A geological disaster digital twin dynamic simulation deduction system based on big data

Abstract

The application discloses a kind of geological disaster digital twin dynamic simulation deduction systems based on big data, belong to the cross field of geological disaster prevention and information technology.The system includes: data fusion and twin model dynamic updating module, collects multi-source heterogeneous geological big data and carries out space-time alignment and semantic fusion, constructs three-dimensional digital twin initial model and based on real-time monitoring data flow continuously updates model state;Multi-physical field coupling simulation module, based on the updated twin model constructs seepage field-stress field-displacement field sequential coupling simulation model, numerical calculation obtains stress distribution, seepage state and deformation trend;Multi-disaster chain deduction module, with directed graph representing disaster chain evolution rule, deduces the chain evolution path from initial disaster species via intermediate transition state to derivative disaster species.The application realizes the "monitoring-simulation-deduction-decision-correction" intelligent closed loop, improves the timeliness, accuracy and decision support capability of geological disaster deduction.

Description

Technical Field

[0001] This invention relates to the intersection of geological disaster prevention and control and information technology, and in particular to a dynamic simulation and deduction system for geological disaster digital twins based on big data. Background Technology

[0002] In recent years, digital twin technology has been introduced into the field of geological disaster prevention and control, providing new technical means for disaster early warning and risk assessment.

[0003] In the prior art, Chinese patent CN117332664B discloses a method for displaying a digital twin of geological hazards based on physical calculation results. This method constructs a landslide geological model and performs finite element calculations. After obtaining the crack system through a crack extraction module, it is displayed on the real-world landslide model. However, this method focuses on the visualization of physical calculation results and lacks the ability to dynamically update the model based on real-time monitoring data. It also cannot perform forward-looking predictions of the disaster evolution path and can only present static results under the current calculation conditions.

[0004] Chinese patent CN117787102B discloses a disaster digital twin simulation method combining a refined dynamic geological model with GEOAI. This method establishes a refined dynamic geological disaster simulation model and performs numerical simulations, then uses GEOAI to train, infer, and calibrate the model to realize the evolution and inversion of mine geological disasters. However, this method is mainly geared towards mine geological disaster scenarios, only involving the simulation of a single disaster type, and does not consider the chain evolution relationship between multiple geological disasters, nor does it provide the ability to perform parallel extrapolation and comparison of different intervention schemes.

[0005] Chinese patent CN119649549A discloses a geological disaster monitoring and early warning method and system based on digital twin technology. This method identifies risk sources by constructing a three-dimensional digital model and issues an early warning when the rate of change of monitored indicators exceeds a threshold. However, this method only provides passive early warning based on threshold judgments of monitoring data, lacking the ability to extrapolate disaster evolution based on multi-physics coupled simulation. It cannot simulate the disaster development trend under different intervention measures, nor can it adaptively correct the simulation model based on extrapolation deviations.

[0006] In summary, the existing technologies have the following shortcomings: (1) The digital twin model lacks a dynamic update mechanism based on real-time monitoring data, and it is difficult to maintain continuous synchronization between the model and the physical entity; (2) The simulation is only for a single type of disaster and cannot depict the chain evolution relationship between multiple geological disasters; (3) It does not have the ability to compare and simulate different intervention schemes in parallel, and it is difficult to support scientific decision-making; (4) There is a lack of feedback loop between the simulation results and the measured data, and the model parameters cannot be adaptively corrected according to the simulation deviation, and the simulation accuracy gradually decreases over time. Summary of the Invention

[0007] Purpose of the invention: To address the problems in existing technologies, such as the lack of dynamic updates in digital twin models, the inability to extrapolate the chain evolution of multiple hazards, the lack of ability to compare and extrapolate multiple schemes in parallel, and the lack of feedback loop between extrapolation results and measured data, this invention provides a dynamic simulation and extrapolation system for geological disaster digital twins based on big data. This system enables real-time dynamic updates of digital twin models, multi-physics coupled simulation, chain evolution extrapolation of multiple hazards, parallel extrapolation and scheme optimization in multiple scenarios, and adaptive correction of models driven by extrapolation feedback, forming an intelligent closed loop of "monitoring-simulation-extrapolation-decision-correction".

[0008] Technical solution: A dynamic simulation and extrapolation system for geological disaster digital twins based on big data, comprising:

[0009] The data fusion and twin model dynamic update module is used to collect multi-source heterogeneous geological big data of the target area and perform spatiotemporal alignment and semantic fusion on it to construct a three-dimensional digital twin initial model containing surface topography, underground geological structure and monitoring and sensing data, and continuously update the state of the digital twin model based on real-time monitoring data stream.

