Grouting simulation method and system based on time sequence interaction and adaptive grid reconstruction

Abstract

The invention provides a grouting simulation method and system based on time sequence interaction and self-adaptive grid reconstruction, and relates to the technical field of grouting engineering.The grouting simulation method comprises the steps that monitoring data of the grouting construction process are obtained, the monitoring data are input into a grouting diffusion prediction model, a grouting pressure prediction value is output, and a series of prediction responses are conducted based on the grouting pressure prediction value. Obtaining a grouting pressure predicted value, a boundary point set of the grouting diffusion area, the area and the volume of the grouting diffusion area and a diffusion response result of the pressure distribution diagram; constructing a geometric model of the grouting area based on the diffusion response result, and performing grid unit division and grouting diffusion area mapping on the geometric model; and performing local grid encryption and reconstruction on the grid units according to a density control rule to obtain a structured grid data model, adaptively updating the structured grid data model by using real-time monitoring data, and performing real-time visual grouting optimization control. According to the invention, the high-precision prediction response of grouting is improved.

Description

Technical Field

[0001] This disclosure relates to the field of grouting engineering technology, specifically to a grouting simulation method and system based on temporal interaction and adaptive mesh reconstruction. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] During tunnel construction, grouting reinforcement serves as a crucial means of controlling stratum deformation, seepage, and stabilizing surrounding rock. It is characterized by strong nonlinearity, numerous influencing factors, and complex spatiotemporal variations. Traditional grouting parameter design and response prediction rely on expert experience and simple models, making it difficult to dynamically reflect the evolution of grout diffusion during the grouting process. Furthermore, existing mesh generation strategies often employ static structures, which struggle to adapt to the dynamic changes in geological and flow parameters during grouting, impacting prediction accuracy and control efficiency.

[0004] With the rapid development of deep learning technology, its capabilities in time series modeling, nonlinear mapping and dynamic system prediction are becoming increasingly significant, and it is particularly suitable for data modeling and intelligent decision-making in complex geological grouting environments. Existing solutions mainly focus on the following model structures: First, time series prediction models based on Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs) can predict the grouting pressure change trend in multiple steps, but they only model single-point time series and are difficult to express spatial diffusion characteristics; Second, feature extraction methods represented by Convolutional Neural Networks (CNNs) are suitable for spatial correlation analysis of two-dimensional monitoring matrices, but their ability to capture temporal information is limited and their ability to express pressure diffusion boundaries is insufficient; Third, fusion models (such as CNN-LSTM and Attention-LSTM) combine spatial-temporal hybrid feature expressions, but they mostly use points as basic units and ignore topological changes and response gradients within the grouting area; Fourth, graph neural network (GNN) models establish spatial dependencies by constructing a point adjacency graph and attempt to model the graph structure of the grouting area. However, this type of method relies on prior graph structures and has poor adaptability when facing dynamic diffusion processes and variable grids.

[0005] In summary, while the existing methods can perform data fitting and short-term prediction of the grouting process to a certain extent, they still have the following shortcomings in practical engineering applications: First, most existing models are "black box" learning structures, lacking consideration of physical mechanisms (such as seepage laws and boundary control), making the prediction results difficult to interpret; second, most models are based on fixed points or static grid designs, failing to achieve local densification and dynamic adjustment of spatial granularity in the response region; third, the prediction results and the system do not form a data feedback mechanism, making it difficult to achieve adaptive optimization and control; finally, the model outputs are mostly scalar sequences, lacking the ability to structurally represent continuous spatial pressure fields and diffusion boundaries. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a grouting simulation method and system based on temporal interaction and adaptive mesh reconstruction. By constructing a CNN-BiLSTM deep fusion model, the sensor data during the grouting process is analyzed and predicted in real time, achieving high-precision response modeling of the grouting diffusion range. Adaptive mesh reconstruction is performed based on the predicted diffusion boundary change trend to improve the decision-making accuracy and engineering intelligence level during the grouting process.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions: Grouting simulation methods based on temporal interaction and adaptive mesh reconstruction include: Acquire monitoring data during the grouting process and perform preprocessing; The preprocessed monitoring data is input into the grouting diffusion prediction model, and the predicted grouting pressure value for the future time step is output. Based on the predicted grouting pressure value, the spatial continuous diffusion pressure field is reconstructed to obtain the pressure distribution map of the grouting area. The grouting diffusion boundary is extracted from the pressure distribution map to determine the grouting diffusion area, and the area and volume of the grouting diffusion area are calculated. The predicted grouting pressure, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map are integrated with the diffusion response results to form structured data. Based on structured data, a geometric model of the grouting area is constructed and boundary conditions are determined. The geometric model is then divided into grid cells, the grouting diffusion area is mapped to the grid cells, and the distance between the grid cells and the diffusion boundary is determined. Based on the distance between the grid cells and the diffusion boundary, density control rules are designed. The grid cells are then locally densified and reconstructed according to the density control rules. The reconstructed grid is used to update the topology and boundary conditions between the grid cells to obtain a structured grid data model. The structured grid data model is continuously updated adaptively using real-time grouting monitoring data to achieve real-time visualized grouting optimization control.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions: The grouting simulation system based on temporal interaction and adaptive mesh reconstruction includes: a data acquisition module, used to acquire monitoring data of the grouting construction process and perform preprocessing; The diffusion prediction module is used to input the preprocessed monitoring data into the grouting diffusion prediction model, output the predicted grouting pressure value for the future time step, reconstruct the spatial continuous diffusion pressure field based on the predicted grouting pressure value, obtain the pressure distribution map of the grouting area, extract the grouting diffusion boundary from the pressure distribution map, determine the grouting diffusion area, and calculate the area and volume of the grouting diffusion area. The digital model building module integrates the grouting pressure prediction, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into diffusion response results to form structured data. Based on the structured data, it constructs a geometric model of the grouting area and determines the boundary conditions. The geometric model is then divided into grid cells, and the grouting diffusion area is mapped to the grid cells. The distance between the grid cells and the diffusion boundary is determined. Based on the distance between the grid cells and the diffusion boundary, density control rules are designed. According to the density control rules, the grid cells are locally densified and reconstructed. The reconstructed grid is used to update the topology and boundary conditions between the grid cells to obtain a structured grid data model. The feedback correction module is used to continuously and adaptively update the structured grid data model using real-time grouting monitoring data, enabling real-time visualized grouting optimization control.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements the grouting simulation method based on temporal interaction and adaptive mesh reconstruction.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the grouting simulation method based on time-series interaction and adaptive mesh reconstruction.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the grouting simulation method based on time-series interaction and adaptive mesh reconstruction.

