Fine intelligent grouting method and system based on multi-dimensional parameter fusion in drilling process

Abstract

The application belongs to the technical field of intelligent grouting, and particularly discloses a fine intelligent grouting method and system based on multi-dimensional parameter fusion in a drilling process, which comprises the following steps: obtaining drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data respectively; extracting speed fluctuation, torque-speed ratio, vibration frequency spectrum characteristic features, fracture density, texture change features, sand content change rate, water inflow gradient features, and hydrological dynamic correlation features respectively; calculating space-time coupling representation weights based on the extracted features; inputting the extracted features into a comprehensive prediction model to obtain collapse sand grade, water inflow intensity, spatial distribution and time change trend, and risk probability prediction results; and adaptively adjusting grouting parameters based on the prediction results to generate a fine grouting scheme. The application innovatively introduces a space-time feature extraction and dynamic weight adjustment mechanism, combines deep analysis of multi-source data, and improves the prediction accuracy of complex stratum risks.

Description

Technical Field

[0001] This invention relates to the field of intelligent grouting technology, and in particular to a refined intelligent grouting method and system based on the fusion of multi-dimensional parameters in the drilling process. Background Technology

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

[0003] In tunnel construction in water-rich sandy gravel strata, the geological complexity and high water content make sand inrush and water inrush two common and frequent geological hazards, directly threatening construction safety, progress, and structural stability. Sand inrush refers to the sudden influx of sand, gravel, and water into the tunnel under pressure differentials. Especially during excavation, the uneven size and distribution of sand and gravel particles, coupled with poor stratum stability, easily lead to sand instability and collapse, causing construction interruptions, equipment damage, and even casualties. Water inrush is frequent due to the high permeability of water-rich sandy gravel strata and the dynamic changes in groundwater flow. Water inrush not only erodes the stratum structure, triggering geological hazards (such as ground subsidence and tunnel collapse), but can also lead to equipment failure, construction delays, and safety accidents. Furthermore, both hazards often occur simultaneously in such strata. Therefore, accurately predicting the risks of sand inrush and water inrush, and implementing effective grouting reinforcement measures, is crucial to ensuring the safety and efficiency of tunnel construction.

[0004] Traditional methods for predicting sand runoff and water inrush mainly rely on geological exploration and numerical simulation. However, geological exploration methods are limited by the number and distribution of exploration points, resulting in insufficient spatial coverage of geological data and difficulty in fully reflecting the heterogeneity and risk distribution of strata. Especially in complex geological environments, the specific location and scale of sand runoff and water inrush are difficult to pinpoint accurately. Numerical simulation methods simulate sand runoff and water inrush processes by establishing strata mechanics models. Although they can reflect the physical behavior of strata to some extent, their prediction accuracy is limited by the accuracy of model parameters and computational resources.

[0005] Existing technologies improve prediction accuracy through deep learning methods; however, sand runoff prediction relies heavily on drilling parameters and geological data, lacking dynamic analysis of hydrodynamic and hydrochemical characteristics. While water inrush prediction considers hydrodynamics and hydrochemistry, it does not fully integrate surrounding rock images and drilling parameters, resulting in incomplete risk assessment. Furthermore, existing methods typically model sand runoff or water inrush separately, lacking a comprehensive assessment of both risks, making it difficult to handle complex scenarios involving the coupled effects of sand runoff and water inrush in water-rich sandy gravel formations. Moreover, existing prediction models are mostly based on static or historical data, failing to reflect the dynamic changes in formation and hydrological conditions during construction in real time, thus failing to meet the needs for real-time risk early warning. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a refined intelligent grouting method and system based on the fusion of multi-dimensional parameters during the drilling process. This method comprehensively acquires drilling parameters, geological and hydrological data, surrounding rock images, hydrodynamic and hydrochemical data, and constructs a comprehensive prediction model for sand inrush prediction and water inrush prediction, thereby enabling intelligent prediction of sand inrush grade, water inrush intensity, spatial distribution, and temporal variation trends.

[0007] In some implementations, the following technical solutions are adopted:

[0008] A refined intelligent grouting method based on multi-dimensional parameter fusion in the drilling process includes:

[0009] Drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data were acquired separately and preprocessed separately; each type of data was then merged after preprocessing.

[0010] Based on drilling parameters, velocity fluctuation, torque-velocity ratio, and vibration frequency spectrum characteristics are extracted respectively; based on surrounding rock images, fracture density and texture variation characteristics are extracted respectively; based on geological and hydrological parameters, sand content change rate and water inflow gradient characteristics are extracted respectively; and based on hydrodynamic and hydrochemical data, hydrological dynamic correlation characteristics are extracted.

[0011] A dynamic weight based on risk sensitivity is introduced, and the spatiotemporal coupling representation weight is calculated based on the extracted features;

[0012] Construct a comprehensive prediction model consisting of a sand runoff prediction sub-model and a water inrush prediction sub-model;

[0013] Input the sand content change rate, velocity fluctuation characteristics, torque-velocity ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient and spatiotemporal coupling characterization weights, as well as drilling parameters, into the sand collapse prediction sub-model; input the hydrodynamic data, hydrochemical data, hydrological correlation characteristics, water inflow gradient and spatiotemporal coupling characterization weights into the water inflow prediction sub-model.

[0014] Finally, the predicted results of sand erosion level, water inflow intensity, spatial distribution and temporal variation trend and risk probability are obtained; based on the predicted results, the grouting parameters are adaptively adjusted to generate a refined grouting scheme.

