Machine learning-based intelligent prediction method and system for grouting amount

Abstract

This invention discloses a machine learning-based intelligent prediction method and system for grouting volume. The method includes: acquiring historical grouting project data and performing preprocessing such as cleaning, missing value imputation, and feature standardization to construct a training dataset; constructing an initial prediction model using the XGBoost regression algorithm, and automatically finding the optimal hyperparameter combination and training it to obtain the final prediction model using a Bayesian optimizer; serializing the model and preprocessing rules and encapsulating them into a RESTful API prediction service; receiving requests containing the features to be predicted and unique identifiers, preprocessing them, and inputting them into the model to obtain predicted values; binding the predicted values ​​with the unique identifiers and returning them to achieve reverse indexing and traceability of the results; the system includes a data preprocessing module, a model training and optimization module, a prediction service API module, and a reverse index management module; this invention can solve the defects of traditional prediction methods such as low accuracy, difficulty in engineering deployment, and lack of traceability of results, thereby improving prediction accuracy and engineering applicability.

Description

Technical Field

[0001] This invention belongs to the field of geotechnical engineering grouting construction technology, and more specifically, relates to a machine learning-based intelligent prediction method and system for grouting volume. Background Technology

[0002] In the field of geotechnical engineering construction, such as water conservancy and hydropower, tunnels, and subways, grouting is a core construction procedure for achieving foundation reinforcement, seepage prevention and plugging, and surrounding rock stability. The grouting volume, as a key control parameter in grouting construction, directly affects the control of construction material costs, construction schedule planning, project quality and safety, and optimization of construction techniques.

[0003] Traditional grouting volume prediction mainly relies on engineering technicians to make manual estimates based on construction specifications, empirical formulas, and on-site geological conditions. This method is overly dependent on subjective experience, has large prediction errors and poor consistency, is difficult to adapt to complex and ever-changing geological and construction conditions, and cannot achieve standardized and large-scale application.

[0004] With the development of machine learning technology, the industry is gradually introducing intelligent algorithms into the field of grouting volume prediction. Existing technologies mostly adopt multi-layer ensemble learning models such as Stacking, which combine multiple base learners and meta-learners to build a prediction system. Although this improves prediction accuracy to some extent, it still has significant technical shortcomings: First, the model structure is complex, the training process is extremely time-consuming, and it consumes a lot of computing resources. The multiple inference levels lead to low prediction efficiency, which cannot meet the real-time prediction needs of engineering sites. Second, the model is difficult to deploy, difficult to lightweight encapsulate, lacks standardized external calling interfaces, and cannot be effectively integrated with existing engineering construction management systems and monitoring systems, forming "model silos" with extremely poor engineering implementation capabilities.

[0005] In summary, existing grouting volume prediction technologies cannot simultaneously meet the practical engineering needs of high prediction accuracy, high operational efficiency, easy deployment and integration, and traceable management. The industry urgently needs a lightweight, efficient, high-precision, and easily applied intelligent grouting volume prediction technology solution to address the inherent shortcomings of traditional methods and existing machine learning methods. Summary of the Invention

[0006] To address the problems of traditional grouting volume prediction, which relies on empirical estimation, suffers from low accuracy and poor consistency, and lacks standardized interfaces and result traceability mechanisms for engineering deployment, this invention provides a machine learning-based intelligent grouting volume prediction method and system. This method constructs a high-quality training dataset by cleaning, imputing missing values, and standardizing features from historical grouting project data. It then uses the XGBoost regression algorithm combined with a Bayesian optimizer to automatically find the optimal hyperparameter combination with the goal of minimizing the mean square error of the prediction, training a high-precision grouting volume prediction model. The model and preprocessing rules are serialized and encapsulated into a RESTful API prediction service supporting HTTP remote calls. A unique mapping relationship is established between the predicted grouting volume value and the original prediction request through a unique identifier, enabling reverse indexing and full traceability of the prediction results. This method boasts significant advantages, including high prediction accuracy, lightweight model, convenient engineering deployment, strong system compatibility, and data traceability, effectively meeting the intelligent prediction and refined management needs of various geotechnical engineering grouting construction projects.