[0010] The multiphysics coupling simulation module is connected to the data fusion and twin model dynamic update module. It is used to construct a multiphysics coupling simulation model of seepage field-stress field-displacement field based on the updated digital twin model, and to perform numerical calculations on the stress distribution, seepage state and deformation trend of the target area to obtain multiphysics coupling simulation results.

[0011] The multi-hazard chain simulation module is connected to the multi-physics field coupled simulation module. It is used to receive external triggering conditions and, based on the multi-physics field coupled simulation results, to deduce the chain evolution path from the initial hazard through the intermediate transition state to the derived hazard according to the preset hazard chain evolution rules, and generate time series data of hazard chain evolution.

[0012] The multi-scenario parallel simulation and scheme selection module is connected to the multi-hazard chain simulation module. It is used to set multiple intervention scheme parameters for the same triggering condition, perform chain simulation in parallel for each intervention scheme, quantitatively evaluate each scheme based on the key disaster indicators in the simulation results, and select the intervention scheme that optimizes the key disaster indicators as the recommended scheme.

[0013] The extrapolation feedback and model adaptive correction module is connected to the data fusion and twin model dynamic update module and the multiphysics coupling simulation module. It is used to compare the time series data of the chain extrapolation with the measured monitoring data in time series, calculate the extrapolation deviation, and adaptively correct the key calculation parameters in the coupling simulation model according to the extrapolation deviation.

[0014] Furthermore, in the data fusion and twin model dynamic update module, the multi-source heterogeneous geological big data includes remote sensing image data, geological exploration data, meteorological observation data, and real-time data collected by ground monitoring stations. The spatiotemporal alignment method is as follows: using a unified spatial reference coordinate system as a reference, spatial registration and temporal interpolation are performed on each source data respectively, so that data with different spatial resolutions and sampling frequencies are unified into the same spatiotemporal grid.

[0015] Furthermore, in the multiphysics coupling simulation module, the seepage field-stress field-displacement field multiphysics coupling simulation model adopts a sequential coupling solution strategy. First, the seepage field is solved to obtain the pore water pressure distribution. Then, the pore water pressure is substituted as a body load into the stress field to obtain the stress distribution and the range of the plastic zone. Finally, the stress field calculation result is used as the boundary condition of the displacement field to obtain the deformation displacement field. The three physical fields achieve coupling convergence through iterative calculation.

[0016] Furthermore, in the multi-hazard chain-based deduction module, the hazard chain evolution rules are represented by a directed graph. The nodes in the graph represent hazard states, and the directed edges represent the transformation conditions between hazard states. When the key physical quantity of a hazard state exceeds the threshold corresponding to the transformation condition, the deduction moves along the directed edge to the next hazard node to continue the evolution. The initial hazard states include rainfall infiltration and reservoir water level changes, and the derived hazard states include landslides, collapses, debris flows, and barrier lakes.

[0017] Furthermore, in the multi-scenario parallel simulation and scheme optimization module, the key disaster indicators include the maximum deformation rate, cumulative deformation, potential slip volume, and area of ​​influence. The quantitative evaluation adopts a weighted comprehensive scoring method, which calculates a comprehensive score for each key disaster indicator according to a preset weight. The intervention scheme with the lowest comprehensive score is the recommended scheme that optimizes the key disaster indicators.

[0018] Furthermore, in the inference feedback and model adaptive correction module, the time-series predicted values ​​of target monitoring point displacement obtained by inference are compared with the measured time-series values ​​of displacement at each time step to obtain a residual sequence. The root mean square error of the residual sequence is calculated as a measure of inference deviation. With minimizing the inference deviation as the objective function, the gradient descent method is used to perform reverse optimization of the geotechnical parameters and permeability coefficient in the coupled simulation model.

[0019] Furthermore, it also includes a graded early warning and information push module, which is connected to the multi-hazard chain simulation module and the multi-scenario parallel simulation and scheme optimization module. It is used to generate four-level early warning signals (blue, yellow, orange, and red) based on the comparison results of the key physical quantities of each hazard node in the chain simulation results and the preset graded thresholds, and push the early warning signals and the recommended schemes to the designated terminals.

[0020] Furthermore, it also includes a 3D visualization interaction module, which is connected to the data fusion and twin model dynamic update module, the multiphysics coupling simulation module, and the multi-hazard chain inference module. This module is used to display the digital twin model, multiphysics coupling simulation results, and disaster chain evolution path in a 3D visualization manner, and supports users to interactively edit intervention plan parameters and view the inference process in a 3D scene.

