A grouting diffusion prediction method and system with complex geological multi-attribute constraints

Abstract

The present invention discloses a method and system for predicting grouting diffusion for complex geological conditions with multi-attribute constraints. The prediction method comprises: obtaining multi-source attribute information of a complex geological body, the multi-source attribute information including fracture characteristics, pore characteristics, and water inflow characteristics; inputting the multi-source attribute information into a trained dynamic prediction model for grouting diffusion, and outputting a prediction result for the complex geological body, the grouting diffusion result including permeability, pressure distribution, and boundary conditions. During the training process of the dynamic prediction model for grouting diffusion, the present invention combines a deep learning model with a real-time monitoring system, continuously performs predictions based on monitoring data during the grouting process, and adjusts the prediction results through a feedback mechanism, optimizing the prediction model and improving the accuracy and reliability of the prediction results.

Description

Technical Field

[0001] The present invention relates to the technical field of underground engineering grouting diffusion simulation, and in particular to a grouting diffusion prediction method and system under complex geological multi-attribute constraints. Background Art

[0002] The problem of grouting diffusion in complex geological environments has always been a key technical challenge in engineering investigation and underground construction. Traditional grouting diffusion prediction methods usually focus on a single physical parameter (such as permeability or porosity), often ignoring the complex multi-attribute coupling effects within the geological body. However, the physical parameters such as cracks, pores, and water inflow in actual geological bodies have significant spatial and temporal variations, which makes it impossible for single-attribute models to fully and accurately simulate the diffusion behavior of slurry. With the increasing complexity of geological bodies, the accuracy and reliability of traditional models in practical applications are limited. In recent years, researchers have tried to compensate for these shortcomings through multi-source data fusion and dynamic coupling models, especially in the coupling analysis of multiple factors such as cracks, pores and water inflow. Multi-attribute coupling models are based on multi-source data such as resistivity and porosity, and can more realistically reflect the diffusion process of slurry in complex geological bodies, but they still face technical difficulties such as real-time monitoring and dynamic feedback control.

[0003] Existing grouting diffusion simulations are mostly based on simplified physical models. While computationally efficient, these approaches cannot meet the requirements for accurate simulation and real-time performance in complex geological structures. The rise of deep learning technology has provided a new solution for processing and predicting multi-source data. Deep learning methods, such as neural networks, can fuse data from various sensors (such as CT imaging, laser scanning, and seismic waves) to dynamically calculate and predict physical properties in complex geological structures. This approach offers significant advantages, particularly in accurately predicting slurry diffusion paths. However, current deep learning-based grouting diffusion models still face challenges such as data acquisition difficulties, insufficient training samples, and difficulties in real-time monitoring, and have yet to be widely adopted in industrial practice. In terms of experimental platforms and monitoring systems, recent research has constructed a variety of geological models and monitored the grouting diffusion process using real-time monitoring equipment (such as pressure sensors, resistivity imaging, and flow meters). These experimental setups provide a more intuitive simulation environment for research, but they still have limitations, such as insufficient fusion of multi-source data and difficulty in fully simulating complex geological structures. Therefore, despite certain technical accumulation, the existing grouting diffusion prediction methods and experimental platforms still need to be further improved, especially in terms of data fusion, model optimization and real-time feedback control. Summary of the Invention

[0004] In response to the problems existing in the existing technology, the present invention provides a grouting diffusion prediction method and system with complex geological multi-attribute constraints. Through deep learning and multi-source data fusion methods, the simulation results can more accurately reflect the slurry diffusion process in complex geological bodies.

[0005] The technical solutions of the present invention are as follows:

[0006] In a first aspect of the present invention, a method for grouting diffusion prediction under complex geological multi-attribute constraints is provided, which is characterized by comprising:

[0007] Acquiring multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics;

[0008] Inputting multi-source attribute information into a trained grouting diffusion dynamic prediction model to output grouting diffusion prediction results for complex geological bodies, the grouting diffusion results including permeability, pressure distribution, and boundary conditions;

[0009] The establishment and training process of the grouting diffusion dynamic prediction model is as follows:

[0010] Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions;

[0011] Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model;

[0012] Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model;

[0013] After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment;

[0014] During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through a feedback mechanism based on the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model.

[0015] In the second aspect of the present invention, a grouting diffusion prediction system with complex geological multi-attribute constraints is provided.

[0016] The multi-attribute information acquisition module is configured to: acquire multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics;

[0017] The grouting diffusion prediction module is configured to: input multi-source attribute information into a trained grouting diffusion dynamic prediction model, and output grouting diffusion prediction results for complex geological bodies, wherein the grouting diffusion results include permeability, pressure distribution, and boundary conditions;

[0018] The establishment and training process of the grouting diffusion dynamic prediction model is as follows:

[0019] Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions;

[0020] Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model;

[0021] Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model;

[0022] After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment;

[0023] During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through a feedback mechanism based on the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model.