[0007] To achieve the above objectives, one aspect of the present invention provides a machine learning-based intelligent prediction method for grouting volume, comprising the following steps: S1: Obtain historical grouting project data, clean the historical grouting project data, fill in missing values ​​and perform feature standardization preprocessing to construct a training dataset; S2: The XGBoost regression algorithm is used to construct an initial prediction model for grouting volume. A Bayesian optimizer is introduced to automatically iterate and search within the preset hyperparameter search space with the objective function of minimizing the prediction error, to obtain the optimal hyperparameter combination. The XGBoost regression model is trained based on the optimal hyperparameter combination to obtain the final XGBoost grouting volume prediction model. S3: Serialize and persist the trained XGBoost grouting volume prediction model and its corresponding preprocessing rules, load the serialized and saved model and preprocessing rules using a web service framework, and encapsulate the grouting volume prediction function into a RESTful API prediction service that supports remote network calls. S4: Receive a prediction request containing the grouting construction feature data to be predicted and a unique identifier through the RESTful API prediction service. Perform the same preprocessing as in step S1 on the grouting construction feature data to be predicted and input it into the trained XGBoost grouting volume prediction model for inference calculation to obtain the corresponding grouting volume prediction value. S5. Associate the predicted grouting volume with the corresponding unique identifier to establish a unique mapping relationship between the prediction result and the original prediction request; encapsulate the completed association and binding response data into a standard data format and return it to the requester through the RESTful API prediction service to realize the reverse index and full traceability between the prediction result and the original prediction request.

[0008] Furthermore, the historical grouting project data mentioned in step S1 includes hole sequence, hole depth, section length, hole diameter, permeability, initial water-cement ratio, grouting pressure characteristics, and actual grouting volume labels.

[0009] Further, in step S1, the missing value imputation is performed using at least one of mean imputation, median imputation, or K-nearest neighbor imputation.

[0010] Furthermore, in step S1, the feature standardization adopts Z-Score standardization or Min-Max standardization; The Z-Score standardization formula is: ,in, These are the original eigenvalues. The characteristic mean, The characteristic standard deviation; The Min-Max standardization formula is: ,in, These are the minimum and maximum values ​​of the feature, respectively.

[0011] Furthermore, the preset hyperparameter search space mentioned in step S2 includes the hyperparameter search space including the number of trees, learning rate, and maximum depth; The objective function mentioned in step S2 is: in, For the hyperparameter combination of the XGBoost model, For the first The actual grouting volume of each sample For hyperparameters The corresponding model predicts the grouting volume. The number of samples; The prediction error to be minimized is denoted as .

[0012] Furthermore, step S2 also includes iteratively selecting the combination of hyperparameters to be evaluated through the expected improvement (EI) acquisition function; The desired improved expression is: in, In hyperparameter combination The expected improvement value is as follows; For mathematical expectation operators; This is for the operation of retrieving the maximum value; This represents the smallest prediction error observed so far. Hyperparameter combination The corresponding objective function value; To improve the quantity, if Then it is positive, if Then it is negative or 0.

[0013] Furthermore, the final XGBoost grouting volume prediction model expression in step S2 is: in, For the first Predicted grouting volume for each sample to be predicted. The optimal number of decision trees determined by Bayesian optimization. For the first The output of each decision tree For the first Grouting feature vectors of samples after standardized preprocessing.

[0014] Further, in step S3, the Web service framework is Flask or Django, and the RESTful API service includes a single-sample prediction interface and a batch prediction interface; the batch prediction interface is used to process multiple samples to be predicted at one time, and each sample to be predicted corresponds to an independent unique identifier; the unique identifier is defined by the requester and is used to uniquely identify a single prediction request or a single prediction sample.

[0015] A second aspect of the present invention provides a machine learning-based intelligent prediction system for grouting volume, characterized in that the method for implementing the machine learning-based intelligent prediction method for grouting volume includes: The data preprocessing module is used to obtain historical grouting project data from the grouting construction monitoring system or engineering database; perform data cleaning processing on the obtained raw data by removing outliers and duplicate samples; perform missing value imputation processing on missing items in the feature data; perform feature standardization processing on the imputed data to eliminate dimensional differences; and finally output a standard training dataset. The model training and optimization module is used to construct an initial prediction model for grout volume using the XGBoost regression algorithm. It sets a hyperparameter search space including the number of trees, learning rate, maximum depth, subsample ratio, and column sampling ratio. A Bayesian optimizer is introduced to automatically iterate and optimize the model with the goal of minimizing the prediction mean square error. The optimal combination of hyperparameters is obtained and trained based on this combination to obtain the final XGBoost grout volume prediction model. At the same time, the trained model and the corresponding preprocessing rules are serialized and persistently saved to form a model file that can be directly loaded and run. The prediction service API module is used to load the serialized and saved XGBoost grouting volume prediction model and preprocessing rules. It encapsulates the prediction function into a RESTful API prediction service that supports HTTP remote calls through a web service framework. It provides single-sample prediction interface, batch prediction interface and service health detection interface to the outside world. It receives JSON format prediction requests submitted by external systems, performs preprocessing on the data to be predicted and calls the model to complete inference calculations, and organizes and outputs prediction result data in a standard format. The reverse index management module is used to extract and verify the custom unique identifier in the prediction request when it is received. After the model inference is completed, the predicted grouting volume value is associated and bound to the unique identifier one by one to establish a unique mapping relationship between the prediction result and the original prediction request. The response data after the association and binding is encapsulated into a standard data format and returned to the requester through the RESTful API prediction service. This realizes the reverse indexing, accurate positioning and full traceability of the prediction result. In the batch prediction scenario, an independent identifier mapping relationship is maintained for each sample to ensure that the results can be traced separately and do not interfere with each other.