[0021] This invention also provides a method for dynamic simulation and extrapolation of geological disaster digital twins based on big data, which employs the above-mentioned system execution: collecting multi-source heterogeneous geological big data, performing spatiotemporal alignment and semantic fusion, constructing and continuously updating a three-dimensional digital twin model; constructing a multi-physics coupled simulation model based on the updated twin model and performing numerical calculations; receiving external triggering conditions and performing multi-hazard chain extrapolation according to the disaster chain evolution rules; setting parameters for multiple intervention schemes for parallel extrapolation and selecting the optimal scheme based on key disaster indicators; comparing the extrapolation results with measured data to calculate the deviation and adaptively correcting the key parameters of the simulation model.

[0022] Beneficial effects:

[0023] (1) Continuous dynamic update of digital twin model: The state of digital twin model is continuously updated based on real-time monitoring data stream, so that the virtual model and physical entity are kept in continuous synchronization, overcoming the shortcomings of existing technologies where the model is basically no longer updated after it is built.

[0024] (2) Multi-hazard chain simulation capability: The disaster chain evolution rules are depicted by directed graphs, and the chain evolution path from the initial disaster type to the derived disaster type is simulated. It can simulate complex disaster scenarios such as landslide → barrier lake → flood outburst, which makes up for the limitation of existing technologies that only target a single disaster type.

[0025] (3) Parallel simulation and scheme selection for multiple scenarios: Multiple intervention schemes are simulated in parallel for the same triggering condition. The optimal scheme is selected based on quantitative evaluation of key disaster indicators, providing a scientific basis for disaster prevention and control decision-making and overcoming the shortcomings of existing technologies in lacking the ability to compare multiple schemes.

[0026] (4) Model adaptive correction driven by deduction feedback: The deduction results are compared with the measured data, and the simulation model parameters are adaptively corrected to form a feedback loop of "deduction-comparison-correction". The deduction accuracy continues to improve with the running time, which solves the problem of the deduction accuracy of the existing technology decreasing with time. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall architecture of a big data-based digital twin dynamic simulation and deduction system for geological disasters according to the present invention.

[0028] Figure 2 This is a schematic diagram of the disaster chain evolution directed graph of the multi-hazard chain inference module of the present invention;

[0029] Figure 3 This is a schematic diagram illustrating the workflow of the deduction feedback and model adaptive correction module of this invention. Detailed Implementation

[0030] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1

[0032] like Figure 1 As shown, a dynamic simulation and extrapolation system for geological disaster digital twins based on big data includes a data fusion and twin model dynamic update module, a multi-physics field coupled simulation module, a multi-hazard chain extrapolation module, a multi-scenario parallel extrapolation and scheme optimization module, an extrapolation feedback and model adaptive correction module, a graded early warning and information push module, and a three-dimensional visualization interaction module.

[0033] The working process of the data fusion and twin model dynamic update module is as follows: First, multi-source heterogeneous geological big data of the target area is collected. The data sources include: (a) remote sensing image data, such as satellite InSAR deformation monitoring data and UAV oblique photogrammetry 3D modeling data; (b) geological exploration data, such as borehole columnar sections, geological profiles and rock and soil mechanics test reports; (c) meteorological observation data, such as hourly rainfall, temperature and humidity data of regional automatic weather stations; (d) real-time data collected by ground monitoring stations, such as surface displacement data of GNSS displacement monitoring stations, underground displacement data of deep inclinometers, seepage pressure data of pore water pressure gauges and soil moisture content sensor data. Spatiotemporal alignment was performed on the aforementioned multi-source heterogeneous data: using the WGS-84 coordinate system as a unified spatial reference coordinate system, orthorectification and geometric registration were performed on the remote sensing image data, and borehole coordinates in the geological exploration data were uniformly transformed to the WGS-84 coordinate system. The spatial locations of ground monitoring stations were marked in the reference coordinate system. In the temporal dimension, linear interpolation was performed on data sources with inconsistent sampling frequencies to unify them to the same time step, such as aligning 5-minute resolution rainfall data and 1-hour resolution GNSS displacement data to a 1-hour time step. In terms of semantic fusion, descriptions of the same geological entity in different data sources were uniformly encoded, such as associating and matching stratigraphic divisions in borehole columnar sections with stratigraphic boundaries in geological profiles to form semantically consistent geological entity models. Finally, a three-dimensional digital twin initial model was constructed, including surface topography (DEM), subsurface geological structure (3D geological model), and monitoring and sensing data (real-time sensor data overlay). During system operation, the data stream collected in real time by the ground monitoring station is continuously input into the module. The module incrementally updates the state of the digital twin model according to the preset update cycle (such as every 6 hours), including updating dynamic layers such as the surface displacement field, groundwater level and pore water pressure field.

