section passes through the slope unit, whether the influence range of the anti-slide pile body covers the slope unit, and whether the grid of the lattice beam covers the slope unit. Drainage connectivity relationships are determined by the water collection unit of the interception and drainage structure, the direction of the ditch, and the drainage destination. Monitoring association relationships are determined by the slope unit where the monitoring point is installed or the anchor bolt, anti-slide pile, lattice beam, and interception and drainage structure directly bound to the monitoring point. This forms a set of twin basic objects that includes slope unit nodes, anchor bolt nodes, anti-slide pile nodes, lattice beam nodes, interception and drainage structure nodes, monitoring point nodes, and the above connection relationships.

[0064] In this specific embodiment, S2 includes:

[0065] Based on the twin-based object set, a unique data channel number is established for each anchor node, anti-slide pile node, lattice beam node, drainage structure node, slope unit node, and monitoring point node, and these are arranged according to a unified timestamp sequence. Collect continuous data from multiple sources, among which, Indicates the first A unified sampling time, Indicates a unified sampling time number and a unified sampling period. The data collected over 1 hour includes multi-source continuous data such as anchor bolt axial force, anchor bolt displacement, anti-slide pile top displacement, anti-slide pile body strain, lattice beam crack width, drainage structure flow rate, drainage structure water level, slope temperature, slope moisture content, pore water pressure, salt content, electrical conductivity, wind speed, snowfall, slope surface point cloud elevation, and deep inclinometer displacement. During data collection, the anchor bolt axial force and anchor bolt displacement are bound to the corresponding anchor bolt nodes, and the anti-slide pile top displacement and anti-slide pile body strain are bound to the corresponding anti-slide pile top displacement and anti-slide pile body strain. For each pile node, the width of the crack in the lattice beam is bound to the corresponding lattice beam node; the flow rate and water level of the intercepting and drainage structure are bound to the corresponding intercepting and drainage structure node; the slope temperature, slope moisture content, pore water pressure, salt content, and electrical conductivity are bound to the corresponding slope unit node or monitoring point node; the wind speed and snowfall are bound to the slope environmental monitoring point node; the slope surface point cloud elevation is bound to the corresponding slope surface monitoring point node; and the deep inclinometer displacement is bound to the corresponding deep displacement monitoring point node.

[0066] During the coordinate registration process, the three-dimensional coordinate system established in step S1 is used to transform the installation coordinates of each elevation point in the slope point cloud, each measuring point in the deep inclinometer hole, and each monitoring point to the three-dimensional coordinate system. Based on the spatial range of the slope unit, the coverage range of the component, or the binding relationship of the monitoring point, the elevation of the slope point cloud, the displacement of the deep inclinometer hole, and each monitoring data are mapped to the corresponding slope unit node, component node, or monitoring point node.

[0067] During time alignment, all data channels are sampled according to a uniform period. Perform resampling at a frequency higher than [previous frequency]. The data channels calculate the mean, peak, trough, and rate of change within the same time window, with a sampling frequency lower than [missing information]. And the interval between adjacent valid sampling points is no greater than The data channels are completed using linear interpolation, with a sampling interval greater than [missing information]. The data channels are marked as missing channels and processed in subsequent data reliability processing. The same time window is defined as... The end time and length are The time interval, wherein the rate of change is the difference between the end value and the beginning value of the time window divided by [the value of the change]. ;

[0068] After time alignment is completed, each node will be placed in a unified timestamp sequence. The mean, peak, valley and rate of change are combined into node features, and all node features, node number, data channel number and timestamp are combined into a spatiotemporally aligned dataset;

[0069] After forming the spatiotemporally aligned dataset, data confidence is calculated for each data channel, and the data confidence is calculated according to... It is confirmed that, among them, Indicates the first The data channel is in the first The reliability of data within a time window corresponding to a unified sampling time. Indicates the data channel number. Indicates the unified sampling time number. Indicates the first The data channel is in the first The proportion of online time within a time window corresponding to a unified sampling time. Indicates the first The normalized value of the deviation between the channel value of each data channel and the median value of adjacent monitoring points. Indicates the first The normalized value of the deviation between the channel values ​​of each data channel and the predicted values ​​of the physical model. 0.3 and 0.3 represent the weights of the online duration ratio, the adjacent monitoring point deviation term, and the physical model deviation term, respectively;

[0070] The adjacent monitoring points are those that are located in the same slope unit or adjacent slope unit as the current monitoring point and have the same monitoring type. The median value of the adjacent monitoring points is the median of the effective channel values ​​of the adjacent monitoring points within the same time window.

