Slope digital twin self-updating and early warning method based on satellite InSAR and ground displacement joint constraint

Abstract

The present application relates to the field of geological disaster monitoring and warning and geotechnical engineering informatization, and discloses a kind of satellite InSAR and ground displacement joint restraint's slope digital twinborn self-update and early warning method, solve the existing slope monitoring technology There are problems such as difficult multi-source data fusion, stability analysis cannot dynamically update parameters and sliding surface geometry, lack of digital twinborn model self-update capability and single early warning index.The present application fuses satellite InSAR surface deformation, ground GNSS point deformation, deep profile deformation data under unified space-time and coordinate datum, realizes the real-time self-update of model by constructing twinborn model augmented state containing geotechnical parameters and sliding surface geometric parameters, combined with hierarchical consistency constraint objective function, then fuses static stability index of limit equilibrium method and dynamic deformation index of proxy model, forms comprehensive hierarchical early warning system considering static stability and dynamic deformation characteristics of slope.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning and geotechnical engineering informatization, specifically to a method for self-updating and early warning of slope digital twins constrained by satellite InSAR (Interferometric Synthetic Aperture Radar) and ground displacement. Background Technology

[0002] Slope instability is a typical sudden geological disaster in scenarios such as open-pit mines, transportation and water conservancy projects, and reservoir banks. Its root cause lies in the combined effects of factors such as the combination of internal structural surfaces of the slope, softening of the soil and rock mass, and changes in pore water pressure, which trigger the formation of potential sliding zones and the evolution of sliding surfaces. If the accelerated deformation process and the formation of deep shear zones cannot be identified in time, it can easily lead to slope collapse or overall landslides, causing significant safety and economic losses. Therefore, the engineering field urgently needs technical means to monitor slope deformation in real time, dynamically assess stability, and automatically update analysis results.

[0003] Currently, slope deformation three-dimensional monitoring and stability analysis technologies are mainly divided into three categories:

[0004] The first category is non-contact deformation monitoring technologies, such as UAV-borne radar LiDAR and synthetic aperture radar interferometry (InSAR). UAV-borne radar technology can identify cracks in vegetation-covered slopes, but its use is limited by airspace conditions and requires regular on-site operations, resulting in low efficiency and high cost. InSAR technology can acquire long-term slope surface deformation data and is suitable for large-scale monitoring; however, under complex terrain and climatic conditions, it may be affected by atmospheric delay and terrain distortion, leading to data gaps and uncertainties, impacting the accuracy and reliability of its engineering applications.

[0005] The second category is contact deformation monitoring technologies, such as ground-based GNSS monitoring stations and deep deformation monitoring instruments. Among them, ground-based GNSS provides high-precision three-dimensional point displacement, which is suitable for a small number of monitoring points, but cannot fully reflect the areas of accelerated deformation on the slope surface. Deep deformation monitoring can provide deep displacement data, but its spatial coverage is limited, and it cannot directly monitor surface deformation dynamics.

[0006] The third category is traditional stability analysis techniques. The limit equilibrium method (LEM) is one of the most widely used methods. It has high computational efficiency and can output the slope safety factor (FoS) and the location of potential slip surfaces. However, this method relies on manual calculation and cannot automatically update the soil and rock parameters and slip surface geometry based on real-time monitoring data. It suffers from problems such as analysis lag and untimely evaluation.

[0007] To overcome the limitations of single monitoring technologies, existing technologies have begun to combine non-contact and contact monitoring methods with stability analysis. For example, time-series InSAR technology is used to acquire large-scale deformation data of the slope surface, and high-precision data from a small number of GNSS monitoring points are used to calibrate the InSAR monitoring results, reducing the error of the InSAR data. Then, the limit equilibrium method is used to calculate the slope safety factor based on the fixed soil and rock parameters obtained from previous engineering surveys, thus combining slope deformation monitoring with stability assessment. However, this approach still has not broken through the bottlenecks of traditional technologies. It is still in an offline, manually iterative operation state, unable to achieve real-time assimilation and fusion of multi-source monitoring data. Soil and rock parameters and slip surface geometry still rely on manual settings and cannot be adaptively updated with dynamic slope deformation data. Furthermore, a digital twin model of the slope is not constructed, naturally lacking the model's continuous self-updating capability. At the same time, this approach can only output static safety factor results and cannot combine dynamic slope deformation characteristics to achieve real-time early warning. It is difficult to cope with the actual engineering situation of continuously changing slope deformation and cannot meet the core needs of real-time monitoring, dynamic assessment, and timely early warning of slope stability on site.

[0008] Furthermore, the current multi-source data fusion of satellite InSAR surface deformation, ground GNSS point deformation, and deep slip point deformation is still in the stage of separate back-end processing and calibration. There is a lack of a unified fusion framework to achieve collaborative analysis of the three types of data. In addition, there are problems such as difficulty in spatiotemporal alignment and unreasonable weight allocation during the data fusion process, which limits the accuracy and reliability of the overall slope monitoring results. Moreover, the existing slope monitoring and stability assessment technology requires manual operation and analysis by professionals throughout the process, with low automation, poor field applicability, and inability to achieve intelligent real-time early warning management. Summary of the Invention

[0009] The technical problem to be solved by this invention is to provide a method for self-updating and early warning of digital twins of slopes with joint constraints of satellite InSAR and ground displacement, which solves the problems of existing slope monitoring technology, such as difficulty in multi-source data fusion, inability to dynamically update parameters and slip surface geometry in stability analysis, lack of self-updating capability of digital twin models, and single early warning indicators.