[0034] The multiphysics coupled simulation module receives the updated digital twin model output from the data fusion and twin model dynamic update module, and constructs a multiphysics coupled simulation model of seepage field, stress field, and displacement field. Specifically, based on the updated 3D geological model, meshed geometric information is extracted, and the target area is meshed using finite element methods. The mesh scale is adaptively determined according to the geological complexity, with a mesh scale of 5-10 meters for key areas (such as known landslides) and 20-50 meters for general areas. The seepage field uses the Richards equation to describe the unsaturated-saturated seepage process, with boundary conditions being the surface rainfall infiltration boundary and the groundwater level boundary. The stress field uses the Mohr-Coulomb elastoplastic constitutive model, and the geotechnical parameters include the internal friction angle, cohesion, elastic modulus, and Poisson's ratio, with initial values ​​determined by geotechnical test results from geological exploration data. The displacement field uses the nodal forces calculated from the stress field as loads to solve the elasticity equations. The three physical fields are solved using a sequential coupling strategy: First, the seepage field is solved based on the current rainfall and water level boundary conditions to obtain the spatial distribution of pore water pressure. Second, the pore water pressure is substituted as a body load into the stress field solution to obtain the stress distribution and the range of the plastic zone. Third, the stress field calculation results are used as the displacement field boundary conditions to obtain the deformation displacement field. Fourth, the influence of the displacement field calculation results on the seepage field and stress field is checked for convergence. If convergence is not achieved, the process returns to the first step and iterates again until the displacement increment between two adjacent iterations is less than the preset convergence tolerance (e.g., 1×10⁻⁶). -6 (meters). After iterative convergence, the multiphysics coupling simulation results are output, including the stress distribution cloud map, plastic zone range, seepage head distribution, and displacement field deformation vector at the current moment.

[0035] The multi-hazard chain simulation module receives external triggering conditions and multiphysics coupled simulation results, and performs chain simulations according to the hazard chain evolution rules. External triggering conditions include meteorological forecasts of rainfall, reservoir water level changes, or seismic motion parameters, etc. See also Figure 2The evolution rules of the disaster chain are represented by a directed graph, where nodes represent disaster states, including: node A (rainfall infiltration / reservoir water infiltration), node B (accelerated slope deformation), node C (landslide initiation), node D (collapse occurrence), node E (debris flow formation), node F (landslide dam formation), and node G (outburst flood). Directed edges represent the transformation conditions between disaster types: the transformation condition from A to B is that the pore water pressure increases beyond the critical water level threshold; the transformation condition from B to C is that the cumulative plastic strain exceeds the sliding surface penetration threshold; the transformation condition from C to D is that the leading edge of the slope is exposed and the tensile crack depth at the trailing edge exceeds the critical value; the transformation condition from C to E is that the landslide enters the gully and the longitudinal slope of the gully exceeds the debris flow initiation threshold; the transformation condition from E to F is that the debris flow blocks the main gully, forming a landslide dam; and the transformation condition from F to G is that the water level of the landslide dam exceeds the elevation of the top surface of the landslide dam. During the simulation, starting from the currently activated disaster node, it is determined whether the corresponding key physical quantities in the multiphysics coupling simulation results meet the transformation conditions. If they do, the simulation is transferred along the directed edge to the next disaster node, and the corresponding physical model is called for calculation. This process is repeated node by node to generate time series data of the disaster chain evolution process, including the activation time, duration, and key physical quantity values ​​of each disaster node.

[0036] The multi-scenario parallel simulation and scheme selection module sets multiple intervention scheme parameters for the same triggering condition and performs chain simulations on each scheme in parallel. The intervention scheme parameters include: (a) no intervention scheme (baseline scheme), i.e., no engineering measures are taken; (b) drainage and pressure reduction scheme, i.e., setting up underground drainage corridors to lower the groundwater level by a preset amount; (c) slope cutting and load reduction scheme, i.e., removing a preset amount of soil and rock from the upper part of the slope; (d) anti-slide pile reinforcement scheme, i.e., setting a preset number and spacing of anti-slide piles in the middle of the slope. For each scheme, the corresponding parameters in the digital twin model are modified (e.g., the seepage boundary conditions are modified for the drainage scheme, the geometric boundary and initial stress field conditions are modified for the slope cutting scheme, and pile-soil contact elements are added to the stress field for the anti-slide pile scheme). Then, multiphysics coupling simulation and chain simulation are performed independently to obtain the time series results of the disaster chain simulation under each scheme. Key disaster indicators include: maximum deformation rate (mm / day), cumulative deformation (mm), potential sliding volume (10,000 cubic meters), and affected area (10,000 square meters). The quantitative evaluation adopts a weighted comprehensive scoring method, with the weights of each indicator set as w1, w2, w3, and w4 (e.g., w1=0.3, w2=0.2, w3=0.3, w4=0.2). For each scheme, the indicators are normalized and then summed according to their weights to obtain a comprehensive score. The scheme with the lowest comprehensive score is the recommended scheme.