[0012] Compared with the prior art, the beneficial effects of this disclosure are as follows: This disclosure presents a grouting simulation method based on temporal interaction and adaptive mesh reconstruction. It constructs a grouting diffusion prediction model based on CNN-BiLSTM, outputting the grouting pressure distribution and diffusion profile at several future time points. The prediction results are output in the form of a three-dimensional point cloud or a mesh matrix. According to the predicted diffusion boundary change trend, the mesh density of the simulation area is dynamically adjusted. A high-density fine mesh is used near the diffusion front, while a coarse mesh is used in the stable area to achieve a balance between spatial resolution and computational efficiency. Furthermore, a feedback control and iterative optimization mechanism is constructed to achieve continuous optimization and error control, thereby improving the decision-making accuracy and engineering intelligence level in the grouting process.

[0013] This disclosed grouting simulation method, based on temporal interaction and adaptive grid reconstruction, effectively extracts nonlinear temporal features and complex spatiotemporal correlations during the grouting process by constructing a deep neural network model integrating CNN and BiLSTM. This enables accurate prediction of future grouting pressure, pore pressure variation trends, and grout diffusion states. Compared to traditional methods based on empirical formulas or single physical models, the prediction accuracy is significantly improved, and it is better adapted to geological complexity and data volatility.

[0014] The grouting simulation method disclosed herein, based on temporal interaction and adaptive mesh reconstruction, adopts a dynamic mesh partitioning mechanism guided by prediction results. Combined with boundary distance function and density control rules, it automatically completes local densification and global mesh reconstruction of key diffusion areas, thereby ensuring that the modeling accuracy of key areas is effectively improved while maintaining overall computational efficiency. It is suitable for grouting modeling and simulation needs under different scales and geological conditions.

[0015] The grouting simulation method disclosed herein, based on temporal interaction and adaptive mesh reconstruction, possesses a feedback correction mechanism. It can continuously receive real-time monitoring data from the site and dynamically compare it with the prediction results. When the system detects that the prediction error or boundary offset exceeds the threshold, it can automatically determine the source of the error and trigger the corresponding feedback path, such as model fine-tuning, structural retraining, or local mesh update, thereby constructing an intelligent closed-loop control process of "prediction-feedback-optimization" to improve system stability and reliability. Attached Figure Description

[0016] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0017] Figure 1 This is a flowchart illustrating the grouting simulation method based on temporal interaction and adaptive mesh reconstruction, according to an embodiment of this disclosure. Detailed Implementation

[0018] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0020] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0021] Example 1 One embodiment of this disclosure provides a grouting simulation method based on temporal interaction and adaptive mesh reconstruction, the steps of which include: Step 1: Obtain monitoring data during the grouting process and perform preprocessing; Step 2: Input the preprocessed monitoring data into the grouting diffusion prediction model, and output the predicted grouting pressure value for the future time step. Based on the predicted grouting pressure value, reconstruct the spatial continuity diffusion pressure field to obtain the pressure distribution map of the grouting area. Extract the grouting diffusion boundary from the pressure distribution map to determine the grouting diffusion area, and calculate the area and volume of the grouting diffusion area. Step 3: Integrate the predicted grouting pressure, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into diffusion response results to form structured data; Step 4: Based on structured data, construct a geometric model of the grouting area and determine the boundary conditions. Divide the geometric model into grid cells, map the grouting diffusion area to the grid cells, and determine the distance between the grid cells and the diffusion boundary. Step 5: Based on the distance between the grid cells and the diffusion boundary, design density control rules, perform local grid densification and reconstruction on the grid cells according to the density control rules, update the topology and boundary conditions between grid cells using the reconstructed grid, obtain a structured grid data model, and continuously use real-time grouting monitoring data to adaptively update the structured grid data model for real-time visualized grouting optimization control.

[0022] As one embodiment, the grouting simulation method based on temporal interaction and adaptive mesh reconstruction disclosed herein constructs a grouting diffusion prediction model that deeply integrates a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM). This model performs real-time analysis and prediction of sensor data during the grouting process, achieving high-precision response modeling of the grouting diffusion range. Based on the prediction results, it automatically reconstructs the mesh, thereby improving the decision-making accuracy and engineering intelligence level during the grouting process. The specific implementation process is as follows: Step 1: Obtain monitoring data during the grouting process and perform preprocessing; Specifically, in the actual construction process, different types of sensors (such as pressure sensors, flow meters, radar, pore pressure gauges, etc.) are used to collect grouting construction monitoring data, and the collected monitoring data is preprocessed to provide reliable input data for subsequent deep learning models.

