Tunnel grouting material application effect prediction method and system based on multi-scale layering

Abstract

The invention provides a tunnel grouting material application effect prediction method and system based on multi-scale layering, and relates to the technical field of grouting treatment material prediction. Performing weighted integration on the multi-source data to form a key data table; hierarchically dividing the data to form geological global structure layer data, geological local structure layer data and semantic layer data of grouting material and hydrological data; geological global structure layer data, geological local structure layer data and semantic layer data are input into a multi-task prediction model, the geological global structure layer data and the geological local structure layer data are input into a GNN network structure to extract geological topological structure features, and the semantic layer data are input into a DNN network structure to extract high-dimensional semantic features; and extracting a shared feature map from the geological topological structure features and the high-dimensional semantic features in a shared feature layer, and outputting different target prediction results through a multi-source output regression model based on the shared feature map.

Description

Technical Field

[0001] The present disclosure relates to the technical field of prediction of grouting treatment materials, and particularly to a method and system for predicting the application effect of tunnel grouting materials based on multi-scale stratification. Background Art

[0002] The statements in this section merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] Tunnel water inrush is a common construction risk. Especially when crossing complex geological or high water pressure areas, the sudden inflow of water into the tunnel excavation face will not only increase the construction difficulty and prolong the construction period, but may even cause safety accidents and threaten the lives of construction workers. Therefore, the reasonable application and effect prediction of grouting materials have become the key in the treatment of water inrush.

[0004] However, there are still deficiencies in the existing technology for predicting grouting effects, and it is difficult to accurately predict key performance indicators such as the impermeability, diffusion range, and consolidation strength of materials. This is mainly because the commonly used grouting treatment methods at the present stage mostly rely on empirical design and static models, lacking the ability of dynamic prediction and feedback optimization, and it is difficult to respond in real time to the complex geological conditions and sudden change of hydrological environment during construction. This results in the inability to adjust the grouting material ratio and grouting parameters in real time, affecting the actual treatment effect of grouting materials. In addition, the grouting effect is affected by multi-dimensional factors such as geology, hydrology, materials, and construction parameters, and there are complex non-linear correlations among these factors, while traditional methods are difficult to capture these relationships and effectively predict and optimize them during the treatment process. Summary of the Invention

[0005] In order to solve the above problems, the present disclosure proposes a method and system for predicting the application effect of tunnel grouting materials based on multi-scale stratification. Through multi-level data stratification and scale division, an independent multi-output layer is constructed, and fine modeling is carried out for multi-task objectives such as impermeability performance, diffusion range, and consolidation strength. Each output layer independently optimizes parameters according to different task requirements, making each prediction have high-precision independence. At the same time, in order to achieve the coordinated optimization of multi-objective tasks, a weighted loss function and an adaptive weight allocation strategy are adopted to ensure the balance of the model among multiple task objectives, significantly improving the comprehensiveness and stability of the prediction model.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] A method for predicting the application effect of tunnel grouting materials based on multi-scale stratification includes:

[0008] Obtain multi-source data of tunnel geological data, water inrush and gushing water hydrological data, grouting material data, and construction parameters, and preprocess them; weight and integrate the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0009] Divide the data in the key data table into levels according to the multi-scale hierarchical dimension to form geological global structure layer data, geological local structure layer data, and semantic layer data of grouting materials and hydrological data;

[0010] Input the geological global structure layer data, geological local structure layer data, and semantic layer data into a multi-task prediction model. In the multi-task prediction model, input the geological global structure layer data and geological local structure layer data into the GNN network structure to extract geological topological structure features, input the semantic layer data into the DNN network structure to extract high-dimensional semantic features, extract a shared feature map of the geological topological structure features and high-dimensional semantic features in the shared feature layer, and output the target prediction result through multi-source output regression based on the shared feature map.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions:

[0012] A prediction system for the application effect of tunnel grouting materials based on multi-scale stratification, including:

[0013] A data acquisition module for obtaining multi-source data of tunnel geological data, water inrush and gushing water hydrological data, grouting material data, and construction parameters, and preprocessing them; weight and integrate the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0014] A stratification module for dividing the data in the key data table into levels according to the multi-scale hierarchical dimension to form geological global structure layer data, geological local structure layer data, and semantic layer data of grouting materials and hydrological data;

[0015] A prediction module for inputting the geological global structure layer data, geological local structure layer data, and semantic layer data into a multi-task prediction model. In the multi-task prediction model, input the geological global structure layer data and geological local structure layer data into the GNN network structure to extract geological topological structure features, input the semantic layer data into the DNN network structure to extract high-dimensional semantic features, extract a shared feature map of the geological topological structure features and high-dimensional semantic features in the shared feature layer, and output the target prediction result through multi-source output regression based on the shared feature map.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] A non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] 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 runs, the processor executes the computer program stored in the memory so that the electronic device executes the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification.