[0016] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the machine learning-based intelligent prediction method for grout volume.

[0017] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention can improve the accuracy and stability of grouting volume prediction and solve the technical defects of large error and strong subjectivity of traditional experience estimation: the data preprocessing module cleans, fills missing values ​​and standardizes features of historical grouting project data to build a high-quality training dataset; the XGBoost regression algorithm is used to build an initial prediction model, and the Bayesian optimizer automatically iterates and optimizes within the preset hyperparameter search space with the goal of minimizing the prediction mean square error, obtains the optimal hyperparameter combination and trains the final prediction model. Through standardized data processing and intelligent optimization, the nonlinear relationship between grouting construction characteristics and grouting volume is effectively captured, significantly improving the prediction accuracy and model generalization ability, and reducing the prediction error.

[0018] (2) This invention can realize the engineering deployment and cross-system integration of prediction services, and solve the problems of existing technologies being difficult to implement and unable to coordinate with construction management systems: The trained XGBoost grouting volume prediction model and preprocessing rules are serialized and persistently saved through the model training and optimization module. Then, the prediction service API module uses the Web service framework to encapsulate the prediction function into a RESTful API prediction service that supports HTTP remote calls. It provides single-sample, batch prediction and health detection interfaces to the outside world, and uses JSON format for data interaction. It can be directly and seamlessly integrated with existing construction monitoring systems, third-party management platforms and mobile terminals to realize online real-time prediction of grouting volume, which greatly improves construction efficiency and engineering implementation capabilities.

[0019] (3) The present invention can realize the full traceability of prediction results, meet the needs of engineering quality inspection and data management, and solve the defects of existing technology in that the prediction results cannot be reverse indexed: by working together with the reverse index management module and the prediction service API module, the unique identifier in the prediction request is extracted, and the grouting volume prediction value obtained by model reasoning is accurately associated and bound with the unique identifier, and a unique mapping relationship between the prediction result and the original prediction request is established. In the batch prediction scenario, an independent mapping is maintained for each sample. Through this technical means, the reverse index, accurate positioning and full traceability of the prediction results are realized, which facilitates engineering quality inspection, anomaly analysis and construction data traceability management, and improves the standardization and refinement of grouting construction. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an intelligent prediction method for grouting volume based on machine learning, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a machine learning-based intelligent prediction system for grouting volume according to 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] like Figure 1 As shown, one aspect of the present invention provides a machine learning-based intelligent prediction method for grouting volume, comprising the following steps: S1: Obtain historical grouting project data, clean the historical grouting project data, fill in missing values ​​and perform feature standardization preprocessing to construct a training dataset; S2: The XGBoost regression algorithm is used to construct an initial prediction model for grouting volume. A hyperparameter search space including the number of trees, learning rate, and maximum depth is set. A Bayesian optimizer is introduced to automatically iterate within the hyperparameter search space with the goal of minimizing the prediction error, and obtain the optimal hyperparameter combination. The XGBoost regression model is trained based on the optimal hyperparameter combination to obtain the final XGBoost grouting volume prediction model. S3: Serialize and persist the trained XGBoost grouting volume prediction model and its corresponding preprocessing rules, load the serialized and saved model and preprocessing rules using a web service framework, and encapsulate the grouting volume prediction function into a RESTful API prediction service that supports remote network calls. S4: Receive a prediction request containing the grouting construction feature data to be predicted and a unique identifier through the RESTful API prediction service. Perform the same preprocessing as in step S1 on the grouting construction feature data to be predicted and input it into the trained XGBoost grouting volume prediction model for inference calculation to obtain the corresponding grouting volume prediction value. S5. Associate the predicted grouting volume with the corresponding unique identifier to establish a unique mapping relationship between the prediction result and the original prediction request; encapsulate the completed association and binding response data into a standard data format and return it to the requester through the RESTful API prediction service to realize the reverse index and full traceability between the prediction result and the original prediction request.