[0015] As a further embodiment, the drilling parameters include drilling speed, drilling torque, drilling position, and drill pipe vibration frequency data;

[0016] The geological and hydrological parameter data include sand content and water inflow data;

[0017] The hydrodynamic data includes: water level, flow rate, flow velocity, and data pressure; the hydrochemical data includes ion concentration, pH value, and redox potential.

[0018] As a further solution, the texture variation feature specifically includes:

[0019] ;

[0020] in, For texture variation features, For the first Texture contrast of frame images For the first Texture contrast of frame images The number of frames.

[0021] As a further solution, the hydrological dynamic correlation characteristics are specifically as follows:

[0022] ;

[0023] in, The correlation coefficient between water flow rate and ion concentration. For a moment The amount of water flowing out, For a moment ion concentration, and These represent the average values ​​of water flow rate and ion concentration, respectively. This represents the size of the time window.

[0024] As a further approach, spatiotemporal coupling representation weights are calculated based on the extracted features, specifically as follows:

[0025]

[0026] in, For at any time and three-dimensional coordinates The comprehensive weighting reflects the risk distribution within the three-dimensional space of the tunnel. and These represent the coordinates of the tunnel surface and cross-section, respectively. Indicates depth, For three-dimensional space Normalized fracture density at the location, For three-dimensional space Normalized sand content at the location, For three-dimensional space Normalized inflow rate at the location For a moment Dynamic texture change characteristics, These are the weighting coefficients.

[0027] As a further option, the weighting coefficient is a dynamic weight based on risk sensitivity, specifically:

[0028] ;

[0029] in, For the updated weights, For the first k The initial weights of each feature, For the first Risk sensitivity of each characteristic The number of features, k=1,2,3,4;

[0030] No. Risk sensitivity of each feature Specifically:

[0031] ;

[0032] in, As a risk index, For the first 1 eigenvalue, This is the mean of all features.

[0033] As a further embodiment, the inrush prediction sub-model couples a hydrodynamic model and a hydrochemical model; wherein, the hydrodynamic model specifically comprises:

[0034] ;

[0035] The water chemistry model is as follows:

[0036] ;

[0037] in, For seepage flow, Permeability coefficient, For head gradient, For the water head, The divergence of the seepage flow. The water storage coefficient, The rate of change of water head over time; For chemical component concentration, The rate of change of concentration over time, ▽•(qC) is the convection term. For seepage velocity, Where is the diffusion coefficient. For diffusion term, This is the reaction term.

[0038] As a further approach, the permeability and diffusion coefficients are dynamically and adaptively adjusted, specifically as follows:

[0039] ;

[0040] ;

[0041] in, For a moment The permeability coefficient, The initial permeability coefficient, To adjust the coefficient, The rate of change of water inflow. This represents the average flow rate. For a moment diffusion coefficient, The initial diffusion coefficient is . To adjust the coefficient, This represents the rate of change in ion concentration. This represents the average ion concentration.

[0042] As a further solution, the loss function of the comprehensive prediction model is specifically as follows:

[0043] ;

[0044] ;

[0045] ;

[0046] in, Losses due to the sandstorm mission, Losses due to water inrush. , The weights for the sand-breaking task and the water-rushing task are respectively. Let the gradient norm be the loss of the sand-bursting task. The gradient norm of the water inrush task loss.

[0047] In other implementations, the following approach is adopted:

[0048] A refined intelligent grouting system based on the fusion of multi-dimensional parameters in the drilling process includes:

[0049] The data acquisition module is configured to acquire drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data respectively, and preprocess them respectively; then merge the preprocessed data of each type.

[0050] The feature extraction module is configured to extract velocity fluctuation, torque-velocity ratio, and vibration frequency spectrum characteristics by combining drilling parameters; extract fracture density and texture variation characteristics by combining surrounding rock images; extract sand content change rate and water inflow gradient characteristics by combining geological and hydrological parameters; and extract hydrological dynamic correlation characteristics by combining hydrodynamic and hydrochemical data.

[0051] The spatiotemporal coupling representation weight calculation module is configured to introduce dynamic weights based on risk sensitivity and calculate spatiotemporal coupling representation weights based on extracted features.

[0052] The prediction model building module is configured to build a comprehensive prediction model consisting of a sand inrush prediction sub-model and a water inrush prediction sub-model.

[0053] The prediction module is configured to input the sand content change rate, velocity fluctuation characteristics, torque-velocity ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient and comprehensive weight, as well as drilling parameters into the sand inrush prediction sub-model; and input hydrodynamic data, hydrochemical data, hydrological correlation characteristics, water inflow gradient and comprehensive weight into the water inflow prediction sub-model; finally, it obtains the prediction results of sand inrush level, water inflow intensity, spatial distribution and temporal variation trend, and risk probability.

[0054] The grouting parameter adjustment module is configured to adaptively adjust the grouting parameters based on the prediction results to generate a refined grouting scheme.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] (1) This invention proposes a refined intelligent grouting method based on multi-dimensional fusion of drilling parameters. Through innovative multi-dimensional spatiotemporal feature fusion and dynamic characterization methods, it integrates drilling parameters, geological and hydrological data, surrounding rock images, hydrodynamic and hydrochemical data to construct a comprehensive prediction model and dynamically assess the risks of sand breaching and water inrush. It innovatively introduces spatiotemporal feature extraction and dynamic weight adjustment mechanisms, combined with in-depth analysis of multi-source data, which significantly improves the prediction accuracy of risks in complex formations.

[0057] (2) This invention proposes a refined intelligent grouting method based on multi-dimensional fusion of drilling parameters. It adopts a dynamic parameter adaptive and real-time monitoring feedback mechanism. Through real-time analysis of hydrodynamic and hydrochemical data, it dynamically adjusts grouting parameters (such as grouting volume and pressure) and evaluates the effect by combining changes in water pressure and inflow. This invention innovatively combines multi-dimensional data fusion with physical law constraints to ensure the adaptability and reliability of grouting control.