[0023] Furthermore, the monitoring data collected by different types of sensors includes grouting pressure. Grouting volume pore water pressure Grouting time Ground-penetrating radar or ultrasonic reflected signals Formation description parameters (permeability coefficient) Porosity internal friction angle ).

[0024] As one embodiment, the monitoring data is preprocessed and a multidimensional time-series data sequence is constructed. The specific process includes: S1-1: Data Synchronization and Interpolation: In actual construction, different types of sensors (such as pressure sensors, flow meters, radar, pore pressure gauges, etc.) have different sampling frequencies and data delays. Data of different frequencies are time-aligned using methods such as linear interpolation and spline interpolation to construct a unified time step. The data sequence.

[0025] S1-2: Anomaly Removal and Normalization: Data often contains instantaneous anomalies (such as sensor jumps and electrical interference), as well as problems with large differences in dimensions and scales. Therefore, cleaning and normalization processing are required.

[0026] First, statistical analysis methods were used to utilize 3 In principle, preliminary anomaly detection should be performed:

[0027] in, To calculate the mean of the sequence; Standard deviation, For multidimensional time series data, Output for anomaly detection.

[0028] To improve model convergence speed and avoid the impact of scale differences on training, the variables are then normalized (Min-Max method):

[0029] in: , For each feature, the minimum and maximum values ​​observed in the historical samples are given. It is compatible with the input requirements of most deep learning models.

[0030] After data synchronization, anomaly removal, and normalization, a unified, time-aligned, and cleaned multidimensional time-series data sequence is obtained. This provides high-quality input for subsequent deep learning modeling.

[0031] Step 2: Construct a fusion deep learning model and train it to obtain a grouting diffusion prediction model; Specifically, based on the time-synchronized and preprocessed grouting data obtained in step 1, a deep learning model integrating a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM) was constructed to achieve high-precision modeling and prediction of grouting diffusion response. During training, this model can effectively extract local features and capture global time dependencies, making it suitable for processing multi-source, time-series grouting monitoring data. The training process includes the following steps: S2-1: Input layer inputs multidimensional time-series data sequences: ; Where R is the set of real numbers, For time steps (such as monitoring data from the past 60 minutes); The characteristic dimensions for each time point (e.g., grouting pressure, grouting volume, pore pressure, geological indicators, etc.).

[0032] S2-2: Convolutional Neural Networks (CNNs) include a Convolutional Feature Layer (CNN), which extracts feature patterns within a local time frame (e.g., local morphology of pressure fluctuations, abrupt change signals). It primarily provides one-dimensional convolution (1D-CNN) for feature extraction along the time dimension.

[0033] in: These are local convolutional features; For activation functions, ReLU is usually chosen; The kernel is a convolution kernel with a size of . ; The input sequence is a segment within the convolution window; The bias term is a trainable scalar.

[0034] The CNN feature extraction layer can learn local change patterns, such as sudden pressurization during grouting and rapid rise in pore pressure. This layer provides input data for the subsequent BiLSTM time series modeling layer. ,in, Effective time step; This represents the number of convolution kernels (i.e., the number of output channels).

[0035] S2-3: The Bidirectional LSTM layer captures the long-range dependencies of grouting dynamics by modeling in both forward and reverse time directions. It takes the feature sequence output from the previous CNN layer... The hidden state is calculated along the forward and reverse time directions, respectively, from the input:

[0036]

[0037] in, In time step The input feature vector; , The forward and backward LSTM units are respectively in time , The hidden state at any given moment; , These represent the forward LSTM unit and the backward LSTM unit, respectively. The forward LSTM extracts the influence of the past on the current state, while the backward LSTM captures the potential feedback of the current state to the future. The final outputs are concatenated as follows:

[0038] Output shape This is a mathematical representation of the shape of the output tensor of the BiLSTM network, representing the feature vector generated by the model at each time step of the input sequence. This refers to the number of time steps (i.e., sequence length) of the input sequence after CNN feature extraction (or sliding window preprocessing). The hidden layer dimension of a unidirectional LSTM is the feature dimension of the output in each direction (forward or backward). This refers to the output dimension of the BiLSTM, because a BiLSTM consists of two parts: a forward LSTM and a backward LSTM, each outputting a dimension. Dimensional vector.

[0039] This output tensor Essentially, it's the result of a deep model modeling the dynamic response of an input grouting time series (such as pressure, pore pressure, grouting volume, etc.) at each time step. It represents the deep feature representation formed at the current moment under the influence of previous and subsequent historical factors. It includes the following features: (1) Time characteristics: The trend of physical parameter values ​​at the current moment (such as the rate of pressure rise / fall) is taken into account. (2) Memory characteristics: including long-term memory of the evolutionary trajectory over a past period of time; (3) Feedback characteristics: The backward LSTM includes the feedback effect of the future state on the current state (such as hysteresis response). (4) Context dependency: It integrates context information throughout the sequence, which helps to determine the phased changes in the grouting response process.