[0023] The steps of the method of the present invention will be described in detail below.

[0024] (1) Data preparation and preprocessing Structured historical grouting project data is extracted from grouting construction monitoring systems, engineering databases, or on-site acquisition equipment. The data includes at least: borehole sequence, borehole depth, grouting section length, borehole diameter, and other borehole location structural characteristics; geological characteristics such as permeability and rock mass integrity coefficient; construction process characteristics such as initial water-cement ratio, grouting pressure, and grout flow rate; and actual grouting volume as label data for model supervised learning.

[0025] Outlier detection and removal are performed on the original data, such as deleting noisy data like extreme pressure values ​​and negative grouting volumes that clearly exceed the engineering specifications; missing feature values ​​are filled using at least one of the following methods: mean imputation, median imputation, or K-nearest neighbor imputation, to ensure the integrity of the dataset; duplicate samples are deduplicated to avoid data redundancy interfering with model training. Z-Score standardization or Min-Max standardization is used to unify the scale of the feature data; Z-Score standardization transforms the features into a distribution with a mean of 0 and a variance of 1, and the formula is: , in, These are the original eigenvalues. The characteristic mean, The characteristic standard deviation; Min-Max standardization maps features linearly to the [0,1] interval, as shown in the formula: ,in, These are the minimum and maximum values ​​of the feature, respectively; Standardized data It can eliminate differences in the units of different features and improve the training stability and convergence speed of XGBoost models; The preprocessed data is divided into training and validation sets in an 8:2 or 7:3 ratio. The training set is used for model parameter learning, and the validation set is used for hyperparameter optimization and model performance evaluation to ensure the consistency of dataset distribution and generalization ability. (2) Model building and Bayesian optimization training An initial prediction model was constructed using the XGBoost regression algorithm. This model achieves gradient boosting by integrating multiple decision trees, which can effectively capture nonlinear features and interaction relationships, and is suitable for handling complex engineering regression problems such as grouting volume prediction. Grouting volume is predicted by weighted combination of multiple decision trees. The regularized objective function expression is as follows: in, This is the model loss term, used to characterize the error between the actual grouting volume and the predicted grouting volume; This is a regularization term for the tree model, used to control model complexity and prevent overfitting. The specific expression for the regularization term is: In the formula, The penalty coefficient for leaf nodes. The number of leaf nodes. These are the weighting coefficients. Output weights for the leaf nodes; Define the hyperparameter search space, including the number of decision trees, learning rate, maximum tree depth, subsample ratio, and column sampling ratio. The number of decision trees (n_estimators) controls the number of ensemble trees, typically ranging from 100 to 1000. The learning rate controls the correction magnitude of each tree to the residuals, typically ranging from 0.01 to 0.3. The maximum tree depth (max_depth) limits the maximum depth of a single tree to prevent overfitting, typically ranging from 3 to 10. The subsample ratio controls the proportion of samples used during training of each tree, typically ranging from 0.6 to 1.0. The column sampling ratio (colsample_bytree) controls the proportion of features used during training of each tree, typically ranging from 0.6 to 1.0. A Bayesian optimizer is introduced, with the minimization of the verification set mean squared error (MSE) as the objective function, to automatically iterate and optimize within the hyperparameter search space; the objective function is: in, For the hyperparameter combination of the XGBoost model, For the first The actual grouting volume of each sample For hyperparameters The corresponding model predicts the grouting volume. The number of samples; The prediction error to be minimized; The Bayesian optimizer constructs a surrogate model based on a Gaussian process to approximate the mapping relationship between the objective function and hyperparameters; The desired improvement (EI) acquisition function iteratively selects the combination of hyperparameters to be evaluated; the desired improvement (EI) is used to automatically select the next set of hyperparameters, achieving efficient optimization; the desired improvement expression is: in, In hyperparameter combination The expected improvement value is used to guide Bayesian optimization in selecting the next set of hyperparameters to be evaluated. The larger the value, the more likely the combination of hyperparameters is to bring better model performance. For mathematical expectation operators, it means that in the current surrogate model (such as a Gaussian process), the expectation is expressed as a function of the expectation operator. Based on the understanding of distribution, the average expectation of the amount of improvement; To maximize the result, only the improvements that are "better than the current best result" are counted. The improvement is 0, and no positive contribution is made. The minimum prediction error observed so far (i.e., the current optimal objective function value) corresponds to the minimum mean square error on the validation set in this invention. ; Hyperparameter combination The corresponding objective function value corresponds to the hyperparameter in this invention. The mean square error of the XGBoost model on the validation set ; To improve the quantity, if Then it is positive (indicating hyperparameter). (better than the current best), if Then it is negative or 0 (no improvement); Iteratively execute the process of training the model, evaluating performance, and updating the surrogate model until the preset number of iterations or error convergence threshold is reached, and output the optimal combination of hyperparameters. Based on optimal hyperparameter combination The XGBoost regression model was retrained on the complete training set to obtain the final XGBoost grouting volume prediction model with high accuracy, strong generalization ability and fast inference speed. This model can stably adapt to the grouting volume prediction needs under different working conditions. The final XGBoost grouting volume prediction model expression is: in, For the first Predicted grouting volume for each sample to be predicted. The optimal number of decision trees determined by Bayesian optimization. For the first The output of each decision tree For the first Grouting feature vectors of each sample after standardized preprocessing; The final XGBoost grouting volume prediction model, trained based on the optimal hyperparameter combination, uses the sum of the integrated outputs of multiple decision trees as the grouting volume prediction model.

