A method and system for predicting unit consumption of cement based on multi-source data of grouting hole section

Abstract

This invention discloses a method and system for predicting the unit ash consumption of grouting borehole sections based on multi-source data. The method includes: using the grouting borehole section as the prediction and evaluation unit, constructing a borehole section data unit covering spatial location and basic attributes, and collecting multi-source related data. After data preprocessing, features are extracted to construct a borehole section-level multi-source feature parameter set, containing the fracture number index (FNI) and fracture density index (FDI) obtained from drilling monitoring data, as well as spatial-temporal correlation features generated by combining data from preceding and adjacent borehole sections. A training sample set is constructed by combining historical borehole section features and measured unit ash consumption to train an initial model. The optimal prediction model is obtained through automatic hyperparameter optimization and iterative parameter tuning. The features of the borehole section to be predicted are input into the model to obtain the predicted value of unit ash consumption, which is used as the basis for generating corresponding construction auxiliary decisions. This provides reliable technical support for optimizing grouting construction planning.

Description

Technical Field

[0001] This invention belongs to the field of grouting engineering technology, and more specifically, relates to a method and system for predicting the unit ash consumption of grouting hole sections based on multi-source data. Background Technology

[0002] Grouting is a core construction technique for reinforcing strata and sealing seepage in geotechnical engineering, water conservancy engineering, and other fields. The quality of grouting directly affects the overall structural safety and operational stability of the project. In grouting operations, the unit grout consumption is a key technical indicator. Its value directly reflects the grout absorption capacity of the grouting section, the development state of rock fractures, the groutability of the strata, and the actual consumption of grout material. It also serves as an important reference for grout material preparation, dynamic adjustment of construction parameters, and comprehensive evaluation of construction effects.

[0003] In actual field construction, the unit grout consumption often varies significantly among different grouting sections. This indicator is not determined by a single condition, but is influenced by a combination of complex factors, including geological elements such as natural geological conditions, the degree of rock mass fracture development, and water pressure test results, as well as human-set parameters such as grouting design schemes and on-site construction techniques. At the same time, construction feedback from already constructed sections within the same borehole and from adjacent borehole sections also affects the grouting effect and grout consumption of the target section.

[0004] Various influencing factors belong to different data systems, with significant differences in data types and collection scales. Furthermore, these factors exhibit complex nonlinear relationships and spatial linkages. Currently, the industry generally relies on a few basic parameters combined with the experience of construction personnel to predict unit ash consumption. This traditional method is highly subjective, unable to comprehensively integrate multi-dimensional and effective information, and struggles to grasp the changing patterns of geological formations and construction conditions, resulting in significant prediction deviations.

[0005] Insufficient prediction accuracy can lead to a series of engineering problems, such as unreasonable grouting material reserves and improper construction parameter settings. This not only increases material waste and delays construction progress but also makes it difficult to identify abnormal grouting sections in a timely manner, hindering on-site construction organization optimization and overall quality control. Given this industry situation, there is an urgent need for a scientific and feasible technical solution to integrate and process multi-source information related to grouting sections, construct reliable prediction methods, and achieve prediction of unit grout consumption. This would provide strong support for on-site grout preparation, grouting parameter selection, construction anomaly detection, and overall construction planning optimization. Summary of the Invention

[0006] This invention aims to integrate multi-source data from grouting projects, uncover the correlation patterns between strata, construction methods, and borehole sections, and construct a predictive model for unit grout consumption. This solves the problem of insufficient accuracy in traditional predictions relying on manual experience, enabling advance prediction of grout consumption and providing technical support for on-site grout material preparation, construction parameter adjustment, anomaly identification, and construction organization optimization.

[0007] To address the aforementioned deficiencies or improvement needs of existing technologies, as a first aspect of this invention, the present invention provides a method for predicting the unit ash consumption of grouting borehole sections based on multi-source data, comprising: S1. Using the grouting borehole segment as the basic prediction and evaluation unit, establish a borehole segment data unit that includes the spatial location and basic attributes of the borehole segment; collect multi-source data around the grouting borehole segment; S2. The collected multi-source data is preprocessed, and features are extracted based on the preprocessed multi-source data to construct a multi-source feature parameter set at the borehole segment level; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments. S3. Obtain the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, construct an initial prediction model, use the training sample set to train the initial prediction model, and introduce an automatic hyperparameter optimization mechanism, use the preset model evaluation index as the objective function to perform parameter iterative optimization, and obtain the trained unit ash consumption prediction model. S4. Obtain the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; generate construction auxiliary decision based on the unit ash consumption prediction value.

[0008] Furthermore, the multi-source data in S1 includes at least geological exploration data, drilling monitoring data, water pressure test data, grouting design data, construction process data, as well as historical characteristic data of previous borehole sections and spatial correlation characteristic data of adjacent borehole sections.

[0009] Furthermore, the preprocessing in S2 includes at least outlier handling, normalization, and alignment mapping of multi-source heterogeneous data according to the hole segment data units.