[0040] S2-4: The fully connected layer (FC) can map the features of the BiLSTM output to the prediction target space, that is, it maps the BiLSTM output results... Perform a linear transformation to obtain the predicted grouting pressure or diffusion index at a future time or spatial point. :

[0041] in, For mapping convolution kernels, For trainable bias; S2-5: The output layer structure employs a fully connected (dense) mapping structure, mapping the context feature tensor output by the BiLSTM to a sequence of predicted results for the target variable. Let the prediction stride be... The number of target variables is The final output will be:

[0042] in, These represent the model's predictions for future times. Grouting pressure at the location pore pressure seepage velocity Predicted values ​​of physical quantities.

[0043] As one example, the output can support two prediction modes: Univariate prediction pattern: when At that time, the model only outputs the time series prediction value of grouting pressure, which is suitable for grouting engineering scenarios with pressure control as the core. Multivariate joint prediction model: when At the same time, the model can output multiple key response indicators (such as pressure, pore pressure, seepage, etc.), realizing comprehensive and global prediction modeling of the grouting process status.

[0044] Furthermore, a weighted mean squared error loss function is used to measure the difference between the true and predicted values:

[0045] in, The importance weight of each prediction point can be set based on geological risk or monitoring area. N For data volume, For predicted values, This is the actual value.

[0046] The above process is continuously iterated until the difference between the true value and the predicted value is minimized, thus completing the training of the fused deep learning model and obtaining the grouting diffusion prediction model.

[0047] Step 3: After training the deep fusion model, the practical application stage of grouting diffusion response prediction will begin. This stage utilizes real-time or historical construction monitoring data, inputting it into the trained grouting diffusion prediction model to predict the grouting pressure change trend, diffusion boundary evolution, and three-dimensional diffusion response of the grouting zone. Specifically, this includes the following steps: S3-1: Process the real-time or historical construction monitoring data according to the preprocessing operation in step 1 to obtain standardized multi-source time-series data, and construct the input sequence required for the grouting diffusion prediction model using a time sliding window method:

[0048] in, The time step required for the model to predict; The characteristic dimensions for each time point typically include: grouting pressure. Grouting volume pore water pressure Geological parameters Ground-penetrating radar response wait; For the first The multidimensional feature vector at time step.

[0049] S3-2: Input sequence The data is input into the grouting diffusion prediction model, and forward propagation is performed to obtain predicted values ​​of grouting pressure or diffusion indicators at future times. The calculation process is as follows:

[0050] in, The model predicts the first Time step (future) grouting pressure or diffusion response value; These are the training parameters for the model; This refers to the predicted future timeframe.

[0051] S3-3: Spatial diffusion pressure field reconstruction. For scenarios requiring spatial pressure field prediction, the grouting pressure values ​​at representative points can be output based on the grouting diffusion prediction model. Spatial continuous diffuse pressure field reconstruction is performed using an interpolation algorithm:

[0052] in, Representative sensor locations; For weights, bilinear interpolation, inverse distance weighting (IDW), or Kriging methods can be used; The number of representative points.

[0053] After reconstruction, a two-dimensional or three-dimensional pressure distribution map of the grouting area can be generated.

[0054] S3-4: Grouting diffusion boundary extraction, setting diffusion threshold pressure (e.g., 0.05 MPa), extract the region in the predicted pressure field that is higher than this value, and determine the grouting diffusion range (region):

[0055] The boundary of this area is the predicted grouting diffusion boundary, which can be used for the next step of mesh reconstruction and construction control.

[0056] S3-5: After obtaining the diffusion zone, the effective area of ​​grout diffusion can be further calculated. and volume Used for quantitative analysis of grouting coverage effect:

[0057]

[0058] in, The effective thickness of the grouting layer is usually provided by the construction design.

[0059] S3-6: Finally, the obtained diffusion response results are integrated and output to form structured data for subsequent mesh generation modules to use. The content includes, but is not limited to, the predicted pressure values ​​at each time step. ; Set of boundary points of the diffusion region Predict the diffusion area and volume; spatial pressure distribution map and visualization data formats (such as point cloud, raster matrix, etc.).

[0060] Step 4: By combining structured data from grouting diffusion prediction, adaptive mesh adjustment of the grouting area is achieved, improving the spatial accuracy and computational efficiency of numerical simulation, grouting control, and diffusion modeling. The mesh is dynamically divided based on changes in the diffusion boundary, with localized refinement implemented in the diffusion-adjacent region and a coarse mesh layout retained in stable regions, thereby optimizing the allocation of computational resources. This includes the following steps: S4-1: Establish a two-dimensional or three-dimensional geometric model of the grouting area And determine the boundary conditions based on the stratigraphic profile and construction design. For the region... Perform initial, equally spaced mesh generation to form a set of mesh cells:

[0061] in, Indicates the first One grid cell; ; Preset threshold for response density It can be determined based on engineering experience and used as a benchmark for subsequent adjustments.

[0062] S4-2: Mapping the diffusion response to the mesh region, resulting in the grouting diffusion region. Mapping to the grid space, determining each grid cell To determine the spatial relationship between the grid cells and the diffusion boundary, a distance function is defined to determine the distance between the grid cells and the diffusion boundary.

[0063] in, Indicates the diffusion boundary; This is the shortest Euclidean distance.

[0064] S4-3: Adaptive mesh density control rules, which set density control rules based on the distance between each mesh cell and the diffusion boundary:

[0065] in, For the first Target mesh density of the cell; , The distance threshold between high-density and low-density areas; For example, the accuracy difference is more than three times.