[0010] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0011] In a first aspect, the present invention provides a method for self-updating and early warning of slope digital twins constrained by satellite InSAR and ground displacement, comprising the following steps:

[0012] S1. Acquire multi-source monitoring data of slope deformation, including surface line-of-sight deformation data obtained by satellite InSAR, point deformation data obtained by ground GNSS, and deep displacement profile deformation data obtained by embedded monitoring devices.

[0013] S2. Perform spatiotemporal alignment and coordinate benchmark unification processing on the multi-source monitoring data to form an equivalent observation set;

[0014] S3. Construct an initial digital twin model of the slope, parameterize the updatable objects of the model, and form an augmented twin model state containing inversion parameter vectors and slip surface geometric dimension parameter vectors;

[0015] S4. Construct a multi-source monitoring mapping operator and error weight system, establish a unified monitoring equation, and map the deformation field output by the twin model into a form that matches each monitoring data to obtain the equivalent covariance matrix and corresponding weight matrix of InSAR, GNSS, and deep monitoring.

[0016] S5. Construct a hierarchical consistency constraint objective function and perform joint inversion solution to update the inversion parameter vector and slip surface geometric dimension parameter vector of the twin model, thereby realizing the self-updating of the slope digital twin model;

[0017] S6. Based on the updated twin model, the slope safety factor is calculated using the limit equilibrium method in combination with external working conditions. The surface and deep deformation fields of the slope are output using the surrogate model, and the maximum deformation rate and maximum acceleration are obtained. Comprehensive graded early warning is carried out based on the safety factor, maximum deformation rate, and maximum acceleration.

[0018] In this scheme, by constructing an augmented state of a twin model and performing joint inversion, adaptive updates of slope soil and rock parameters and sliding surface geometry are achieved, solving the problems of fixed parameters and analysis lag in traditional stability analysis. By combining the safety factor of the limit equilibrium method with the deformation rate and acceleration of the surrogate model, a multi-indicator comprehensive early warning is achieved, avoiding the one-sidedness of a single early warning indicator and significantly improving the comprehensiveness of slope monitoring, the dynamism of stability assessment, and the accuracy of early warning.

[0019] Furthermore, in step S2, spatiotemporal alignment of the multi-source monitoring data is performed, including:

[0020] With the first The next update time Based on the time window The multi-source monitoring data is resampled to form an equivalent monitoring vector aligned with the time series:

[0021] ;

[0022] in, Indicates alignment to The equivalent monitoring vector at any given time; Indicates the width of the time window; Indicates the monitoring type index. Indicates the InSAR monitoring type. Indicates the type of GNSS monitoring. Indicates the type of deep monitoring; Representation type The original monitoring vector sequence; Indicates the time alignment operator;

[0023] At the same time, a time decay weight is introduced:

[0024] ;

[0025] in, Indicates the time decay weight; Represents the time decay constant; Indicates the sampling time of the original monitoring data;

[0026] Finally, the equivalent monitoring vector of the time series alignment is multiplied by the time decay weight to obtain the final spatiotemporal alignment result.

[0027] In this scheme, the differences in sampling frequency and sampling time of multi-source monitoring data are eliminated by spatiotemporal alignment, which improves the comparability and fusion accuracy of multi-source data. At the same time, by introducing time decay weight, the influence of outdated monitoring data is reduced, and the latest slope deformation data accounts for a higher proportion in the model update, making the model update more consistent with the real-time deformation state of the slope.

[0028] Furthermore, in step S3, the augmented state of the twin model is represented as follows:

[0029] ;

[0030] in, for The augmented state of the slope twin model at the update time; for Update the time-inversion parameter vector; for Update the surface geometry parameter vector at each time step;

[0031] The inversion parameter vector should include at least the cohesion, internal friction angle, and pore water pressure of each layer of soil and rock on the slope.

[0032] The augmented state of the twin model also pre-defines the upper and lower numerical boundaries of the inversion parameters, the prior value vector, the prior covariance matrix, and the feasible range of the surface geometry.

[0033] This solution clarifies the core updatable parameters of the twin model, making model updates more targeted. At the same time, by presetting parameter boundaries and feasible regions, it avoids unconstrained drift during parameter inversion, ensuring the engineering rationality and scientific validity of the updated model.

[0034] Furthermore, in step S4, the unified monitoring equation is expressed as:

[0035] ;

[0036] in, Indicates monitoring type exist Update the monitoring vector at each time step; Representation type Mapping operators; Indicates monitoring error;

[0037] The process of mapping the deformation field output by the twin model into a form that matches each monitoring data point includes:

[0038] InSAR mapping: Project the surface displacement vector of the twin model into the InSAR line-of-sight displacement, considering the east-west, south-north, and vertical components, and combine the InSAR deformation LOS unit vector calculation model to predict the LOS displacement.

[0039] GNSS mapping: The displacement field of the twin model is sampled at the GNSS monitoring points, and the GNSS reference zero bias parameter is introduced to match the GNSS three-dimensional point deformation data;

[0040] Deep monitoring mapping: Construct a relative displacement mapping along the depth direction to convert the deep displacement field of the twin model into the relative displacement of the profile and match the deep monitoring data.