[0010] Furthermore, in S2, the fracture quantity index FNI and fracture density index FDI, which are quantified based on drilling monitoring data, specifically include: Based on the drilling speed, drilling torque, and specific energy while drilling data, calculate the specific mechanical energy while drilling sequence; The sliding window algorithm was used to detect abrupt change points in the mechanical energy sequence during drilling, and the energy drop points were identified as fracture characterization points. The fracture number index (FNI) is calculated based on the fracture characterization points, using the following formula: ,in This represents the total number of fracture characterization points identified within the current borehole segment. This represents the length of the current hole segment; The energy drop amplitude at each fracture characterization point is extracted as the equivalent fracture aperture, and the fracture density index (FDI) is calculated using the following formula: ,in For the first Crack opening.

[0011] Furthermore, in S2, the spatial-temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments specifically include: Obtain the measured unit ash consumption and permeability of the previous drilled sections located above the current drilled section within the same borehole, calculate their weighted average and coefficient of variation, and use them as longitudinal temporal correlation features; Using the spatial three-dimensional coordinates of the current borehole segment as the center, set the search radius and filter out adjacent borehole segments whose depth difference is within a preset threshold. Calculate the spatial interpolation estimate of the measured unit ash consumption of the adjacent hole segment and the local spatial autocorrelation Moran's I index, and use them as the lateral spatial correlation feature; The longitudinal temporal correlation feature and the lateral spatial correlation feature are concatenated to form the spatial temporal correlation feature.

[0012] Furthermore, the step of aligning and mapping the multi-source heterogeneous data according to the aperture segment data units specifically includes: For discrete geological exploration data and grouting design data, attribute binding is performed directly according to the borehole segment number; For continuous time series of drilling monitoring data and construction process data, based on the mapping relationship between borehole depth and timestamp, the time series data is resampled and aggregated into statistical feature values ​​of the corresponding borehole depth using the depth interval integral averaging method or the depth weighted sampling method. For point-depth pressure test data, the nearest principle or linear interpolation method is used to map them to the corresponding borehole depth range, so as to achieve unified alignment of multi-source heterogeneous data in the borehole spatial dimension.

[0013] Furthermore, S3 introduces an automatic hyperparameter optimization mechanism, using a preset model evaluation index as the objective function to perform iterative parameter optimization, specifically including: The multi-source data features corresponding to each sample in the validation set are obtained, and the local geological complexity index of each sample is calculated. The local geological complexity index characterizes the degree of spatial variation of the multi-source data features in the neighborhood of the sample. The model evaluation index is constructed as a weighted sum of prediction errors. The prediction error weight of each sample is dynamically determined by the corresponding local geological complexity index and the current iteration round: in the initial preset iteration rounds, the prediction error weight is negatively correlated with the local geological complexity index; in the later preset iteration rounds, the prediction error weight is positively correlated with the local geological complexity index. With minimizing the model evaluation index as the optimization objective, the parameters are iteratively optimized within the preset hyperparameter search space until the preset iteration termination condition is reached, and the optimal hyperparameter combination is output.

[0014] Furthermore, in step S4, generating construction auxiliary decisions based on the predicted unit ash consumption specifically includes: The predicted value of unit ash consumption The theoretical ash consumption is obtained by multiplying the designed grouting volume of the borehole section, and then the predicted value per unit ash consumption is calculated. The dynamic loss coefficient is adaptively matched within the numerical range, and the theoretical ash consumption is multiplied by the dynamic loss coefficient to obtain the estimated value of slurry consumption; The predicted value of unit ash consumption Compare with the preset ash consumption threshold, when When the grout consumption exceeds the high ash consumption threshold, recommended instructions are generated to reduce grouting pressure and adjust the initial water-cement ratio; when When the grout consumption threshold is less than the low ash consumption threshold, a recommended instruction to increase the grouting pressure and adjust the initial water-cement ratio is generated. Obtain the measured unit ash consumption after the construction of the target borehole segment, and calculate its value compared with the predicted unit ash consumption. The relative deviation rate is used to trigger an early warning when the relative deviation rate exceeds a preset deviation threshold, and the deviation is classified and identified as a high-consumption abnormal hole segment or a low-consumption abnormal hole segment according to the positive or negative direction of the deviation. Statistical analysis of the predicted unit ash consumption for multiple adjacent borehole sections within the current construction area. The spatial distribution characteristics, when a predetermined number of consecutive adjacent hole segments When all values ​​exceed the preset encryption threshold, construction organization adjustment suggestions are generated to increase hole spacing or add reinforcing holes; Calculate the measured unit ash consumption of the constructed borehole section and the predicted unit ash consumption. The goodness of fit is used as a quantitative evaluation index of the degree of control over the grouting construction process.

[0015] As a second aspect of the present invention, a system for predicting the unit ash consumption of grouting borehole sections based on multi-source data is also provided, comprising: The borehole segment data construction unit is used to establish a borehole segment data unit containing the spatial location and basic attributes of the borehole segment, using the grouting borehole segment as the basic prediction and evaluation unit; and to collect multi-source data around the grouting borehole segment. The multi-source feature parameter extraction unit is used to preprocess the collected multi-source data, extract features based on the preprocessed multi-source data, and construct a borehole segment-level multi-source feature parameter set; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments; The prediction model training unit is used to acquire the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, it constructs an initial prediction model, uses the training sample set to train the initial prediction model, and introduces an automatic hyperparameter optimization mechanism to perform parameter iterative optimization with a preset model evaluation index as the objective function, so as to obtain the trained unit ash consumption prediction model. The auxiliary decision-making unit is used to acquire the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; and generate construction auxiliary decisions based on the unit ash consumption prediction value.