[0020] Compared with the prior art, the beneficial effects of the present disclosure are:

[0021] The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification of the present disclosure realizes cross-scale feature extraction of geological, hydrological, material, and construction data through data stratification and scale division at three levels: macro, meso, and micro. At the macro level, it identifies the stability of the global structure of the large-scale geology. At the meso level, it evaluates the characteristics of the local surrounding rock. At the micro level, it deeply analyzes the diffusion and condensation characteristics of the grouting material in the water inrush environment. This multi-scale feature extraction method greatly enhances the feature expression ability of the model and lays a solid foundation for accurate prediction under complex tunnel conditions.

[0022] The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification of the present disclosure combines DNN and GNN to construct a multi-task prediction model. It captures the spatial correlation of geological and hydrological features through GNN and processes high-dimensional grouting material and construction parameter data through DNN to form a more refined multi-objective prediction model. The GNN part is responsible for converting geological structures such as faults and water flow paths into topological structures, and the DNN is responsible for learning the physical and chemical properties of the grouting material. This combined model makes full use of the advantages of different algorithms to accurately describe the multi-dimensional data feature relationship in a complex tunnel environment.

[0023] The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification of the present disclosure introduces a shared feature layer to fuse the features extracted by GNN and DNN in the same feature space, improves the information sharing efficiency, adopts a weighted aggregation strategy to ensure the effectiveness of different data features, and mitigates the problem of gradient disappearance caused by too deep layers through residual connection (ResNet). The shared layer integrates multi-source information, enhances the common feature extraction ability of the model, and improves the prediction accuracy and data sharing effect.

[0024] The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification in the present disclosure constructs a multi-source output regression model and performs fine modeling for the task objectives of impermeability performance, diffusion range, and consolidation strength. The model independently optimizes parameters according to different task requirements, enabling each prediction to have high-precision independence. At the same time, in order to achieve the coordinated optimization of multi-objective tasks, a weighted loss function and an adaptive weight allocation strategy are adopted to ensure the balance of the model among multiple task objectives, significantly improving the comprehensiveness and stability of the prediction model.

[0025] The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification in the present disclosure receives the latest data during construction (such as the mix ratio and rheological properties of the grouting materials) in real time and performs dynamic prediction based on the current geological and hydrological conditions. If the predicted effect deviates from the target standard, the system will automatically trigger a feedback optimization mechanism to adjust the composition ratio of the grouting materials and the configuration of grouting parameters to ensure the stability of the treatment effect. This real-time feedback and optimization function ensures that various parameters during construction can be dynamically adjusted according to environmental changes, enhancing the responsiveness and treatment reliability of the system in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings forming a part of this disclosure are used to provide a further understanding of the disclosure. The schematic embodiments and descriptions thereof of the disclosure are used to explain the disclosure and do not constitute an improper limitation of the disclosure.

[0027] Figure 1 It is a process architecture implementation diagram for the application of the method in the embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0030] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, 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.

[0031] Embodiment 1

[0032] In one embodiment of the present disclosure, a method for predicting the application effect of tunnel grouting materials based on multi-scale stratification is provided, including the following steps:

[0033] Step 1: Obtain multi-source data of tunnel geological data, water inrush and gushing water hydrological data, grouting material data, and construction parameters, and preprocess them; weight and integrate the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0034] Step 2: Divide the data in the key data table into levels according to the multi-scale hierarchical dimension to form geological global structure layer data, geological local structure layer data, and semantic layer data of grouting materials and hydrological data;

[0035] Step 3: Input the geological global structure layer data, geological local structure layer data, and semantic layer data into a multi-task prediction model. In the multi-task prediction model, input the geological global structure layer data and geological local structure layer data into the GNN network structure to extract geological topological structure features, input the semantic layer data into the DNN network structure to extract high-dimensional semantic features, extract a shared feature map of the geological topological structure features and high-dimensional semantic features in the shared feature layer, and output the target prediction result through multi-source output regression based on the shared feature map.

[0036] As an embodiment, the specific implementation process of a prediction method for the application effect of tunnel grouting materials based on multi-scale stratification of the present disclosure is as follows:

[0037] Step 1: Obtain multi-source data of tunnel geological data, water inrush and gushing water hydrological data, grouting material data, and construction parameters, and preprocess them;