[0024] One or more technical solutions of the present invention have the following beneficial effects:

[0025] (1) The present invention conducts coupling analysis based on multi-source information (crack characteristics, pore characteristics and water gushing characteristics) to dynamically simulate and predict changes in permeability and diffusion state during the grouting diffusion process. In the adopted grouting dynamic prediction model, a multi-attribute coupling permeability model and a dynamic equation are embedded as physical constraints to ensure that the output of the prediction model conforms to the laws of fluid mechanics. In addition, the dynamic changes in the permeability of the constructed experimental system over time and space will be monitored in real time, and updated to the input layer of the prediction model through a feedback mechanism to further optimize the spatiotemporal coupling characteristics and accurately simulate the diffusion path and speed of the slurry. In this way, the multi-attribute coupling permeability model provides physical consistency constraints and real-time update support for dynamic prediction, thereby improving the accuracy and reliability of the prediction results. The prediction model can accurately predict permeability changes and slurry diffusion behavior, and can be widely used in geological multi-attribute simulation and grouting process optimization in geological exploration, tunnel construction, scientific research experiments and other fields, with good scalability, adaptability and stability.

[0026] (2) The multi-attribute coupled permeability model proposed in this invention constructs a dynamic coupled permeability prediction model by introducing multiple physical parameters such as fracture characteristics, pore characteristics, and water inflow characteristics. These parameters are integrated and used as key constraints and dynamic input features in the prediction model. The model establishes a quantitative relationship between multi-source attributes and permeability by dynamically coupling fracture characteristics, pore characteristics, and water inflow characteristics, revealing the mechanism of their influence on permeability changes. In the prediction model, the output of the coupled permeability model—permeability data—is used as a physical constraint and embedded in the physical constraint integration layer of the grouting diffusion dynamic prediction model to ensure that the model output conforms to the laws of fluid mechanics.

[0027] (3) The dynamic grouting diffusion simulation model proposed in the present invention integrates multiple physical properties and uses deep learning and multi-source data fusion methods to enable the simulation results to more accurately reflect the slurry diffusion process in complex geological bodies. At the same time, this model can be dynamically updated and optimized with the input of real-time data, overcoming the static and simplified nature of traditional models. The dynamic grouting diffusion simulation model can provide accurate slurry diffusion predictions, can adapt to different geological conditions and changes in different physical parameters, and can adjust and optimize the diffusion process in real time. In the case of missing data or insufficient real-time monitoring data, the model can still fill in the missing information through small-sample learning and prediction to ensure the continuity and stability of the simulation process.

[0028] (4) The dynamic grouting diffusion simulation model proposed in the present invention compares the actual monitoring data (such as pressure, flow, and concentration) with the prediction results through a real-time feedback mechanism, reversely adjusts the input parameters or weights of the permeability model, and optimizes the accuracy of the permeability prediction. This two-way coupling enables the permeability model and the diffusion prediction model to support each other, and jointly improves the dynamic simulation and prediction accuracy of the slurry diffusion behavior in complex geological bodies. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart for establishing, training and optimizing the grouting diffusion dynamic prediction model in Example 1 of the present invention;

[0030] Figure 2 Schematic diagram of the grouting diffusion simulation experimental system in Example 2 of the present invention;

[0031] In the figure: 1. Complex geological simulation unit; 2. Water injection and grouting unit. DETAILED DESCRIPTION

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] Example 1

[0034] In a typical embodiment of the present invention, a grouting diffusion prediction method with complex geological multi-attribute constraints is proposed, comprising:

[0035] Step 1: Acquire multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics.

[0036] When obtaining crack characteristics, a 3D laser scanner and a cave scanner are used to measure the crack characteristics of the rock mass, generate 3D point cloud data of the cracks, and use a handheld scanner to perform additional scanning of local details;

[0037] When acquiring pore characteristics, a microscopic CT imager is used to obtain porosity data and establish a microscopic model of pore distribution. In combination with a pore water pressure gauge, the pressure changes of the fluid inside the pores are monitored in real time to obtain correlation data between pores and fractures.

[0038] When obtaining the characteristics of water inflow, resistivity equipment and ground-penetrating radar are used to measure the distribution information of stratum moisture and water inflow area, and the permeability of different layers is measured through the groundwater stratification test sampling system. The overall strength and crushing structure characteristics of the rock mass are obtained in combination with the elasticity meter. Flow meters and pressure sensors are arranged at the tunnel face and inside the model tunnel to obtain dynamic permeability data.

[0039] Step 2: Input the multi-source attribute information into the trained grouting diffusion dynamic prediction model to output the grouting diffusion prediction results of the complex geological body, wherein the grouting diffusion results include the permeability, pressure distribution and boundary conditions;

[0040] The establishment and training process of the grouting diffusion dynamic prediction model is as follows:

[0041] Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions;

[0042] Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model;

[0043] Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model;

[0044] After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment;

[0045] During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through a feedback mechanism based on the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model.

[0046] Specifically, when establishing a multi-attribute coupled permeability model, statistical and machine learning methods are used to model the coupling relationship between multiple attribute information to reveal the influence mechanism of each physical attribute on permeability.

