Tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method and system

Abstract

The invention discloses a tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method and system, and the method comprises the steps: collecting multi-source heterogeneous data in real time through intelligent drilling equipment, a distributed optical fiber sensing network and a mobile scanning device which are disposed on a tunnel construction surface; inputting the multi-source heterogeneous data into a space-time diagram convolutional network, constructing a space-time correlation model of geological features and construction disturbance parameters, and outputting a coupling feature matrix; based on the coupling characteristic matrix, simulating the dynamic response of the surrounding rock-support system through a streaming voxelization engine and a multi-physics coupling engine, and generating a risk evolution three-dimensional thermodynamic diagram; and inputting the risk evolution three-dimensional thermodynamic diagram into the deep reinforcement learning model, and dynamically optimizing to generate a construction parameter regulation and control instruction meeting a safety constraint condition. According to the technology, a full-link channel of'data perception-mechanical analysis-decision execution 'is opened in a closed-loop mode, and therefore the double risks of'data distortion misjudgment' and'response lag out-of-control 'of a traditional method are avoided.

Description

Technical Field

[0001] This application relates to the technical field of tunnel engineering, and particularly to an intelligent prediction and control method and system for tunnel multi-source fusion dynamic twin surrounding rock. Background Art

[0002] In the process of tunnel engineering intelligentization, the dynamic perception of surrounding rock state and the precise control of construction have always been core problems. As tunnels extend into deep and complex geological environments, traditional methods have gradually revealed systematic limitations: on the one hand, due to the strong coupling characteristics of multi-source monitoring data (such as acoustic waves, deformation, and vibration signals) with geological structures and construction disturbances, it is difficult to achieve effective fusion in dynamic scenarios, resulting in the accumulation of deviation in the interpretation of the mechanical state of the surrounding rock; on the other hand, existing numerical models are limited by static calculation frameworks and cannot synchronously map the non-linear response of the surrounding rock caused by the operation of tunneling machines, resulting in significant lags in support decision-making and risk prevention and control. Summary of the Invention

[0003] To solve the above problems, an embodiment of the present invention provides an intelligent prediction and control method for tunnel multi-source fusion dynamic twin surrounding rock, and the method includes: Real-time collect multi-source heterogeneous data through intelligent drilling equipment, distributed optical fiber sensing networks, and mobile scanning devices deployed on the tunnel construction surface; Input the multi-source heterogeneous data into a spatio-temporal graph convolutional network to construct a spatio-temporal correlation model of geological features and construction disturbance parameters, and output a coupled feature matrix; Based on the coupled feature matrix, simulate the dynamic response of the surrounding rock-support system through a streaming voxelization engine and a multi-physics field coupling engine to generate a three-dimensional heat map of risk evolution; Input the three-dimensional heat map of risk evolution into a deep reinforcement learning model to dynamically optimize and generate construction parameter control instructions that meet safety constraint conditions.

[0004] Furthermore, the intelligent drilling equipment includes an axial thrust detection module and a rock acoustic wave feature acquisition module; the distributed optical fiber sensing network is arranged along the tunnel excavation direction and includes a temperature-compensated strain sensing unit; the mobile scanning device uses a multi-line lidar to realize the three-dimensional point cloud reconstruction of the construction surface.

[0005] Furthermore, the construction method of the spatio-temporal graph convolutional network includes: Map the distribution of structural planes in geological exploration data into the spatial topological relationship of the graph structure; Use a time convolutional layer to extract the temporal propagation characteristics of construction machinery vibration parameters; Fuse the correlation weight distribution of geological parameters and construction parameters through a multi-head attention mechanism.

[0006] Furthermore, the streaming voxelization engine dynamically adjusts the computational grid resolution according to the surrounding rock stress gradient; the multi-physics coupling engine includes an interaction interface between the discrete element module and the fluid dynamics module; the three-dimensional thermal map of risk evolution includes a plastic strain distribution layer and a support stress warning layer.

[0007] Furthermore, the real-time calibration operation includes: Dynamically comparing the surrounding rock strain data monitored by the distributed optical fiber sensing network with the simulated strain values in the three-dimensional thermal map of risk evolution; When the deviation between the monitored data and the simulated data exceeds the preset threshold, trigger the weight retraining mechanism of the spatio-temporal graph convolutional network; Based on the updated coupled feature matrix, correct the constitutive model parameters of the multi-physics coupling engine.