[0071] The physical model predictions are obtained from a physical calculation model constructed based on the twin-based object set in step S1. Specifically, the anchor axial force and anti-slide pile displacement are predicted using a linear elastic force calculation model established based on component design stiffness, node connection relationships, and slope element deformation. The flow rate and water level of the intercepting and drainage structure are predicted using a hydraulic connectivity calculation model established based on ditch cross-sectional dimensions, ditch direction, catchment units, and design drainage capacity. The deep inclinometer displacement is predicted using a limit equilibrium displacement trend model established based on soil-rock layer boundaries, slope element geometry, and slope stability calculation parameters. Slope temperature, slope moisture content, pore water pressure, salt content, and electrical conductivity are predicted using a combination of historical time series data from the same slope element and spatial interpolation from adjacent slope elements. When the data reliability... When the value is less than the preset confidence threshold of 0.6, the node features of the corresponding data channel within that time window are replaced with the fusion value of the weighted interpolation of adjacent nodes and the prediction value of the physical model. The weighted interpolation of adjacent nodes is normalized and weighted according to the reciprocal of the spatial distance between the adjacent node and the current node. During fusion, the weight of the weighted interpolation of adjacent nodes is 0.6, and the weight of the prediction value of the physical model is 0.4. The edge weights of the force transmission edge, seepage drainage edge, or deep slope coupling edge related to the data channel are multiplied by the aforementioned data confidence level. This yields the replaced node features and adjusted edge weights;

[0072] When the data credibility When the confidence level is not less than the preset confidence threshold of 0.6, the node features and related edge weights of the corresponding data channel within the time window are retained.

[0073] Finally, the unified timestamp sequence, the replaced node features, the retained node features, the adjusted edge weights, and the unadjusted edge weights are used together as the spatiotemporal aligned dataset output in step S2, and serve as the data basis for feature extraction in subsequent step S3 and model evaluation in step S5.

[0074] In this specific embodiment, S3 includes:

[0075] Based on spatiotemporally aligned datasets, to unify sampling times To evaluate the moment, and from to The time interval is used as the current evaluation time window. For each slope unit node, anchor node, anti-slide pile node, lattice beam node, drainage structure node and monitoring point node, the characteristics of the cold and drought environment and the component response characteristics are extracted.

[0076] Among them, the freeze-thaw deterioration characteristic is caused by the slope temperature crossing the freezing point within the current evaluation time window. The number of times and the temperature are lower than The duration of freezing and the temperature are higher than The melting duration and peak slope moisture content are considered. The number of times the freezing point is crossed is counted based on the number of times the slope temperature sign changes at adjacent unified sampling times. The freezing duration is defined as the slope temperature within the current evaluation time window being less than [a certain value]. The cumulative duration, wherein the melting duration is defined as the slope temperature within the current evaluation time window being greater than or equal to... The cumulative duration, the peak slope moisture content is the maximum slope moisture content within the current evaluation time window; the wind erosion characteristics consist of the slope surface point cloud elevation difference, cumulative wind speed, and the proportion of exposed slope surface area. The slope surface point cloud elevation difference is the absolute value of the difference between the slope surface point cloud elevation at the end of the current evaluation time window and the slope surface point cloud elevation at the end of the previous evaluation time window for the same slope surface monitoring point node. The cumulative wind speed is the average wind speed within the current evaluation time window sampled according to a uniform sampling period. The cumulative wind action is obtained by summing up the slope surface exposed area ratio, which is the ratio of the slope surface area not covered by grid beams, sprayed layer or vegetation protection to the total slope surface area of ​​the corresponding slope unit.

[0077] Salt erosion characteristics consist of salt content, electrical conductivity, and pore water pressure changes. The salt content is the average salt content of the corresponding slope unit node within the current evaluation time window. The electrical conductivity is the average electrical conductivity of the corresponding slope unit node within the current evaluation time window. The pore water pressure change is the absolute value of the difference between the pore water pressure at the end of the current evaluation time window and the pore water pressure at the beginning of the current evaluation time window. Drainage failure characteristics consist of intercepting and drainage structure flow attenuation rate, water level retention time, and ditch siltation ratio. The intercepting and drainage structure flow attenuation rate is the non-negative value obtained by dividing the difference between the designed drainage capacity and the average measured flow rate within the current evaluation time window by the designed drainage capacity. The water level retention time is the cumulative time during which the water level in the intercepting and drainage structure is higher than the designed warning water level. The ditch siltation ratio is the ratio of the silted cross-sectional area within the intercepting and drainage structure to the designed water passage cross-sectional area.

[0078] After obtaining the above features, each feature is normalized according to the minimum and maximum values ​​in the historical calibration samples. The normalized feature values ​​are restricted to a closed interval between 0 and 1. The historical calibration samples are cold and arid slope engineering samples that have completed the verification of detection, monitoring and maintenance records. When normalizing, values ​​less than the historical minimum value are taken as 0, and values ​​greater than the historical maximum value are taken as 1.