[0016] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor as described in any one of the claims, a method for predicting the unit ash consumption of a grouting hole segment based on multi-source data.

[0017] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The present invention provides a method for predicting the unit ash consumption of grouting borehole sections based on multi-source data. This method constructs a data system using grouting borehole sections as the basic unit and collects various types of raw data, including geological, construction, and monitoring data. Simultaneously, it completes data preprocessing and heterogeneous data alignment and mapping. This method unifies the statistical caliber and spatial dimension of data from different sources, eliminates interference caused by outliers and dimensional differences, and transforms scattered multi-source information into a standardized and usable data foundation. Based on the standardized raw data, the method further extracts quantitative indicators related to rock mass fractures and the spatiotemporal correlation characteristics of borehole sections. It transforms the degree of fracture development, which is difficult to quantify directly, and the mutual influence between borehole sections into calculable parameters, fully preserving objective information about geological characteristics and the construction environment. This provides comprehensive and practical feature support for subsequent model analysis.

[0018] 2. The method for predicting unit ash consumption of grouting borehole sections based on multi-source data of the present invention constructs a training sample set by combining historical borehole section characteristic parameters and measured unit ash consumption, builds a prediction model, and conducts training. Simultaneously, an automatic hyperparameter optimization mechanism is introduced to complete iterative parameter optimization. During the training process, model evaluation indicators are used as the optimization direction, and sample weights are dynamically adjusted according to geological complexity, allowing the model to gradually learn the intrinsic relationship between characteristic data and unit ash consumption. This method can adjust model parameter configuration according to different geological conditions, balance the model's fitting effect in simple and complex strata, improve the overall generalization ability of the model, reduce prediction bias caused by single parameter configuration, and enable the trained model to adapt to diverse geological conditions in the field, ensuring the stability of prediction results.

[0019] 3. The method for predicting the unit ash consumption of grouting borehole sections based on multi-source data of the present invention obtains the corresponding predicted value of unit ash consumption by inputting the characteristic parameters of the borehole section to be predicted into a trained model, and generates corresponding construction auxiliary decisions based on the prediction results. The predicted values ​​output by the model can objectively reflect the grout absorption capacity and grout consumption level of the target borehole section, replacing the traditional manual experience-based judgment method. Various construction guidance contents derived from the prediction results can be directly applied to on-site grouting operations, helping staff to reasonably control the amount of grout used, adjust construction process parameters, promptly identify construction anomalies and optimize construction layout, so that all aspects of grouting construction can be promoted based on quantitative analysis results, and standardize on-site construction management processes. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for predicting the unit ash consumption of a grouting hole section based on multi-source data, according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the multi-source data structure of an embodiment of the present invention; Figure 3 This is a schematic diagram of the prediction model structure according to an embodiment of the present invention; Figure 4 This is a schematic diagram showing the relationship between the unit ash consumption and permeability of the grouting hole section in an embodiment of the present invention. Figure 5 This is a schematic diagram of the system units in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] Example 1 Please refer to Figure 1 This embodiment 1 provides a method for predicting the unit ash consumption of grouting hole sections based on multi-source data, including: S1. Using the grouting borehole segment as the basic prediction and evaluation unit, establish a borehole segment data unit that includes the spatial location and basic attributes of the borehole segment; collect multi-source data around the grouting borehole segment; S2. The collected multi-source data is preprocessed, and features are extracted based on the preprocessed multi-source data to construct a multi-source feature parameter set at the borehole segment level; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments. S3. Obtain the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, construct an initial prediction model, use the training sample set to train the initial prediction model, and introduce an automatic hyperparameter optimization mechanism, use the preset model evaluation index as the objective function to perform parameter iterative optimization, and obtain the trained unit ash consumption prediction model. S4. Obtain the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; generate construction auxiliary decision based on the unit ash consumption prediction value.

[0023] Please refer to Figure 2 as well as Figure 3 This embodiment 1 further elaborates on the above steps.

[0024] (1) Construction of borehole segment data In this embodiment, continuous boreholes are divided into multiple discrete grouting segments, which are then used as basic prediction and evaluation units to establish borehole segment data units. Each borehole segment data unit is assigned a unique identifier and its three-dimensional spatial coordinates (including horizontal, vertical, and elevation coordinates) are recorded. The top and bottom depths of the borehole segment are also recorded to define its spatial extent, thus forming a borehole segment data unit that includes the spatial location and basic attributes of the borehole segment.

[0025] Based on the established borehole segment data units, geological survey data, grouting design data, and construction process data are extracted from the engineering geological database and construction management system. Geological survey data includes lithology, weathering degree, fracture development characteristics, rock mass integrity, groundwater conditions, burial depth or elevation, etc., to characterize the original geological conditions; grouting design data includes borehole diameter, borehole spacing, row spacing, segment length, borehole sequence, sub-sequence, design pressure, grout mix ratio, final grouting standard, etc., to clarify construction control objectives; construction process data includes grouting pressure, grouting flow rate, grout density, water-cement ratio, cumulative mortar consumption, grouting duration, stage transition information, etc.