[0066] S4-4: Local Mesh Refinement and Reconstruction. Based on density control rules, local mesh refinement and reconstruction are performed on the mesh cells. The operation process is as follows: (a) Mark the grid cells that need to be encrypted, for those that meet the requirements. The units are further subdivided. A preset threshold for response density (often an empirical value or a dynamically updated value) is used to classify regions where response changes exceed this threshold as critical regions requiring higher precision representation. The specific subdivision process is as follows: 1. Divide the unit into smaller cells, such as 2×2 or 4×4; 2. Update the grid node coordinates and topology connectivity; 3. Recalculate the response values ​​on the refined unit (interpolation or re-prediction); 4. Retain the original cell number index for result back-mapping.

[0067] This process can be performed recursively until all subdivisions are completed. Or reach the maximum refinement level limit.

[0068] (ii) Employing a grid encryption algorithm to complete the response density function Calculation and threshold After comparison, for all that satisfy The system automatically triggers a mesh refinement process when the grid cells are selected. This process involves constructing a locally refined mesh in areas of sudden response changes to improve prediction accuracy and resolution within those areas.

[0069] Specifically, for any grid cell that satisfies the encryption conditions Its boundary coordinates can be expressed as:

[0070] Put it in shaft and The axial direction is divided into equal parts. and , generate total There are 10 sub-grid cells. The spatial resolution of each sub-grid cell is:

[0071] in and It can be set to a fixed value (such as 2, 4, 8), or based on local... and The relative size is adaptively set. During the subdivision process, the system retains the original cell index and topology, and numbers each newly generated subgrid cell as follows: ,in , This ensures that subgrids can be uniquely located during structural reconstruction and data mapping. Each sub-cell inherits the response block attributes of the original cell and initializes its corresponding predicted physical quantities (such as pressure, pore pressure, seepage velocity, etc.) using linear interpolation or local inference methods based on the prediction model, ensuring the spatial continuity and physical rationality of the model results.

[0072] (III) Generating the reconstructed mesh set It ensures mesh continuity, controls element shape quality (avoiding long strips and twisted elements), and is compatible with simulation tools such as finite element method and CFD.

[0073] After completing the above encryption operations, the system enters the new grid set generation phase. This phase aims to merge the original grid set with all encrypted sub-grids to form the current time frame. The latest dynamic mesh collection The specific process includes the following steps: 1. Mesh Merging: Merges the original global mesh set. Unencrypted cells and all sub-cells generated after encryption They are integrated together to form a new set of hybrid precision meshes; 2. Node and Boundary Reconstruction: Update the list of mesh node coordinates At the same time, it automatically adjusts the topological connections between grid cells (such as shared edges, adjacency, boundary mapping, etc.) to maintain the integrity and connectivity of the grid structure; 3. Attribute remapping: Reassign predicted physical quantities and response parameters to all new elements and update the mapping function. This ensures that the predicted values ​​are accurately mapped to the new grid structure; 4. Time step label synchronization: Assign time step labels to the new grid set. This facilitates data retrieval and historical comparison in subsequent predictions or iterations by the model; 5. Grid level annotation: Assign a refinement level to each cell. It is used to control the maximum number of refinement layers or for visualization.

[0074] Ultimately, the newly generated mesh set It possesses characteristics such as non-uniform distribution, adaptive density, and densification of the prediction sensitive interval, which can be used for the system to enter the feedback control stage or for the next step of response evolution simulation.

[0075] S4-5: Mesh connectivity and boundary condition update. This update reconstructs the topology between meshes (such as cell connectivity and boundary adjacency information) and updates boundary conditions for boundary cells, including: pressure boundaries; fixed boundaries; and mapping of formation properties (such as permeability and porosity assigned to the mesh). This provides a structured mesh foundation for subsequent feedback calculations (subsequent error assessment and iterative optimization).

[0076] S4-6: Output a structured grid data model, in the following formats: point set + element connectivity; grid file (e.g., Gmsh, VTK, Fluent, etc.); and a grid density field visualization map to show the grouting diffusion influence area. The grid information will be used in subsequent numerical simulations, construction scheduling systems, or 3D visualization platforms.

[0077] Step 5: By monitoring data in real time and comparing it with the prediction integration results, a feedback control mechanism is constructed to dynamically evaluate model errors and diffusion boundary deviations, thereby achieving model fine-tuning, adaptive mesh reconstruction, or control parameter optimization. This step provides the system with the ability to continuously learn and self-update, and is a key component in realizing "intelligent grouting response prediction." Specifically, it includes the following steps: S5-1: Real-time acquisition and registration of monitoring data. During the grouting process, the system continuously collects the following real-time monitoring data: grouting pressure. Grouting volume pore water pressure Data from diffusion boundary monitoring points (such as radar images or diffusion front sensors), and pressure at space monitoring points. Compare the current observation value with the predicted value from step 3. , Perform time and space alignment to build a comparison pair.

[0078] S5-2: Error index calculation, setting the error index function, commonly used indices include: (1) Mean Square Error (MSE):

[0079] (2) Boundary offset error (IoU or profile offset):

[0080] And set dynamic thresholds When the error exceeds the threshold, the model or mesh update mechanism is triggered.