[0079] The first, second, and third weight tables are all obtained by fitting historical calibration samples and are fixed as linear weighted tables in this specific embodiment. The first weight table includes a freeze-thaw degradation feature weight of 0.35, a salt erosion feature weight of 0.25, and a deep displacement increment weight of 0.40. The second weight table includes a wind erosion feature weight of 0.55 and a lattice beam crack change weight of 0.45. The third weight table includes a drainage failure feature weight of 0.65 and a slope moisture content increment weight of 0.35. The deep displacement increment is the absolute value of the difference between the deep inclinometer displacement at the end of the current evaluation time window and the deep inclinometer displacement at the beginning of the current evaluation time window. The lattice beam crack change is the absolute value of the difference between the lattice beam crack width at the end of the current evaluation time window and the lattice beam crack width at the beginning of the current evaluation time window. The slope moisture content increment is the positive value of the difference between the slope moisture content at the end of the current evaluation time window and the slope moisture content at the beginning of the current evaluation time window.

[0080] The slope mechanical deterioration coefficient, the slope surface protection reduction coefficient, the drainage failure coefficient, and the physical residual vector are calculated according to:

[0081] ,

[0082] ,

[0083] ,

[0084] ;

[0085] in, Indicates the first The slope element node at the first The slope mechanical deterioration coefficient at a uniform sampling time. Indicates the first The slope element node at the first Slope protection reduction factor at a uniform sampling time Indicates the first The slope element node at the first The drainage failure coefficient at a uniform sampling time. Indicates the node number of the slope element. Indicates the unified sampling time number. This represents the normalized freeze-thaw degradation characteristics. This represents the normalized characteristics of salt erosion. This represents the normalized increment of deep displacement. This represents the normalized characteristics of wind erosion. This represents the normalized change in crack size in the lattice beam. This represents the normalized drainage failure characteristics. This represents the normalized increase in slope moisture content. Indicates the first The slope element node at the first The physical residual vector at a uniform sampling time. This represents the monitored displacement obtained by mapping deep inclinometer displacement and slope surface monitoring displacement. This represents the displacement calculated based on physical stability using the limit equilibrium displacement trend model. This indicates the monitored drainage flow rate at the corresponding interception and drainage structure node. This represents the hydraulically calculated flow rate obtained by the hydraulic connectivity calculation model based on the ditch cross-sectional dimensions, ditch orientation, catchment units, and design drainage capacity. This indicates the internal forces of the monitored components corresponding to anchor bolt nodes, anti-slide pile nodes, or lattice beam nodes. This represents the internal forces calculated by the linear elastic force calculation model based on the component design stiffness, connection relationships, and slope element deformation. T represents the vector transpose.

[0086] When calculating the physical residual vector, the monitored displacement, monitored drainage flow and monitored component internal force are all taken from the original node features with a confidence level of not less than 0.6 in step S2 or the node features that have been replaced after the confidence level is less than 0.6. The physical stability calculation displacement, hydraulic calculation flow and structural calculation internal force are all taken from the output results of the same physical calculation model in step S2, so that the physical residual vector and the edge weight correction in the subsequent step S4 keep the same data source.

[0087] The freeze-thaw degradation characteristics, wind erosion characteristics, salt erosion characteristics, drainage failure characteristics, slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector corresponding to each slope unit node are written into the node attribute field of the spatiotemporal alignment dataset and used as input for S4 to construct heterogeneous spatiotemporal graphs and update weight matrices.

[0088] In this specific embodiment, S4 includes:

[0089] Based on the twin-based object set and the freeze-thaw degradation characteristics, wind erosion characteristics, salt erosion characteristics, drainage failure characteristics, slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector written into the node attribute fields in step S3, in the first... A heterogeneous spatiotemporal graph is constructed using a unified sampling time, and the heterogeneous spatiotemporal graph is denoted as […]. ,in, Indicates the first A heterogeneous spatiotemporal graph with a unified sampling time. Represents a set of nodes. Denotes the set of edges. Indicates the first The weight matrix at a uniform sampling time;

[0090] The set of nodes This includes slope unit nodes, anchor nodes, anti-slide pile nodes, lattice beam nodes, drainage structure nodes, and monitoring point nodes. The slope unit nodes bear the soil and rock layer type, spatial range, freeze-thaw degradation characteristics, salt erosion characteristics, drainage failure characteristics, slope mechanical degradation coefficient, drainage failure coefficient, and physical residual vector. The anchor nodes bear the length, inclination angle, axial force design value, anchorage section location, anchor axial force, and anchor displacement. The anti-slide pile nodes bear the pile location, pile length, cross-sectional dimensions, pile top displacement design limit, pile top displacement, and pile strain. The lattice beam nodes bear the beam grid location, cross-sectional dimensions, crack monitoring location, and lattice beam crack width. The drainage structure nodes bear the ditch direction, cross-sectional dimensions, catchment unit, design drainage capacity, drainage structure flow rate, and drainage structure water level. The monitoring point nodes bear the monitoring type, installation coordinates, sampling channel, and node characteristics formed in step S2.