[0026] To obtain dynamic response information reflecting the internal physical and mechanical state of the rock mass, real-time monitoring data is acquired through a sensor network mounted on the drilling rig, combined with water pressure test data obtained from hydrogeological testing. The monitoring data includes drilling speed, rotational speed, torque, feed pressure, specific energy, etc. , etc., among which Used to characterize the number of fractures It is used to characterize the cumulative thickness of cracks and the density of cracks; the water pressure test data includes water permeability, water pressure test pressure, water pressure test flow rate, and segmented water pressure test results.

[0027] Furthermore, based on the obtained basic drilling monitoring data and water pressure test data, this embodiment further recognizes that the unit ash consumption is not only related to the geological conditions of the current borehole section, but is also significantly constrained by the overall structural characteristics of the rock mass and the construction process. Please refer to... Figure 4 A clear power function relationship exists between permeability and unit ash consumption, which verifies the effectiveness of permeability as a key characterization parameter. It also indicates that the prediction of a single borehole segment needs to consider the spatially continuous variation of geological conditions. Therefore, to more accurately predict the unit ash consumption of a grouting segment, it is necessary to incorporate historical characteristic data of previous borehole segments and spatial correlation characteristic data of adjacent borehole segments into a multi-source data system to comprehensively reflect the spatial distribution characteristics of rock mass fracture development and the temporal evolution of the construction process.

[0028] The historical characteristic data of the preceding borehole segments are obtained by extracting the measured unit ash consumption and measured permeability of borehole segments above the current segment that have completed grouting within the same borehole, and calculating their weighted average and coefficient of variation to characterize the vertical evolution of formation properties along the borehole depth. The spatial correlation characteristic data of adjacent borehole segments are obtained by setting a spatial search radius centered on the three-dimensional spatial coordinates of the current borehole segment, extracting the measured unit ash consumption of adjacent borehole segments at similar depths that have been grouted within this radius, calculating spatial interpolation estimates using the inverse distance weighting method, and calculating the local spatial autocorrelation index to quantify the lateral distribution characteristics and clustering effect of ash consumption in three-dimensional space, ultimately completing the comprehensive acquisition and aggregation of multi-source data.

[0029] (2) Extraction of multi-source feature parameters In this embodiment, rigorous data cleaning and spatial alignment are required for the collected multi-source data. Due to significant differences in sampling frequency and spatial granularity among different data sources, direct fusion can lead to feature misalignment. Therefore, outliers must be removed and normalized to eliminate the influence of dimensions. Based on this, the multi-source data is aligned and mapped according to borehole segment data units. For discrete geological exploration data and grouting design data, attribute binding is directly performed according to the borehole segment number. For continuous time-series drilling monitoring data and construction process data, considering that fluctuations in drilling speed in different rock strata can lead to uneven time sampling, resampling is required based on the mapping relationship between borehole depth and timestamp. Let the depth range of the current borehole segment be... Continuous time series data in depth The mapping value at that location is The statistical characteristic values ​​of this borehole section were calculated using the depth interval integral averaging method. Its calculation expression is: In the formula, This represents the top depth of the current hole segment. This represents the bottom depth of the current hole segment. This represents the integral variable within the depth range. For point-depth pressure test data, to compensate for its spatial sampling sparsity, the nearest neighbor principle or linear interpolation method is used to map it to the corresponding borehole segment depth range, thereby achieving unified alignment of multi-source data in the borehole segment spatial dimension and constructing a basic borehole segment-level multi-source feature parameter set.

[0030] After acquiring the aligned monitoring-while-drilling (WWD) data, to quantify the development of hidden fractures within the rock mass, it is necessary to extract drilling speed, drilling torque, and WWD specific energy, and calculate the WWD mechanical specific energy sequence accordingly. Let the drilling speed be... Drilling torque is The specific energy while drilling is Calculate the individual energy value in the specific energy sequence of the machine while drilling. Its calculation expression is: In the formula, This represents the current energy value in the MWD sequence. Due to the physical phenomenon of a sudden decrease in resistance when the drill bit passes through fractures or weak interlayers, a sliding window algorithm is used to detect abrupt change points in the MWD sequence. Let the length of the sliding window be... The mean of the first segment of the sequence within the window is The mean of the latter part of the sequence is Calculate the energy difference statistics within the window. Its calculation expression is: In the formula, This represents the energy difference statistic within the sliding window. When... When the energy drop exceeds a preset threshold and manifests as energy decay, the location is identified as an energy drop point and used as a fracture characterization point. Based on the identification results, a fracture number index is calculated. Its calculation expression is: In the formula, This represents the total number of crack characterization points identified within the current borehole segment. This represents the length of the current borehole segment. Simultaneously, the energy drop amplitude at each fracture characterization point is extracted as the equivalent fracture aperture, and then the fracture density index is calculated. Its calculation expression is: In the formula, Representing the The equivalent fracture aperture corresponding to each fracture characterization point The serial number represents the point representing the fracture.