[0047] The multi-attribute coupled permeability model uses key parameters such as fracture characteristics, pore characteristics, and water inflow characteristics to establish a constraint relationship between fractures, pores, and permeability through the dynamic coupling of multi-attribute information. This model integrates various geological information (such as fracture characteristics, pore characteristics, and water inflow characteristics) to reflect their combined impact on formation permeability.

[0048] The process of establishing a multi-attribute coupled permeability model:

[0049] Data collection and attribute selection: Collect multiple attribute data of geological bodies (such as fracture characteristics, pore characteristics and water inflow characteristics, etc.).

[0050] Coupling analysis: Use statistical and machine learning methods (such as regression analysis and support vector machines) to model the coupling relationship between multiple attributes and reveal the impact mechanism of each physical attribute on permeability.

[0051] Dynamic evolution model: Build a permeability prediction model based on time changes (multi-attribute coupled permeability model), use real-time data monitoring to make dynamic adjustments, and solve the problem of permeability changes of geological bodies under different working conditions.

[0052] The final model:

[0053] This permeability prediction model provides permeability estimates under varying conditions and can adjust predicted permeability values ​​in real time, enabling more accurate prediction of slurry diffusion paths and velocities within complex geological structures. This permeability prediction model can be used not only for static permeability analysis but also for dynamic optimization and adjustment based on real-time data input.

[0054] The multi-attribute coupled permeability model is integrated into the grouting diffusion dynamic prediction model as a key constraint and dynamic input feature. By dynamically coupling fracture characteristics, pore characteristics, and water inflow characteristics, the multi-attribute coupled permeability model establishes a quantitative relationship between multiple source attributes and permeability, revealing the mechanisms that influence permeability changes. In the grouting diffusion dynamic prediction model, the output of the multi-attribute coupled permeability model—permeability data—is embedded within the model's physical constraint integration layer as a physical constraint, ensuring that the model output conforms to the laws of fluid mechanics (such as Darcy's law and the diffusion equation).

[0055] In this embodiment, the structure of the grouting diffusion dynamic prediction model includes two parts: a physical model and a deep learning model.

[0056] Physical model part: Establish physical equations to describe the grouting diffusion process, consider multiple physical properties (such as permeability, pressure distribution, porosity, fracture structure, etc.), and provide constraints for the deep learning model.

[0057] Deep learning model part: Based on neural network architecture (such as convolutional neural network CNN and long short-term memory network LSTM, etc.), it processes multi-source data from different sensors (such as pressure, temperature, resistivity, etc.) and combines physical constraints for learning and prediction.

[0058] In the physical model, permeability, pressure distribution, and diffusion boundaries are key physical properties that directly affect the slurry diffusion process. Dynamic constraint equations are used to limit the range of neural network output results through the physical model and ensure that they conform to fluid mechanics and geological principles. Specific constraint equations include:

[0059] Darcy's law: The relationship between the velocity and pressure gradient that limit the flow of a slurry.

[0060] Fluid dynamics equations: coupling of limiting pressure distribution and permeability.

[0061] Diffusion equation: describes the relationship between the diffusion rate of slurry and the boundary.

[0062] Multiphysics coupling: Combine multiple properties such as permeation rate, pressure distribution, and boundary conditions to dynamically update physical constraints.

[0063] These physical properties are coupled with dynamic constraint equations, and physical constraints such as Darcy's law, fluid dynamics equations, and diffusion equations are coupled with deep learning models: physical equations are used as constraints for the neural network and embedded into the training process through physical constraint loss functions to ensure that the prediction results conform to physical laws; physical variables (such as permeation rate, pressure distribution, and boundary conditions) are passed as input features to the neural network and participate in learning and prediction together with other data.

[0064] Specifically, the permeation rate is coupled with Darcy's law, the pressure distribution is coupled with the fluid mechanics equation, and the diffusion boundary is coupled with the diffusion equation, including:

[0065] 1. Coupling of permeation rate and Darcy’s law

[0066] The expression of Darcy's law is:

[0067]

[0068] Where: q: permeation rate (flow); k: permeability; Pressure gradient; μ: fluid viscosity.

[0069] Specific coupling methods include:

[0070] (1) Physical Constraint Layer

[0071] In the deep learning model, a "physical constraint layer" is introduced to embed Darcy's law into the network's training loss function. The penetration rate q predicted by the neural network is pred and the permeation rate q calculated according to Darcy's law phys The error between them gives the first physical loss function:

[0072]

[0073] This first loss function affects the network weight update and ensures that the permeation rate of the network output complies with Darcy's law.

[0074] (2) Input feature enhancement:

[0075] The permeability k and pressure gradient Directly use it as input features and add it to the input layer of the deep learning model.

[0076] The network is trained using these input features along with other geological attributes (fractures, porosity, temperature, etc.) to achieve more accurate predictions.

[0077] 2. Coupling of pressure distribution and fluid dynamics

[0078] The fluid dynamics equation (pressure field equation) is as follows:

[0079]

[0080] Specific coupling method:

[0081] (1) Physical consistency loss:

[0082] The pressure field P output by the network pred The above equation needs to be satisfied and is implemented through the second physical constraint loss function:

[0083]

[0084] Where, P pred represents the pressure field predicted by the neural network;

[0085] During the network training process, the second physical constraint loss function enables the model to learn the pressure distribution that conforms to the laws of fluid dynamics.