[0091] The set of edges According to the edge type, it is divided into force transmission edge, seepage drainage edge, and slope surface deep coupling edge. The force transmission edge connects the slope unit node with the anchor node, anti-slide pile node and grid beam node, and is used to represent the transmission relationship between slope deformation and the force of the reinforcement component. The seepage drainage edge connects the slope unit node with the intercepting and drainage structure node, and is used to represent the connectivity relationship between the slope unit water collection, seepage and drainage structure discharge. The slope surface deep coupling edge connects the slope surface monitoring point node, slope unit node and deep displacement monitoring point node, and is used to represent the coupling relationship between slope surface erosion, slope unit deterioration and deep displacement change. When generating the force transmission edge, the slope unit through which the anchor bolt anchoring section passes, the slope unit covered by the influence range of the anti-slide pile body and the slope unit covered by the grid beam are respectively connected to the corresponding component node. The influence range of the anti-slide pile body is the spatial range covered by three times the width of the pile cross section on both sides of the pile centerline and along the pile length direction.

[0092] When generating seepage drainage edges, the slope units are connected to the corresponding intercepting and drainage structure nodes according to the water catchment unit affiliation relationship, and the edge direction is determined according to the ditch direction and drainage destination; when generating slope surface deep coupling edges, the slope surface monitoring point nodes, slope unit nodes and deep displacement monitoring point nodes located within the same slope unit spatial range are connected, and the edge direction is made from the slope surface monitoring point node to the slope unit node and from the slope unit node to the deep displacement monitoring point node;

[0093] The initial edge weight is determined by the spatial distance between nodes, the design connection stiffness, and the hydraulic connection direction. The smaller the spatial distance between nodes, the larger the initial edge weight. The larger the design connection stiffness, the larger the initial edge weight of the force transmission edge. If the hydraulic connection direction is consistent with the ditch direction, the initial edge weight of the seepage drainage edge remains unchanged. If the hydraulic connection direction is opposite to the ditch direction, the initial edge weight of the seepage drainage edge is reset to 0.

[0094] In the At a unified sampling time, the slope mechanical deterioration coefficient, slope surface protection reduction coefficient, and drainage failure coefficient are mapped to a closed interval of 0 to 1 using a closed interval truncation method. The displacement residual, drainage flow residual, and component internal force residual in the physical residual vector are normalized according to the maximum absolute residual in the historical calibration samples, and the edge weights are updated according to the edge type. The update relationship is as follows:

[0095] ;

[0096] in, Indicates the first A unified sampling time node With nodes The type between After updating the edge weight, Indicates the starting node number of the edge. Indicates the terminal node number of the edge. Indicates the unified sampling time number. Indicates the edge type. Indicates the edge on which the force is transmitted. Indicates the seepage drainage edge. This indicates the deep coupling edge of the slope surface. Represents a node With nodes The type between The initial edge weights of the edges, The type is The corresponding degradation correction factor for the edge is taken as follows: for the force transmission edge, the seepage drainage edge, and the deep coupling edge of the slope surface, respectively. and Indicates the first The slope element node at the first A unified sampling time and type are The degraded input corresponding to the edge, when hour Take the slope mechanical deterioration coefficient, when hour Take the drainage failure coefficient, when hour The arithmetic mean of the slope surface protection reduction coefficient and the slope mechanical deterioration coefficient is taken. Represents nodes or node The node numbers of the connected slope unit. The type is The residual correction factor corresponding to the edge is taken as follows: force transmission edge, seepage drainage edge, and deep coupling edge of slope surface are all taken as follows. Indicates the first The slope element node at the first A unified sampling time and type are The normalized physical residual components corresponding to the edges, when hour Take the normalized internal force residual of the component, when hour Take the normalized drainage flow residual, when hour Take the normalized displacement residual, This means that results with values ​​less than 0 will be assigned 0, results with values ​​greater than 1 will be assigned 1, and results within the closed interval between 0 and 1 will remain unchanged.