[0031] Considering the strong continuity of groundwater seepage and grout diffusion in three-dimensional space during grouting projects, and that the construction sequence significantly affects formation stress and permeability, further extraction of spatial-temporal correlation features is needed. In the vertical temporal dimension, the measured unit grout consumption and permeability of the previously constructed borehole segments located above the current borehole segment are obtained. Let the number of previous borehole segments be... , No. The measured attribute values ​​of each preceding hole segment are The corresponding depth-distance weight is Calculate the weighted average value of the longitudinal temporal correlation features. Its calculation expression is: In the formula, The weighted average of the measured properties of the preceding borehole segment. This represents the sequence number of the preceding borehole segment. Based on this, the coefficient of variation of the measured properties of the preceding borehole segment is calculated. Its calculation expression is: In the formula, The coefficient of variation represents the measured properties of the preceding borehole segment. The standard deviation represents the measured properties of the preceding borehole segment, thus characterizing the evolution of formation properties during the top-down construction process. In the lateral spatial dimension, a search radius is set with the current borehole segment's three-dimensional spatial coordinates as the center, filtering out adjacent borehole segments whose depth difference is within a preset threshold. Let the number of filtered adjacent borehole segments be... , No. The spatial distance between the current borehole segment and the adjacent borehole segments is The actual unit ash consumption is The spatial interpolation estimate is calculated using the inverse distance weighting method. Its calculation expression is: In the formula, The spatial interpolation estimate representing the measured unit ash consumption of adjacent borehole sections. This represents the sequence number of the adjacent aperture segment. Simultaneously, the local spatial autocorrelation Moran's I index is calculated. Its calculation expression is: In the formula, Moran's I index represents local spatial autocorrelation. This represents the measured unit ash consumption of the current borehole section. This represents the global average of the measured unit ash consumption of all adjacent borehole segments and the current borehole segment. Represents the current hole segment and the first Spatial weights between neighboring borehole segments. The longitudinal temporal correlation features and the lateral spatial correlation features are vector-concatenated to form a complete spatial temporal correlation feature, which is then included in the borehole segment-level multi-source feature parameter set along with the aforementioned fracture features.

[0032] (3) Prediction model training Before starting model training, it is necessary to retrieve historical grouting borehole records from the engineering database, extract the multi-source feature parameter sets at the borehole level constructed in the previous steps, and match them with their corresponding measured unit ash consumption. The multi-source feature parameter set of each historical grouting borehole segment is used as the input feature vector, and its corresponding measured unit ash consumption is used as the target label, performing a one-to-one mapping and binding to construct a training sample set containing a complete input-output mapping relationship. To ensure the objectivity of model evaluation, the training sample set is divided into a training set and a validation set. The training set is used for updating the model's basic weights, and the validation set is used for performance evaluation in the automatic hyperparameter optimization mechanism.

[0033] Based on the pre-divided dataset, an initial prediction model needs to be constructed and its basic network trained using the training set. Let the total number of samples in the training set be... , No. The multi-source feature parameter set of each training set sample is The corresponding measured unit ash consumption is ,in Let be the sample index of the training set. Construct a neural network containing an input layer, several hidden layers, and an output layer as the initial prediction model. Let the set of trainable weight parameters within the model be . .Will Input the initial prediction model, perform forward propagation calculations using the activation functions of each hidden layer, and obtain the... Predicted unit ash consumption per training sample To measure the overall prediction bias of the model on the training set, a training loss function is constructed. Its calculation expression is: Based on the training loss function For the set of weight parameters Calculate the partial derivatives to obtain gradient information, and set the learning rate and control parameters to update the step size. The gradient descent algorithm is used to update the set of weight parameters. Its update expression is In the formula Represents the set of weight parameters The gradient operator, through multiple forward and backward propagation iterations, enables the model to initially grasp the nonlinear mapping relationship between multi-source features and ash consumption.

[0034] While the model is undergoing basic training, to fully evaluate its generalization ability under different geological conditions and to introduce an automatic hyperparameter optimization mechanism, it is necessary to obtain the multi-source data features corresponding to each sample in the validation set and calculate the local geological complexity index for each sample. Let the total number of samples in the validation set be... , No. The multi-source data feature vectors of the validation set samples are: ,in The sample index is used for the validation set. To quantify the spatial variability of multi-source data features within the sample's neighborhood, the index is determined. The set of samples within the neighborhood of each sample And count the number of neighborhood samples in this set. Extract the first Within the neighborhood of the sample, the first Feature vector of each sample ,in Let be the neighborhood sample index. Considering that multi-source data features contain physical quantities with multiple dimensions, let the dimension of the feature vector be . Calculate the first Within the neighborhood of the sample, the first Variance of dimensional features ,in Let be the feature dimension index. Based on the sum of the variances of each dimension, calculate the _th _. Local geological complexity index of each sample Its calculation expression is: .