[0081] S5-3: Error source identification and feedback path selection. Based on the error type, conduct source analysis and select the appropriate feedback control path. (1) When MSE and MAPE exceed the standard, that is, when the model prediction error is large, start model fine-tuning (S5-4). (2) When the IoU deviation is significant, i.e. the spatial boundary error is large, start mesh reconstruction (S5-5). (3) When data changes suddenly or anomalies are detected, i.e. when there is a sudden event interference (formation change), parameter adjustment and alarm mechanism (S5-6) shall be implemented.

[0082] S5-4: Deep model fine-tuning update. When the prediction error is identified as being caused by model mismatch, a lightweight online model update is performed. The operation process is as follows: Extract the most recent sliding window sample from the latest observation data. Use a smaller learning rate For model parameters Perform gradient descent update: Elastic Weight Consolidation (EWC) or Replay Memory is used to avoid catastrophic forgetting; the updated model is used for the next round of prediction to maintain continuity.

[0083] S5-5: Adaptive mesh re-division. If the feedback error mainly originates from changes in the diffusion boundary exceeding the accuracy range of the original mesh, then local mesh reconstruction is triggered: the current predicted boundary is re-divided. With the observation boundary The boundary offset values ​​are compared and calculated; a new high-density grid is generated for the area adjacent to the boundary of the diffusion region; the grid optimization strategy can reuse the grid refinement module of S4; the new grid is used for the next step of numerical simulation and data prediction to improve local accuracy.

[0084] S5-6: Automatic adjustment of control parameters and abnormal response. If the system detects a "sudden shift" or high-frequency abnormal fluctuation in the prediction error, and determines it to be a formation change or external disturbance event, the control strategy can be automatically adjusted: dynamically lowering the upper limit of grouting pressure. Grouting rate Automatic control is paused and switched to manual intervention mode; construction alarm logs are generated and synchronized to the visualization platform; the current data segment is marked as "abnormal data" and the model exclusion strategy is activated.

[0085] S5-7: System closed-loop optimization and continuous learning integrates the above-mentioned feedback mechanisms to form a stable closed-loop prediction and control system: continuously updates the model and mesh, maintaining the working mechanism of "predicting and optimizing at the same time"; continuously expands the model training set using accumulated feedback data to improve generalization performance; the system has the intelligent characteristics of "self-learning and self-evolution" in long-term operation, realizing intelligent modeling, prediction and control of the entire grouting construction process.

[0086] The feedback correction mechanism disclosed herein continuously receives real-time monitoring data from the construction site and compares it with the output results of the response prediction module. The system evaluates the prediction accuracy in real time based on indicators such as mean square error (MSE), mean absolute error (MAE), or intersection-over-union ratio (IoU). When the error exceeds a preset threshold, the system will automatically determine the source of the error and select the corresponding feedback path. When the deviation originates from inaccurate model predictions, the system will use the latest sample data to trigger the model's fine-tuning mechanism, updating the network parameters through small sample increments. When the deviation originates from drastic changes in the diffusion area, the system will automatically trigger local mesh refinement and reconstruction to maintain the accuracy of spatial modeling. If special events such as sudden changes or surges occur in the monitoring data, the system will automatically adjust control parameters (such as the upper limit of grouting pressure) and trigger an early warning mechanism to achieve rapid response to construction risks.

[0087] Example 2 One embodiment of this disclosure provides a grouting simulation system based on temporal interaction and adaptive mesh reconstruction, including: The data acquisition module is used to acquire monitoring data during the grouting construction process and perform preprocessing. The diffusion prediction module is used to input the preprocessed monitoring data into the grouting diffusion prediction model, output the predicted grouting pressure value for the future time step, reconstruct the spatial continuous diffusion pressure field based on the predicted grouting pressure value, obtain the pressure distribution map of the grouting area, extract the grouting diffusion boundary from the pressure distribution map, determine the grouting diffusion area, and calculate the area and volume of the grouting diffusion area. The digital model building module integrates the grouting pressure prediction, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into diffusion response results to form structured data. Based on the structured data, it constructs a geometric model of the grouting area and determines the boundary conditions. The geometric model is then divided into grid cells, and the grouting diffusion area is mapped to the grid cells. The distance between the grid cells and the diffusion boundary is determined. Based on the distance between the grid cells and the diffusion boundary, density control rules are designed. According to the density control rules, the grid cells are locally densified and reconstructed. The reconstructed grid is used to update the topology and boundary conditions between the grid cells to obtain a structured grid data model. The feedback correction module is used to continuously and adaptively update the structured grid data model using real-time grouting monitoring data, enabling real-time visualized grouting optimization control.

[0088] As one embodiment, the data acquisition module is used to collect and aggregate raw data generated by various sensing devices from the grouting construction site, including but not limited to grouting pressure, grouting flow rate, pore water pressure, ground-penetrating radar information, and other auxiliary sensor signals, ensuring data integrity, high timeliness, and a unified timestamp alignment mechanism. The collected raw data is then cleaned and standardized, outliers are removed, missing items are filled, and a unified data structure is formed through normalization and sliding window reconstruction methods. Simultaneously, a multi-dimensional time-series feature vector is constructed for modeling.

[0089] The diffusion prediction module is used to deploy a deep learning model combining a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM) to perform spatiotemporal feature extraction, pattern recognition, and prediction model training or online inference tasks related to grout diffusion behavior. Based on the current input features and the output of the deep fusion model, it predicts the grout diffusion state over a future period during the grout diffusion process, including key response indicators such as pressure change trends, diffusion boundary contours, and diffusion range evolution.