[0038] Specifically, (1) Source and acquisition of data

[0039] Firstly, geological exploration techniques and on-site monitoring systems are utilized to collect geological structure information of the tunnel area. Specifically, the system obtains the distribution of rock formations through methods such as drilling and ultrasonic imaging, and simultaneously identifies and records special geological structures such as fault structures and karst cave structures. Through in-depth understanding of the regional geological conditions, a reliable geological basis can be provided for subsequent data analysis and model establishment; Hydrological data is obtained through real-time monitoring. Monitoring devices such as flow meters, pressure gauges, and temperature sensors are deployed at the tunnel construction site to collect key hydrological parameters of water inrush, including water pressure, water level, flow direction, and temperature. The monitoring devices transmit data through networking, continuously update the dynamic change information of water inrush, and send these real-time data to the data center. Through dynamic monitoring of hydrological data, the system can capture the change trend of water inrush in the tunnel area at different time periods, providing key hydrological inputs for the intelligent prediction model; Establishment of grouting material data: To accurately analyze the adaptability of different grouting materials, a database of the physical and chemical properties of grouting materials is established through laboratory tests and historical project data. Specifically, through laboratory experiments, parameter indicators such as construction density, viscosity, setting time, and impermeability performance of common grouting materials (such as cement slurry, chemical slurry, etc.) are measured, and the ingredient ratios and types of admixtures of different materials are systematically recorded. At the same time, by sorting out and summarizing the grouting treatments of historical projects, reference is provided for the application effects of different grouting materials under various geological conditions.

[0040] Furthermore, all the collected data is effectively stored and managed through the database module of the system, establishing a unified data structure and access interface. Datasets such as geological data, hydrological data, grouting material characteristics, and construction parameters are all stored in the database for subsequent calling and analysis. At the same time, the database has a data backup and recovery mechanism to ensure the security and integrity of the data, providing long-term support for the continuous update and optimization of the model.

[0041] Step 2: Weightedly integrate the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0042] Specifically, all the obtained data is respectively cleaned, complemented, and noise and outliers are removed, and standardized and normalized to complete the preprocessing process to ensure the integrity, effectiveness, and consistency of the model input data, including:

[0043] Missing values are complemented using interpolation or mean substitution methods. For time series data, such as water pressure and water level of hydrological parameters, linear interpolation methods are preferentially used for supplementation to maintain the time continuity and rationality of the data; for non-time series data, such as material characteristic parameters, mean substitution is used to fill in the missing items to ensure the stability of the data distribution.

[0044] Adopt anomaly detection algorithms to identify and eliminate noise data and outliers. For geological parameter data, use clustering algorithms to identify outliers with large deviations from the mainstream distribution; for time series data, adopt the sliding window method to detect mutation data that does not conform to the trend. Through outlier elimination, the system can clear invalid information and provide more real and reliable data support for model training.

[0045] Standardize and normalize geological, hydrological, material properties, and construction parameters. Geological parameter and material property data need to be standardized due to different units and dimensions. Convert each data feature into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate dimensional differences. Hydrological data and construction parameters are normalized to compress the data into the [0,1] interval to improve the convergence speed of the model.

[0046] Use filtering algorithms to denoise time series data such as hydrological and construction parameters. Specifically, for the volatility of hydrological parameters such as water pressure and flow rate, smooth the data curve through low-pass filtering or Kalman filtering algorithms to remove the interference of high-frequency noise. For instantaneous abnormal peaks in construction parameters, use mean filtering to eliminate abnormal fluctuations and retain the main trend information of the data.

[0047] Among them, the standardization formula is:

[0048]

[0049] Among them, X is the original data, μ is the mean, and σ is the standard deviation.

[0050] Step 3: Weightedly integrate the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0051] Specifically, align the time and space of the preprocessed multi-source data, and then hierarchically process the aligned multi-source data in the time dimension. Construct a comprehensive data table according to the stability of the geological structure, the dynamic changes of hydrological characteristics, the physical and chemical properties of material characteristics, and the operating conditions of construction parameters. Among them, fuse the structural parameters and hydrological characteristics of the tunnel geological environment hierarchically, and at the same time correlate the chemical composition of the grouting material with the construction parameters to form a sequence in the network input format. Then, preliminarily weightedly integrate the data according to the correlation between data features to obtain a key data table that has an important impact on predicting the grouting effect.

[0052] Among them, to ensure the consistency of geological, hydrological, material properties, and construction data in time and space, first, the spatio-temporal attributes of all data are aligned. The alignment process matches data from different sources according to timestamps and spatial positions, so that each data record can correspond at the same time period and the same location. For geological data, this involves matching spatial information such as faults and rock layer distributions with the collected hydrological parameters (such as water pressure and water level); for time series data, such as the flow rate, temperature of hydrological monitoring, and construction parameters (such as grouting speed and pressure), the alignment step precisely matches these data according to timestamps to ensure the dynamic consistency of the data.

[0053] Furthermore, after the data alignment is completed, the present disclosure performs hierarchical fusion processing on the multi-source data according to the data sources and contents. Specifically, data such as geological structures, hydrological characteristics, material properties, and construction parameters that have been standardized and spatio-temporally aligned are processed in layers in the time dimension. According to the stability of the geological structure, the dynamic changes of hydrological characteristics, the physicochemical properties of material properties, and the operating conditions of construction parameters, a comprehensive data table is constructed. In this process, the system fuses the macroscopic parameters (such as fault positions and karst cave distributions) of the tunnel geological environment with the hydrological conditions (such as underground water flow direction and water level fluctuations) hierarchically, and at the same time correlates the chemical composition of the grouting material with the construction process parameters to form the basic structure (sequence) of the network input format, providing consistent and comprehensive data input for the model.