[0058] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0059] Figure 1 This is a flowchart of a refined intelligent grouting method based on the fusion of multi-dimensional parameters in the drilling process, as described in an embodiment of the present invention. Detailed Implementation

[0060] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] 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 scope of exemplary embodiments according to the invention. 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.

[0062] Example 1

[0063] In one or more embodiments, a refined intelligent grouting method based on the fusion of multi-dimensional parameters of the drilling process is disclosed, combined with... Figure 1 Specifically, it includes the following process:

[0064] S101: Acquire drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data respectively, and preprocess them respectively; merge each type of data after preprocessing.

[0065] In this embodiment, the collection of multi-source data is specifically as follows:

[0066] (1) For the acquisition of drilling parameters, during the construction of tunnels in water-rich sandy and gravelly strata, multi-dimensional drilling parameters, including drilling speed (m / s), drilling torque (N·m), drilling position (m) and drill rod vibration frequency (Hz), are collected in real time by installing equipment such as strain sensors and frequency meters on the drill rod. Sensors are installed at multiple positions on the drill rod to ensure that the data covers the entire drilling process.

[0067] (2) For the acquisition of surrounding rock images, a miniature camera (equipped with LED supplementary light and waterproof shell) is installed on the inner wall of the top of the drill rod. The surrounding rock is acquired in real time with 360° by pre-set holes and the rotation of the drill rod. The global image of the surrounding rock is obtained and the crack, texture and color features are recorded.

[0068] (3) For the collection of geological and hydrological parameters, multi-source geological and hydrological parameters are mainly obtained through core drilling and field observation.

[0069] Geological parameters are determined through sampling and analysis, such as stratigraphic layering, rock properties, and sand content (%); hydrological parameters are obtained through turbidity analysis, such as water inflow (…). And water quality (turbidity, ion concentration), etc.

[0070] (4) For the acquisition of hydrodynamic and hydrochemical data, a dense sensor network is deployed in and around the tunnel construction area (aquifer, fault zone, karst channel), including water level sensor, flow meter, velocity meter, pressure sensor (for collecting water level, water flow, flow velocity, water pressure) and chemical sensor (for collecting ion concentration, pH value, redox potential), and the sensor network covers potential risk areas.

[0071] The collected data were preprocessed as follows:

[0072] (1) Preprocessing of drilling parameter data:

[0073] Weighted average filtering is performed on each drilling parameter data to remove vibration interference:

[0074] ;

[0075] in, For smoothing signals, For the number of sensors, Indicates the first The raw signals from each sensor, This indicates filtering. This formula can remove vibration interference and ensure the smoothness of data such as drilling speed and torque.

[0076] The filtered data is then normalized to eliminate dimensional differences. The formula is as follows:

[0077] ;

[0078] in, and This refers to the range of the corresponding data (e.g., drilling speed range [0,5] m / s).

[0079] (2) Preprocessing of surrounding rock image data: Median filtering is used to remove image noise, and histogram equalization is used to adjust image brightness to ensure that crack and texture features are clear.

[0080] (3) Preprocessing of geological and hydrological parameter data:

[0081] The normalization formula for geological and hydrological parameters is as follows:

[0082]

[0083] in, These are the original parameter values ​​(such as sand content). and For ranges (e.g., sand content [0, 100] %).

[0084] (4) Preprocessing of hydrodynamic and hydrochemical data:

[0085] The moving average method is used to smooth the data of water level, flow velocity, water pressure, and ion concentration, with a window size of 5.

[0086] ;

[0087] in, Let be the smoothed value of time t. This is the original data.

[0088] Then, the preprocessed multi-source data is merged, and the specific process is as follows:

[0089] Taking sand content data as an example, for sand content data, firstly, multi-sensor collaborative acquisition and weighted average fusion are adopted: Sand content is a key parameter for assessing the risk of formation sand collapse, reflecting the proportion of sand and gravel particles. High sand content is usually associated with loose formations and high permeability, and is prone to sand collapse. High-precision data is obtained through borehole core analysis (once per hour), real-time data is obtained through borehole outflow turbid water analysis (once per minute), and continuous data is obtained through surrounding rock image inference (once per minute). Since core analysis has high precision but low frequency, turbid water analysis is real-time but noisy, and image inference is continuous but indirect, weighted average fusion is used to generate comprehensive sand content.

[0090] The formula for weighted average fusion of sand content data is as follows:

[0091] ;

[0092] in, The overall sand content (%) after fusion. For the first The weight of each sensor, For the first Sand content obtained from each sensor, The number of sensors is determined by the accuracy and real-time nature of the data source.

[0093] Similarly, for other parameters, such as the results obtained from multiple sensors, such as water flow rate or water pressure, a weighted average fusion can be performed, while checking the data logic and removing faulty data.

[0094] ;

[0095] in, For the time corresponding to the parameter type The fusion result For the first The weight of each sensor, For the first Each sensor at time The preprocessed data, This is the bias term (usually set to 0). This represents the number of sensors.

[0096] S102: Combine drilling parameters to extract velocity fluctuation, torque-velocity ratio, and vibration frequency spectrum characteristics; combine surrounding rock images to extract fracture density and texture variation characteristics; combine geological and hydrological parameters to extract sand content change rate and water inflow gradient characteristics; and combine hydrodynamic and hydrochemical data to extract hydrological dynamic correlation characteristics.

[0097] This embodiment generates high-precision borehole formation characterization results through feature extraction from multidimensional data and identifies areas at risk of sand erosion and water inrush. The specific process for extracting spatiotemporal features from multidimensional data is as follows:

[0098] (1) For drilling data, calculate the speed fluctuation, torque-speed ratio and frequency spectrum characteristics of drilling parameters respectively.