[0035] After clarifying the geological complexity of each sample, it is necessary to construct model evaluation indicators to guide the iterative optimization of hyperparameters. Let the current iteration be... The preset total number of iterations is Calculate the current iteration progress percentage Its calculation expression is: To avoid the model failing to converge due to noise interference from complex geological environments in the early stages of training, and to encourage the model to focus on difficult samples in the later stages of training to improve prediction robustness, the prediction error weights for each sample are dynamically determined by the corresponding local geological complexity index and the current iteration round. To reasonably control the magnitude of weight adjustment, the weight adjustment coefficients must first be calculated. Obtain the local geological complexity index for all samples in the validation set, and extract the maximum value. and minimum value And set the maximum weight adjustment factor allowed for complexity differences. This multiple reflects the maximum difference in the model's focus on difficult and easy samples. Based on the above parameters, the weight adjustment coefficient is calculated. Its calculation expression is: In the formula, Let represent the natural logarithm function. Based on this, calculate the ... The sample at the th Prediction error weights during round iteration Its calculation expression is: When the iteration is in its early stages, The smaller value results in a negative exponential component, and the prediction error weight is negatively correlated with the local geological complexity index, guiding the model to prioritize fitting smooth samples with simple geological conditions to grasp the basic physical laws; as the iteration enters the later stages... The magnitude is relatively large, the exponential portion turns positive, and the prediction error weight is positively correlated with the local geological complexity index, forcing the model to shift its optimization focus to high-complexity, difficult samples. (The remaining text appears to be unrelated and possibly machine-generated.) The true unit ash consumption of each sample And the predicted unit ash consumption output by the model under the current hyperparameters. Calculate the first Prediction error per sample Its calculation expression is: The prediction error of each sample is multiplied by its corresponding dynamic weight and summed to construct the weighted sum of prediction errors, which serves as the model evaluation metric. Its calculation expression is: .

[0036] The optimization objective is to minimize the evaluation index of the model within a predefined hyperparameter search space. Parameter iterative optimization is performed internally. In each iteration, the current weighted prediction error is used as the basis for the optimization. The gradient direction of hyperparameters is calculated or the fitness of different hyperparameter combinations is evaluated, and then the hyperparameter values ​​are updated. During this process, the iteration status is monitored in real time, and when a preset iteration termination condition is reached, such as the current iteration round... Equal to the total number of iterations Or the sum of weighted prediction errors in multiple consecutive iterations. When the decrease in error is less than the preset convergence threshold, the iterative optimization process stops. Finally, the hyperparameter configuration that minimizes the total weighted prediction error or satisfies the convergence requirement is extracted. Output the optimal combination of hyperparameters The prediction model is then retrained based on the optimal hyperparameter combination to obtain the trained prediction model for unit ash consumption. .

[0037] (4) Decision support Complete the unit ash consumption prediction model After construction and training, the multi-source feature parameter set of the target pore segment to be predicted is obtained and input into the trained unit ash consumption prediction model. In the output, the predicted unit ash consumption of the target hole section is calculated. To guide on-site material preparation, the predicted unit ash consumption value will be used. Grouting volume of the borehole section design Multiply to obtain the theoretical ash consumption Its calculation expression is: Considering the differences in slurry loss and pipeline residue under different geological conditions, based on the predicted value of unit ash consumption... The numerical range in which the dynamic loss coefficient is adaptively matched Theoretical ash consumption With dynamic loss coefficient Multiply to obtain the estimated value of pulp consumption. Its calculation expression is: This provides a basis for material preparation for on-site slurry mixing.

[0038] Based on the calculated predicted value of unit ash consumption Further dynamic adjustment instructions for grouting process parameters need to be generated. The predicted unit ash consumption value... With the preset high ash consumption threshold and preset low ash consumption threshold Compare the values. When the predicted unit ash consumption... Greater than the high ash consumption threshold When this occurs, it indicates that the formation has extremely strong grout absorption capacity or large fractures exist. In this case, recommended instructions are generated to reduce grouting pressure and adjust the initial water-cement ratio to prevent excessive grout diffusion and waste. When the predicted unit ash consumption value... Less than the low ash consumption threshold When this occurs, it indicates that the formation is dense or has poorly developed micro-fractures. At this time, recommended instructions are generated to increase the grouting pressure and adjust the initial water-cement ratio to enhance the penetration and fracturing ability of the grout.

[0039] After grouting is completed in the target borehole section, the predicted results need to be verified and anomalies identified using measured data. This involves obtaining the measured unit grout consumption after grouting of the target borehole section. Calculate its relationship with the predicted value of unit ash consumption. relative deviation rate Its calculation expression is: When the relative deviation rate Exceeding the preset deviation threshold An early warning is triggered, and the deviation is categorized and identified based on its direction: if the measured unit ash consumption... Greater than the predicted value of unit ash consumption If the measured unit ash consumption is abnormally high, it is identified as a high-consumption abnormal pore section, indicating the possible existence of unexplored hidden karst caves or grout leakage channels; if the measured unit ash consumption is... Less than the predicted value of unit ash consumption If it is identified as a low-consumption abnormal borehole section, it indicates that there may be construction quality problems such as cross-contamination of grout or blockage of grouting pipeline.

[0040] To optimize the overall construction organization design from a macro perspective, it is necessary to statistically analyze the predicted unit ash consumption of multiple adjacent borehole sections within the current construction area. Spatial distribution characteristics. Assume the total number of adjacent borehole segments within the current construction area is... , No. The predicted unit ash consumption of each adjacent hole segment is: ,in This refers to the sequence number of adjacent hole segments. When a preset number of consecutive segments is used... Predicted unit ash consumption of adjacent hole sections All are greater than the preset encryption threshold. This indicates the presence of a large area of ​​weak or highly permeable strata, prompting suggestions to adjust the construction organization by increasing the borehole spacing or adding reinforcing boreholes.