[0054] Further, after the multi-source data is fused, the data is initially weighted and integrated according to the correlation between the data characteristics. By comparing the correlations of historical project data such as geological conditions, hydrological environments, and construction effects, the system automatically identifies the key characteristics that affect the treatment effect of water inrush and weights these characteristics. For example, in areas with significant water pressure changes, the system assigns higher weights to hydrological parameters; in geological environments with relatively dense fault distributions, higher weights are assigned to fault distributions and rock layer type characteristics to ensure that the data characteristics that have an important impact on the prediction of grouting effects can be reasonably presented during the data fusion process. Further, the high-quality data that has completed hierarchical fusion and weighted integration is stored as a multi-dimensional data set of the system to support the subsequent training and analysis of the model.

[0055] Among them, the weighted average formula is:

[0056]

[0057] Among them, f i is the i-th scale feature, w i is its weight, and m is the total amount of scale features.

[0058] Step 4: Classify the data in the key data table according to the multi-scale hierarchical dimension to form the data of the geological global structure layer, the data of the geological local structure layer, and the semantic layer data of the grouting material and the hydrogeological data;

[0059] Specifically, first, the data is classified into three levels: macro, meso, and micro according to the spatial scope and semantic information of the data. The specific classification principle is as follows: The macro level includes the global geological structure characteristics of the tunnel area, which are used to identify the stability of the large-scale geological environment; the meso level focuses on the local geological structure characteristics of the tunnel surrounding rock to evaluate the impact of local geological conditions on the grouting effect; the micro level is the semantic information of the refined physical and chemical properties of the grouting material and the hydrogeological parameters of the water inrush, which is used to analyze the diffusion and condensation behavior of the material in the microenvironment.

[0060] Among them, the macro level mainly covers the large-scale geological global structure characteristics of the tunnel construction area, collects and integrates key geological information such as faults, folds, karst caves, and rock layer boundaries to build a foundation for the stability assessment of the entire tunnel area. At the macro level, the data is mainly used to identify the changing trend of large-scale geological conditions and predict the possible paths of water inrush. The macro data also includes the dynamic change information of the regional geological environment, such as the change of the regional water level and the activity status of the fault, to evaluate the overall safety of the tunnel.

[0061] The meso level mainly focuses on the local geological characteristics of the tunnel surrounding rock. The data covers information such as the porosity, density, rock type, and water content of the surrounding rock during the tunnel excavation process. The main goal of the meso-level data is to evaluate the direct impact of the local geological conditions of the surrounding rock on the grouting effect, which is convenient for analyzing the diffusion path and condensation behavior of the grouting material in the local environment.

[0062] The micro level mainly covers the physical and chemical properties of the grouting material and the hydrogeological parameters of the water inrush to accurately analyze the diffusion, penetration, and condensation characteristics of the material in the microenvironment. The data includes parameters such as the particle distribution, chemical composition, setting speed, and anti-seepage ability of the grouting material, as well as the water pressure, water temperature, and flow rate of the water inrush.

[0063] Finally, after the data level classification is completed, the system stores the macro, meso, and micro three-layer data in a structured manner to form a multi-level data set. By tagging and indexing the data at different levels, the system can flexibly call each layer of data, which is convenient for realizing the collaborative analysis of macro trends, meso details, and micro characteristics in different analysis modules.

[0064] Step 5: Input the geological global structure layer data, geological local structure layer data, and semantic layer data into the multi-task prediction model. In the multi-task prediction model, input the geological global structure layer data and geological local structure layer data into the GNN network structure to extract geological topological structure features, and input the semantic layer data into the DNN network structure to extract high-dimensional semantic features. Extract the shared feature map of the geological topological structure features and high-dimensional semantic features in the shared feature layer, and output the target prediction result through multi-source output regression based on the shared feature map.

[0065] Specifically, S1: Construction and pre-training of the multi-task prediction model

[0066] First, the multi-task prediction model includes a dual-channel network structure. Adopting a multi-task learning framework and combining the advantages of the deep neural network (DNN) and the graph neural network (GNN), the dual-channel network structure is the CNN network structure and the GNN network structure respectively, and a multi-object prediction model suitable for complex tunnel environments is constructed.

[0067] In the model structure, the GNN network structure part is mainly used to capture the topological structure and non-linear spatial relationship of geological and underground structure features, while the DNN network structure part is used to process high-dimensional and continuous hydrogeological parameters, grouting materials, and construction parameters. The specific construction steps are as follows:

[0068] 1) Construct the GNN network structure: First, convert the features such as faults, fractures, and water flow paths in the geological structure into a graph structure, where the nodes represent spatial positions and the edges represent the spatial correlation between nodes (such as water flow paths, fracture extension directions, etc.). Then, assign an initial feature vector to each node, and the vector content includes the geological features (water content, porosity, density, etc.) and hydrogeological characteristics (such as flow velocity, pressure) of the area corresponding to the node.