[0037] The extrapolation feedback and model adaptive correction module compares the time-series data from the chain extrapolation with the measured monitoring data. Specifically, it selects the grid nodes in the digital twin model corresponding to the GNSS monitoring station and the deep inclinometer, extracts the displacement time-series prediction values ​​obtained from the extrapolation, and calculates the difference between these values ​​and the measured displacement time-series values ​​within the same time window, obtaining the residual sequence e(t). The root mean square error (RMSE) of the residual sequence is then calculated as a measure of extrapolation bias. When the RMSE exceeds a preset threshold (e.g., 5 mm), adaptive correction is triggered: with minimizing the extrapolation bias as the objective function, the gradient descent method is used to perform reverse optimization of the key parameters in the coupled simulation model. The optimization parameters include geotechnical parameters (internal friction angle φ, cohesion c, elastic modulus E) and permeability coefficient K. The specific steps of the gradient descent method are as follows: Apply a small perturbation Δp to each parameter to be tuned, re-execute the simulation to calculate the new RMSE, calculate the differential gradient of the objective function with respect to the parameters ∂RMSE / ∂p, update the parameter values ​​along the negative gradient direction with a preset learning rate (e.g., 0.01), and iterate until the RMSE drops below the threshold or the maximum number of iterations is reached. The corrected parameters are synchronously updated to the data fusion and twin model dynamic update module and the multiphysics coupling simulation module, thereby improving the accuracy of subsequent inferences.

[0038] The tiered early warning and information push module generates early warning signals based on the chain-like simulation results. A blue early warning signal is generated when the simulation results show the target area is only at node A (rainfall infiltration); a yellow early warning signal is generated when it is at node B (accelerated slope deformation); an orange early warning signal is generated when the simulation reaches node C or D (landslide initiation or collapse) and there are residential areas within the expected impact range; and a red early warning signal is generated when the simulation reaches node E or above (debris flow, landslide dam, outburst flood) or the deformation rate exceeds the red threshold. The early warning signals and recommended solutions are pushed to designated terminals (such as the emergency management command platform and the mobile phones of village / group leaders).

[0039] The 3D visualization interactive module displays the digital twin model, multiphysics coupling simulation results, and disaster chain evolution path in a 3D visualization manner. Users can interactively edit intervention plan parameters in the 3D scene, such as dragging the location of drainage corridors, adjusting the slope cutting range, and modifying the layout of anti-slide piles. After editing, the system executes the simulation in real time and displays the results.

[0040] Example 2

[0041] This embodiment uses a landslide on the bank of a reservoir in a mountainous area as a specific application scenario to illustrate the system's multi-hazard chain simulation process. (See [link to documentation]). Figure 2 .

[0042] The target area is a large ancient landslide on the right bank of a reservoir, covering an area of ​​approximately 25,000 square meters and with a volume of approximately 1.5 million cubic meters. The leading edge of the landslide is below the normal water level of the reservoir, while the trailing edge is at an elevation of approximately 520 meters. The data fusion and twin model dynamic update module collected InSAR deformation data, geological exploration data from 12 boreholes, rainfall data from regional automatic weather stations, and real-time data from 6 GNSS displacement monitoring stations, 4 deep inclinometers, and 3 pore water pressure gauges deployed on the landslide. After spatiotemporal alignment and semantic fusion, a three-dimensional digital twin model was constructed.

[0043] During a flood season in a certain year, weather forecasts indicated a heavy rainfall event with accumulated precipitation exceeding 200 mm within the next 72 hours. A multiphysics coupled simulation module, based on a current digital twin model, used the predicted rainfall event as the upper boundary condition of the seepage field to perform coupled simulations of the seepage field, stress field, and displacement field. Simulation results showed that 24 hours after rainfall, the pore water pressure at the slip surface increased by approximately 35 kPa; 48 hours after rainfall, tensile cracks with a depth of approximately 3 meters appeared at the rear edge of the slope, and the plastic zone expanded from the front edge of the slope towards the middle; 72 hours after rainfall, the slip surface penetration rate exceeded 80%, and the maximum displacement rate of the slope reached 15 mm / day.