[0041] After completing single-hole verification and regional spatial analysis, the degree of control over the overall construction process needs to be quantitatively evaluated. This involves collecting measured unit ash consumption data for all constructed borehole sections within the current construction area. Corresponding predicted unit ash consumption Calculate the goodness of fit between the two. Goodness of fit As a quantitative evaluation index of the degree of control over the grouting construction process, when the goodness of fit is... When the value approaches the theoretical maximum, it indicates that the actual construction conditions are highly consistent with the previous geological predictions, and the construction process is under strict control; when the goodness of fit is close to the theoretical maximum, it indicates that the actual construction conditions are highly consistent with the previous geological predictions, and the construction process is under strict control; If the value is too low, it indicates that the on-site construction parameters are not being implemented properly or that there is a significant deviation in the preliminary geological survey data. Management personnel need to intervene in a timely manner to investigate the cause and adjust the subsequent construction strategy.

[0042] Example 2 Please refer to Figure 5 This embodiment 2 provides a grouting hole section unit ash consumption prediction system based on multi-source data, including: The borehole segment data construction unit is used to establish a borehole segment data unit containing the spatial location and basic attributes of the borehole segment, using the grouting borehole segment as the basic prediction and evaluation unit; and to collect multi-source data around the grouting borehole segment. The multi-source feature parameter extraction unit is used to preprocess the collected multi-source data, extract features based on the preprocessed multi-source data, and construct a borehole segment-level multi-source feature parameter set; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments; The prediction model training unit is used to acquire the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, it constructs an initial prediction model, uses the training sample set to train the initial prediction model, and introduces an automatic hyperparameter optimization mechanism to perform parameter iterative optimization with a preset model evaluation index as the objective function, so as to obtain the trained unit ash consumption prediction model. The auxiliary decision-making unit is used to acquire the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; and generate construction auxiliary decisions based on the unit ash consumption prediction value.

[0043] Example 3 This embodiment 3 also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement any step of a method for predicting the unit ash consumption of a grouting hole section based on multi-source data.

[0044] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0045] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0046] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the unit ash consumption of grouting hole sections based on multi-source data, characterized in that, include: S1. Using the grouting borehole segment as the basic prediction and evaluation unit, establish a borehole segment data unit that includes the spatial location and basic attributes of the borehole segment; Multi-source data were collected around the grouting hole section; S2. The collected multi-source data is preprocessed, and features are extracted based on the preprocessed multi-source data to construct a multi-source feature parameter set at the borehole segment level; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments. S3. Obtain the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, construct an initial prediction model, use the training sample set to train the initial prediction model, and introduce a hyperparameter automatic optimization mechanism, use the preset model evaluation index as the objective function to perform parameter iterative optimization, and obtain the trained unit ash consumption prediction model. S4. Obtain the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; generate construction auxiliary decision based on the unit ash consumption prediction value.

2. The method for predicting the unit ash consumption of grouting hole sections based on multi-source data according to claim 1, characterized in that, The multi-source data in S1 includes at least geological exploration data, drilling monitoring data, water pressure test data, grouting design data, construction process data, as well as historical characteristic data of previous borehole sections and spatial correlation characteristic data of adjacent borehole sections.

3. The method for predicting the unit ash consumption of grouting hole sections based on multi-source data according to claim 1, characterized in that, The preprocessing in S2 includes at least outlier handling, normalization, and alignment mapping of multi-source heterogeneous data according to the aperture segment data units.

4. The method for predicting the unit ash consumption of grouting hole sections based on multi-source data according to claim 1, characterized in that, In S2, the fracture quantity index FNI and fracture density index FDI, which are quantified based on drilling monitoring data, specifically include: Based on the drilling speed, drilling torque, and specific energy while drilling data, calculate the specific mechanical energy while drilling sequence; The sliding window algorithm was used to detect abrupt change points in the mechanical energy sequence during drilling, and the energy drop points were identified as fracture characterization points. The fracture number index (FNI) is calculated based on the fracture characterization points, using the following formula: ,in This represents the total number of fracture characterization points identified within the current borehole segment. This represents the length of the current hole segment; The energy drop amplitude at each fracture characterization point is extracted as the equivalent fracture aperture, and the fracture density index (FDI) is calculated using the following formula: ,in For the first Crack opening.

5. The method for predicting the unit ash consumption of grouting hole sections based on multi-source data according to claim 1, characterized in that, In step S2, the spatial-temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments specifically include: Obtain the measured unit ash consumption and permeability of the previous drilled sections located above the current drilled section within the same borehole, calculate their weighted average and coefficient of variation, and use them as longitudinal temporal correlation features; Using the spatial three-dimensional coordinates of the current borehole segment as the center, set the search radius and filter out adjacent borehole segments whose depth difference is within a preset threshold. Calculate the spatial interpolation estimate of the measured unit ash consumption of the adjacent hole segment and the local spatial autocorrelation Moran's I index, and use them as the lateral spatial correlation feature; The longitudinal temporal correlation feature and the lateral spatial correlation feature are concatenated to form the spatial temporal correlation feature.