[0097] After completing the edge weight update, the updated edge weights of all force transfer edges, seepage drainage edges, and deep slope surface coupling edges are written into the weight matrix in node number order. Weight matrix The row corresponds to the starting node of the edge, the column corresponds to the ending node of the edge, the matrix elements are the updated weights of the corresponding edges, and the matrix elements with no connection are set to 0. The node set is then... Edge set and weight matrix Together, they are output as an updated heterogeneous spacetime graph to S5.

[0098] In this specific embodiment, S5 includes:

[0099] The updated heterogeneous spacetime graph The evaluation model is continuously input at a uniform sampling time. The evaluation model is composed of a temporal coding layer, a graph attention propagation layer, a physical stability calibration layer, and a multi-task output layer connected in sequence. The evaluation is performed with fixed parameters after the model training is completed.

[0100] The timing coding layer is composed of a single-layer gated loop unit, and the sliding time window length is taken as... , corresponding to from to The 24-hour node feature sequence has an input dimension equal to the total number of node feature fields formed in steps S2 and S3, and a hidden dimension of 64. The node feature fields include the mean, peak value, valley value, rate of change, freeze-thaw degradation characteristics, wind erosion characteristics, salt erosion characteristics, drainage failure characteristics, slope mechanical degradation coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector of multi-source continuous data. The temporal coding layer encodes the node feature sequence of each node within the sliding time window to obtain the node temporal hidden vector.

[0101] The graph attention propagation layer employs a four-head attention structure that distinguishes parameters by edge type. Edge types include force transmission edges, seepage drainage edges, and slope surface deep coupling edges. The edge type embedding dimension is 8, and the output dimension of each attention head is 16. The four attention heads are concatenated to obtain a node state vector with a dimension of 64. During calculation, the temporal latent vectors of adjacent nodes, edge type embeddings, and weight matrices are included. The corresponding edge weights are concatenated and input into the attention scoring function, and normalized according to the same edge type of the same target node, so that the force transmission edge, seepage drainage edge and deep slope coupling edge form independent attention distributions respectively.

[0102] The physical stability calibration layer generates a calibration vector based on the limit equilibrium safety factor, pile-anchor force utilization rate, and drainage capacity ratio. The limit equilibrium safety factor is calculated by the physical calculation model in step S2 based on the soil-rock layer boundary, slope element geometry, slope water content, and pore water pressure. The pile-anchor force utilization rate is the ratio of the monitored value of the anchor axial force or the internal force of the anti-slide pile to the corresponding design bearing capacity. The drainage capacity ratio is the ratio of the measured flow rate of the intercepting and drainage structure to the design drainage capacity. The calibration vector and the node state vector are fused by a fully connected fusion layer to obtain the calibration state vector. The fully connected fusion layer has an input dimension of 67, an output dimension of 64, and uses a modified linear unit as the activation function.

[0103] The physical stability calibration layer also generates deformation attribute markers, with the number of consecutive sliding time windows taking [value missing]. This involves calculating the proportion of deep displacement reduction after freeze-thaw recovery within seven consecutive sliding time windows, and calculating the duration of the residual axial force of anchor bolts, the residual displacement of anti-slide piles, and the residual crack change of lattice beams. Freeze-thaw recovery refers to the temperature of the corresponding slope unit remaining at or above a certain level for 6 consecutive hours. The displacement fallback ratio is the ratio of the deep inclinometer displacement fallback after the freeze-thaw temperature recovers to the deep inclinometer displacement increment before the freeze-thaw temperature recovers. The preset recovery threshold is 0.50, and the preset duration threshold is 72h. When the displacement fallback ratio is not less than 0.50 and the duration is less than 72h, the corresponding change is marked as seasonal reversible deformation and assigned a value of 0. When the displacement fallback ratio is less than 0.50 and the duration is not less than 72h, the corresponding change is marked as irreversible reinforcement damage and assigned a value of 1. Other changes are marked as deformation to be continuously observed and assigned a value of 0.5. The deformation attribute marking is input into the weight correction branch of the deep coupling edge of the slope surface and the comprehensive maintenance priority calculation branch.

[0104] The multi-task output layer consists of a component status classification head, a slope surface spalling risk regression head, a deep slippage trend regression head, a drainage failure category classification head, and a maintenance priority classification head. The component status classification head outputs four types of component service status: normal, watch out, warning, and failure. The slope surface spalling risk regression head outputs slope surface spalling risk with a value range of 0 to 1. The deep slippage trend regression head outputs the predicted value of deep displacement increment for the next evaluation time window. The drainage failure category classification head outputs four types of drainage failure: normal drainage, siltation failure, freezing and water retention failure, and insufficient cross-sectional capacity. The maintenance priority classification head outputs the first maintenance priority, the second maintenance priority, and the third maintenance priority.