[0086] (2) Input feature integration:

[0087] The permeability k and other parameters in the fluid dynamics equation (such as boundary conditions and initial pressure distribution) are used as input features and learned in combination with historical spatiotemporal data.

[0088] (3) Numerical simulation correction:

[0089] Use CFD (computational fluid dynamics) and other methods to solve fluid dynamics equations and generate high-quality training samples to guide the supervised learning of the network.

[0090] 3. Coupling of diffusion boundaries and spatial physical constraints

[0091] The diffusion equation (Fick's law) is:

[0092]

[0093] Where: C: concentration field; u: velocity field; D: diffusion coefficient.

[0094] Specific coupling method:

[0095] (1) Concentration conservation constraint:

[0096] During network training, a third physical constraint loss is introduced Ensure that the concentration predictions satisfy the diffusion equation:

[0097]

[0098] Where C pred represents the concentration field predicted by the neural network.

[0099] (2) Boundary conditions and initial condition constraints:

[0100] The velocity field u and diffusion coefficient D are used as network input features to dynamically simulate the relationship between the diffusion process and spatial constraints. During training, boundary conditions (such as concentration boundaries and velocity boundaries) and initial conditions (initial concentration distribution) are used as constraints to ensure that the network output results are reasonable.

[0101] By defining the aforementioned physical loss functions (such as Darcy's law, fluid dynamics equations, and diffusion equations), physical laws are embedded in the network training process, ensuring that the model output conforms to physical constraints. Key parameters in the physical equations (such as permeability, pressure gradient, and concentration field) are directly used as network inputs and learned together with other geological attribute data. Numerical simulations (such as CFD and FEM) are used to solve the physical equations and generate training samples, providing supervised learning data for deep learning.

[0102] In the grouting diffusion dynamic prediction model, the dynamic coupling between the permeation rate, pressure distribution and diffusion boundary is achieved through the following mechanisms:

[0103] Physical constraints embedded in neural networks: The inputs to deep learning models include multiple physical properties, such as permeation rate, pressure distribution, and diffusion boundaries. These physical properties are adjusted through the physical constraint layer within the network. For example, for each prediction, the network not only relies on historical data but also makes corrections based on the permeation rate, pressure distribution, and diffusion boundaries calculated by the physical model to ensure that the prediction is consistent with physical laws.

[0104] Dynamic adjustment of time and space: Based on network structures such as LSTM, the model can process time and space sequence data, dynamically adjust the pressure distribution and permeation rate at different time and space nodes, and predict the path and speed of slurry diffusion in real time.

[0105] Feedback mechanism: After real-time data (such as sensor data) is input, the model will adjust the predicted values ​​of infiltration rate, pressure field and diffusion boundary according to the new data feedback, and automatically adjust the grouting parameters through optimization algorithm to ensure the accuracy of the simulation process.

[0106] By coupling permeability, pressure distribution, and diffusion boundaries, the model accurately simulates and predicts slurry diffusion processes in complex geological structures. Dynamically coupled equations not only ensure physical consistency but also, through deep learning algorithms, achieve spatiotemporal dependencies and real-time optimization based on multi-source data, improving model accuracy and computational efficiency. This model can provide powerful technical support for underground construction, grouting optimization, and other applications.

[0107] In order to meet the above-mentioned needs for prediction and feedback of grouting diffusion process in complex geological environments, the grouting diffusion dynamic prediction model provided in this embodiment has a new deep learning model structure - multimodal spatiotemporal coupled adaptive neural network (MTSCNN). This model combines the multimodal data fusion, spatiotemporal dynamic adjustment and feedback mechanism of deep learning, and can efficiently handle grouting diffusion simulation problems in complex geological environments.

[0108] MTSCNN (Multimodal Temporal-Spatial Coupled Neural Network) is a multimodal spatiotemporal coupled adaptive neural network architecture. It combines multi-source sensor data (such as pressure sensors, temperature sensors, resistivity imaging, fluid mechanics simulation data, etc.) with the spatial characteristics of geological bodies (such as cracks, porosity, permeability, etc.), and realizes real-time prediction and dynamic feedback of complex underground grouting processes through time series modeling and feedback mechanisms. Specifically, by combining physical properties such as permeability rate, pressure distribution, diffusion boundary with dynamic coupling equations, and combining deep learning methods for prediction and optimization, the multiple influencing factors of slurry diffusion in complex geological bodies are solved. The core idea of ​​the model is to combine physical constraints with the predictive ability of deep learning to perform more accurate simulations and dynamically optimize key parameters in the grouting process.