[0026] (3) Model serialization and service encapsulation The trained XGBoost grouting volume prediction model and the corresponding feature standardization rules (such as mean, standard deviation, extreme values, etc.) in step S1 are persistently saved as binary files using joblib or pickle serialization tools to achieve the portability and reusability of the model and preprocessing logic. Using lightweight web service frameworks such as Flask or Django, the service process is started and the serialized model and preprocessing rules are loaded, and the model inference logic is encapsulated into a callable function interface; The grouting volume prediction function is encapsulated as a RESTful API service, providing a standardized interface to the outside world. Single-sample prediction interface ( / predict): handles requests for a single sample to be predicted; Batch prediction interface ( / batch_predict): Handles batch requests for multiple samples to be predicted; Health monitoring interface ( / health): Used to verify the service's running status, facilitating operation and maintenance monitoring; The interface uses the HTTP protocol for communication, and the data exchange format is JSON, supporting cross-platform and cross-system remote calls; The batch prediction interface is used to receive a batch prediction request containing multiple grouting construction feature samples to be predicted, assign an independent unique identifier to each grouting construction feature sample to be predicted, perform preprocessing consistent with step S1 on each sample and input it into the XGBoost grouting volume prediction model for inference, bind the grouting volume prediction value corresponding to each sample with its own unique identifier and return it, maintaining an independent mapping relationship between each sample and the prediction result. (4) Predicting request reception and inference calculation The RESTful API prediction service receives POST requests from external systems (such as grouting construction management platforms and mobile apps). Perform the same preprocessing operations as step S1 on the feature data to be predicted in the request: Handle any missing features according to the missing value imputation rules from the training phase; The same standardization formula is used to scale the features to ensure that the distribution of the input data is consistent with that of the training set. The preprocessed data is converted into an array format acceptable to the model and input into the XGBoost grout volume prediction model. The model performs forward reasoning on the preprocessed feature data and outputs a continuous grouting volume prediction value through weighted summation of integrated decision trees. This value is the estimated result of the expected grouting volume under the corresponding construction conditions. (5) Result association encapsulation and traceability The predicted grouting volume obtained through reasoning is associated with the unique identifier (key) in the request one by one to establish a mapping relationship between the prediction results and the original request, ensuring that each prediction result can be accurately traced back to the corresponding construction sample. The bound results are encapsulated in a standard JSON format, which includes a unique identifier, predicted value and timestamp, making it easy for the requester to store and analyze the data. The RESTful API prediction service returns the encapsulated results to the requester, while storing the associated log of "key-prediction result-original request" on the server side, realizing full lifecycle traceability of prediction results and supporting engineering quality verification, anomaly analysis and data backtracking requirements; like Figure 2 As shown, a second aspect of the present invention provides a machine learning-based intelligent prediction system for grouting volume, used to implement the above prediction method, including a data preprocessing module, a model training and optimization module, a prediction service API module, and an inverse index management module. The data preprocessing module is used to acquire historical grouting project data, including hole sequence, hole depth, section length, hole diameter, permeability, initial water-cement ratio, grouting pressure, and actual grouting volume; perform outlier detection and noise removal on the raw data, removing abnormal and duplicate samples that exceed the reasonable engineering range; complete missing feature values ​​using at least one of mean imputation, median imputation, or K-nearest neighbor imputation; perform dimensionless processing on the feature data using Z-Score standardization or Min-Max standardization to unify the data distribution scale; divide the preprocessed data into training and validation sets according to a preset ratio, construct a standard training dataset, and transmit it to the model training and optimization module; The model training and optimization module is used to construct an initial prediction model for grout volume using the XGBoost regression algorithm. It pre-defines a hyperparameter search space including the number of decision trees, learning rate, maximum tree depth, subsample ratio, and column sampling ratio. A Bayesian optimizer is introduced, with the objective function of minimizing the mean squared error of the prediction, to perform automatic iterative search within the hyperparameter search space to obtain the optimal combination of hyperparameters that minimizes the model's prediction error. Based on this optimal hyperparameter combination, the XGBoost regression model is fully trained to obtain a final XGBoost grout volume prediction model with strong generalization ability and high prediction accuracy. The trained XGBoost grout volume prediction model, along with the preprocessing rules and standardized parameters from the data preprocessing module, are serialized and persistently saved to form a portable and loadable model file for use by the prediction service API module. The prediction service API module loads the XGBoost grouting volume prediction model, along with its corresponding preprocessing rules and standardized parameters, serialized and saved by the model training and optimization module. It uses a Web service framework to start a background service process, encapsulating the grouting volume prediction function as a RESTful API prediction service supporting remote HTTP calls. It provides standardized interfaces including at least single-sample prediction, batch prediction, and service health checks. Through these interfaces, it receives JSON-formatted prediction requests from external construction management systems, user terminals, or third-party applications, parses the request content to extract the grouting construction feature data and unique identifier to be predicted, performs preprocessing operations identical to those in the data preprocessing module, inputs the processed feature vectors into the loaded XGBoost grouting volume prediction model for inference calculations, and obtains the predicted grouting volume value. In collaboration with the reverse index management module, it associates and binds the prediction results with the unique identifier and encapsulates them into standard JSON-formatted response data, ultimately returning it to the requester, thus achieving a cross-platform, cross-system, and efficient intelligent grouting volume prediction service. The reverse index management module works in conjunction with the prediction service API module. Upon receiving a prediction request, it extracts and verifies the user-defined unique identifier in the request. After the model completes inference calculations to obtain the predicted grouting volume, it precisely associates and binds the predicted grouting volume with the corresponding unique identifier, establishing a one-to-one mapping relationship between the prediction result and the original prediction request. The associated prediction result and unique identifier are encapsulated into standard response data and returned to the requester by the prediction service API module. It supports reverse index queries, result traceability, and data verification of prediction records. In batch prediction scenarios, it maintains an independent identifier mapping relationship for each sample to be predicted, ensuring that the prediction results of multiple samples do not interfere with each other and can be independently located and traced.