[0008] Furthermore, the action space of the deep reinforcement learning model is defined as a set of continuous construction control parameters, mapping the construction control parameters into a high-dimensional continuous action vector; compressed into the standardized space of [-1, 1] through the construction parameter regulation instruction, and the double delayed deep deterministic policy gradient algorithm is adopted; The safety constraint conditions include the surrounding rock convergence threshold and the bearing safety factor of the support structure; The construction parameter regulation instructions include the tunneling rate adjustment instruction and the support timing optimization instruction; The generation method of the construction parameter regulation instruction includes: Based on the coupled feature matrix, the output layer of the policy network adopts a branch structure to generate the construction parameter regulation instruction. The generation method of the construction parameter regulation instruction includes: The tunneling rate instruction is processed by the time series smoothing algorithm to limit the speed adjustment amplitude within adjacent decision cycles; The support timing instruction is realized through the event trigger mechanism. When the rock mass damage factor exceeds the critical value, start the multi-level support collaborative deployment plan; the multi-level support collaborative deployment plan includes primary support, secondary support and tertiary support.

[0009] Furthermore, it also includes: Filter and denoise the multi-source heterogeneous data collected by the edge computing node; Synchronously update the dynamic mapping relationship between the physical tunnel and the virtual model based on the digital twin platform; Overlay and display the deviation analysis of the real-time monitoring data and the simulation prediction results in the visualization interface.

[0010] Tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control system, the system includes: Synchronous sensing module, which collects multi-source heterogeneous data in real time through intelligent drilling equipment, distributed optical fiber sensor networks and mobile scanning devices deployed on the tunnel construction surface; Feature extraction module: The feature extraction module inputs multi-source heterogeneous data into the spatiotemporal graph convolutional network, constructs a spatiotemporal correlation model between geological features and construction disturbance parameters, and outputs a coupling feature matrix; Dynamic simulation module: Based on the coupling characteristic matrix, the dynamic simulation module simulates the dynamic response of the surrounding rock-support system through the flow voxelization engine and the multi-physics field coupling engine to generate a three-dimensional thermal map of risk evolution; Intelligent decision-making module: The intelligent decision-making module inputs the three-dimensional heat map of risk evolution into the deep reinforcement learning model, and dynamically optimizes and generates construction parameter control instructions that meet safety constraints.

[0011] The technical effects and advantages of the tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method and system provided by the present invention are as follows: The technical closed loop of the present invention opens up the full-link real-time channel of "data perception-mechanical analysis-decision execution", providing an intelligent control paradigm with both robustness and sensitivity for complex geological tunnel construction, and fundamentally avoiding the dual risks of "data distortion and misjudgment" and "response lag and loss of control" in traditional methods; the present invention aims at the feature interference problem under the coupling of geology and construction, and uses a dynamic feature decoupling algorithm to separate construction disturbance and geological intrinsic response in multimodal data such as sound wave attenuation, deformation rate, vibration spectrum, etc., to solve the feature confusion caused by time dislocation and spatial superposition in traditional methods, and improve the accuracy of surrounding rock fracture development judgment; based on the dynamic numerical modeling technology driven by streaming data, a millisecond-level bidirectional mapping channel of surrounding rock stress-seepage field and excavation parameters is established, and real-time working condition parameters such as tool penetration and support reaction force are synchronously integrated, and the minute-level delay of traditional offline modeling is compressed to the construction beat synchronization level, so as to achieve seamless connection between advanced warning and dynamic parameter adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of the tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method in Example 1; Figure 2 This is a flow chart of the tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method in Example 2; Figure 3 This is a connection diagram of the tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control system in Example 3. DETAILED DESCRIPTION

[0013] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0014] Embodiment 1: Please refer to Figure 1 As shown, the embodiment of the present invention provides a method for intelligent prediction and control of tunnel multi-source fusion dynamic twin surrounding rock, and the method includes: Real-time collection of multi-source heterogeneous data through intelligent drilling equipment, distributed optical fiber sensing network and mobile scanning device deployed on the tunnel construction surface; Input the multi-source heterogeneous data into the spatio-temporal graph convolutional network, construct a spatio-temporal correlation model of geological features and construction disturbance parameters, and output a coupled feature matrix; Based on the coupled feature matrix, simulate the dynamic response of the surrounding rock-support system through a streaming voxelization engine and a multi-physics field coupling engine, and generate a three-dimensional thermal map of risk evolution; Input the three-dimensional thermal map of risk evolution into the deep reinforcement learning model, and dynamically optimize and generate construction parameter regulation instructions that meet the safety constraint conditions.