[0099] The velocity fluctuation characteristics of drilling parameters are calculated as follows:

[0100] ;

[0101] in, The velocity fluctuation characteristic (m / s) is shown. For a moment Drilling speed (m / s). The average velocity (m / s) within the time window. The time window size is denoted by s. This formula extracts the dynamic fluctuations in drilling speed, reflecting changes in formation mechanical response, and subsequently serves as input features for step modeling to predict sand breach risk.

[0102] The torque-speed ratio for calculating drilling parameters is given by the following formula:

[0103] ;

[0104] in, Torque-speed ratio For a moment Drilling torque, For a moment The drilling speed. This formula characterizes formation resistance and is subsequently used as a model input feature to predict sand breach risk.

[0105] The frequency spectrum characteristics of drilling parameters are calculated, and the dominant frequency is extracted using Fast Fourier Transform (FFT):

[0106] ;

[0107] in, Main frequency, For a moment The drill pipe vibration frequency, The Fourier transform of the vibration frequency. The frequency corresponding to the maximum value is used. The dominant frequency reflects the frequency characteristics of the formation's mechanical response and is subsequently used as a feature input to the model.

[0108] (2) For the surrounding rock image, extract the fracture density and texture variation features respectively.

[0109] For the preprocessed surrounding rock image, an edge detection method is used to identify linear fracture features in the image. Threshold segmentation is used to separate fracture regions from the background. The detection results generate a binary image of the fractures, with fracture regions marked as 1 and non-fracture regions marked as 0. The number of fractures per unit area or unit length, or the total length, is calculated, and then the fracture density is calculated. :

[0110] ;

[0111] in, Representing three-dimensional space The number of cracks in the image. Represents unit area, Representing three-dimensional space The total length of the crack, Indicates unit length.

[0112] Texture contrast is used to quantify the texture complexity of surrounding rock images, reflecting the heterogeneity of sand and gravel particle distribution, surface roughness, and pore structure characteristics. High contrast may be associated with uneven sand and gravel particle distribution and strong permeability, affecting the risk of water inrush and sand runoff. Texture contrast is calculated by using the gray-level co-occurrence matrix (GLCM) on the preprocessed surrounding rock image. formula:

[0113] ;

[0114] in, Represents the gray values ​​in the gray-level co-occurrence matrix. and The co-occurrence probability represents the grayscale distribution of pixel pairs (adjacent or at a specified distance). This represents the grayscale difference, reflecting the intensity of local grayscale changes. Indicates the first The texture contrast of a frame image; the larger the value, the coarser the texture and the more uneven the grain distribution.

[0115] For each frame of the image, calculate the texture contrast based on the gray-level co-occurrence matrix (GLCM). Typically, co-occurrence matrices in multiple directions (such as 0°, 45°, 90°, 135°) are considered, and the average value is taken to improve robustness.

[0116] In this embodiment, dynamic texture change features are introduced to reflect the dynamic evolution of the surrounding rock texture over time or during construction. This can capture real-time changes in the formation texture and is highly correlated with the dynamic evolution of sand inrush and water inrush risks.

[0117] Dynamic texture change features are calculated using the following formula:

[0118] ;

[0119] in, For texture variation features, For the first Texture contrast of frame images For the first Contrast of frame images The number of frames.

[0120] (3) Extract the characteristics of sand content change rate and water inflow gradient for geological and hydrological parameter data;

[0121] The extraction of the sand content change rate characteristics is as follows:

[0122] ;

[0123] in, The rate of change of sand content, For a moment The overall sand content, For a moment Three-dimensional space The overall sand content, The time interval is defined by this formula. This formula extracts the dynamic changes in sand content, reflecting the changes in the degree of formation loosening, and is subsequently used as input features for the model.

[0124] To extract the gradient features of the inflow rate, based on the obtained inflow rate data, the following calculations are performed:

[0125] ;

[0126] in, For the inflow gradient, and For three-dimensional space At the same time in three-dimensional space The water inflow at the location, Depth at the same time The water inflow at the location, The depth interval is used. This formula extracts the spatial variation of water inflow, reflecting hydrological dynamics, and is subsequently used as input features for the model.

[0127] (4) For hydrodynamic and hydrochemical data, extract hydrological dynamic correlation features, specifically:

[0128] ;

[0129] in, The correlation coefficient between water flow rate and ion concentration. For a moment Three-dimensional space The water inflow at the location, For a moment ion concentration, and These represent the average values ​​of water flow rate and ion concentration, respectively. This represents the size of the time window. This formula extracts the coupling relationship between hydrodynamics and hydrochemistry, which is then used as input features for the model.

[0130] S103: Introduce dynamic weights based on risk sensitivity, and calculate spatiotemporal coupling representation weights based on extracted features.

[0131] In this embodiment, a comprehensive index is generated by weighted fusion of fracture density, sand content, water inflow, and dynamic texture variation characteristics. This index, namely the spatiotemporal coupling characterization weight, is used to generate a comprehensive risk characterization at a specific depth and time, thereby quantifying the formation risk level at a specific depth z and time t.

[0132] The specific formula for calculating the spatiotemporal coupling representation weights is as follows:

[0133] ;

[0134] in, For at any time and three-dimensional coordinates The comprehensive weighting reflects the risk distribution within the three-dimensional space of the tunnel. and These represent the coordinates of the tunnel surface and cross-section, respectively. Indicates depth, For three-dimensional space Normalized fracture density at the location, For three-dimensional space Normalized sand content at the location, For three-dimensional space Normalized inflow rate at the location For a moment Dynamic texture change characteristics, These are the weighting coefficients.