[0069] Then, through the message passing mechanism, the GNN passes information between each node and its neighbor nodes to update the feature representation of the nodes. Finally, the GNN gradually captures the spatial correlation between nodes through multiple layers of iteration, integrates the information of surrounding nodes into the feature representation of each node, and thus learns the non-linear spatial relationship of geological structure features.

[0070] Among them, the formula of the message passing mechanism is:

[0071]

[0072] Among them, represents the message aggregation of node v at the k-th layer, N(v) is the set of neighbor nodes of node v, is the feature vector of neighbor node u, and e uv represents the edge feature between node u and node v; is the feature vector of node v.

[0073] Furthermore, the node update formula is:

[0074]

[0075] Where, is the feature representation of node v at the k+1-th layer, W (k) is the learned weight matrix, σ is the activation function, and b (k) is the bias vector at the k-th layer.

[0076] Furthermore, the input of the GNN model is the topological graph structure formed by tunnel geology and hydrological flow, and the output end is the embedding vector of geology and hydrological dynamics, that is, the geological topological structure feature. The update formula is:

[0077]

[0078] Where, is the feature vector of node v at the k+1-th layer, W (k) is the learned weight matrix, N(v) is the neighbor node, and c vu is the normalization constant.

[0079] 2) Construct the DNN network structure. The input end of the DNN model is the standardized hydrological parameters, material properties, and construction parameters, and the output end is the high-dimensional semantic feature. The calculation formula is:

[0080]

[0081] Where, y is the output high-dimensional semantic feature, x i is the input data, w i is the weight, b is the bias constant, and f is the activation function.

[0082] Furthermore, the forward propagation formula of the hidden layer of the DNN model is:

[0083] h (l) = f(W (l-1) h (l-1) + b (l-1) )

[0084] Where, h (l) is the output of the neuron at the l-th layer, W (l-1) and b (l-1) are the weight matrix and bias from the (l-1)-th layer to the l-th layer, and f is the activation function.

[0085] 3) Construction of the shared feature layer: A shared feature layer is set after the CNN network structure and the GNN network structure to extract the common information of geological, hydrological, material, and construction parameters, so as to improve the information sharing efficiency among multiple tasks. The shared feature layer fuses the feature outputs h of GNN and DNN combined , integrating the topological structure of geological and hydrological information with the high-dimensional features of material and construction parameters in the same feature space. The feature fusion strategy of the shared layer adopts a weighted aggregation method, ensuring the effectiveness of various data features during the fusion process. At the same time, the system uses the residual connection (ResNet) method to mitigate the problem of gradient disappearance caused by overly deep layers. After weighted aggregation and residual connection processing, the final shared feature h shared contains the spatial topological features of geological and hydrological information and the high-dimensional features of grouting materials and construction parameters. The output of this shared feature layer will be used as the input for multi-task prediction, providing common feature support for each task (such as impermeability prediction, diffusion range prediction, and consolidation strength prediction).

[0086] Assume that the feature vector output by GNN is h GNN , and the feature vector output by DNN is h DNN . The comprehensive feature vector h combined is:

[0087] h combined = [h GNN , h DNN

[0088] Furthermore, the weighted aggregation formula is:

[0089] h shared = α GNN · h GNN + α DNN · h DNN

[0090] where α GNN and α DNN are the weight coefficients of GNN and DNN features respectively, satisfying α GNN + α DNN = 1.

[0091] Furthermore, the residual connection formula is:

[0092] h residual = h combined + F(h combined )

[0093] where F(h combined ) represents the non-linear transformation of the fused features, such as linear transformation and activation through one or more layers of neural networks.​

[0094] 4) Construction of multi-source output regression model: To achieve multi-objective prediction, a multi-source output regression model is set up after the shared feature layer in the model. This model combines a dynamic weighting mechanism and a task-specific input adjustment strategy, and designs independent output modules focusing on different input data characteristics for each task. Each output module corresponds to a different prediction task, including impermeability performance, diffusion range, and consolidation strength.

[0095] Among them, the impermeability performance prediction mainly focuses on the impermeability effect of the material, with the tunnel water pressure and the penetration parameters of the grouting material as inputs; the diffusion range prediction predicts the diffusion area of the material in the water inrush based on data such as grouting pressure, viscosity, and water flow rate; the consolidation strength prediction predicts the consolidation effect based on information such as the setting time of the grouting material and the construction temperature. Each output layer conducts independent parameter optimization and error feedback for its respective task objectives.

[0096] Furthermore, before entering each output layer, the model assigns different weights to the input features according to the task requirements to generate task-specific feature subsets. The weighted feature of task t is:

[0097] h t =Weighting(h shared ,W t )

[0098] where W t is the feature weighting matrix of task t, dynamically adjusting the contribution degree of each feature.