[0109] The structure of the grouting diffusion dynamic prediction model includes:

[0110] a. Input layer: multimodal data fusion

[0111] Multi-source data input: Model input includes multiple modal data sources, such as: Geological data: including physical properties such as fractures, porosity, and permeability, as static input; Sensor data: including real-time pressure fields, temperature fields, concentration fields, flow fields, etc., as dynamic input; Numerical simulation data: Fluid mechanics-related data (such as velocity fields and pressure fields) obtained through numerical simulations such as CFD models, as auxiliary information input.

[0112] Each type of data goes through a separate preprocessing layer, such as normalization, standardization, and convolutional feature extraction of image data. Data from different modalities are fused through a feature fusion layer to convert different types of data into a unified feature representation.

[0113] b. Spatiotemporal modeling layer:

[0114] Temporal modeling: LSTM (Long Short-Term Memory) networks are used to process the time series information in the input data. This part of the model is responsible for capturing the dynamic evolution of the grouting process, such as the diffusion process of the slurry, pressure changes, and temperature changes.

[0115] Spatial modeling: 3D convolutional neural networks (3D CNNs) are used to extract spatial features. 3D CNNs can effectively capture spatial information within geological bodies and analyze the physical property changes at different locations, such as cracks and pores, and their impact on slurry diffusion.

[0116] Spatiotemporal Coupling Layer: Combining LSTM and 3DCNN, a spatiotemporal coupling layer is embedded in the model. This layer links temporal and spatial features to better understand how the slurry diffusion process changes over time and space. Specifically, the spatiotemporal coupling layer combines the dynamic evolution of the time series with the distribution information of spatial features for joint modeling.

[0117] c. Physical Constraint Integration Layer:

[0118] Physical Model Embedding: To ensure that the deep learning model adheres to the laws of fluid dynamics, this layer embeds physical laws into the neural network through a "physical constraint loss function" (such as Darcy's law and the continuity equation). This physical constraint loss function strengthens the model's adaptability to real-world physical processes, ensuring that its predictions adhere to actual physical laws.

[0119] d. Feedback mechanism and adaptive adjustment:

[0120] Feedback layer: After each prediction, the feedback layer compares real-time monitoring data (such as pressure, temperature, resistivity, and other sensor data) with the predicted results, calculates the error, and corrects the error through the backpropagation algorithm. The purpose of the feedback mechanism is to use actual field data to adaptively adjust model parameters to optimize the subsequent prediction process.

[0121] Adaptive learning rate adjustment: Based on real-time feedback data, the model's learning rate is dynamically adjusted to improve responsiveness to new data and prediction accuracy. This adaptive learning rate helps the model quickly adjust and converge, avoids overfitting, and ensures real-time predictions.

[0122] e. Output layer:

[0123] Prediction results: The output layer will generate prediction results of grouting diffusion based on the predicted spatiotemporal characteristics, including:

[0124] Diffusion path and concentration distribution: Predict the diffusion path and concentration distribution of slurry in geological bodies over time.

[0125] Pressure and velocity fields: Predict changes in pressure and velocity during grouting.

[0126] Diffusion Boundary: Predict the boundary of slurry diffusion and identify the farthest location the slurry may reach.

[0127] The training and optimization process of the grouting diffusion dynamic prediction model is as follows:

[0128] a. Training data:

[0129] The training data sources include

[0130] Field experimental data: experimental data from different geological conditions, including real-time monitoring data from multiple sensors.

[0131] Numerical simulation data: Based on existing fluid mechanics simulation results, generate training samples that match the actual scenario.

[0132] Historical grouting data: including monitoring data and prediction results of all previous grouting processes, used for model training and error correction.

[0133] b. Loss function design:

[0134] The loss function combines a weighted combination of physical constraint losses (such as Darcy's law and fluid dynamics), prediction error losses (such as MSE or MAE), and feedback adjustment losses (such as error correction). By comprehensively optimizing these loss functions, we ensure that the model not only meets the required accuracy, but also adheres to physical laws and is capable of adaptive adjustment.

[0135] Technical implementation:

[0136] Deep learning framework: Use mainstream deep learning frameworks, such as TensorFlow or PyTorch, to build and train models.

[0137] Multimodal data interface: Through a unified data processing interface, data from different sensors are integrated and real-time data transmission and processing are ensured.

[0138] Physical model embedding: Use physics-based loss functions (such as fluid dynamics constraints, boundary conditions, etc.) to optimize the neural network training process and ensure that the model output is consistent with physical laws.

[0139] c. Optimization method:

[0140] The model is optimized using the Adam optimizer, an adaptive learning rate, and a dynamic update strategy. Based on real-time feedback, the optimizer dynamically adjusts the learning rate to ensure the model's stability and efficiency during real-time predictions.

[0141] During the training and optimization process, the implementation of prediction and feedback mechanisms is key to ensuring the model efficiently and accurately reflects the actual geological behavior. To achieve this goal, multi-source data fusion, deep learning models, spatiotemporal dynamic adjustment, and feedback mechanisms work together to ensure dynamic correction and optimization during the simulation process.