[0041] This solution achieves accurate mapping between the deformation field of the twin model and different types of monitoring data, enabling the model output to be effectively compared with the actual monitoring data.

[0042] Furthermore, in step S4, the equivalent covariance matrices of InSAR, GNSS, and deep monitoring are obtained by superimposing the monitoring error covariance matrices of InSAR, GNSS, and deep monitoring with their respective model error covariance matrices.

[0043] Among them, the monitoring error covariance matrix of InSAR is composed of the pixel standard deviation, which is calculated based on the pixel coherence; the monitoring error covariance matrix of GNSS is composed of the monitoring variance / covariance of each monitoring point in the east-west, north-south, and vertical directions, which is directly calculated from the monitoring data; the monitoring error covariance matrix of deep monitoring is calculated by combining the borehole standard deviation with the error growth coefficient along the depth.

[0044] In this scheme, corresponding equivalent covariance calculation methods are designed for the characteristics of InSAR, GNSS and deep monitoring, so that the weight allocation of various monitoring data is more in line with their actual monitoring accuracy, and further improves the rationality of multi-source data fusion and the accuracy of model updates.

[0045] Furthermore, in step S5, the objective function of the hierarchical consistency constraint is expressed as:

[0046]

[0047] in, For the first The objective function value of the hierarchical consistency constraint at the next update time; To unify monitoring vectors; To monitor the mapping vector; This is the weight matrix; This represents the error vector of the digital twin model. The inversion parameter vector; This is the initial parameter vector; The parameter is the covariance matrix; For parameter weights; For parameter constraint functions; For the geometric weights of the smooth surface; This is a geometric range constraint function; Weights for the sliding shear band; This is the feedback distance for the sliding shear band.

[0048] In this scheme, multi-dimensional objective function constraints are used to achieve comprehensive control over model updates, ensuring the rationality, accuracy, and matching with deep shear zones of the updated inversion parameters and slip surface geometric parameters, making the twin model more consistent with the actual geological conditions of the slope.

[0049] Furthermore, in step S6, the external working conditions include at least one of rainfall, earthquake, and construction; the surrogate model includes any one of neural network, Gaussian process, random forest, and support vector regression.

[0050] This solution can adapt to the external working conditions of different engineering scenarios, and provides a variety of proxy models to choose from. It can be flexibly configured according to the actual needs of the project and data conditions, which improves the engineering practicality and flexibility of the method.

[0051] Furthermore, in step S6, the comprehensive hierarchical early warning includes:

[0052] The slope condition is divided into four levels: Level I safe, Level II low risk, Level III medium risk, and Level IV high risk. If the conditions of any higher risk level are met, the slope will be classified into the corresponding risk level.

[0053] Different monitoring, updating, and response strategies are adopted for different warning levels: Level I is safe and the current monitoring and updating cycle is maintained; Level II is low risk and the monitoring and updating cycle is shortened; Level III is medium risk and updates are intensified with a focus on key monitoring areas; Level IV is high risk and warning information is output and emergency mode is activated.

[0054] This plan classifies different risk levels based on the slope condition and designs differentiated monitoring and response strategies for different warning levels. Regular monitoring is maintained for safe conditions, while emergency mode is immediately activated for high-risk conditions, making the monitoring work more targeted. This approach can effectively prevent and control the risk of slope instability while controlling monitoring costs and improving the actual operability on the engineering site.

[0055] Secondly, the present invention also provides a computer device, including a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described above.

[0056] Thirdly, the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described above.

[0057] The beneficial effects of this invention are:

[0058] (1) The multi-source monitoring information is richer and the fusion reliability is higher:

[0059] By comprehensively integrating satellite InSAR surface deformation, ground GNSS local point deformation, and deep profile deformation data, three-dimensional deformation monitoring of slopes from surface to depth and from surface field to local area has been achieved, breaking through the limitations of single monitoring technology. Through spatiotemporal alignment, coordinate unification, and consistency constraints of surface-point-depth data, the problem of multi-source data fusion has been effectively solved, significantly improving the degree of collaborative fusion evaluation of monitoring data and making the monitoring results more comprehensive and accurate.

[0060] (2) Digital twin models have adaptive and self-updating capabilities:

[0061] The core parameters of the soil and rock mass and the geometric parameters of the slip surface are constructed into an augmented state of the digital twin model. The joint inversion update is achieved through the hierarchical consistency constraint objective function, which enables the model to dynamically adjust the parameters and slip surface geometry according to real-time monitoring data. This solves the problems of fixed parameters and lagging analysis in traditional models, and ensures that the digital twin model always keeps in line with the actual state of the slope.

[0062] (3) More accurate stability assessment and early warning:

[0063] By combining the safety factor (static stability index) of the limit equilibrium method with the deformation rate and acceleration (dynamic deformation index) of the surrogate model, a multi-indicator comprehensive graded early warning is carried out, avoiding the one-sidedness and omission problems of single-indicator early warning; by adopting the conservative principle of "taking the higher value and not the lower value", the reliability of the early warning is greatly improved, and the potential instability risk of slope can be identified in a timely manner.