[0099] Furthermore, the regression model for each task independently designs the model structure and hyperparameters according to the task characteristics to achieve non-linear modeling of specific features. Specifically as follows:

[0100] The impermeability performance prediction uses a multi-layer perceptron (MLP) to extract the non-linear relationship of features layer by layer:

[0101] y 抗渗 =f 抗渗 (W 抗渗 ·h 抗渗 +b 抗渗 )

[0102] where f 抗渗 is a non-linear activation function designed based on the impermeability characteristic prediction, h 抗渗 is the weighted impermeability prediction feature set, and W 抗渗 and b 抗渗 are the weight and bias parameters of the prediction model respectively.

[0103] The diffusion range prediction uses a convolutional neural network (CNN) to capture the local correlation between the characteristics of the grouting material and the hydrogeological parameters:

[0104] y 扩散 = CNN(h 扩散 ; θ 扩散 )

[0105] where h 扩散 is the weighted diffusion performance prediction feature set, and θ 扩散 are the parameters of the diffusion performance prediction task model.

[0106] The consolidation strength prediction uses a Recurrent Neural Network (RNN) to analyze the dynamic changes in the consolidation process in combination with time series data:

[0107] y 固结 = RNN(h 固结 ; θ 固结 )

[0108] where h 固结 is the weighted consolidation strength performance prediction feature set, and θ 固结 are the parameters of the consolidation strength performance prediction task model.

[0109] 5) Model Optimization

[0110] To balance the losses between different task objectives, the model adopts a weighted loss function, and assigns loss weights according to the importance of each task and the convergence situation during training. The initial weights are set based on the priority of the tasks and the initial performance of the model for each task. During the training process, the model dynamically adjusts the weights to achieve coordinated optimization of each task. The adaptive weight allocation strategy adaptively adjusts through a weight adjustment function according to the loss feedback of each task to ensure that the model can dynamically respond to the convergence situation of each task in multi-task learning. Specifically, when the error of a certain task is large, the system increases the weight of this task to enhance the attention during training, so that the model can achieve better coordination between different objectives.

[0111] Furthermore, the calculation formula of the multi-task weighted loss function is:

[0112]

[0113] where is the loss of the t-th task, and α t is its weight.

[0114] Furthermore, during the model training process, the Stochastic Gradient Descent (SGD) algorithm is used to optimize the parameters. To improve the efficiency and accuracy of model training, the system uses a mini-batch training strategy and updates the model parameters for each batch of data to gradually reduce the loss function and approach the global optimal solution. To further improve the convergence speed and prediction accuracy of the model, the Bayesian optimization algorithm is adopted to automatically select the best combination of hyperparameters. Bayesian optimization intelligently selects the new round of hyperparameters based on the training results of historical hyperparameter combinations to avoid redundant parameter tuning processes and enable the model to quickly achieve the best training effect.

[0115] Furthermore, the formula for the Stochastic Gradient Descent (SGD) optimization algorithm is as follows:

[0116]

[0117] where θ is the model parameter and η is the learning rate.

[0118] Furthermore, after the training is completed, the model evaluates the multi-objective prediction performance through the validation set, and uses the Mean Absolute Error (MAE) to quantitatively analyze the prediction results of anti-seepage performance, diffusion range, consolidation strength, etc. The system fine-tunes the model based on the performance of the validation set to ensure the stability and accuracy of the model among different task objectives. Finally, the system saves the best model parameters for each task for model loading and rapid deployment in subsequent applications.

[0119] The Mean Squared Error (MSE) is calculated through the validation set to evaluate the model performance, and the calculation formula is as follows:

[0120]

[0121] where y i is the actual value, is the predicted value, and n is the number of samples in the validation set.

[0122] As an embodiment, the method of the present disclosure is used for real-time prediction and feedback optimization. By establishing a real-time prediction and feedback optimization mechanism and combining the real-time data during the grouting process, dynamic adjustment of the material ratio is realized. The specific operation is as follows:

[0123] First, during the grouting construction process, the system collects key data in real time from sensors and monitoring devices, including: the mixing ratio parameters of grouting materials (such as the ratio of cement, water, additives, etc.), the rheological properties of materials (such as viscosity, fluidity, etc.), geological and hydrological conditions (parameters such as fault and fracture distribution, water flow direction and velocity in the current construction section). All data is transmitted to the central database of the system through the network and serves as the real-time input of the model.

[0124] Next, after receiving the latest data, the system makes predictions based on a multi-task learning model and a multi-objective optimization algorithm. The DNN part of the model analyzes the high-dimensional characteristics of the material (such as viscosity and rheology). The GNN part is responsible for modeling the topological relationships of geological structures and hydrogeological features and generating hydrogeological feature vectors. The outputs of the DNN and GNN are fused in a shared feature layer, concatenated to generate a comprehensive feature vector, and three targets are predicted respectively through a multi-source output regression model: anti-seepage effect prediction, diffusion range prediction, and consolidation strength prediction. The prediction results are weighted and synthesized through a multi-objective optimization algorithm to judge the applicability of the current material ratio and output prediction values for subsequent feedback optimization.