[0027] The output of the data preprocessing module serves as the input to the model training and optimization module, providing a standard training dataset for model training. The model training and optimization module outputs the serialized model file and preprocessing rules to the prediction service API module. The inverse index management module works in conjunction with the prediction service API module to complete the extraction of unique identifiers, binding of predicted values, encapsulation of results, and return. External systems call the prediction service API module via the HTTP protocol to realize online prediction of grouting volume and result traceability.

[0028] It should be noted that the machine learning-based intelligent prediction system for grout volume provided in this embodiment can be a computer program (including program code) running on a computer device. For example, the machine learning-based intelligent prediction system for grout volume is an application software. The machine learning-based intelligent prediction system for grout volume can be used to execute the corresponding steps in the methods provided in the embodiments of this application.

[0029] This invention introduces a Bayesian optimizer to automate and intelligently tune the hyperparameters of the XGBoost grouting volume prediction model. It pre-defines a hyperparameter search space including core parameters such as the number of decision trees, learning rate, and maximum tree depth. Using the minimization of the mean squared error (MSE) as the objective function, it automatically obtains the optimal hyperparameter combination through iterative search. Compared to traditional prediction models with manual parameter tuning and fixed parameter settings, this invention effectively improves the model's ability to capture the nonlinear correlation between grouting construction characteristics and grouting volume, significantly improving prediction accuracy (measured by MSE) by 15%-25%. Furthermore, this invention abandons the complex multi-layered Stacking ensemble model architecture of existing technologies, employing a single and efficient XGBoost regression algorithm to construct the prediction model. While maintaining prediction accuracy, it significantly simplifies the model structure, reducing training time by more than 60%, achieving a dual improvement in prediction accuracy and training / inference efficiency, thus meeting the engineering needs of real-time prediction of grouting construction.

[0030] This invention encapsulates the trained and serialized persistent XGBoost grouting volume prediction model and its corresponding preprocessing rules into a standardized RESTful API microservice through a prediction service API module. Data interaction is conducted via HTTP protocol, with data format uniformly set to JSON. Standardized interfaces for single-sample prediction, batch prediction, and service health checks are provided. This design simplifies and speeds up system deployment, eliminating the need for complex runtime environment configurations and multi-model coupling operations. Ordinary technical personnel can complete the deployment process, including model loading and service startup, significantly reducing the technical threshold and deployment cost for engineering applications. Furthermore, the standard HTTP interface design ensures excellent compatibility, allowing the system to be easily integrated with any third-party software that supports remote network calls (such as engineering construction management platforms, mobile monitoring applications, and on-site data acquisition terminals) without additional interface adaptation development, effectively improving the system's engineering feasibility and scenario adaptability.