[0015] The multi-source heterogeneous data includes geological structure parameters, surrounding rock deformation data and three-dimensional spatial form data; The intelligent drilling equipment includes an axial thrust detection module and a rock acoustic wave feature acquisition module; the distributed optical fiber sensing network is arranged along the tunnel excavation direction and includes a temperature compensation type strain sensing unit; the mobile scanning device uses a multi-line lidar to realize the three-dimensional point cloud reconstruction of the construction surface.

[0016] The data collection link realizes collaborative perception through three types of core devices; As the core device for geological structure detection, the intelligent drilling equipment's built-in axial thrust detection module can real-time monitor the dynamic resistance change during the drill pipe propulsion process. At the same time, through the rock acoustic wave feature acquisition module, time-frequency domain analysis is carried out on the acoustic wave signals generated during the drilling operation. Based on the acoustic wave propagation speed, energy attenuation characteristics and main frequency shift law, comprehensively judge the distribution of rock mass fissures and the structural stability; The distributed optical fiber sensing network is continuously arranged along the tunnel construction surface excavation direction. Its temperature compensation type strain sensing unit adopts a dual-core optical fiber architecture. Among them, the main core optical fiber is used to measure the surrounding rock strain distribution, and the auxiliary core optical fiber eliminates the influence of environmental temperature fluctuations on the measurement results through thermal isolation packaging technology, and realizes the deformation monitoring accuracy at the micro-strain level; The mobile scanning device is equipped with a multi-line lidar module, which performs three-dimensional spatial scanning of the construction surface based on the time-of-flight ranging principle. High-density point cloud data is obtained through a non-uniformly distributed laser beam array and a high-frequency scanning mechanism. Combining with the SLAM (Simultaneous Localization and Mapping) algorithm, the dynamic reconstruction of the three-dimensional shape of the construction surface is realized, providing a sub-centimeter-level accurate spatial reference for the surrounding rock convergence analysis.

[0017] The three types of devices achieve the alignment of data acquisition timings through a unified time-domain synchronization protocol, ensuring the spatio-temporal consistency of geological parameters, deformation data, and spatial morphological information.

[0018] The construction method of the spatio-temporal graph convolutional network includes: Mapping the distribution of structural planes in geological exploration data into the spatial topological relationship of the graph structure; Using a temporal convolutional layer to extract the temporal propagation characteristics of the vibration parameters of construction machinery; Fusing the correlation weight distribution of geological parameters and construction parameters through a multi-head attention mechanism.

[0019] In the construction of the spatio-temporal correlation model, first, based on the distribution characteristics of structural planes (including fissures, joints, and bedding planes) in geological exploration data, a spatial topological graph structure is established.

[0020] Specifically, taking the spatial proximity and attitude similarity between adjacent structural planes as edge weights, and the geometric center of the structural plane as the graph node, an initial graph model reflecting the spatial correlation of geological structures is constructed. On this basis, for dynamic disturbance parameters such as construction machinery vibration and support stress loading, a stacked temporal convolutional layer is used to extract features from the temporal data, and a dilated convolutional kernel cross-layer connection structure is used to capture the long-range temporal dependence relationships of parameters such as vibration acceleration and energy spectral density.

[0021] Furthermore, a multi-head attention mechanism is introduced to map geological parameters (such as rock mass strength and structural plane density) and construction parameters (such as tunneling speed and grouting pressure) to multiple groups of independent feature subspaces. By parallel computing the mutual information correlation degree between different parameters, a geological-construction coupling feature weight matrix is generated.

[0022] Finally, through graph convolution operations, the spatio-temporal features and the weight matrix are fused, and a coupling feature matrix including the evolution law of geological conditions and the propagation characteristics of construction disturbances is output, providing a high-dimensional feature representation for subsequent mechanical response simulation.