[0142] The feedback process ensures that the model can be continuously optimized in actual applications, dynamically adjusting grouting parameters and improving simulation accuracy. The feedback mechanism continuously adjusts the model prediction results based on real-time sensor data and changes in the grouting process to achieve more accurate simulation and control. Specifically, it includes:

[0143] 1. Real-time data collection and input:

[0144] Sensor feedback: Various sensors installed on site (such as pressure sensors, resistivity sensors, temperature sensors, etc.) will provide real-time feedback on the data of underground slurry diffusion process, including slurry flow rate, diffusion range, pressure changes, etc.

[0145] Real-time input of experimental data and field monitoring data: These data are continuously input into the deep learning model as new input features to ensure that the model is constantly updated and optimized.

[0146] 2. Dynamic adjustment and optimization:

[0147] Physical constraints and model correction:

[0148] By monitoring data in real time, the model can dynamically adjust physical parameters (such as permeability, pressure distribution, etc.), and these physical parameters will be used as constraints to participate in the next prediction calculation.

[0149] Feedback correction of deep learning models: After receiving real-time data, the model will provide feedback corrections to its predictions. For example, if the actual slurry diffusion path deviates from the predicted path, the model will adjust its prediction at the next moment based on the newly input monitoring data to correct the slurry diffusion path.

[0150] Feedback mechanism implementation:

[0151] Backpropagation and weight updating: When there's a discrepancy between the model's predictions and actual data, the model uses backpropagation algorithms (such as gradient descent) to adjust network weights, gradually approximating the actual diffusion process. Through continuous iterative training, the model learns more precise physical laws and updates in real time.

[0152] Real-time optimization: The model optimizes in real time based on feedback data, adjusting predictions to match the current state of the geological entity. For example, permeability can be adjusted based on feedback from pressure distribution, or the velocity and path of slurry flow can be adjusted based on feedback from diffusion boundaries.

[0153] 3. Feedback and optimization loop:

[0154] An alternating process of feedback and prediction: During the actual grouting process, real-time monitoring data is periodically fed into the model, with each input generating new predictions. The model's adjustments and feedback to these predictions form a closed-loop optimization system. Whenever a variable in the grouting process changes (such as abnormal pressure or unexpected slurry diffusion), the model automatically adjusts and optimizes to generate a new prediction path and grouting strategy.

[0155] Technical implementation of feedback optimization:

[0156] Adaptive learning rate: To better perform real-time optimization, the feedback mechanism can be combined with an adaptive learning rate, allowing the model to dynamically adjust the learning speed according to changes in real-time data during training, thereby improving response efficiency.

[0157] Deep Reinforcement Learning (DRL): In more complex scenarios, reinforcement learning methods can be used to optimize grouting strategies. DRL can dynamically adjust grouting parameters based on the reward function according to the environmental state (such as pressure field, slurry diffusion state, etc.) to achieve the long-term optimal strategy.

[0158] The MTSCNN model architecture integrates multi-source data (such as sensor data, geological data, and numerical simulation data), spatiotemporal modeling (using LSTM and 3D CNN), physical constraint embedding (such as fluid dynamics equations), and feedback mechanisms (error feedback and adaptive adjustment) to provide efficient and accurate grouting diffusion prediction and dynamic optimization solutions. Through real-time feedback and adaptive learning, the model can continuously adjust and optimize prediction results, addressing the limitations of traditional methods in simulation accuracy and real-time performance, and providing intelligent support for underground engineering projects in complex geological environments.

[0159] Example 2

[0160] In a typical embodiment of the present invention, a grouting diffusion prediction system with complex geological multi-attribute constraints is proposed, comprising:

[0161] The attribute information acquisition module is configured to: acquire multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics;

[0162] The grouting diffusion prediction module is configured to: input multi-source attribute information into a trained grouting diffusion dynamic prediction model, and output grouting diffusion prediction results for complex geological bodies, wherein the grouting diffusion results include permeability, pressure distribution, and boundary conditions;

[0163] The establishment and training process of the grouting diffusion dynamic prediction model is as follows:

[0164] Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions;

[0165] Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model;

[0166] Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model;

[0167] After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment;

[0168] During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through a feedback mechanism based on the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model.

[0169] Furthermore, the grouting diffusion simulation experimental system, such as Figure 2 Shown, including:

[0170] The complex geological simulation unit 1 is used to simulate complex geology, including a model box filled with various rock layers, sand layers and soil layers, and embedded with different geological features such as horizontal and vertical cracks, sand layers, gravel layers, fault fracture zones and karst caves to simulate complex geological structures, such as:

[0171] Rock Layer Structure: Simulates a variety of rock, sand, and soil layers, including stratum cover, fracture settings, and cave presets. The model materials are a new type of lightweight cement and water-resistant resin, and the surface is sprayed with water-resistant pigment to simulate the texture of different rock layers.

[0172] Tunnel and structure penetration design: The transparent acrylic tunnel penetration model shows the rock strata, water seepage conditions, crack distribution, and water inflow path of the tunnel penetration area.

[0173] Geological layer simulation: Different geological features such as horizontal and vertical cracks, sand layers, gravel layers, fault fracture zones, and caves are embedded in the model to simulate various complex structures in the natural geological environment.