[0031] This invention adopts a modular design approach, dividing the entire prediction system into a data preprocessing module, a model training and optimization module, a prediction service API module, and a reverse index management module. Each module has clear functional boundaries and well-defined responsibilities, with low coupling between modules. This avoids the problems of multi-model interdependence and complex maintenance found in existing integrated model groups. Compared to maintaining multi-layered integrated model groups, the maintenance difficulty of a single XGBoost model is significantly reduced, effectively decreasing maintenance costs and workload during long-term system operation. Simultaneously, through the collaborative work of the reverse index management module and the prediction service API module, this invention extracts a unique identifier from each prediction request and accurately associates the predicted grouting volume obtained from model inference with this unique identifier, establishing a unique mapping relationship between the prediction result and the original prediction request. In batch prediction scenarios, an independent mapping relationship is maintained for each sample to be predicted, ensuring that each prediction result can be accurately associated with the corresponding original request data and construction feature parameters. This greatly facilitates engineering quality analysis, prediction result anomaly verification, construction data statistics, and construction process optimization, significantly improving the system's practical value, reliability, and engineering applicability.

[0032] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement... Figure 1 The methods provided in each step are detailed in the implementation methods provided in the above steps, and will not be repeated here.

[0033] 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 machine learning-based intelligent prediction method for grouting quantity, characterized in that, Includes the following steps: S1: Obtain historical grouting project data, clean the historical grouting project data, fill in missing values ​​and perform feature standardization preprocessing to construct a training dataset; S2: The XGBoost regression algorithm is used to construct an initial prediction model for grouting volume. A Bayesian optimizer is introduced to automatically iterate and search within the preset hyperparameter search space with the objective function of minimizing the prediction error, to obtain the optimal hyperparameter combination. The XGBoost regression model is trained based on the optimal hyperparameter combination to obtain the final XGBoost grouting volume prediction model. S3: Serialize and persist the trained XGBoost grouting volume prediction model and its corresponding preprocessing rules, load the serialized and saved model and preprocessing rules using a web service framework, and encapsulate the grouting volume prediction function into a RESTful API prediction service that supports remote network calls. S4: Receive a prediction request containing the grouting construction feature data to be predicted and a unique identifier through the RESTful API prediction service. Perform the same preprocessing as in step S1 on the grouting construction feature data to be predicted and input it into the trained XGBoost grouting volume prediction model for inference calculation to obtain the corresponding grouting volume prediction value. S5. Associate the predicted grouting volume with the corresponding unique identifier to establish a unique mapping relationship between the prediction result and the original prediction request; The response data that completes the association and binding is encapsulated into a standard data format and returned to the requester through the RESTful API prediction service, realizing a reverse index and full traceability between the prediction result and the original prediction request. 2.The method of claim 1, wherein: The historical grouting project data mentioned in step S1 includes hole sequence, hole depth, section length, hole diameter, permeability, initial water-cement ratio, grouting pressure characteristics, and actual grouting volume label. 3.The method of claim 2, wherein: In step S1, the missing value imputation is performed using at least one of mean imputation, median imputation, or K-nearest neighbor imputation. 4.The machine learning based intelligent prediction method of grouting quantity according to any one of claims 1-3, characterized in that: In step S1, the feature standardization adopts Z-Score standardization or Min-Max standardization; The Z-Score standardization formula is: wherein, is the original feature value, is the feature mean value, is the feature standard deviation; The Min-Max standardization formula is: ,in, These are the minimum and maximum values ​​of the feature, respectively.