[0023] In the process of generating the three-dimensional thermal map of risk evolution, the streaming voxelization engine dynamically adjusts the resolution of the calculation grid according to the surrounding rock stress gradient; The multi-physics field coupling engine includes the interaction interface between the discrete element module and the fluid dynamics module; The three-dimensional thermal map of risk evolution includes a plastic strain distribution layer and a support stress warning layer.

[0024] The streaming voxelization engine dynamically reconstructs the computational grid using an octree structure based on the real-time updated surrounding rock stress gradient field. Specifically, when the local stress gradient value exceeds a preset threshold (e.g., > 2 MPa / m), the grid encryption mechanism is triggered, and the basic grid is recursively subdivided to a resolution of 0.125 m³. In the area where the stress gradient change rate exceeds 5% / min, the anisotropic subdivision mode is enabled to ensure that the aspect ratio of the grid in the fracture propagation direction is controlled within 3:1. At the same time, a virtual buffer layer is used to achieve a smooth transition between grids of different resolutions and avoid numerical oscillations.

[0025] The multi-physics coupling engine realizes cross-scale interaction through a discrete element - fluid dynamics two-way coupling interface. The discrete element module uses the Hertz-Mindlin contact model to simulate the motion of jointed rock mass particles and outputs the porosity change matrix in the fracture zone in real time. The fluid dynamics module calculates the fracture seepage field based on the Brinkman equation and feeds back the permeability tensor to the discrete element system through a custom field mapping protocol. The two adopt an asynchronous time-step coupling strategy and achieve cross-time-scale data synchronization through the momentum interpolation algorithm.

[0026] The three-dimensional thermal map of risk evolution integrates the plastic strain field and support monitoring data through a feature fusion algorithm: the plastic strain distribution layer calculates the equivalent plastic strain using the J2 flow criterion and maps the strain values in the HSV color space (red corresponds to the critical failure area > 5%); the support stress warning layer fuses multi-source data such as bolt axial force and lining concrete strain through a Bayesian probability model. When the local support stress exceeds 80% of the material yield strength, a radial warning halo is generated, and the halo radius R and the overrun degree Δσ satisfy: R = 0.2Δσ 1.5 (unit: m), realizing the visual expression of the spatio-temporal cumulative effect of the risk situation.

[0027] During the generation process of the three-dimensional thermal map of risk evolution, real-time calibration operations are required, including: Dynamically comparing the surrounding rock strain data real-time monitored by the distributed fiber optic sensing network with the simulated strain values in the three-dimensional thermal map of risk evolution; Through the streaming data processing pipeline, spatio-temporal registration is performed on the surrounding rock strain time series signal collected by the distributed fiber optic sensing network and the simulated strain field output by the three-dimensional thermal map of risk evolution; the feature space projection algorithm is used to calculate the vector similarity between the multi-dimensional strain tensors (including axial strain and shear strain components) at the monitoring points and the simulated values of the corresponding spatial coordinates, generating a difference field distribution matrix with time scale alignment characteristics.

[0028] When the deviation between the monitoring data and the simulation data exceeds the preset threshold, the weight re-training mechanism of the spatio-temporal graph convolutional network is triggered; When the statistical deviation (including root mean square error and peak-valley offset) between the monitoring value and the simulation value exceeds the preset dynamic threshold, the online learning module of the spatio-temporal graph convolutional network is activated. By constructing a mixed training set including real-time monitoring data augmented samples, an incremental retraining is carried out on the multi-head attention weights and graph convolutional kernel parameters in the network using a sliding time window strategy, and the correlation dimension of the geological-construction coupling feature matrix is updated synchronously; Modify the constitutive model parameters of the multi-physical field coupling engine based on the updated coupling feature matrix; Input the updated coupling feature matrix into the parameter optimizer of the multi-physical field coupling engine. Based on the tensor decomposition method, extract the rock mass damage evolution characteristics and the fluid-solid coupling action mode. By constructing an implicit mapping relationship between the constitutive equation parameters and the high-dimensional features, adopt a non-intrusive correction strategy to dynamically adjust the key parameters such as the hardening modulus and permeability coupling coefficient in the elastoplastic constitutive model, and form a monitoring data-driven model self-evolution mechanism.

[0029] Achieve millisecond-level response through a microservice architecture to ensure the dynamic consistency between the numerical model and the physical reality.