[0090] The digital model building module is used to build or update the spatial grid model of the grouting area based on the diffusion boundary and pressure distribution results output by the response prediction module, dynamically adjust the grid density, and achieve local densification and modeling accuracy optimization in the diffusion proximity zone.

[0091] The feedback correction module is used to receive real-time monitoring data from the construction site, compare and analyze it with the system prediction results, evaluate model errors, and then trigger feedback control operations such as fine-tuning of model parameters, retraining of the structure, or re-division of the mesh to achieve closed-loop optimization of the system.

[0092] The system architecture modules disclosed herein encompass functions such as data acquisition, data preprocessing, deep fusion modeling, response prediction, mesh generation and updating, and feedback correction. The aim is to achieve intelligent modeling, prediction, and control of the entire grouting construction process, as described in detail below: The system first acquires multi-source sensor information from the grouting construction site through a data acquisition module. This module comprises various hardware components, including pressure sensors, flow meters, pore pressure monitoring equipment, diffusion front detection devices (such as piezometers and ground-penetrating radar), and formation parameter acquisition units, to collect physical data closely related to diffusion behavior during the grouting process in real time. All acquired data is uniformly synchronized to a central data channel and timestamped to lay the foundation for subsequent in-depth modeling.

[0093] Next, the system cleans, normalizes, and constructs features from the raw data stream. First, outliers and signal jumps are removed, and missing data is filled using interpolation or moving averages to ensure the stability and continuity of the input data. Then, the system uses min-max normalization or Z-score normalization to unify all physical variables to the same scale. Based on this, a sliding window strategy is used to construct time-series samples, simultaneously fusing multiple dimensions of static and dynamic features, such as grouting pressure, flow rate, pore pressure, geological layer information, and radar impedance changes, ultimately forming a tensor structure that satisfies the input format of a deep neural network.

[0094] The diffusion prediction module employs a spatiotemporal fusion structure based on a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM). In the model structure, the CNN part extracts dynamic change features within local time windows, capturing nonlinear trends and short-term fluctuations; the BiLSTM network learns long-term dependencies, modeling sequence patterns in both forward and backward time directions to achieve a precise understanding of the complex evolutionary behavior of the construction process. Finally, the predicted values ​​of key variables for grout diffusion are output through fully connected layers. The system supports both offline training and online inference, and the model can continuously and adaptively update based on actual construction data. The model output is transformed into a grout diffusion state prediction in a physical field sense. Specifically, after the model outputs a sequence of grouting pressure or pore pressure for several future time steps, the system uses spatial interpolation methods (such as RBF, Kriging, or inverse distance weighting) to restore the predicted values ​​at the points into a continuous two-dimensional or three-dimensional diffusion pressure field. Further, the system sets a diffusion threshold, segments the pressure field into regions, extracts the predicted grout diffusion boundary contour, and calculates indicators such as diffusion area and volume accordingly. The prediction results can be used not only for real-time display, but also for spatial model adjustment by the grid division module.

[0095] The digital model building module transforms response prediction results into a foundation for spatial structure modeling. Through adaptive control of grid density, it achieves high-precision modeling of key diffusion areas. This module first establishes an initial uniform grid model in the construction area and calculates the shortest distance between each element and the diffusion boundary. Based on this, different grid density grading strategies are set according to the distance, with denser grids applied to areas adjacent to the diffusion boundary and coarser grids maintained for areas further away. Delaunay triangulation, quadtree, or octree partitioning algorithms are used to reconstruct the grid of the region, generating a new set of structured grids. All new grids maintain topological continuity and are bound to geological attributes for subsequent numerical simulations or prediction iterations.

[0096] To achieve closed-loop optimization of the system, the feedback correction module continuously receives real-time monitoring data from the construction site and compares it with the output of the response prediction module. The system evaluates prediction accuracy in real time based on indicators such as mean square error (MSE), mean absolute error (MAE), or intersection-over-union ratio (IoU). When the error exceeds a preset threshold, the system automatically identifies the source of the error and selects the appropriate feedback path. When the deviation stems from inaccurate model predictions, the system triggers a model fine-tuning mechanism using the latest sample data, updating network parameters through incremental learning with small samples. When the deviation originates from drastic changes in the diffusion area, the system automatically triggers local mesh refinement and reconstruction to maintain spatial modeling accuracy. If special events such as sudden changes or surges occur in the monitoring data, the system automatically adjusts control parameters (such as the upper limit of grouting pressure) and triggers an early warning mechanism to achieve rapid response to construction risks.

[0097] This system achieves intelligent processing of grouting data throughout its entire lifecycle by constructing an intelligent closed-loop architecture of "perception-prediction-response-feedback". It not only has high-precision diffusion modeling and prediction capabilities, but also good scalability, real-time performance and engineering adaptability, making it particularly suitable for the intelligent control needs of grouting projects under complex geological conditions.

[0098] Example 3 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the grouting simulation method based on time-series interaction and adaptive mesh reconstruction.

[0099] Example 4 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the grouting simulation method based on time-series interaction and adaptive mesh reconstruction.

[0100] Example 5 One embodiment of this disclosure provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the grouting simulation method based on time-series interaction and adaptive mesh reconstruction.