[0135] In this embodiment, for the weighting coefficient A dynamic weighting based on risk sensitivity is adopted, specifically:

[0136] ;

[0137] in, For the updated weights, Let be the initial weights for the k-th feature. For the first Risk sensitivity of each characteristic The number of features, k=1,2,3,4;

[0138] No. Risk sensitivity of each feature Specifically:

[0139] ;

[0140] in, As a risk index, For the first 1 eigenvalue, This is the mean of all features.

[0141] This embodiment optimizes dynamic weights in real time by assessing risk sensitivity, enabling it to focus on key features, adapt to changes in operating conditions, and optimize resource allocation, thereby improving the accuracy and robustness of predictions.

[0142] Through the above Data, as three-dimensional observations of known points, records their spatial coordinates. and time Calculate the empirical variability function among all observation points to determine the spatial correlation structure:

[0143] ;

[0144] in, They are respectively Distance in the direction, Distance The point logarithm; Represents a known observation point in three-dimensional space Observations Represents a known observation point in three-dimensional space The observed values.

[0145] A three-dimensional theoretical variation model was fitted, using an isotropic spherical model:

[0146] ;

[0147] in, Value of a nugget. For sill values, For variable range, the initial values ​​are as follows: , , =3 meters.

[0148] Finally, solve the three-dimensional linear equation system. ,in, They are Lagrange multipliers, satisfying the unbiasedness constraint. Seeking .

[0149] in, Represents the variogram in spatial distance The value at this point is used to quantify the spatial variability or similarity differences between observations, and is usually calculated based on empirical variability. Represent two known points and The variogram values ​​between them reflect their autocorrelation in three-dimensional space and are used to construct the coefficient matrix of the linear equation system;

[0150] ;

[0151] in, , , They are spatial points The three-dimensional coordinates, , , They are spatial points The three-dimensional coordinates.

[0152] Indicates the point to be interpolated and known points The variogram values ​​between these values ​​are used as the right-hand side terms of the system of equations to calculate the weights. ;

[0153] ;

[0154] in , , They are spatial points The three-dimensional coordinates, , , They are spatial points The three-dimensional coordinates.

[0155] For each point to be interpolated Selecting nearby points, the ordinary kriging formula is as follows:

[0156] ;

[0157] in, The predicted value of the interpolation point represents the three-dimensional space. The overall risk value at the location, Given points The weight, Given points The observed values, This represents the number of neighboring points.

[0158] calculate The formula generates a continuous 3D characterization surface and divides risk and non-risk areas by setting a threshold (e.g., 0.7), outputting the 3D spatial distribution result. This formula generates the continuous characterization surface for risk area identification, which is then used as input for the spatial distribution.

[0159] S104: Construct a comprehensive prediction model consisting of a sand runoff prediction sub-model and a water inrush prediction sub-model.

[0160] Input the following data into the sand breakage prediction sub-model: sand content change rate, velocity fluctuation characteristics, torque-velocity ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient and comprehensive weight, as well as drilling parameters; input the following data into the water inflow prediction sub-model: hydrodynamic data, hydrochemical data, hydrological correlation characteristics, water inflow gradient and comprehensive weight.

[0161] In this embodiment, the comprehensive prediction model includes a sand inrush prediction sub-model and a water inrush prediction sub-model. Through multi-task learning and joint optimization, the model shares a feature extraction layer. The training process includes data partitioning, optimization, and prediction. The model outputs sand inrush level, water inrush intensity, spatial distribution, and temporal variation trend, and introduces risk probability assessment.

[0162] Specifically, the sand inrush prediction sub-model employs a spatiotemporal attention neural network to capture the spatial and temporal dependencies of features. The input layer receives the extracted feature data, including: sand content change rate, velocity fluctuation characteristics, torque-velocity ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient, comprehensive weights for all regions, drilling speed, drilling torque, and drill pipe vibration frequency; the input tensor dimension is [batch_size, seq_len, number of features], where batch_size is the number of samples, seq_len is the time series length, and the number of features is 12.

[0163] First, spatial and temporal weights of features are calculated through a spatiotemporal attention layer: spatial attention focuses on features at different depths. The correlation on time, the attention features focus on over time The variation pattern is then analyzed; subsequently, a nonlinear mapping is performed through a fully connected layer; finally, the output layer uses the softmax function to generate sand erosion levels (high, medium, low), with corresponding values ​​of [3, 2, 1].

[0164] The formula for spatial attention calculation is as follows:

[0165] ;

[0166] in, This is the output for spatial attention (dimensions are [batch_size, seq_len, number of features]). The spatial features are defined by dimensions [batch_size, seq_len, number of features], such as the rate of change of sand content with depth. Conv is a 1×1 convolution operation (used for linear transformation), and softmax is a normalization function. This formula calculates spatial correlation, ensuring that the model focuses on feature changes along the depth direction.

[0167] The formula for calculating temporal attention is as follows:

[0168] ;

[0169] in, For temporal attention output (dimensions are [batch_size, seq_len, number of features]), For time features (dimensions of [batch_size, seq_len, number of features], such as velocity fluctuations over time). (Change), LSTM is a Long Short-Term Memory network (used to extract time dependencies). This formula calculates time correlation, ensuring that the model pays attention to the dynamic changes of features over time.