[0105] The comprehensive maintenance priority is also constrained by a rule-based scoring branch. This branch converts the component's service status into a service status code, slope spalling risk into a spalling risk value, deep slippage trend into a slippage trend value, and drainage failure category into a drainage category code. The service status code, spalling risk value, slippage trend value, drainage category code, the norm of the physical residual vector, and the deformation attribute label are input into a priority weight table to obtain a comprehensive score. The weights of the service status code, spalling risk value, slippage trend value, drainage category code, physical residual vector norm, and deformation attribute label in the priority weight table are sequentially assigned... And 0.10, the first scoring threshold is taken as the comprehensive score of historical maintenance samples. quantile value, the second scoring threshold is taken from the comprehensive score of historical maintenance samples. The quantile value is used to output the first maintenance priority when the comprehensive score is not less than the first score threshold, the second maintenance priority when the comprehensive score is less than the first score threshold but not less than the second score threshold, and the third maintenance priority when the comprehensive score is less than the second score threshold.

[0106] The above evaluation process satisfies:

[0107] ,

[0108] ,

[0109] ,

[0110] ;

[0111] in, Represents a node In the The node temporal latent vector at a uniform sampling time. Indicates the number of any node. This represents a gated loop unit with a hidden dimension of 64. Represents a node from to The node feature sequence, Indicates the first A unified sampling time node To node The type is The edge attention coefficient, Indicates the starting node number of the edge. Indicates the terminal node number of the edge. Indicates the edge type. The type is The attention parameter vector, where T represents the vector transpose. Represents a node The node-time latent vector, Represents a node The node-time latent vector, The type is Edge type embedding, Represents a node To node The type is edge weights, This represents vector concatenation. This indicates that normalization is performed based on the same target node and the same edge type. Represents a node The node state vector, This indicates a modified linear unit activation function. Indicates that through type The edges connect to the nodes The set of adjacent nodes, The type is The linear transformation matrix, Represents a node The calibration state vector, This represents the fully connected fusion matrix of the physical stability calibration layer. Represents a node The corresponding limit equilibrium safety factor, Represents a node The corresponding pile-anchor stress utilization rate Represents a node The corresponding drainage capacity ratio;

[0112] The evaluation model is trained using historical cold and arid slope engineering samples. The training samples include a spatiotemporally aligned dataset of historical cold and arid slope engineering projects, component detection records, slope surface spalling records, deep slippage monitoring records, drainage maintenance records, and maintenance records. Following the same processing methods as steps S3 and S4, a heterogeneous spatiotemporal map and a training weight matrix are generated. Component detection records generate component status classification labels, slope surface spalling records generate slope surface spalling risk regression labels, deep slippage monitoring records generate deep slippage trend regression labels, drainage maintenance records generate drainage category classification labels, and maintenance records generate maintenance priority classification labels. The training objective is composed of a weighted average of component status classification loss, slope surface spalling regression loss, deep slippage trend regression loss, drainage category classification loss, maintenance priority classification loss, edge weight smoothing loss, and physical residual consistency loss, with the corresponding weights sequentially selected. The classification loss uses cross-entropy loss, the regression loss uses mean absolute error loss, the edge weight smoothing loss constrains the change in the same edge weight at adjacent unified sampling times, and the physical residual consistency loss constrains the output of the calibration state vector to be consistent with the change direction of the physical residual vector. The number of training rounds is 200, the batch size is 32, the learning rate is 0.001, and the optimizer uses an adaptive moment estimation optimizer. After training, the parameters of the temporal coding layer, graph attention propagation layer, physical stability calibration layer, and multi-task output layer are fixed, and the component service status, slope surface spalling risk, deep slippage trend, drainage failure category, and comprehensive maintenance priority are output at each unified sampling time.

[0113] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0114] This invention utilizes a continuous processing relationship between a three-dimensional digital twin model, a spatiotemporally aligned dataset, cold and drought degradation characteristics, heterogeneous spatiotemporal maps, and an evaluation model. This enables the state changes of anchor bolts, anti-slide piles, lattice beams, drainage structures, and monitoring points to correspond with freeze-thaw cycles, wind erosion, salt erosion, drainage, and deep displacement responses. This addresses the problem of continuously monitoring the state of slope reinforcement in cold and drought-prone areas by outputting state, risk, and priority results that can be used for maintenance assessment.

[0115] The combined module of cold and drought degradation and physical residuals uses freeze-thaw cycles, moisture content, salinity, electrical conductivity, wind erosion, drainage flow attenuation, anchor axial force, anti-slide pile displacement, and lattice beam crack changes for coefficient and residual calculation. Based on this, the weights of the force transmission edge, seepage drainage edge, and deep slope coupling edge are dynamically adjusted to make the evaluation model more suitable for distinguishing between seasonal reversible deformation and irreversible reinforcement damage.