[0064] (4) Adaptive monitoring and scheduling to reduce costs:

[0065] It can utilize publicly available / commercial satellite data sources to achieve short-cycle area monitoring, and combine refined data from ground GNSS and deep monitoring to achieve adaptive scheduling of monitoring frequencies. In the event of emergencies, the monitoring frequency can be increased to improve monitoring efficiency. At the same time, the warning level is linked with the monitoring and response strategy to rationally allocate monitoring resources, reduce monitoring costs while ensuring safety, and the method has a high degree of automation, requiring no large amount of manual intervention and strong field applicability. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the overall process of the slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement in this embodiment of the invention.

[0067] Figure 2 This is a schematic diagram of the displacement field and stability output of the updated twin model in an embodiment of the present invention. Detailed Implementation

[0068] This invention aims to provide a self-updating and early warning method for slope digital twins based on joint constraints of satellite InSAR and ground displacement, addressing the problems of existing slope monitoring technologies, such as difficulties in multi-source data fusion, inability to dynamically update parameters and slip surface geometry in stability analysis, lack of self-updating capability of digital twin models, and limited early warning indicators. Its core idea is: based on joint constraints of multi-source monitoring data, with adaptive self-updating of the slope digital twin model as the core, and with comprehensive hierarchical early warning based on multiple indicators as the goal, it fuses satellite InSAR planar deformation, ground GNSS point deformation, and deep profile deformation data under a unified spatiotemporal and coordinate reference. By constructing an augmented state of the twin model containing soil and rock mass parameters and slip surface geometric parameters, and combining it with a hierarchical consistency constraint objective function, the model achieves real-time self-updating. Furthermore, it integrates the static stability indicators of the limit equilibrium method and the dynamic deformation indicators of the surrogate model to form a comprehensive hierarchical early warning system that considers both the static stability and dynamic deformation characteristics of the slope, ultimately realizing real-time monitoring, dynamic stability assessment, and intelligent early warning management of slopes. The present invention is applicable to the long-term monitoring, stability assessment and early warning management of medium and large slopes with periodic remote sensing and on-site monitoring conditions, such as open-pit mine slopes, transportation and water conservancy project slopes, reservoir banks and landslide bodies.

[0069] Example:

[0070] This embodiment provides a method for self-updating and early warning of slope digital twins constrained by satellite InSAR and ground displacement. (See also...) Figure 1 It includes the following implementation process:

[0071] S1. Acquisition of multi-source monitoring data:

[0072] This step acquires three-dimensional deformation monitoring data of the slope from the surface to the depths and from the surface field to the local area, providing basic observation data for the construction and updating of the digital twin model, thus overcoming the information limitations of single monitoring technology.

[0073] In one exemplary implementation, at least three types of monitoring data are acquired, including surface line-of-sight deformation obtained by satellite InSAR, point deformation obtained by ground GNSS, and deep displacement profile deformation monitoring obtained by embedded monitoring devices. Specific details are as follows:

[0074] The first type of data, InSAR deformation, needs to be obtained based on SAR satellite data and using typical time-series InSAR methods. Ultimately, the line-of-sight deformation values ​​for all sampling points within the slope area can be obtained.

[0075] ;

[0076] in, Indicates the first One sampling point; Indicates the first The time corresponding to the SAR satellite data; Indicates sampling point exist The deformation value along the line of sight at any given time; This indicates the number of sampling points involved in constraining the InSAR deformation mass of the slope.

[0077] The first type of data obtained above is used to characterize the overall planar deformation features of the slope surface.

[0078] The second type of data is deformation monitoring data from GNSS monitoring points on the slope surface:

[0079] ;

[0080] in, Indicates the number of GNSS monitoring points. Indicates the first The deformation sequence of GNSS monitoring points.

[0081] The second type of data obtained above is used to characterize the high-precision deformation features of local points on the slope.

[0082] The third type of data is obtained using side-mounted inclinometer holes or deformation holes. Obtain deep profile deformation monitoring data:

[0083] ;

[0084] in, Indicates the first depth sampling nodes ( (Indicates the location of the deep monitoring borehole). Indicates deep monitoring well In depth Relative deformation monitoring records; This indicates the number of sampling nodes along the depth direction.

[0085] The third type of data obtained above is used to characterize the deformation features of the deep slope along the profile.

[0086] Based on the above methods, satellite InSAR can achieve large-scale area monitoring, making up for the problem of insufficient coverage of ground monitoring points; ground GNSS can achieve high-precision monitoring of local points, providing benchmark calibration for InSAR data; deep deformation monitoring can achieve deep profile monitoring, capturing the deformation characteristics of potential shear zones of slopes. The combination of these three types of data forms a three-dimensional monitoring system of surface-deep and area-point, which can comprehensively reflect the overall deformation state of the slope and provide data support for the accurate construction of models.

[0087] S2. Data spatiotemporal alignment and coordinate benchmark unification:

[0088] This step eliminates the inherent differences in sampling frequency, sampling time, and coordinate reference of multi-source monitoring data, forming an equivalent observation set that can be used for joint constraints. This ensures the comparability and reasonableness of multi-source data fusion, laying the foundation for subsequent model mapping and joint inversion.

[0089] In one exemplary implementation, coordinate benchmark unification and spatiotemporal alignment are achieved as follows:

[0090] Unified coordinate reference: InSAR pixel coordinates, GNSS point coordinates and deep borehole coordinates are uniformly transformed to the same engineering coordinate system (such as the local ENU coordinate system) to ensure the consistency and comparability of the spatial locations of each observation.