5. A machine learning-based intelligent prediction method for grouting volume according to any one of claims 1-3, characterized in that: The preset hyperparameter search space mentioned in step S2 includes the hyperparameter search space including the number of trees, learning rate, and maximum depth; The objective function mentioned in step S2 is: in, For the hyperparameter combination of the XGBoost model, For the first The actual grouting volume of each sample For hyperparameters The corresponding model predicts the grouting volume. The number of samples; The prediction error to be minimized is denoted as .

6. The intelligent prediction method for grouting volume based on machine learning according to claim 5, characterized in that: Step S2 also includes iteratively selecting the combination of hyperparameters to be evaluated through the desired improvement (EI) acquisition function; The desired improved expression is: in, In hyperparameter combination The expected improvement value is as follows; For mathematical expectation operators; This is for the operation of retrieving the maximum value; This represents the smallest prediction error observed so far. Hyperparameter combination The corresponding objective function value; To improve the quantity, if Then it is positive, if Then it is negative or 0.

7. The intelligent prediction method for grouting volume based on machine learning according to claim 6, characterized in that: The final XGBoost grouting volume prediction model expression in step S2 is: in, For the first Predicted grouting volume for each sample to be predicted. The optimal number of decision trees determined by Bayesian optimization. For the first The output of each decision tree For the first Grouting feature vectors of samples after standardized preprocessing.

8. A machine learning-based intelligent prediction method for grouting volume according to any one of claims 1-3, characterized in that: In step S3, the Web service framework is Flask or Django, and the RESTful API service includes a single-sample prediction interface and a batch prediction interface. The batch prediction interface is used to process multiple samples to be predicted at once, and each sample to be predicted corresponds to an independent unique identifier. The unique identifier is defined by the requester and is used to uniquely identify a single prediction request or a single prediction sample.

9. A machine learning-based intelligent prediction system for grouting volume, characterized in that, The method for implementing the machine learning-based intelligent prediction of grouting volume as described in any one of claims 1-8 includes: The data preprocessing module is used to obtain historical grouting project data from the grouting construction monitoring system or engineering database; perform data cleaning processing on the obtained raw data by removing outliers and duplicate samples; perform missing value imputation processing on missing items in the feature data; perform feature standardization processing on the imputed data to eliminate dimensional differences; and finally output a standard training dataset. The model training and optimization module is used to construct an initial prediction model for grout volume using the XGBoost regression algorithm. It sets a hyperparameter search space including the number of trees, learning rate, maximum depth, subsample ratio, and column sampling ratio. A Bayesian optimizer is introduced to automatically iterate and optimize the model with the goal of minimizing the prediction mean square error. The optimal combination of hyperparameters is obtained and trained based on this combination to obtain the final XGBoost grout volume prediction model. At the same time, the trained model and the corresponding preprocessing rules are serialized and persistently saved to form a model file that can be directly loaded and run. The prediction service API module is used to load the serialized and saved XGBoost grouting volume prediction model and preprocessing rules. It encapsulates the prediction function into a RESTful API prediction service that supports HTTP remote calls through a web service framework. It provides single-sample prediction interface, batch prediction interface and service health detection interface to the outside world. It receives JSON format prediction requests submitted by external systems, performs preprocessing on the data to be predicted and calls the model to complete inference calculations, and organizes and outputs prediction result data in a standard format. The reverse index management module is used to extract and verify the custom unique identifier in the prediction request when it is received. After the model inference is completed, the predicted grouting volume value is associated and bound to the unique identifier one by one to establish a unique mapping relationship between the prediction result and the original prediction request. The response data after the association and binding is encapsulated into a standard data format and returned to the requester through the RESTful API prediction service. This realizes the reverse indexing, accurate positioning and full traceability of the prediction result. In the batch prediction scenario, an independent identifier mapping relationship is maintained for each sample to ensure that the results can be traced separately and do not interfere with each other.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the machine learning-based intelligent prediction method for grouting volume as described in any one of claims 1 to 8.

Images