[0170] The inrush prediction sub-model couples hydrodynamic and hydrochemical models to predict inrush intensity and spatial distribution. Input data include: hydrodynamic data (water level, flow velocity, water pressure) and hydrochemical data (ion concentration, pH value, redox potential), as well as hydrological correlation characteristics, inrush gradient, and comprehensive weights for all regions;

[0171] The hydrodynamic model, based on Darcy's law and the continuity equation, calculates the head distribution; the hydrochemical model, based on the convection-diffusion-reaction equation, calculates the ion concentration distribution. An innovative dynamic parameter adaptation is introduced, adjusting the permeability and diffusion coefficients using real-time data. The model is solved using the finite element method to generate head and concentration predictions, and then calculates the inrush intensity.

[0172] For the hydrodynamic model, the formula is as follows:

[0173] ;

[0174] in, For seepage flow, Permeability coefficient, For head gradient, For the water head, The divergence of the seepage flow. The water storage coefficient, The formula is used to calculate the head distribution, which is then used to predict the inrush intensity, representing the rate of change of the head over time.

[0175] For the water chemistry model, the formula is as follows:

[0176] ;

[0177] in, For chemical component concentration, The rate of change of concentration over time, ▽•(qC) is the convection term. For seepage velocity, Where is the diffusion coefficient. For diffusion term, This is the reaction term. This formula calculates the ion concentration distribution, which is subsequently used for predicting inrush intensity.

[0178] For adaptive dynamic parameters, the formula is as follows:

[0179] ;

[0180] in, For a moment The permeability coefficient, The initial permeability coefficient, To adjust the coefficient, The rate of change of water inflow. This represents the average flow rate. For a moment diffusion coefficient, The initial diffusion coefficient is . To adjust the coefficient, This represents the rate of change in ion concentration. This represents the average ion concentration. This formula dynamically adjusts the permeability and diffusion coefficients to ensure the model adapts to real-time changes, and is subsequently used to solve hydrodynamic and hydrochemical models.

[0181] The integrated prediction model jointly optimizes the prediction of sand runoff and water inrush through multi-task learning. It shares a feature extraction layer (MLP, 5 layers, 128 neurons per layer, SiLU activation). The input includes all the features from step S3. The shared feature extraction layer performs nonlinear mapping on these inputs to generate a unified feature representation with dimensions [batch_size, seq_len, 128], where 128 is the feature dimension.

[0182] The training / test set ratio is 70% / 30%, meaning 70% is used for training and 30% for validation. The training data consists of multiple samples, each representing a time series of length seq_len (e.g., 60 minutes). The test data is used to validate model performance and ensure generalization ability.

[0183] The loss function comprises the classification loss for the sand runoff prediction task and the regression loss for the water inrush prediction task, which are jointly optimized through dynamic weight adjustment. The sand runoff task loss uses cross-entropy loss, as shown in the following formula:

[0184] ;

[0185] in, The losses due to the sandstorm mission, For the sample size, The weighted true sand erosion level of the g-th sample (high, medium, low, coded as [3, 2, 1]). Let g be the predicted probability of the g-th sample. This formula calculates the classification error to ensure that the model accurately predicts the sand runoff level.

[0186] The mean squared error (MSE) is used to calculate the loss from water inrush, as shown in the following formula:

[0187] ;

[0188] in, For the losses of the water surge mission, For the sample size, For the g-th sample, the actual inrush intensity is... Let g be the predicted inrush intensity for the g-th sample. This formula calculates the regression error to ensure the model accurately predicts the inrush intensity.

[0189] The total loss is jointly optimized through dynamic weight adjustment, as shown in the following formula:

[0190] ;

[0191] in, For the total loss, For a moment The weight of the sand collapse task, For a moment The weight of the water surge task, Losses due to the sandstorm mission, Losses due to water inrush.

[0192] The dynamic weight adjustment formula is as follows:

[0193]

[0194] in, Let the gradient norm be the loss of the sand-bursting task. Let be the gradient norm of the loss from the water inrush task. This formula dynamically adjusts the weights to ensure a balance between the losses of the two tasks.

[0195] Optimize the total loss using the Adam optimizer. During training, the loss curve is recorded to ensure model convergence. In each iteration, the input data is processed through a shared feature extraction layer to generate a unified feature representation, which is then input into the sand inrush and water inrush sub-models respectively to calculate the loss and update the model parameters.

[0196] S106: The final results are the sand runoff level, water inrush intensity, spatial distribution and temporal variation trend, and risk probability prediction.

[0197] Sand inrush grade prediction: The sand inrush sub-model outputs a sand inrush grade (high, medium, low), corresponding to values ​​[3, 2, 1]. The prediction results are based on a comprehensive analysis of input features, such as the rate of change in sand content. and fracture density When the level is high, the model tends to predict a high-risk level.

[0198] Inrush Intensity Prediction: The inrush sub-model outputs the inrush intensity. The hydrodynamic model calculates the head distribution and seepage velocity, and then combines this with the ion concentration distribution from the hydrochemical model to comprehensively calculate the inrush intensity. Input features include hydrological correlations. and inflow gradient ( This ensures the accuracy of the predictions.

[0199] Spatial Distribution and Temporal Trends: The spatial distribution is generated based on the comprehensive weighting and Kriging interpolation results, resulting in the spatial distribution of risk areas. For example, the comprehensive weighting of high-risk areas is... The temporal trend is generated through time series predictions from the model, such as predicting the inrush intensity for the next hour based on dynamic changes in water level, flow rate, and ion concentration.

[0200] S107: Based on the prediction results, adaptively adjust the grouting parameters to generate a refined grouting scheme.