[0091] Time alignment: based on the time of the k-th update Based on the time window The above monitoring data is resampled to form equivalent monitoring data aligned with the time series:

[0092] ;

[0093] Simultaneously, a time decay weight can be introduced to enhance the contribution of the latest monitoring sequence:

[0094] ;

[0095] in, Indicates the first Next update time. Indicates the width of the time window. Indicates the monitoring type index ( Indicates the InSAR monitoring type. Indicates the type of GNSS monitoring. (Indicates the type of deep monitoring); Representation type The original monitoring vector sequence, Indicates alignment to The equivalent monitoring vector at any given time; This indicates the time alignment operator (which can be obtained through interpolation or weighted averaging). Indicates the time decay weight. This represents the time decay constant.

[0096] Finally, the equivalent monitoring vector aligned to the time series is multiplied by the time decay weight to obtain the final spatiotemporal alignment result.

[0097] This scheme reduces the impact of outdated monitoring data by introducing time decay weights, allowing the latest slope deformation data to account for a higher proportion in model updates, thus making the model updates more closely reflect the real-time deformation state of the slope.

[0098] S3. Twin model initialization and augmented state construction:

[0099] This step establishes an initial digital twin model that matches the actual geological and topographical features of the slope, parameterizes the updatable objects in the model, and forms an augmented state of the twin model that can be dynamically updated, providing a basic framework for subsequent updates of the model.

[0100] In one exemplary implementation, based on the actual topography, geological structure, boundary conditions, and engineering zoning information of the slope, a twin geometric model and a zoning model of the slope are constructed to reconstruct the physical morphology of the slope; the updatable objects of the model are parameterized to construct an augmented twin model state containing inversion parameter variables and slip surface geometric parameter variables, the expression of which is:

[0101] ;

[0102] in, for The augmented state of the slope twin model at the update time (including inversion parameters, slip surface geometry information, etc.); for The updated inversion parameter vector must include at least the soil and rock strength parameters (cohesion) of each layer of the slope. internal friction angle ), pore pressure wait; for The vector of sliding surface geometry parameters at each update time is used to characterize the geometry of the potential sliding surface or shear band.

[0103] If the slope is layered and partitioned, the slope inversion parameter vector can be expanded according to the partition, and the inversion parameter vector is illustrated as follows:

[0104] ;

[0105] Based on geotechnical test data or by referring to existing records, determine the upper and lower numerical boundaries of relevant parameters. Prior value vector and the prior covariance matrix ; Represents the GNSS reference zero bias vector, and the feasible region for predicting the sliding surface geometry. .

[0106] This approach, by pre-setting parameter boundaries and feasible regions, can avoid unreasonable parameters that exceed the physical properties of the soil and rock mass and the actual engineering situation during subsequent inversion processes, thus ensuring the scientific validity of the model.

[0107] S4. Model Mapping and Weight Construction:

[0108] This step aims to achieve accurate adaptation between the twin model deformation field and multi-source monitoring data, construct an error weight system that reflects the actual reliability of each monitoring type, and enable multi-source monitoring data to participate in model updates with reasonable weights, thereby improving the rationality of joint constraints.

[0109] In one exemplary implementation, a unified monitoring equation is established to map the deformation field output by the slope twin model onto different types of monitoring data, mapping the twin displacement to InSAR line-of-sight displacement; sampling the twin displacement at GNSS points and introducing a baseline zero bias; and mapping the twin deep displacement to profile relative displacement. Then, based on quality indicators and uncertainties, an observation covariance and weight matrix are constructed, and the model error covariance is incorporated to form an equivalent covariance.

[0110] The unified monitoring equation expression is:

[0111] ;

[0112] in, Indicates monitoring type exist Update the monitoring vector at each time step; Representation type The mapping operator, These represent the monitoring mapping operators for InSAR, GNSS, and deep deformation, respectively. The slope twin model integrates and weights the observations from different data sources through the mapping operators, so that the contributions of all data sources can be reflected in the slope stability and deformation prediction. This indicates the monitoring error.

[0113] For InSAR monitoring, the monitored deformation needs to be projected and mapped to a twin model deformation. This deformation considers east-west, south-north, and vertical vectors. The InSAR deformation LOS unit vector is defined as follows: Surface displacement vector based on slope twin model It can calculate the LOS displacement using a digital twin model:

[0114] ;

[0115] in, Indicates the first Update time next Predict displacement at each sampling point; , and This represents the components of the unit vector in the LOS direction along the east-west, north-south, and vertical directions; , and This indicates the components of the predicted displacement along the east-west, north-south, and vertical directions.

[0116] For GNSS monitoring, samples were taken from the slope twin model at the GNSS location, and GNSS baseline zero bias parameters were introduced. With monitoring error ;

[0117] ;

[0118] For deep deformation monitoring, a relative displacement mapping along the depth direction is constructed to obtain the relative displacement of the model-predicted profile. ;

[0119] ;

[0120] To achieve comparability and computability of joint constraints for multi-source monitoring data, monitoring error covariance matrices are constructed for each type of monitoring data. These monitoring error covariance matrices include at least the InSAR monitoring covariance matrix. GNSS monitoring covariance matrix and deep monitoring covariance matrix Furthermore, the model error covariance matrix is ​​superimposed onto the monitoring covariance matrix to obtain the equivalent covariance matrix.