[0030] The action space of the deep reinforcement learning model is defined as a set of continuous construction control parameters; Map the construction control parameters to a high-dimensional continuous action vector, including adjustable variables such as the propulsion speed of the tunneling system, the cutterhead torque distribution ratio, and the activation time window of the temporary support; compress the construction parameter regulation instructions to the standardized space of [-1, 1], and adopt the Twin Delayed Deep Deterministic Policy Gradient algorithm (TD3) to achieve action noise suppression and policy stability improvement.

[0031] The safety constraint conditions include the surrounding rock convergence threshold and the bearing safety factor of the support structure; Fuse the real-time monitoring data of the surrounding rock convergence rate and the stress-strain relationship matrix of the support structure in the state observation space to construct a constraint violation degree evaluation function; when the monitoring data approaches the convergence threshold, dynamically adjust the exploration boundary of the policy network by the Lagrange multiplier method, and at the same time introduce a quadratic penalty term in the reward function to suppress the action strategy that may lead to the degradation of the support safety factor.

[0032] The construction parameter regulation instructions include the tunneling rate adjustment instruction and the support timing optimization instruction; The generation method of the construction parameter regulation instructions includes: Based on the coupling feature matrix, the output layer of the policy network adopts a branch structure to generate the construction parameter regulation instructions. The generation method of the construction parameter regulation instructions includes: The tunneling rate instruction is processed by a time series smoothing algorithm to limit the speed adjustment amplitude within adjacent decision cycles; The support timing instruction is implemented through an event-trigger mechanism. When the rock mass damage factor exceeds the critical value, a multi-level support collaborative deployment plan is initiated. The multi-level support collaborative deployment plan includes primary support, secondary support, and tertiary support. Primary support: Flexible support using shotcrete + short rock bolts is rapidly applied within 30 minutes after the excavation face is exposed to control the initial deformation of the surrounding rock (e.g., when the fiber optic monitors 0.15% strain, the robotic arm shotcrete operation is automatically triggered). Secondary support: Rigid support with steel arch frames + long cable bolts is initiated after the face advances 5 times the tunnel diameter. The cable bolt tensile force is dynamically adjusted through prestress monitoring (e.g., when the convergence rate > 2 mm / d, the 200 kN class prestressed cable bolt is activated). Tertiary support: Permanent support with reinforced concrete lining + grouting reinforcement is set, and the optimal construction timing is determined according to the long-term deformation trend pre-acted by the BIM model.

[0033] The multi-level support collaborative deployment includes spatial collaboration, temporal collaboration, and mechanical collaboration. Spatial collaboration: The stiffness matching of different support layers is achieved through the support parameter mapping matrix (e.g., the shotcrete thickness h and the rock bolt spacing s satisfy the stiffness coupling condition of h / s ≤ 0.3). Temporal collaboration: A support activation time window function is established to ensure that the construction interval between adjacent support layers meets the surrounding rock creep equation (critical creep time). Mechanical collaboration: A distributed fiber optic sensing network is used to monitor the internal force redistribution of the support structure in real time, and the load sharing ratio of each support layer is adjusted through a PID controller.

[0034] After the control instruction is generated, it is filtered through a physical constraint verification module to filter out abnormal outputs that violate the equipment operating conditions boundary to ensure the executability of the instruction.

[0035] Example 2: As Figure 2 shown, this example further improves the design based on Example 1. The difference is that in the actual operation of Example 1, it is found that there is high-frequency noise interference in the original monitoring data, resulting in deviation in feature extraction, and there is a time delay effect in the dynamic mapping between the virtual model and the physical entity, which will cause system prediction inaccuracy and decision-making lag, ultimately leading to the failure of surrounding rock deformation control and a doubling of the tunnel construction safety risk. Based on this, the tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method further includes: Filtering and noise reduction processing of the multi-source heterogeneous data (geological structure parameters, surrounding rock deformation data, and three-dimensional spatial form data) collected through edge computing nodes. Synchronously updating the dynamic mapping relationship between the physical tunnel and the virtual model based on the digital twin platform. Overlaying and displaying the deviation analysis of the real-time monitoring data and the simulation prediction results in the visualization interface.