[0174] The water injection and grouting unit 2 includes a dynamic water circulation module and a grouting control module. The dynamic water circulation module is used to simulate the groundwater seepage process, provide constant pressure water flow, and simulate the natural groundwater flow path in the model. The grouting control module realizes precise control of the grouting rate, pressure and flow rate, and monitors the pressure and flow rate data during the grouting process through multi-source sensors.

[0175] Data acquisition and monitoring unit, including pressure sensors, flow sensors, pore water pressure gauges, resistivity imaging systems and ultrasonic sensors;

[0176] The data processing unit is used to acquire and process data. Equipped with a desktop computer and professional control software, it supports real-time data acquisition, processing, and analysis. The software system features data visualization, displaying data from flow meters, pressure sensors, resistivity imaging, and other sources in real time to assist in monitoring and controlling the grouting process. Remote control and monitoring are supported, allowing for remote adjustment of grouting parameters such as pressure and flow through the software interface. Data storage, historical data query, and experimental data export are also supported.

[0177] The steps are as follows:

[0178] 1. Device preparation and system initialization

[0179] Check the integrity of all components of the geological model to ensure the correct pre-set structures for fractures, pores, caves, and sand and gravel formations. Initialize the water injection and grouting system to ensure the proper functioning of the water circulation and grouting equipment. Start the data acquisition system, including the 3D laser scanner, cave scanner, CT imager, and resistivity equipment, and check the connections to ensure smooth signal acquisition. Start the computer and control software and configure the experimental parameters, including initial geological model attributes such as fracture distribution, pore structure, and water inflow area.

[0180] 2. Multi-source attribute measurement

[0181] (1) Crack characteristic measurement:

[0182] 3D laser scanners and cave scanners were used to measure the crack characteristics (width, length, and roughness) of the rock mass, generating 3D point cloud data of the cracks. A handheld scanner was used to supplement the scans of local details to improve the resolution of the crack model.

[0183] (2) Pore structure measurement:

[0184] Porosity data is acquired using a micro-CT imager to establish a microscopic model of pore distribution. Combined with a pore water pressure gauge, real-time monitoring of pressure changes within the pores allows for the acquisition of correlation data between pores and fractures.

[0185] (3) Measurement of water inflow and permeability characteristics:

[0186] Resistivity equipment and ground-penetrating radar were used to measure the distribution of stratum moisture and water inflow areas. A groundwater stratification sampling system was used to measure the permeability of different layers, and elasticity instruments were used to determine the overall strength and fracture structure of the rock mass. Flow meters and pressure sensors were deployed at the tunnel face and within the model tunnel to obtain dynamic permeability data.

[0187] 3. Grouting diffusion process experiment

[0188] (1) Grouting system initialization:

[0189] Set up the water injection and grouting test bench at a specific location within the formation model, adjust the initial grouting flow rate and pressure based on experimental requirements, and configure the grouting liquid (e.g., colored water or slurry) to facilitate monitoring of its diffusion process.

[0190] (2) Real-time monitoring of the diffusion process:

[0191] Start the grouting system, gradually increase the grouting pressure, and monitor the speed and direction of slurry diffusion.

[0192] Parameters such as flow rate, pressure, and diffusion radius are recorded in real time, and the slurry diffusion path is visualized using control software. Permeability changes within the formation are monitored using equipment such as resistivity imaging and pore water pressure gauges to identify trends in pores and fractures during the diffusion process.

[0193] 3. Data processing and coupling analysis

[0194] (1) Multi-source data collection and fusion:

[0195] Import data from various sensors and measurement devices, including 3D fracture models, porosity distribution, permeability, and pressure changes. Preprocess the collected fracture, porosity, and permeability data to unify coordinates and data formats to ensure consistency in data fusion.

[0196] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A grouting diffusion prediction method with complex geological multi-attribute constraints, characterized by: include: Acquiring multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics; Inputting multi-source attribute information into a trained grouting diffusion dynamic prediction model to output grouting diffusion prediction results for complex geological bodies, the grouting diffusion results including permeability, pressure distribution, and boundary conditions; The establishment and training process of the grouting diffusion dynamic prediction model is as follows: Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions; Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model; Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model; After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment; During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through the feedback mechanism according to the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model; The grouting diffusion dynamic prediction model includes an input layer, a spatiotemporal modeling layer, a physical constraint integration layer, a feedback mechanism and adaptive adjustment layer, and an output layer.

2. The grouting diffusion prediction method with complex geological multi-attribute constraints as claimed in claim 1 is characterized in that: The dynamic equations include Darcy's law, fluid mechanics equations and diffusion equations. During the training of the grouting diffusion dynamic prediction model, the permeation rate is coupled with Darcy's law, the pressure distribution is coupled with the fluid mechanics equations, and the diffusion boundary is coupled with the diffusion equation.