[0121] Among them, the InSAR monitoring covariance matrix Composed of the standard deviation of InSAR pixels within the slope surface monitoring range:

[0122] ;

[0123] In the above formula, the InSAR pixel standard deviation is calculated as follows:

[0124] ;

[0125] in, Represents a pixel The monitoring standard deviation of InSAR, Represents a pixel Location InSAR coherence, and These represent the upper and lower limits of the standard deviation, respectively.

[0126] Assume the first The standard deviation of the model error for each pixel is Then the InSAR model error equation matrix for:

[0127] ;

[0128] in, For pixels The standard deviation of the InSAR model error.

[0129] InSAR monitoring equivalent covariance matrix It can be obtained from formulas (10) and (12), in the following form:

[0130] (15);

[0131] Similarly, the equivalent covariance matrix of GNSS monitoring The calculation formula is as follows:

[0132] ;

[0133] in, The overall GNSS monitoring covariance matrix; The error covariance matrix of the GNSS model;

[0134] Let be the monitoring covariance matrix of the Nth GNSS point, and its calculation formula is as follows:

[0135] ;

[0136] in, , and The first The monitoring variance of each GNSS point in the east-west, north-south, and vertical directions is calculated from the monitoring data; , and The first The monitoring covariance of each GNSS point in the east-west, north-south, and vertical directions can be estimated.

[0137] Similarly, the equivalent covariance matrix for deep monitoring The calculation formula is as follows:

[0138] ;

[0139] in, and These are the deep monitoring covariance matrix and the model error matrix, respectively. and They are respectively Standard deviation and model error of deep-sea monitoring at depth; among which It can be approximately calculated from the standard deviation of the borehole at the deep monitoring point. The calculation formula is as follows:

[0140] ;

[0141] in, The standard deviation of the borehole at the deep monitoring point. The error growth factor along the depth can be predicted.

[0142] Finally, based on the constructed equivalent covariance matrices of InSAR monitoring, GNSS monitoring, and deep monitoring... , and Determine the corresponding weight matrix. , , Under the Gaussian error or equivalent least squares principle, monitoring data with lower uncertainty should contribute more. Therefore, the weight of monitoring data decreases as the covariance increases. Missing or low-quality monitoring data can be removed or assigned minimal weights to improve robustness.

[0143] Based on the above methods, the twin model outputs the overall deformation field of the slope. However, the multi-source monitoring data have different forms and dimensions. Without targeted mapping, it is impossible to effectively compare the model output with the monitoring data. The unified monitoring equation realizes the adaptation of the model to multi-source data. Since the monitoring accuracy and reliability of each monitoring type are different, the equivalent covariance matrix can accurately reflect the actual reliability of each monitoring data. The weight system constructed based on this can realize the "weighted fusion" of multi-source data, avoiding the model update deviation caused by fusion with equal weights. Incorporating model errors into the equivalent covariance can compensate for the errors caused by model simplification and sudden changes in working conditions, and improve the robustness of the weight system.

[0144] S5. Digital twin model update for slope:

[0145] This step utilizes the joint constraints of multi-source monitoring data to update the inversion parameters and slip surface geometric parameters of the digital twin model through joint inversion solutions, thereby achieving adaptive self-updating of the digital twin model and ensuring that the model always remains consistent with the actual deformation state and geological characteristics of the slope.

[0146] In one exemplary implementation, the three types of monitoring—InSAR, GNSS, and deep profiling—are first uniformly incorporated into a unified monitoring vector. and monitoring mapping vectors :

[0147] ;

[0148] Secondly, a hierarchical consistency constraint objective function is constructed and a joint inversion solution is performed to obtain the updated twin state. The objective function includes at least a multi-source residual term, a parameter prior term, a regularization term, and a deep shear band feedback term. The multi-source residual term is weighted by equivalent covariance, and the deep shear band feedback term is used to impose constraints on the sliding surface geometry parameters. Under the conditions of satisfying the upper and lower bounds of the parameters and the feasible region of the sliding surface, the optimal solution of the objective function is solved to update the inversion parameters and the sliding surface geometry parameters. The core expression formula of the hierarchical consistency constraint objective function is as follows:

[0149] ;

[0150] The first term is the multi-source monitoring consistency term, which constrains the monitoring deformation from surface to point to depth, and includes a unified monitoring vector. Monitoring mapping vector and overall weight matrix (Assigned by equivalent covariance; when measured covariance is lacking, empirical values ​​or estimations from historical data can be used).

[0151] The second term is the prior constraint term for the inversion parameters, which restricts unconstrained parameter drift and includes the inversion parameter vector. Initial parameter vector And parameter covariance matrix .

[0152] The third term is the inversion parameter constraint regularization term, which constrains the parameters to not exceed upper and lower bounds and includes parameter weights. and parameter constraint functions .

[0153] The fourth term is the geometric regularization term for deep monitoring of the slip surface, which ensures the feasibility of the geometric parameters, including the slip surface geometric weights. and geometric range constraint functions .

[0154] The fifth term is the deep sliding shear band feedback term, which transforms the deep information of the shear band into sliding surface geometric constraints, including the sliding shear band weights. and sliding shear band feedback distance (When the predicted depth of the sliding surface near the hole falls into the shear zone) hour, When the crossing depth deviates from the shear zone interval, take the minimum distance from the predicted failure depth to the boundary of the shear zone interval.