Claims

1. A digital twin-based assessment method for reinforcing cold and arid slopes in Xinjiang, characterized in that, include: S1. Establish a three-dimensional digital twin model of the cold and arid slope, divide the slope into slope units, and register the spatial locations, component attributes, and connection relationships of anchor bolts, anti-slide piles, lattice beams, drainage structures, and monitoring points in the three-dimensional digital twin model to obtain a twin foundation object set; S2. Collect multi-source continuous data corresponding to the twin foundation object set according to a unified timestamp, and perform coordinate registration and time alignment on the multi-source continuous data to obtain a spatiotemporally aligned dataset; S3. Extract freeze-thaw degradation features, wind erosion features, salt erosion features, and drainage failure features from the spatiotemporally aligned dataset, and generate a slope mechanical degradation system from the features. S4. Based on the twin-based object set and the features, construct a heterogeneous spatiotemporal graph, and update the weight matrices of the force transmission edge, seepage drainage edge, and deep coupling edge of the slope according to the slope mechanical deterioration coefficient, slope surface protection reduction coefficient, drainage failure coefficient, and physical residual vector; S5. Input the updated heterogeneous spatiotemporal graph into the evaluation model, and the evaluation model performs time-series encoding, graph attention propagation based on the weight matrix, and physical stability calibration, and outputs the component service status, slope surface spalling risk, deep slippage trend, drainage failure category, and comprehensive maintenance priority.

2. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 1, characterized in that, S1 includes: establishing a three-dimensional coordinate system based on slope design data, topographic survey data, and geological survey data; generating the slope unit using soil-rock layer boundaries, slope grid partitioning, and monitoring profiles as segmentation conditions; registering the anchor length, inclination angle, axial force design value, and anchorage section location as anchor node attributes; registering the anti-slide pile location, pile length, cross-sectional dimensions, and pile top displacement design limit as anti-slide pile node attributes; registering the grid location, cross-sectional dimensions, and crack monitoring location of the grid beam as grid beam node attributes; and registering the ditch direction, cross-sectional dimensions, and water collection unit of the drainage structure as drainage structure node attributes; and generating the connection relationship according to spatial adjacency, component connection relationship, and drainage connectivity relationship.

3. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 1, characterized in that, In step S2, the multi-source continuous data includes anchor bolt axial force, anchor bolt displacement, anti-slide pile top displacement, anti-slide pile body strain, lattice beam crack width, drainage structure flow rate, drainage structure water level, slope temperature, slope moisture content, pore water pressure, salt content, electrical conductivity, wind speed, snowfall, slope surface point cloud elevation, and deep inclinometer displacement; the coordinate registration includes mapping the slope surface point cloud elevation, deep inclinometer displacement, and coordinates of each monitoring point to the three-dimensional coordinate system; the time alignment includes resampling each data channel according to a preset sampling period, and forming the spatiotemporal aligned dataset using the mean, peak value, valley value, and rate of change within the same time window.

4. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 1, characterized in that, S3 includes: generating the freeze-thaw degradation characteristics based on the number of times the slope temperature crosses the freezing point, the duration of freezing, the duration of thawing, and the peak moisture content; generating the wind erosion characteristics based on the slope surface point cloud elevation difference, cumulative wind speed, and the proportion of exposed slope surface area; generating the salt erosion characteristics based on changes in salt content, conductivity, and pore water pressure; generating the drainage failure characteristics based on the flow attenuation rate of the intercepting and drainage structure, water retention time, and the proportion of ditch siltation; and inputting the freeze-thaw degradation characteristics, the salt erosion characteristics, and the deep displacement increment into historical calibration samples for fitting. The first weight table is used to obtain the slope mechanical deterioration coefficient; the wind erosion characteristics and the change in lattice beam cracks are input into the second weight table obtained by fitting historical calibration samples to obtain the slope surface protection reduction coefficient; the drainage failure characteristics and the slope moisture content increment are input into the third weight table obtained by fitting historical calibration samples to obtain the drainage failure coefficient; the physical residual vector is composed of the difference between the monitored displacement and the displacement calculated by physical stability, the difference between the monitored drainage flow rate and the hydraulically calculated flow rate, and the difference between the monitored component internal force and the structural calculated internal force.

5. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 1, characterized in that, In step S4, the heterogeneous spatiotemporal diagram includes slope unit nodes, anchor nodes, anti-slide pile nodes, lattice beam nodes, drainage structure nodes, and monitoring point nodes; the force transmission edge connects the slope unit nodes with the anchor nodes, anti-slide pile nodes, and lattice beam nodes; the seepage drainage edge connects the slope unit nodes with the drainage structure nodes; and the deep coupling edge of the slope surface connects the slope surface monitoring point nodes, slope unit nodes, and deep displacement monitoring point nodes; the initial edge weights are determined by the spatial distance between nodes, the design connection stiffness, and the hydraulic connection direction; the slope mechanical degradation coefficient, the slope surface protection reduction coefficient, and the drainage failure coefficient are mapped to a closed interval of 0 to 1, and the force transmission edge, the seepage drainage edge, and the deep coupling edge of the slope surface are corrected respectively by combining the normalized components of the physical residual vector to obtain the updated weight matrix.

6. The method for evaluating the reinforcement of cold and arid slopes in Xinjiang based on digital twins according to claim 1, characterized in that, In step S5, the evaluation model includes a temporal encoding layer, a graph attention propagation layer, a physical stability calibration layer, and a multi-task output layer. The temporal encoding layer generates node temporal latent vectors based on the node feature sequence within a sliding time window. The graph attention propagation layer concatenates the temporal latent vectors of adjacent nodes, edge type embeddings, and the weight matrix to calculate edge attention coefficients, and normalizes the edge attention coefficients according to edge type to obtain node state vectors. The physical stability calibration layer generates calibration vectors based on the limit equilibrium safety factor, pile-anchor force utilization rate, and drainage capacity ratio, and fuses the calibration vectors with the node state vectors to obtain a calibration state vector. The multi-task output layer outputs the component service status, slope surface spalling risk, deep slippage trend, drainage failure category, and comprehensive maintenance priority from the calibration state vectors.

7. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 6, characterized in that, The physical stability calibration layer also generates deformation attribute markers, which are determined as follows: within m consecutive sliding time windows, the displacement fallback ratio of deep displacement after the freeze-thaw temperature recovery is calculated, and the duration of the residual axial force of anchor bolts, the residual displacement of anti-slide piles, and the residual crack change of lattice beams is calculated, where m is a preset positive integer; when the displacement fallback ratio is not less than a preset recovery threshold and the duration is less than a preset duration threshold, the corresponding change is marked as seasonal reversible deformation; When the displacement fallback ratio is less than a preset recovery threshold and the duration is not less than a preset duration threshold, the corresponding change will be marked as irreversible reinforcement damage. The deformation attribute markers are used as the weight correction input for the deep coupling edge of the slope surface and the calculation input for the comprehensive maintenance priority.

8. The assessment method for strengthening cold and arid slopes in Xinjiang based on digital twins according to claim 7, characterized in that, The comprehensive maintenance priority is determined as follows: the service status of the component is converted into a service status code, the slope surface spalling risk is converted into a spalling risk value, the deep slippage trend is converted into a slippage trend value, and the drainage failure category is converted into a drainage category code; the service status code, the spalling risk value, the slippage trend value, the drainage category code, the norm of the physical residual vector, and the deformation attribute label are input into a priority weight table to obtain a comprehensive score; when the comprehensive score is not less than a first score threshold, a first maintenance priority is output; when the comprehensive score is less than the first score threshold but not less than a second score threshold, a second maintenance priority is output; when the comprehensive score is less than the second score threshold, a third maintenance priority is output; the first score threshold is greater than the second score threshold, and the first score threshold and the second score threshold are determined by the score percentile values ​​of historical maintenance samples.

9. The method for evaluating the reinforcement of cold and arid slopes in Xinjiang based on digital twins according to claim 6, characterized in that, The evaluation model is trained as follows: A spatiotemporally aligned dataset of historical cold and arid slope engineering projects, component detection records, slope surface spalling records, deep slippage monitoring records, drainage maintenance records, and maintenance records are acquired; a heterogeneous spatiotemporal map and a weight matrix for training are generated using the same processing methods as steps S3 and S4; training labels are generated using the component detection records, slope surface spalling records, deep slippage monitoring records, drainage maintenance records, and maintenance records. The evaluation model is obtained by using the weighted sum of component state classification loss, slope surface spalling regression loss, deep slippage trend regression loss, drainage category classification loss, maintenance priority classification loss, edge weight smoothing loss, and physical residual consistency loss as the training objective.

10. The method for evaluating the reinforcement of cold and arid slopes in Xinjiang based on digital twins according to claim 3, characterized in that, After forming the spatiotemporal aligned dataset, the data credibility is calculated for each data channel. The data credibility is obtained by normalizing and weighting the online duration ratio, the deviation from the median value of adjacent monitoring points, and the deviation from the physical model prediction value. When the data credibility is less than a preset credibility threshold, the node features of the corresponding data channel are replaced with the fusion value of the weighted interpolation of adjacent nodes and the physical model prediction value, and the edge weights related to the corresponding data channel are reduced. The replaced node features and the adjusted edge weights are then input into the evaluation model.

Images