[0036] In the specific implementation process of the tunnel intelligent construction system, through the deep integration of edge computing and digital twin technology, a full-chain closed-loop control system from data collection to decision support was constructed. The core implementation process of the system is as follows: In the data preprocessing stage, edge computing devices (a miniaturized computing unit deployed at the construction site) are arranged at key nodes along the tunnel to clean the multi-source heterogeneous data collected by the shield machine sensors in real time. Adaptive filtering algorithms are used to eliminate signal noise caused by mechanical vibration. For example, for high-frequency interference in the cutter head torque monitoring data, effective signal components are separated by wavelet transform technology. At the same time, a sliding time window statistical detection mechanism is set up. When a statistically abnormal fluctuation in the pressure sensor reading is identified (exceeding the range of three times the standard deviation), the data re-collection process is automatically triggered to ensure the reliability of the input data.

[0037] In the model dynamic mapping stage, a virtual-real synchronization mechanism for tunnel engineering is established based on digital twin technology (a technology that reflects the state of physical entities through virtual models); a high-fidelity three-dimensional model including surrounding rock structure, support system, and construction machinery is constructed in virtual space, in which the rock and soil body adopts an elastic-plastic constitutive model to describe its stress-strain relationship, and the support structure simulates its collaborative deformation characteristics through beam-shell coupling units. The system receives on-site preprocessing data at specific intervals and dynamically corrects the geomechanical parameters in the model through the inverse parameter identification algorithm. When the monitoring data deviates significantly from the simulated prediction value, the grid adaptive encryption technology is used to reconstruct the local model of the key area. For example, when constructing in a soft rock area, the virtual model automatically adjusts the rock mass Poisson's ratio parameters to improve the settlement prediction accuracy to within the allowable error range of the project.

[0038] During the visualization interaction stage, a three-dimensional visualization platform was developed to realize the spatial integrated display of multi-source information. The platform aligns the real-time monitoring data stream (such as shield thrust pressure, segment displacement, etc.) with the digital twin simulation results in time and space, and uses chromaticity mapping technology to generate a superimposed display of thermal maps of the surrounding rock stress field; the system has a built-in deviation analysis module, which calculates the relative error matrix between the measured values ​​and the simulated values, and marks abnormal areas with pulsed halos on the surface of the tunnel model; engineers can use interactive cutting tools to compare the actual deformation curve of any section with the predicted trend line. When a systematic deviation in the stress distribution of the support structure is identified (for example, a monitoring found that the deviation in the vault area continued to exceed 25%), the system automatically pushes a grouting reinforcement recommendation plan.

[0039] Embodiment 3: Figure 3As shown, based on the same inventive concept as the intelligent prediction and control method for tunnel multi-source fusion dynamic twin surrounding rock in the foregoing embodiments, the present application provides a tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control system. The system and method embodiments in the present application are based on the same inventive concept. Among them, the system includes: A synchronous perception module, which collects multi-source heterogeneous data in real time through intelligent drilling equipment, distributed optical fiber sensing networks, and mobile scanning devices deployed on the tunnel construction surface; A feature extraction module, which inputs the multi-source heterogeneous data into a spatio-temporal graph convolutional network, constructs a spatio-temporal correlation model of geological features and construction disturbance parameters, and outputs a coupled feature matrix; A dynamic simulation module, which, based on the coupled feature matrix, simulates the dynamic response of the surrounding rock-support system through a streaming voxelization engine and a multi-physics field coupling engine to generate a three-dimensional thermal map of risk evolution; An intelligent decision-making module, which inputs the three-dimensional thermal map of risk evolution into a deep reinforcement learning model and dynamically optimizes and generates construction parameter regulation instructions that meet the safety constraint conditions.

[0040] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

[0041] The above-mentioned are only the preferred specific embodiments of the embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application, according to the technical solution and its concept of the present application, makes equivalent substitutions or changes, and should be covered by the protection scope of the present application.

Claims

1. The tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control method is characterized by: include: Through the intelligent drilling equipment, distributed fiber optic sensor network and mobile scanning devices deployed on the tunnel construction surface, multi-source heterogeneous data can be collected in real time; Input multi-source heterogeneous data into the spatiotemporal graph convolutional network, build a spatiotemporal correlation model between geological characteristics and construction disturbance parameters, and output a coupling feature matrix; Based on the coupling characteristic matrix, the dynamic response of the surrounding rock-support system is simulated through the flow voxelization engine and the multi-physics field coupling engine to generate a three-dimensional thermal map of risk evolution; The three-dimensional heat map of risk evolution is input into the deep reinforcement learning model to dynamically optimize and generate construction parameter control instructions that meet safety constraints.