[0155] Based on the above methods, the components of the hierarchical consistency constraint objective function form a multi-dimensional control over model updates: the multi-source residual term ensures the consistency between the model output and the multi-source monitoring data; the parameter prior term restricts unconstrained parameter drift; the regularization term prevents parameters from exceeding reasonable boundaries; and the deep shear zone feedback term integrates the actual shear zone information from deep monitoring into the slip surface geometry update, making the slip surface geometry prediction more closely match the actual potential instability area of ​​the slope. By jointly inverting to solve for the optimal solution, the multi-source monitoring data achieves "reverse constraint" on the model parameters, allowing the model to dynamically adjust according to the actual monitoring data and truly achieve self-updating.

[0156] S6. Update Results and Warning Output:

[0157] In this step, based on the updated digital twin model, dynamic assessment of slope stability and comprehensive hierarchical early warning of multiple indicators are realized, and differentiated monitoring and disposal strategies are formulated according to the early warning level, providing a basis for decision-making for slope safety management at the engineering site.

[0158] In one exemplary implementation, the slope inversion parameter vector is updated. and monitoring surface geometry parameter vector Under certain conditions, based on external working conditions (such as rainfall, earthquakes, and construction), the slope safety factor FoS is calculated using the limit equilibrium method. The surface and deep deformation fields are then output using a surrogate model. The updated displacement field and stability output of the twin model are illustrated below. Figure 2 As shown.

[0159] The general form of the safety factor is as follows:

[0160] ;

[0161] Meanwhile, based on multi-source monitoring deformation data combined with updated parameters and geometry, a slope deformation proxy model is used to output surface deformation and deep deformation. The deformation proxy model takes the following form:

[0162] ;

[0163] in, express Update the slope surface deformation prediction output at the specified time. express Update the deep slope deformation prediction output at the specified time. Represents a slope deformation proxy model; This is the slope inversion parameter vector; This is a vector of geometric parameters for the sliding surface in slope monitoring. This is the operating condition input vector.

[0164] The maximum deformation rate can be obtained by using the difference approximation method based on the surface deformation and deep deformation obtained by formula (20). Maximum acceleration .

[0165] In this embodiment, the updated safety factor is used as the basis. Maximum deformation rate With maximum acceleration In conjunction with the threshold range requirements for slope monitoring sub-items, a comprehensive four-level early warning classification and response system is implemented:

[0166] The safety factor threshold, deformation rate threshold, and deformation acceleration threshold levels are as follows:

[0167] 1) Safety factor threshold level: ;

[0168] 2) Deformation rate threshold level: ;

[0169] 3) Deformation acceleration threshold level: ;

[0170] in, These are the dividing thresholds for high, medium, and low safety levels, respectively. These are the dividing thresholds for high, medium, and low deformation rates, respectively. These are the dividing thresholds for high, medium, and low acceleration levels. The thresholds can be calibrated or configured according to conditions such as slope grade and project type.

[0171] Using the first The next update output Deformation rate and deformation acceleration The warning levels will be categorized according to the following rules (meeting any one of the conditions will result in a higher risk level, following the principle of "higher risk level prevails"):

[0172] when , and At that time, it is a Level I safety state;

[0173] when or or At that time, it was classified as a Level II low-risk state;

[0174] when or or At that time, it was classified as a Level III medium-risk situation;

[0175] when or or At that time, it was classified as a Level IV high-risk state.

[0176] In addition, different warning levels correspond to different follow-up update and handling strategies, as shown below:

[0177] During Level I safety status, maintain the current monitoring and update cycle;

[0178] Level II low-risk status, shorten the monitoring period;

[0179] Level III medium risk status, further intensified updates, with increased focus on key areas (such as GNSS and deep monitoring profiles).

[0180] Level IV high-risk status: Output early warning information to the early warning terminal platform and enter emergency mode.

[0181] Based on the above methods, the limit equilibrium method is a classic approach for slope stability analysis. It boasts high computational efficiency and can output safety factors and the most unfavorable slip surface, making it suitable for rapid on-site assessment. The surrogate model can quickly predict the slope deformation field, avoiding costly numerical simulations and enabling rapid acquisition of deformation rate and acceleration. The combination of these two methods balances the professionalism of stability assessment with the efficiency of deformation prediction. The principle of "choosing the highest indicator over the lowest" for comprehensive multi-indicator early warning is adopted because slope instability is sudden, and an anomaly in any indicator may foreshadow potential instability risk. This conservative approach effectively avoids missed diagnoses and improves engineering safety. Linking early warning levels with response strategies enables the rational allocation of monitoring resources. While ensuring slope safety, it avoids wasting monitoring resources in low-risk situations, improving the method's engineering economy and operability.

[0182] Although embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present invention, and all such changes and alterations shall not depart from the protection scope of the present invention.