[0044] The multi-hazard chain simulation module performs chain simulations based on the above simulation results. The initial triggering condition is a heavy rainfall process, corresponding to node A (rainfall infiltration) in the directed graph of the hazard chain. The simulation process is as follows: 24 hours after rainfall, the pore water pressure rises above the critical water level threshold, satisfying the A→B transformation condition, and the simulation shifts to node B (accelerated slope deformation); 48 hours after rainfall, the cumulative plastic strain reaches the sliding surface penetration threshold, satisfying the B→C transformation condition, and the simulation shifts to node C (landslide initiation); since the leading edge of the landslide is located in a reservoir, after the landslide initiation, part of the landslide body will slide into the reservoir, and the longitudinal slope of the gully at this location is about 25°, satisfying the C→E transformation condition (the landslide body enters the gully and the longitudinal slope exceeds the debris flow initiation threshold of 20°), and the simulation shifts to node E (debris flow formation); the debris flow travels down the gully for about 800 meters before entering the main river, forming a dam in a narrow section, satisfying the E→F transformation condition, and the simulation shifts to node F (dammed lake formation); the water level of the dammed lake continues to rise, and after about 6 hours of calculation, the water level exceeds the elevation of the top surface of the dammed body, satisfying the F→G transformation condition, and the simulation shifts to node G (breakwater flood). This leads to the complete deduction of the six-node disaster chain evolution path: "rainfall infiltration → slope deformation → landslide → debris flow → barrier lake → outburst flood", and generates the activation time and duration of each node.

[0045] The multi-scenario parallel simulation and scheme selection module sets three intervention schemes: (a) baseline scheme (no intervention); (b) underground drainage scheme, setting up two drainage corridors to lower the groundwater level by 5 meters; (c) slope cutting and load reduction scheme, removing 80,000 cubic meters of soil and rock from the upper part of the slope. The parallel simulation results of the three schemes show that: under the baseline scheme, a complete six-node disaster chain is formed; under the drainage scheme, the increase in pore water pressure is reduced, the B→C transformation condition is not met, the disaster chain terminates at node B, and no landslide occurs; under the slope cutting scheme, the sliding surface penetration rate is reduced to about 60%, the C→E transformation condition is not met, the landslide is initiated but no debris flow is formed. Quantitative evaluation of key disaster indicators: the drainage scheme has the lowest comprehensive score (0.25), followed by the slope cutting scheme (0.48), and the baseline scheme has the highest score (0.92). The system recommends the drainage scheme.

[0046] After the rainfall event ended, the simulation feedback and model adaptive correction module compared the simulated displacement time series with the measured data from six GNSS monitoring stations. The RMSE was 3.2 mm, which is below the 5 mm threshold, so there was no need to trigger adaptive correction, indicating that the current model parameters are reliable.

[0047] Example 3

[0048] This embodiment illustrates the working process of the inference feedback and model adaptive correction module. See [link to documentation]. Figure 3 .

[0049] In another landslide monitoring area, the system continuously performs simulations during routine operation outside of the flood season. After each simulation, the simulation feedback and model adaptive correction module automatically extracts the displacement time-series prediction values ​​of the grid nodes corresponding to the GNSS monitoring stations in the digital twin model, and calculates the difference between these values ​​and the measured displacement time-series values ​​of the GNSS stations within the same time window to obtain the residual sequence e(t). The root mean square error (RMSE) of the residual sequence is then calculated.

[0050] After a certain simulation, the RMSE calculation result was 7.8 mm, exceeding the preset threshold of 5 mm, triggering adaptive correction. Parameters to be optimized include the internal friction angle φ (current value 22°), cohesion c (current value 28 kPa), elastic modulus E (current value 800 MPa), and permeability coefficient K (current value 1.2 × 10⁻⁶). -6 (m / s). Small perturbations Δp are applied to each parameter (φ perturbation 0.5°, c perturbation 1 kPa, E perturbation 50 MPa, K perturbation 1 × 10⁻⁶ m / s). -7 (m / s), re-execute the simulation to calculate the new RMSE, and obtain the differential gradient of each parameter. RMSE / p. The gradient calculation results show: RMSE / φ=-0.8 mm / degree RMSE / c = -0.3 mm / kPa RMSE / E=0.01 mm / MPa RMSE / K = 12.5 mm / (m / s). It is evident that the permeability coefficient K and the internal friction angle φ have the greatest impact on the extrapolation deviation. Parameter updates were performed with a learning rate α = 0.01: φ was updated to 22° + 0.01 × 0.8° = 22.008°, c was updated to 28 kPa + 0.01 × 0.3 kPa = 28.003 kPa, and K was updated to 1.2 × 10⁻⁶. -6 -0.01×12.5×10 -7 =1.075×10 -6 The speed is m / s. The gradient of E is extremely small, so no adjustment is needed for now. The simulation is re-executed with the updated parameters. The new RMSE is reduced to 5.1 mm, still slightly above the threshold. The second iteration continues: the new gradient is calculated, and the parameters are updated again. After the second simulation, the RMSE is reduced to 3.6 mm, below the threshold of 5 mm, and the adaptive correction ends. The corrected parameters are φ = 22.02°, c = 28.01 kPa, and K = 1.05 × 10⁻⁶. -6 The speed in meters per second is synchronously updated to the data fusion and twin model dynamic update module and the multiphysics coupling simulation module. Subsequent inferences use the corrected parameters, thus improving the inference accuracy.