3. The grouting diffusion prediction method with complex geological multi-attribute constraints as claimed in claim 2 is characterized in that: The first physical loss function is obtained by coupling the permeation rate with Darcy's law : ； Where, represents the permeation rate predicted by the neural network; represents the permeation rate obtained by numerical simulation; The pressure distribution is coupled with the fluid mechanics equation to obtain the second physical loss function : ； Where, represents the pressure field predicted by the neural network; k represents the permeability; The diffusion boundary is coupled with the diffusion equation to obtain the third physical loss function : ； Where, represents the concentration field predicted by the neural network; u represents the velocity field; and D represents the diffusion coefficient.

4. The grouting diffusion prediction method with complex geological multi-attribute constraints as claimed in claim 1 is characterized in that: The optimization of the grouting diffusion dynamic prediction model includes: Backpropagation and weight updating: When there is a deviation between the model's predictions and actual experimental data, the model adjusts network weights through a backpropagation algorithm to gradually approximate the actual diffusion process. Through continuous iterative training, the model can learn more precise physical laws and update them in real time. Real-time optimization: The model is optimized in real time based on feedback data, adjusting the prediction results to match the real-time geological state.

5. The grouting diffusion prediction method with complex geological multi-attribute constraints as claimed in claim 1, characterized in that: When obtaining crack characteristics, a 3D laser scanner and a cave scanner are used to measure the crack characteristics of the rock mass, generate 3D point cloud data of the cracks, and use a handheld scanner to perform additional scanning of local details; When acquiring pore characteristics, a microscopic CT imager is used to obtain porosity data and establish a microscopic model of pore distribution. In combination with a pore water pressure gauge, the pressure changes of the fluid inside the pores are monitored in real time to obtain correlation data between pores and fractures. When obtaining the characteristics of water inflow, resistivity equipment and ground-penetrating radar are used to measure the distribution information of stratum moisture and water inflow area, and the permeability of different layers is measured through the groundwater stratification test sampling system. The overall strength and crushing structure characteristics of the rock mass are obtained in combination with the elasticity meter. Flow meters and pressure sensors are arranged at the tunnel face and inside the model tunnel to obtain dynamic permeability data.

6. The grouting diffusion prediction method with complex geological multi-attribute constraints as claimed in claim 1, characterized in that: When establishing a multi-attribute coupled permeability model, statistical and machine learning methods are used to model the coupling relationship between multiple attribute information to reveal the influence mechanism of each physical attribute on permeability.

7. A grouting diffusion prediction system with complex geological multi-attribute constraints, characterized by: include: The multi-attribute information acquisition module is configured to: acquire multi-source attribute information of a complex geological body, wherein the multi-source attribute information includes fracture characteristics, pore characteristics, and water inflow characteristics; The grouting diffusion prediction module is configured to: input multi-source attribute information into a trained grouting diffusion dynamic prediction model, and output grouting diffusion prediction results for complex geological bodies, wherein the grouting diffusion results include permeability, pressure distribution, and boundary conditions; The establishment and training process of the grouting diffusion dynamic prediction model is as follows: Build a grouting diffusion simulation experimental system to simulate the grouting diffusion process in complex geological conditions; Acquiring multi-source attribute information and permeability of the simulated formation of the experimental system, establishing a multi-attribute coupled permeability model, and embedding the multi-attribute coupled permeability model into the grouting diffusion dynamic prediction model; at the same time, embedding the dynamic equation as a physical constraint condition into the grouting diffusion dynamic prediction model; Conduct numerical simulations on the experimental system to obtain the permeability rate, pressure distribution, and diffusion boundary during the grouting experiment, and train the grouting diffusion dynamic prediction model; After the training is completed, the pressure, resistivity and concentration of the grouting experiment at the current moment are obtained in real time and input into the grouting diffusion dynamic prediction model to predict the grouting diffusion process at the next moment; During the prediction process, the grouting diffusion dynamic prediction model continuously adjusts the prediction results through the feedback mechanism according to the grouting results of the grouting experiment to optimize the grouting diffusion dynamic prediction model; The grouting diffusion dynamic prediction model includes an input layer, a spatiotemporal modeling layer, a physical constraint integration layer, a feedback mechanism and adaptive adjustment layer, and an output layer.

8. The grouting diffusion prediction system with complex geological multi-attribute constraints as claimed in claim 7, characterized in that: The grouting diffusion simulation experimental system includes: Complex geological simulation unit, used to simulate complex geology; The water injection and grouting unit includes a dynamic water circulation module and a grouting control module. The dynamic water circulation module is used to simulate the groundwater seepage process, provide constant pressure water flow, and simulate the natural groundwater flow path in the model. The grouting control module realizes precise control of the grouting rate, pressure and flow rate, and monitors the pressure and flow rate data during the grouting process through multi-source sensors. Data acquisition and monitoring unit, including pressure sensors, flow sensors, pore water pressure gauges, resistivity imaging systems and ultrasonic sensors; The data processing unit is used to acquire data and perform data processing.

9. The grouting diffusion prediction system with complex geological multi-attribute constraints as claimed in claim 7, characterized in that: The complex geological simulation unit includes a model box, which is filled with various rock layers, sand layers and soil layers, and embedded with different geological features such as horizontal and vertical cracks, sand layers, gravel layers, fault fracture zones and caves to simulate complex geological structures.

Images