Claims

1. A method for self-updating and early warning of slope digital twins constrained by satellite InSAR and ground displacement, characterized in that, Includes the following steps: S1. Acquire multi-source monitoring data of slope deformation, including surface line-of-sight deformation data obtained by satellite InSAR, point deformation data obtained by ground GNSS, and deep displacement profile deformation data obtained by embedded monitoring devices. S2. Perform spatiotemporal alignment and coordinate benchmark unification processing on the multi-source monitoring data to form an equivalent observation set; The coordinate reference unification process includes: unifying the InSAR pixel coordinates, GNSS point coordinates and deep borehole coordinates to the same engineering coordinate system; The spatiotemporal alignment includes: With the first The next update time Based on the time window The multi-source monitoring data is resampled to form an equivalent monitoring vector aligned with the time series: ； in, Indicates alignment to The equivalent monitoring vector at any given time; Indicates the width of the time window; Indicates the monitoring type index. Indicates the InSAR monitoring type. Indicates the type of GNSS monitoring. Indicates the type of deep monitoring; Representation type The original monitoring vector sequence; Indicates the time alignment operator; At the same time, a time decay weight is introduced: ； in, Indicates the time decay weight; Represents the time decay constant; Indicates the sampling time of the original monitoring data; Finally, the equivalent monitoring vector aligned to the time series is multiplied by the time decay weight to obtain the final spatiotemporal alignment result. S3. Construct an initial digital twin model of the slope, parameterize the updatable objects of the model, and form an augmented twin model state containing inversion parameter vectors and slip surface geometric dimension parameter vectors; S4. Construct a multi-source monitoring mapping operator and error weight system, establish a unified monitoring equation, and map the deformation field output by the twin model into a form that matches each monitoring data to obtain the equivalent covariance matrix and corresponding weight matrix of InSAR, GNSS, and deep monitoring. S5. Construct a hierarchical consistency constraint objective function and perform joint inversion solution to update the inversion parameter vector and slip surface geometric dimension parameter vector of the twin model, thereby realizing the self-updating of the slope digital twin model; S6. Based on the updated twin model, the slope safety factor is calculated using the limit equilibrium method in combination with external working conditions. The surface and deep deformation fields of the slope are output using the surrogate model, and the maximum deformation rate and maximum acceleration are obtained. Comprehensive graded early warning is carried out based on the safety factor, maximum deformation rate, and maximum acceleration.

2. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in claim 1, characterized in that, In step S3, the augmented state of the twin model is represented as follows: ； in, for The augmented state of the slope twin model at the update time; for Update the time-inversion parameter vector; for Update the surface geometry parameter vector at each time step; The inversion parameter vector should include at least the cohesion, internal friction angle, and pore water pressure of each layer of soil and rock on the slope. The augmented state of the twin model also pre-defines the upper and lower numerical boundaries of the inversion parameters, the prior value vector, the prior covariance matrix, and the feasible range of the surface geometry.

3. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in claim 1, characterized in that, In step S4, the unified monitoring equation is expressed as: ； in, Indicates monitoring type exist Update the monitoring vector at each time step; Representation type Mapping operators; Indicates monitoring error; The process of mapping the deformation field output by the twin model into a form that matches each monitoring data point includes: InSAR mapping: Project the surface displacement vector of the twin model into the InSAR line-of-sight displacement, considering the east-west, south-north, and vertical components, and combine the InSAR deformation LOS unit vector calculation model to predict the LOS displacement. GNSS mapping: The displacement field of the twin model is sampled at the GNSS monitoring points, and the GNSS reference zero bias parameter is introduced to match the GNSS three-dimensional point deformation data; Deep monitoring mapping: Construct a relative displacement mapping along the depth direction to convert the deep displacement field of the twin model into the relative displacement of the profile and match the deep monitoring data.

4. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in claim 3, characterized in that, In step S4, the equivalent covariance matrices of InSAR, GNSS, and deep monitoring are obtained by superimposing the monitoring error covariance matrices of InSAR, GNSS, and deep monitoring with their respective model error covariance matrices. Among them, the monitoring error covariance matrix of InSAR is composed of the pixel standard deviation, which is calculated based on the pixel coherence; the monitoring error covariance matrix of GNSS is composed of the monitoring variance / covariance of each monitoring point in the east-west, north-south, and vertical directions, which is directly calculated from the monitoring data; the monitoring error covariance matrix of deep monitoring is calculated by combining the borehole standard deviation with the error growth coefficient along the depth.

5. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in claim 1, characterized in that, In step S5, the objective function for the hierarchical consistency constraint is expressed as: ； in, For the first The objective function value of the hierarchical consistency constraint at the next update time; To unify monitoring vectors; To monitor the mapping vector; This is the weight matrix; This represents the error vector of the digital twin model. The inversion parameter vector; This is the initial parameter vector; The parameter is the covariance matrix; For parameter weights; For parameter constraint functions; For the geometric weights of the smooth surface; This is a geometric range constraint function; Weights for the sliding shear band; This is the feedback distance for the sliding shear band.

6. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in claim 1, characterized in that, In step S6, the external working conditions include at least one of rainfall, earthquake, and construction; the surrogate model includes any one of neural network, Gaussian process, random forest, and support vector regression.

7. The slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in any one of claims 1 to 6, characterized in that, In step S6, the comprehensive hierarchical early warning includes: The slope condition is divided into four levels: Level I safe, Level II low risk, Level III medium risk, and Level IV high risk. If the conditions of any higher risk level are met, the slope will be classified into the corresponding risk level. Different monitoring, updating, and response strategies will be adopted for different warning levels: If it is Level I, then the current monitoring and update cycle will be maintained for safety. If it is Level II, then shorten the monitoring update cycle; If it is Level III, then the updates will be encrypted and key monitoring areas will be given special attention; If it is Level IV, a warning message will be issued and emergency mode will be activated.

8. A computer device comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in any one of claims 1 to 7.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the slope digital twin self-updating and early warning method with joint constraints of satellite InSAR and ground displacement as described in any one of claims 1 to 7.

Images