[0051] Example 4

[0052] This embodiment uses a combined landslide-debris flow hazard in the back mountain of a city as an example to illustrate the system's handling of different hazard chain branches in multi-hazard chain simulation. (See also...) Figure 2 .

[0053] The target area is a slope on the back mountain of a city. The upper part of the slope is a steep rock slope with unstable rock masses, while the lower part is a sedimentary layer. A channel at the foot of the slope leads to the urban area. The data fusion and twin model dynamic update module collects LiDAR scan data, geological survey data, rainfall monitoring data, and real-time data from sensors deployed on the unstable rock masses and sedimentary layers in this area to construct a three-dimensional digital twin model.

[0054] Under conditions triggered by heavy rainfall, the multiphysics coupled simulation module showed that the tensile crack depth at the rear edge of the unstable rock mass reached 8 meters, exceeding the critical value of 5 meters; simultaneously, the pore water pressure at the bottom of the deposit layer increased beyond the critical value. Based on this, the multi-hazard chain-like extrapolation module performed the following extrapolation: starting from node A (rainfall infiltration), simultaneously satisfying the A→B and A→D transformation conditions (tensile crack depth at the rear edge of the unstable rock mass exceeding the threshold), the extrapolation split into two branches: Branch 1 follows the A→B→C path, where accelerated deformation of the deposit layer ultimately triggers a landslide; Branch 2 follows the A→D path, where the unstable rock mass collapses. In Branch 1, after the landslide begins, the landslide body enters the gully at the toe of the slope, with a longitudinal slope of approximately 22°, satisfying the C→E transformation condition, forming a debris flow. In Branch 2, the collapsed rocks roll down the slope, some entering the gully to provide a source of loose solid material, further exacerbating the debris flow. The two branches converge at node E (debris flow formation), and the extrapolation results show that the peak debris flow flow is approximately 40% higher than that of a single branch. The debris flow traveled approximately 1.5 kilometers downstream along the gully to the edge of the urban area, posing a threat to the city. The system generated an orange alert signal and pushed it to the urban emergency management department. The recommended solution was to construct a series of dams in the middle reaches of the gully to intercept the solid debris flow, while simultaneously anchoring unstable rock masses to eliminate the risk of collapse. Parallel simulations across multiple scenarios showed that: with only the dams, the debris flow could still reach the urban area, but the peak flow was reduced by 60%; with only the unstable rock masses anchored, the debris flow size was reduced by approximately 40%, but it could still cause a disaster. The dual-measure solution resulted in the debris flow stopping at the dams, achieving the lowest overall score (0.18), and was therefore the recommended solution.

[0055] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A dynamic simulation and extrapolation system for geological disaster digital twins based on big data, characterized in that, include: The data fusion and twin model dynamic update module is used to collect multi-source heterogeneous geological big data of the target area and perform spatiotemporal alignment and semantic fusion on it to construct a three-dimensional digital twin initial model containing surface topography, underground geological structure and monitoring and sensing data, and continuously update the state of the digital twin model based on real-time monitoring data stream. The multiphysics coupling simulation module is connected to the data fusion and twin model dynamic update module. It is used to construct a multiphysics coupling simulation model of seepage field-stress field-displacement field based on the updated digital twin model, and to perform numerical calculations on the stress distribution, seepage state and deformation trend of the target area to obtain multiphysics coupling simulation results. The multi-hazard chain simulation module is connected to the multi-physics field coupled simulation module. It is used to receive external triggering conditions and, based on the multi-physics field coupled simulation results, to deduce the chain evolution path from the initial hazard through the intermediate transition state to the derived hazard according to the preset hazard chain evolution rules, and generate time series data of hazard chain evolution. The multi-scenario parallel simulation and scheme selection module is connected to the multi-hazard chain simulation module. It is used to set multiple intervention scheme parameters for the same triggering condition, perform chain simulation in parallel for each intervention scheme, quantitatively evaluate each scheme based on the key disaster indicators in the simulation results, and select the intervention scheme that optimizes the key disaster indicators as the recommended scheme. The extrapolation feedback and model adaptive correction module is connected to the data fusion and twin model dynamic update module and the multiphysics coupling simulation module. It is used to compare the time series data of the chain extrapolation with the measured monitoring data in time series, calculate the extrapolation deviation, and adaptively correct the key calculation parameters in the coupling simulation model according to the extrapolation deviation.

2. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data as described in claim 1, characterized in that, In the data fusion and twin model dynamic update module, the multi-source heterogeneous geological big data includes remote sensing image data, geological exploration data, meteorological observation data, and real-time data collected by ground monitoring stations. The spatiotemporal alignment method is as follows: using a unified spatial reference coordinate system as a reference, spatial registration and temporal interpolation are performed on each source data respectively, so that data with different spatial resolutions and sampling frequencies are unified into the same spatiotemporal grid.

3. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data as described in claim 1, characterized in that, In the multiphysics coupling simulation module, the seepage field-stress field-displacement field multiphysics coupling simulation model adopts a sequential coupling solution strategy. First, the seepage field is solved to obtain the pore water pressure distribution. Then, the pore water pressure is substituted as a body load into the stress field to obtain the stress distribution and the range of the plastic zone. Finally, the stress field calculation result is used as the boundary condition of the displacement field to obtain the deformation displacement field. The three physical fields achieve coupling convergence through iterative calculation.

4. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data as described in claim 1, characterized in that, In the multi-hazard chain-based deduction module, the hazard chain evolution rules are represented by a directed graph. The nodes in the graph represent hazard states, and the directed edges represent the transformation conditions between hazard states. When the key physical quantity of a hazard state exceeds the threshold corresponding to the transformation condition, the deduction moves along the directed edge to the next hazard node to continue the evolution. The initial hazard states include rainfall infiltration and reservoir water level changes, and the derived hazard states include landslides, collapses, debris flows, and barrier lakes.

5. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data according to claim 1, characterized in that, In the multi-scenario parallel simulation and scheme optimization module, the key disaster indicators include the maximum deformation rate, cumulative deformation, potential slip volume, and area of ​​influence. The quantitative evaluation adopts a weighted comprehensive scoring method, which calculates a comprehensive score for each key disaster indicator according to a preset weight. The intervention scheme with the lowest comprehensive score is the recommended scheme that optimizes the key disaster indicators.

6. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data according to claim 1, characterized in that, In the inference feedback and model adaptive correction module, the time-series predicted values ​​of target monitoring point displacement obtained by inference are compared with the measured time-series values ​​of displacement at each time step to obtain a residual sequence. The root mean square error of the residual sequence is calculated as a measure of inference deviation. With minimizing the inference deviation as the objective function, the gradient descent method is used to perform reverse optimization of the geotechnical mechanical parameters and permeability coefficient in the coupled simulation model.

7. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data according to claim 1, characterized in that, It also includes a graded early warning and information push module, which is connected to the multi-hazard chain simulation module and the multi-scenario parallel simulation and scheme optimization module. It is used to generate four-level early warning signals (blue, yellow, orange, and red) based on the comparison results of the key physical quantities of each hazard node in the chain simulation results and the preset graded thresholds, and push the early warning signals and the recommended schemes to the designated terminal.

8. The dynamic simulation and extrapolation system for geological disaster digital twins based on big data according to claim 1, characterized in that, It also includes a 3D visualization interaction module, which is connected to the data fusion and twin model dynamic update module, the multiphysics coupling simulation module, and the multi-hazard chain inference module. It is used to display the digital twin model, multiphysics coupling simulation results, and disaster chain evolution path in a 3D visualization manner, and supports users to interactively edit intervention plan parameters and view the inference process in a 3D scene.

9. A method for dynamic simulation and extrapolation of geological disaster digital twins based on big data, characterized in that, The system described in any one of claims 1 to 8 is used to perform the following: collecting multi-source heterogeneous geological big data, performing spatiotemporal alignment and semantic fusion, constructing and continuously updating a three-dimensional digital twin model; constructing a multi-physics field coupled simulation model based on the updated twin model and performing numerical calculations; receiving external triggering conditions and performing multi-hazard chain-like deduction according to the disaster chain evolution rules; Multiple intervention scheme parameters are set and simulated in parallel, and the scheme is selected based on key disaster indicators; the simulation results are compared with the measured data to calculate the deviation, and the key parameters of the simulation model are adaptively corrected.

Images