[0201] In this embodiment, intelligent grouting control adaptively adjusts grouting parameters based on the prediction results of S106, generates a refined grouting plan, and dynamically optimizes it through real-time monitoring. First, a comprehensive risk index is calculated, and the risk level is determined based on a weighted combination of sand collapse level and water inflow intensity. Then, the grouting volume and pressure are calculated according to the comprehensive risk to generate a grouting plan. The grouting volume and pressure are increased in high-risk areas, and the grouting volume is reduced in low-risk areas. Finally, the grouting effect is monitored in real time, and the results are evaluated through changes in water pressure and water inflow. If the risk has not decreased, the grouting parameters are adjusted until the risk drops to a safe threshold.

[0202] The formula for calculating the comprehensive risk index is as follows:

[0203] ;

[0204] in, As a comprehensive risk index, Sand erosion level (high=3, medium=2, low=1, no unit, from step S4). For the inrush intensity, , The weight (initially 0.5, dynamically adjusted based on real-time data) is used to determine grouting parameters by taking into account the risks of sand breaching and water inrush.

[0205] The formulas for calculating the grouting volume and pressure are as follows:

[0206] ;

[0207] ;

[0208] in, This refers to the grouting volume. For grouting pressure, , These are weighting coefficients (initial values ​​of 0.3, 0.2, and 0.1 respectively). This formula adjusts the grouting parameters based on the risk index to ensure a refined grouting plan.

[0209] Example 2

[0210] In one or more embodiments, a refined intelligent grouting system based on the fusion of multi-dimensional parameters of the drilling process is disclosed, comprising:

[0211] The data acquisition module is configured to acquire drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data respectively, and preprocess them respectively; then merge the preprocessed data of each type.

[0212] The feature extraction module is configured to extract velocity fluctuation, torque-velocity ratio, and vibration frequency spectrum characteristics by combining drilling parameters; extract fracture density and texture variation characteristics by combining surrounding rock images; extract sand content change rate and water inflow gradient characteristics by combining geological and hydrological parameters; and extract hydrological dynamic correlation characteristics by combining hydrodynamic and hydrochemical data.

[0213] The spatiotemporal coupling representation weight calculation module is configured to introduce dynamic weights based on risk sensitivity and calculate spatiotemporal coupling representation weights based on extracted features.

[0214] The prediction model building module is configured to build a comprehensive prediction model consisting of a sand inrush prediction sub-model and a water inrush prediction sub-model.

[0215] The prediction module is configured to input the sand content change rate, velocity fluctuation characteristics, torque-velocity ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient and comprehensive weight, as well as drilling parameters into the sand inrush prediction sub-model; and input hydrodynamic data, hydrochemical data, hydrological correlation characteristics, water inflow gradient and comprehensive weight into the water inflow prediction sub-model; finally, it obtains the prediction results of sand inrush level, water inflow intensity, spatial distribution and temporal variation trend, and risk probability.

[0216] The grouting parameter adjustment module is configured to adaptively adjust the grouting parameters based on the prediction results to generate a refined grouting scheme.

[0217] It should be noted that the specific implementation methods of the above modules are the same as those in Example 1, and will not be described in detail again.

[0218] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A fine intelligent grouting method based on multi-dimensional parameter fusion of a drilling process, characterized in that, The method comprises the following steps: Respectively acquire drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data, and respectively preprocess them; Merge each type of preprocessed data; Respectively extract speed fluctuation, torque-speed ratio and vibration frequency spectrum characteristics in combination with drilling parameters, respectively extract fracture density and texture change characteristics in combination with surrounding rock images, respectively extract sand content rate change rate and water inflow gradient characteristics in combination with geological and hydrological parameters, and extract hydrological dynamic correlation characteristics in combination with hydrodynamic and hydrochemical data; Introduce dynamic weights based on risk sensitivity, and calculate the spatiotemporal coupling representation weight based on the extracted characteristics; Build a comprehensive prediction model composed of a sand collapse prediction submodel and a water inflow prediction submodel; Input the sand content rate change rate, speed fluctuation characteristics, torque-speed ratio, vibration frequency spectrum characteristics, fracture density, texture change characteristics, hydrological correlation characteristics, water inflow gradient, spatiotemporal coupling representation weight and drilling parameters into the sand collapse prediction submodel; input the hydrodynamic data, hydrochemical data, hydrological correlation characteristics, water inflow gradient and spatiotemporal coupling representation weight into the water inflow prediction submodel; the sand collapse prediction submodel adopts a spatiotemporal attention neural network to capture the spatial and temporal dependence of the characteristics, and the water inflow prediction submodel couples the hydrodynamic and hydrochemical models to predict the water inflow strength and spatial distribution; the comprehensive prediction model jointly optimizes the sand collapse and water inflow prediction through multi-task learning, and shares the feature extraction layer; Finally, obtain the sand collapse grade, water inflow strength, spatial distribution and time change trend and risk probability prediction results; based on the prediction results, adaptively adjust the grouting parameters to generate a refined grouting scheme; Calculate the spatiotemporal coupling representation weight based on the extracted characteristics, specifically as follows: wherein, is the comprehensive weight value of the tunnel at time and three-dimensional coordinates reflects the risk distribution in the three-dimensional space of the tunnel, and respectively represent the coordinates of the tunnel surface and the cross section, represents the depth, is the normalized fracture density at the three-dimensional space , is the normalized sand content at the three-dimensional space , is the normalized water inflow at the three-dimensional space , is the dynamic texture change feature at time , is the weight coefficient; The weight coefficient is a dynamic weight based on risk sensitivity, specifically as follows: ； wherein, is the updated weight, is the initial weight of the th feature, is the risk sensitivity of the th feature, is the number of features, k = 1, 2, 3, 4; The first characteristic is the risk sensitivity of the second characteristic. ； wherein, is the risk index, is the first characteristic value, is the mean of all characteristics; The water inflow prediction submodel couples the hydrodynamic and hydrochemical models; wherein the hydrodynamic model is specifically as follows: ； The hydrochemical model is specifically as follows: ； wherein, is the seepage flow, is the permeability coefficient, is the hydraulic gradient, is the water head, is the divergence of the seepage flow, is the storage coefficient, is the rate of change of the water head with time; is the chemical component concentration, is the rate of change of the concentration with time, is the convection term, is the seepage velocity, is the diffusion coefficient, is the diffusion term, is the reaction term.