[0101] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A grouting simulation method based on temporal interaction and adaptive mesh reconstruction, characterized in that, include: Acquire monitoring data during the grouting process and perform preprocessing; The preprocessed monitoring data is input into the grouting diffusion prediction model, and the predicted grouting pressure value for the future time step is output. Based on the predicted grouting pressure value, the spatial continuous diffusion pressure field is reconstructed to obtain the pressure distribution map of the grouting area. The grouting diffusion boundary is extracted from the pressure distribution map to determine the grouting diffusion area, and the area and volume of the grouting diffusion area are calculated. The predicted grouting pressure, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map are integrated with the diffusion response results to form structured data. Based on structured data, a geometric model of the grouting area is constructed and boundary conditions are determined. The geometric model is then divided into grid cells, the grouting diffusion area is mapped to the grid cells, and the distance between the grid cells and the diffusion boundary is determined. Based on the distance between the grid cells and the diffusion boundary, density control rules are designed. The grid cells are then locally densified and reconstructed according to the density control rules. The reconstructed grid is used to update the topology and boundary conditions between the grid cells to obtain a structured grid data model. The structured grid data model is continuously updated adaptively using real-time grouting monitoring data to achieve real-time visualized grouting optimization control.

2. The grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in claim 1, characterized in that, The monitoring data obtained during the grouting construction process includes grouting pressure, grouting volume, pore water pressure, grouting time, ground-penetrating radar or ultrasonic reflection signals, and formation description parameters. The monitoring data undergoes preprocessing operations such as data synchronization, outlier removal, and normalization, and is then constructed into a multi-dimensional time-series data sequence.

3. The grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in claim 1, characterized in that, The grouting diffusion prediction model is a deep neural network model that integrates a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM). Multidimensional time-series data sequences are input into the grouting diffusion prediction model. The CNN feature extraction layer uses one-dimensional convolution to extract local features within a local time range. These local features are then input into the BiLSTM time-series modeling layer. The BiLSTM time-series modeling layer captures the long-range dependencies of grouting dynamics by modeling in both forward and reverse time directions. A fully connected layer maps the features output by the BiLSTM to the prediction target space. A linear transformation is performed on the BiLSTM output to obtain the predicted grouting pressure value at a future time or spatial point.

4. The grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in claim 1, characterized in that, The process of integrating the predicted grouting pressure, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into structured data includes: reconstructing the spatially continuous diffusion pressure field using an interpolation algorithm based on the predicted grouting pressure values ​​at representative points output by the grouting diffusion prediction model, forming a two-dimensional or three-dimensional pressure distribution map of the grouting area; setting a diffusion threshold pressure; extracting areas in the pressure distribution map that are higher than the diffusion threshold to determine the grouting diffusion area; using the boundary of the grouting diffusion area as the predicted grouting diffusion boundary; further calculating the effective area and volume of grouting diffusion after obtaining the grouting diffusion area; quantitatively analyzing the grouting coverage effect; and integrating the predicted grouting pressure, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into structured data.

5. The grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in claim 1, characterized in that, The determination of the distance between the grid cell and the diffusion boundary includes: establishing a two-dimensional or three-dimensional geometric model of the grouting area based on structured data, determining the boundary conditions according to the geological profile and construction design, dividing the grouting area into an initial equidistant grid to form a set of grid cells, mapping the grouting diffusion area into the grid cell space, defining a distance function, judging the spatial relationship between each grid cell and the diffusion boundary, and determining the distance between the grid cell and the diffusion boundary.

6. The grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in claim 1, characterized in that, Based on the distance between the mesh cells and the diffusion boundary, density control rules are designed. Local mesh refinement and reconstruction are then performed on the mesh cells according to these rules. The steps include: Step 1: Mark the mesh cells that need to be encrypted, and for those that meet the requirements... Subdivide the grid cells. For the first Target mesh density of mesh cells The preset threshold for response density; Step 2: Employ a mesh densification algorithm, using Delaunay triangulation to densify the two-dimensional irregular region, using a quadtree partitioning method to adaptively refine and densify the regular mesh, and using an octree partitioning method to densify the three-dimensional spatial mesh. Step 3: Generate a new mesh set, ensuring mesh continuity, effective control of element shape quality, and compatibility with simulation tools.

7. A grouting simulation system based on temporal interaction and adaptive mesh reconstruction, characterized in that, include: The data acquisition module is used to acquire monitoring data during the grouting construction process and perform preprocessing. The diffusion prediction module is used to input the preprocessed monitoring data into the grouting diffusion prediction model, output the predicted grouting pressure value for the future time step, reconstruct the spatial continuous diffusion pressure field based on the predicted grouting pressure value, obtain the pressure distribution map of the grouting area, extract the grouting diffusion boundary from the pressure distribution map, determine the grouting diffusion area, and calculate the area and volume of the grouting diffusion area. The digital model building module is used to integrate the grouting pressure prediction value, the boundary point set of the grouting diffusion area, the area and volume of the grouting diffusion area, and the pressure distribution map into diffusion response results to form structured data; Based on structured data, a geometric model of the grouting area is constructed and boundary conditions are determined. The geometric model is then divided into grid cells, and the grouting diffusion area is mapped to the grid cells. The distance between the grid cells and the diffusion boundary is determined. Based on the distance between the grid cells and the diffusion boundary, density control rules are designed. The grid cells are then locally densified and reconstructed according to the density control rules. The reconstructed grid is used to update the topology and boundary conditions between the grid cells, resulting in a structured grid data model. The feedback correction module is used to continuously and adaptively update the structured grid data model using real-time grouting monitoring data, enabling real-time visualized grouting optimization control.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the grouting simulation method based on temporal interaction and adaptive mesh reconstruction as described in any one of claims 1-6.

Images