2. The method according to claim 1, characterized in that: The intelligent drilling equipment includes an axial thrust detection module and a rock acoustic wave feature acquisition module; the distributed optical fiber sensor network is arranged along the tunnel excavation direction and includes a temperature-compensated strain sensor unit; the mobile scanning device uses a multi-line laser radar to achieve three-dimensional point cloud reconstruction of the construction surface.

3. The method according to claim 1, characterized in that The construction method of spatiotemporal graph convolutional network includes: Map the structural surface distribution in geological exploration data into the spatial topological relationship of the graph structure; The temporal convolution layer is used to extract the temporal propagation characteristics of the vibration parameters of the construction machinery; The associated weight distribution of geological parameters and construction parameters is fused through a multi-head attention mechanism.

4. The method according to claim 1, characterized in that The fluid voxelization engine dynamically adjusts the computational grid resolution according to the surrounding rock stress gradient; the multi-physics field coupling engine includes an interactive interface between a discrete element module and a fluid dynamics module; and the risk evolution three-dimensional thermal map includes a plastic strain distribution layer and a support stress warning layer.

5. The method according to claim 1, characterized in that Real-time calibration operations include: Dynamically comparing surrounding rock strain data monitored in real time by the distributed optical fiber sensing network with simulated strain values ​​in the risk evolution three-dimensional thermal map; When the deviation between the monitoring data and the simulation data exceeds a preset threshold, a weight retraining mechanism of the spatiotemporal graph convolutional network is triggered; The constitutive model parameters of the multi-physics coupling engine are modified based on the updated coupling characteristic matrix.

6. The method according to claim 1, characterized in that The action space of the deep reinforcement learning model is defined as a set of continuous construction control parameters, and the construction control parameters are mapped into high-dimensional continuous action vectors; the construction parameter control instructions are compressed into a standardized space of [-1, 1], and a double-delayed deep deterministic policy gradient algorithm is adopted; The safety constraint conditions include the surrounding rock convergence threshold and the support structure bearing safety factor; The construction parameter control instructions include excavation rate adjustment instructions and support timing optimization instructions; The method for generating the construction parameter control instruction includes: Based on the coupling feature matrix, the output layer of the strategy network uses a branch structure to generate construction parameter control instructions. The method for generating construction parameter control instructions includes: The tunneling rate command is processed through a time series smoothing algorithm to limit the speed adjustment range within adjacent decision cycles; The support timing instruction is realized through an event trigger mechanism. When the rock damage factor exceeds the critical value, the multi-level support coordinated deployment plan is activated; the multi-level support coordinated deployment plan includes primary support, secondary support and tertiary support.

7. The method according to claim 1, characterized in that Also includes: Filter and reduce noise on the collected multi-source heterogeneous data through edge computing nodes; Synchronously update the dynamic mapping relationship between the physical tunnel and the virtual model based on the digital twin platform; The deviation analysis of real-time monitoring data and simulation prediction results is superimposed and displayed in the visualization interface.

8. The tunnel multi-source fusion dynamic twin surrounding rock intelligent prediction and control system is characterized by: The system includes: Synchronous sensing module, which collects multi-source heterogeneous data in real time through intelligent drilling equipment, distributed optical fiber sensor networks and mobile scanning devices deployed on the tunnel construction surface; Feature extraction module: The feature extraction module inputs multi-source heterogeneous data into the spatiotemporal graph convolutional network, constructs a spatiotemporal correlation model between geological features and construction disturbance parameters, and outputs a coupling feature matrix; Dynamic simulation module: Based on the coupling characteristic matrix, the dynamic simulation module simulates the dynamic response of the surrounding rock-support system through the flow voxelization engine and the multi-physics field coupling engine to generate a three-dimensional thermal map of risk evolution; Intelligent decision-making module: The intelligent decision-making module inputs the three-dimensional heat map of risk evolution into the deep reinforcement learning model, and dynamically optimizes and generates construction parameter control instructions that meet safety constraints.

Images