[0125] Finally, if the real-time prediction effect deviates from the design requirements or standard values, the system triggers a feedback optimization mechanism. First, differential analysis and deviation feedback are carried out: the system compares the predicted values with the target treatment standards and calculates the deviation amounts of anti-seepage performance, diffusion range, and consolidation strength. On the premise of ensuring the balance of anti-seepage, diffusion, and consolidation targets, a multi-objective optimization algorithm is used to adjust the material ratio parameters. The optimization algorithm dynamically adjusts the material composition ratio according to the current actual deviation to achieve the optimization goal. After the model obtains the new material ratio parameters, the system feeds back the optimization results to the construction site, updates the ratio of the grouting material, and continues to monitor the construction situation. If the deviation is effectively corrected, the current ratio is maintained; if the deviation still exists, the feedback optimization process is repeated until the treatment standards are met.

[0126] Furthermore, the core of multi-objective optimization is to construct an objective function, which transforms the deviation amounts of multiple prediction indicators (anti-seepage performance, diffusion range, consolidation strength) into an optimizable mathematical form:

[0127] First, let the ratio parameters of the grouting material be P = [p 1 , p 2 , …, p n , where p i represents the ratio of a certain component (such as cement, water, additive), and the ratio parameters need to satisfy the total ratio of 1: p i ≥ 0. Then, minimize the weighted sum of the deviations of each prediction indicator: where α t is the weight of task t, dynamically adjusted according to the task priority to ensure target balance, is the deviation loss function of task t (anti-seepage, diffusion, consolidation): where is the model prediction value based on the current material ratio parameter p, is the target value of task t.

[0128] Then, the weighted summation method is adopted to synthesize multiple targets through the weight α tOptimize the weighted sum into a single objective to obtain the objective function as follows: Select the initial ratio parameter P 0 , and update the ratio parameter using the gradient descent method: where η is the learning rate, and is the gradient of the objective function.

[0129] Finally, select the globally minimized P optimized by the weighted sum method * and submit it to the construction system as the optimal ratio parameter for the configuration of grouting materials. Update the model prediction according to the actual construction results. If the deviation still exists, repeat the optimization process until the target requirements are met.

[0130] Example 2

[0131] In an embodiment of the present disclosure, a prediction system for the application effect of tunnel grouting materials based on multi-scale stratification is provided, including:

[0132] A data acquisition module for acquiring multi-source data of tunnel geological data, water inrush and outburst hydrological data, grouting material data, and construction parameters, and performing preprocessing; integrating the preprocessed multi-source data according to geological structure, hydrological characteristics, material characteristics, and construction characteristics to form a key data table;

[0133] A stratification module for dividing the data in the key data table into levels according to multi-scale stratification dimensions to form geological global structure layer data, geological local structure layer data, and semantic layer data of grouting materials and hydrological data;

[0134] A prediction module for inputting the geological global structure layer data, geological local structure layer data, and semantic layer data into a multi-task prediction model. In the multi-task prediction model, input the geological global structure layer data and geological local structure layer data into the GNN network structure to extract geological topological structure features, input the semantic layer data into the DNN network structure to extract high-dimensional semantic features, extract a shared feature map of the geological topological structure features and high-dimensional semantic features in the shared feature layer, and output the target prediction result through a multi-source output regression model based on the shared feature map.

[0135] Example 3

[0136] In an embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification is implemented.

[0137] Example 4

[0138] In an embodiment of the present disclosure, an electronic device is provided, including: 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 runs, the processor executes the computer program stored in the memory so that the electronic device executes the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification as described above.

[0139] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one or more flows Figure 1 one or more flows and / or blocks Figure 1 or a plurality of blocks.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate computer-implemented processing, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows Figure 1 one or more flows and / or blocks Figure 1 or a plurality of blocks.

[0141] Although the specific implementation manners of the present disclosure have been described above in conjunction with the accompanying drawings, they do not limit the protection scope of the present disclosure. Those skilled in the art should understand that, based on the technical solutions of the present disclosure, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present disclosure.

Claims

1. A method for predicting the application effect of tunnel grouting materials based on multi-scale stratification, characterized in that: include: Acquire multi-source data including tunnel geological data, sudden water hydrological data, grouting material data and construction parameters, and pre-process them; The pre-processed multi-source data are weighted and integrated according to geological structure, hydrological characteristics, material characteristics and construction characteristics to form a key data table; Divide the data in the key data table into levels according to the multi-scale hierarchical dimensions to form the semantic layer data of geological global structure layer data, geological local structure layer data, and grouting material and hydrological data; The geological global structure layer data, geological local structure layer data and semantic layer data are input into the multi-task prediction model. In the multi-task prediction model, the geological global structure layer data and geological local structure layer data are input into the GNN network structure to extract geological topological structure features, and the semantic layer data is input into the DNN network structure to extract high-dimensional semantic features. The geological topological structure features and high-dimensional semantic features are used to extract shared feature maps in the shared feature layer, and the target prediction results are output through multi-source output regression based on the shared feature maps.

2. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 1, characterized in that: Through drilling and ultrasonic imaging, geological data on rock distribution, fault structure and cave structure are obtained. Hydrological data include water pressure, water level, flow direction and temperature data. Grouting material data include grouting materials, proportions and physical and chemical properties. Construction parameters include density, viscosity, setting time and various performance indicators. All the acquired data are cleaned, completed, and noise and outliers are eliminated. They are standardized and normalized to complete the preprocessing process.

3. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 1, characterized in that: The pre-processed multi-sources are weighted and integrated according to geological structure, hydrological characteristics, material characteristics and construction characteristics to form key data tables, including: The preprocessed multi-source data are aligned in time and space, and then the aligned multi-source data are processed in layers in the time dimension. A comprehensive data table is constructed according to the stability of the geological structure, the dynamic changes of the hydrological characteristics, the physical and chemical properties of the material characteristics, and the operating conditions of the construction parameters. Among them, the structural parameters of the tunnel geological environment and the hydrological characteristics are integrated hierarchically, and the chemical composition of the grouting material is associated with the construction parameters to form a sequence in the network input format. Then, the data are preliminarily weighted and integrated according to the correlation between the data characteristics to obtain a key data table that has an important impact on the prediction of the grouting effect.

4. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 1, characterized in that: The data in the key data table are divided into levels according to the multi-scale hierarchical dimensions to form the global geological structure layer data, local geological structure layer data, and semantic layer data of grouting materials and hydrological data, including: according to the spatial scope and semantic information of the data, the data are divided into three levels: macro, meso and micro. The specific division principles are as follows: the macro level contains the global geological structure characteristics of the tunnel area, which is used to identify the stability of the large-scale geological environment; the meso level is the local geological structure characteristics of the tunnel surrounding rock, which evaluates the influence of local geological conditions on the grouting effect; the micro level is the semantic information of the detailed physical and chemical properties of the grouting material and the hydrological parameters of the sudden water, which is used to analyze the diffusion and condensation behavior of the material in the microenvironment.

5. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 1, characterized in that: The macro, meso and micro data are structured and stored to form multi-level data, and the multi-level data are input into the multi-task prediction model. The multi-task prediction model includes a dual-channel network structure, which is a CNN network structure and a GNN network structure. After the multi-level data are input into the multi-task prediction model, the macro and meso data enter the GNN network structure to capture the geological topological structure and nonlinear spatial relationship, and extract the geological topological structure characteristics; The micro-level data enters the CNN network structure to process high-dimensional and continuous hydrological data, grouting materials and construction parameters, and extract high-dimensional semantic features.

6. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 1, characterized in that: A shared feature layer is set after the CNN network structure and the GNN network structure. The geological topological structure features and high-dimensional semantic features are spliced ​​and combined on the shared feature layer to extract the shared feature map. The model sets a multi-source output regression model in the shared feature layer. Combined with the dynamic weighting mechanism and the task-specific input adjustment strategy, an independent output module focusing on different input data characteristics is designed for each task. Each output module corresponds to a different prediction task.

7. The method for predicting the application effect of tunnel grouting materials based on multi-scale stratification according to claim 6, characterized in that: Each output module corresponds to a different prediction task, including impermeability, diffusion range and consolidation strength.

8. The tunnel grouting material application effect prediction system based on multi-scale stratification is characterized by: include: Data acquisition module, used to acquire multi-source data including tunnel geological data, sudden water hydrological data, grouting material data and construction parameters, and pre-process them; The pre-processed multi-source data are weighted and integrated according to geological structure, hydrological characteristics, material characteristics and construction characteristics to form a key data table; A hierarchical module is used to divide the data in the key data table into hierarchical levels according to multi-scale hierarchical dimensions to form semantic layer data of geological global structural layer data, geological local structural layer data, and grouting material and hydrological data; The prediction module is used to input geological global structure layer data, geological local structure layer data and semantic layer data into a multi-task prediction model. In the multi-task prediction model, the geological global structure layer data and geological local structure layer data are input into the GNN network structure to extract geological topological structure features, and the semantic layer data are input into the DNN network structure to extract high-dimensional semantic features. The geological topological structure features and high-dimensional semantic features are used to extract shared feature maps in the shared feature layer, and the target prediction results are output through multi-source output regression based on the shared feature maps.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification as described in any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: 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 so that the electronic device executes the method for predicting the application effect of tunnel grouting materials based on multi-scale stratification as described in any one of claims 1-7.

Images