2. The method according to claim 1, characterized in that: The drilling parameters include drilling speed, drilling torque, drilling position and drill rod vibration frequency data; The geological and hydrological parameter data include sand content rate and water inflow data; The hydrodynamic data include water level, flow rate, flow velocity and data pressure; the hydrochemical data include ion concentration, pH value and oxidation-reduction potential.

3. The fine intelligent grouting method based on multi-dimensional parameter fusion of drilling process according to claim 1, characterized in that, The texture change characteristics are specifically as follows: ； wherein, is a texture change feature, is a first texture contrast of the frame image, is a first texture contrast of the frame image, is a frame number.

4. The fine intelligent grouting method based on multi-dimensional parameter fusion of drilling process according to claim 1, characterized in that, The hydrological dynamic correlation characteristics are specifically as follows: ； wherein, is a correlation coefficient of the water inflow and the ion concentration, is a time of the water inflow, is a time of the ion concentration, and are mean values of the water inflow and the ion concentration, respectively, is a size of a time window.

5. The fine intelligent grouting method based on multi-dimensional parameter fusion of drilling process according to claim 1, characterized in that, Dynamically and adaptively adjust the permeability coefficient and the diffusion coefficient, specifically as follows: ； ； wherein, is the permeability coefficient at time , is the initial permeability coefficient, is the adjustment coefficient, is the rate of change of water inflow, is the average water inflow; is the diffusion coefficient at time , is the initial diffusion coefficient, is the adjustment coefficient, is the rate of change of ion concentration, is the average ion concentration.

6. The fine intelligent grouting method based on multi-dimensional parameter fusion of drilling process according to claim 1, characterized in that, The loss function of the comprehensive prediction model is specifically as follows: ； ； ； where, is the collapse task loss, is the flood task loss, , are weights for the collapse task and the flood task, respectively, is the gradient norm of the collapse task loss, is the gradient norm of the flood task loss.

7. A fine intelligent grouting system based on multi-dimensional parameter fusion of a drilling process, characterized in that, The method comprises the following steps: The data acquisition module is configured to acquire drilling parameters, surrounding rock images, geological and hydrological parameters, hydrodynamic and hydrochemical data, and respectively preprocess them; Merge each type of preprocessed data; The feature extraction module is configured to extract speed fluctuation, torque-speed ratio, vibration frequency spectrum characteristic features in combination with drilling parameters, respectively; extract fracture density and texture change features in combination with surrounding rock images; extract sand content change rate and water inflow gradient features in combination with geological and hydrological parameters, and extract hydrological dynamic correlation features in combination with hydrodynamic and hydrochemical data; The spatio-temporal coupling representation weight calculation module is configured to introduce a dynamic weight based on risk sensitivity, and calculate spatio-temporal coupling representation weights based on the extracted features; The prediction model construction module is configured to construct a comprehensive prediction model composed of a sand collapse prediction sub-model and a water inflow prediction sub-model; The prediction module is configured to input the sand content change rate, speed fluctuation features, torque-speed ratio, vibration frequency spectrum characteristics, fracture density, texture change features, hydrological correlation features, water inflow gradient and comprehensive weights, and drilling parameters into the sand collapse prediction sub-model; input the hydrodynamic data, hydrochemical data, hydrological correlation features, water inflow gradient and comprehensive weights into the water inflow prediction sub-model; and finally obtain sand collapse grade, water inflow intensity, spatial distribution and time change trend, and risk probability prediction results; the sand collapse prediction sub-model adopts a spatio-temporal attention neural network to capture the spatial and temporal dependencies of features, and the water inflow prediction sub-model couples hydrodynamic and hydrochemical models to predict water inflow intensity and spatial distribution; the comprehensive prediction model jointly optimizes sand collapse and water inflow prediction through multi-task learning, and shares feature extraction layers; The grouting parameter adjustment module is configured to adaptively adjust grouting parameters based on the prediction results to generate a refined grouting scheme; The spatio-temporal coupling representation weight is calculated based on the extracted features, specifically: wherein, is the comprehensive weight value of the tunnel at time and three-dimensional coordinates reflects the risk distribution in the three-dimensional space of the tunnel, and respectively represent the coordinates of the tunnel surface and the cross section, represents the depth, is the normalized fracture density at the three-dimensional space , is the normalized sand content at the three-dimensional space , is the normalized water inflow at the three-dimensional space , is the dynamic texture change feature at time , is the weight coefficient; The weight coefficient is a dynamic weight based on risk sensitivity, specifically: ； wherein, is the updated weight, is the initial weight of the th feature, is the risk sensitivity of the th feature, is the number of features, k = 1, 2, 3, 4; The first characteristic is the risk sensitivity of the second characteristic. ； wherein is the risk index, is the first characteristic value, is the mean of all characteristics; The water inflow prediction sub-model couples hydrodynamic and hydrochemical models; wherein the hydrodynamic model is specifically: ； The hydrochemical model is specifically: ； wherein, is the seepage flow, is the seepage coefficient, is the hydraulic gradient, is the water head, is the divergence of the seepage flow, is the storage coefficient, is the rate of change of the water head with time; is the chemical component concentration, is the rate of change of the concentration with time, is the convection term, is the seepage velocity, is the diffusion coefficient, is the diffusion term, is the reaction term.

Images