6. The method for predicting the unit ash consumption of grouting hole section based on multi-source data according to claim 3, characterized in that, The step of aligning and mapping multi-source heterogeneous data according to the aperture segment data units specifically includes: For discrete geological exploration data and grouting design data, attribute binding is performed directly according to the borehole segment number; For continuous time series of drilling monitoring data and construction process data, based on the mapping relationship between borehole depth and timestamp, the time series data is resampled and aggregated into statistical feature values ​​of the corresponding borehole depth using the depth interval integral averaging method or the depth weighted sampling method. For point-depth pressure test data, the nearest principle or linear interpolation method is used to map them to the corresponding borehole depth range, so as to achieve unified alignment of multi-source heterogeneous data in the borehole spatial dimension.

7. The method for predicting the unit ash consumption of grouting hole sections based on multi-source data according to claim 1, characterized in that, The S3 section introduces an automatic hyperparameter optimization mechanism, which uses a preset model evaluation index as the objective function to perform iterative parameter optimization. Specifically, this includes: The multi-source data features corresponding to each sample in the validation set are obtained, and the local geological complexity index of each sample is calculated. The local geological complexity index characterizes the degree of spatial variation of the multi-source data features in the neighborhood of the sample. The model evaluation index is constructed as a weighted sum of prediction errors. The prediction error weight of each sample is dynamically determined by the corresponding local geological complexity index and the current iteration round: in the initial preset iteration rounds, the prediction error weight is negatively correlated with the local geological complexity index; in the later preset iteration rounds, the prediction error weight is positively correlated with the local geological complexity index. With minimizing the model evaluation index as the optimization objective, the parameters are iteratively optimized within the preset hyperparameter search space until the preset iteration termination condition is reached, and the optimal hyperparameter combination is output.

8. The method for predicting the unit ash consumption of grouting hole section based on multi-source data according to claim 1, characterized in that, In step S4, generating construction auxiliary decisions based on the predicted unit ash consumption specifically includes: The predicted value of unit ash consumption The theoretical ash consumption is obtained by multiplying the grouting volume of the borehole section by the theoretical grouting volume, and then the predicted value per unit ash consumption is calculated. The dynamic loss coefficient is adaptively matched within the numerical range, and the theoretical ash consumption is multiplied by the dynamic loss coefficient to obtain the estimated value of slurry consumption. The predicted value of unit ash consumption Compare with the preset ash consumption threshold, when When the grout consumption exceeds the high ash consumption threshold, recommended instructions are generated to reduce grouting pressure and adjust the initial water-cement ratio; when When the grout consumption threshold is less than the low ash consumption threshold, a recommended instruction to increase the grouting pressure and adjust the initial water-cement ratio is generated. Obtain the measured unit ash consumption after the construction of the target borehole segment, and calculate its value compared with the predicted unit ash consumption. The relative deviation rate is used to trigger an early warning when the relative deviation rate exceeds a preset deviation threshold, and the deviation is classified and identified as a high-consumption abnormal hole segment or a low-consumption abnormal hole segment according to the positive or negative direction of the deviation. Statistical analysis of the predicted unit ash consumption for multiple adjacent borehole sections within the current construction area. The spatial distribution characteristics, when a predetermined number of consecutive adjacent hole segments When all values ​​exceed the preset encryption threshold, construction organization adjustment suggestions are generated to increase hole spacing or add reinforcing holes; Calculate the measured unit ash consumption of the constructed borehole section and the predicted unit ash consumption. The goodness of fit is used as a quantitative evaluation index of the degree of control over the grouting construction process.

9. A system for predicting the unit ash consumption of grouting borehole sections based on multi-source data, characterized in that, include: The borehole segment data construction unit is used to establish a borehole segment data unit that includes the spatial location and basic attributes of the borehole segment, using the grouting borehole segment as the basic prediction and evaluation unit. Multi-source data were collected around the grouting hole section; The multi-source feature parameter extraction unit is used to preprocess the collected multi-source data, extract features based on the preprocessed multi-source data, and construct a borehole segment-level multi-source feature parameter set; wherein, the feature parameter set includes the fracture number index FNI and fracture density index FDI generated based on the quantification of drilling monitoring data, as well as the spatial temporal correlation features extracted based on the historical feature data of the preceding borehole segment and the spatial correlation feature data of adjacent borehole segments; The prediction model training unit is used to acquire the multi-source feature parameter set of the historical grouting hole segment and its corresponding measured unit ash consumption, and construct a training sample set; at the same time, it constructs an initial prediction model, uses the training sample set to train the initial prediction model, and introduces an automatic hyperparameter optimization mechanism to perform parameter iterative optimization with a preset model evaluation index as the objective function, so as to obtain the trained unit ash consumption prediction model. The auxiliary decision-making unit is used to acquire the multi-source feature parameter set of the target hole segment to be predicted, input it into the unit ash consumption prediction model, and output the unit ash consumption prediction value of the target hole segment; and generate construction auxiliary decisions based on the unit ash consumption prediction value.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor as described in any one of claims 1-8: a method for predicting the unit ash consumption of grouting boreholes based on multi-source data.

Images