Construction method of intelligent prediction model for performance of synchronous grouting slurry

Abstract

The invention relates to the technical field of shield tunnel construction, in particular to a construction method of a synchronous grouting slurry performance intelligent prediction model, and mainly solves the technical problems that an existing slurry performance experiment method is poor in accuracy and long in period. The method comprises the following steps: S1, acquiring liquid performance data and solid performance data as an initial data set; s2, expanding the initial data set by adopting a generative adversarial network to generate a target data set; s3, establishing an initial model by adopting four algorithms of an artificial neural network, a random forest, extreme gradient lifting and support vector regression; s4, exploring a hyper-parameter space by adopting a Bayesian optimization agent model and an acquisition function, and training by adopting K-fold cross validation to obtain a target model; and S5, evaluating the performance of the target model, and explaining through the SHAP. The prediction model constructed by the method is high in accuracy and short in test period, and can be used for predicting and optimizing the liquid performance and the solid performance of the slurry under different construction conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of shield tunnel construction, and in particular to a method for constructing an intelligent prediction model for synchronous grouting slurry performance. Background Art

[0002] Shield tunneling involves using a shield machine to excavate a tunnel while ensuring that the excavation face and surrounding rock formations do not collapse. The machine then constructs the tunnel by inserting lining segments onto the excavated surface. This method not only automates excavation, excavation, and lining, but also maintains the integrity of surface transportation infrastructure and underground pipelines. Consequently, shield tunneling is becoming increasingly common in urban underground construction.

[0003] However, due to the uneven geological conditions, this method inevitably disturbs the soft soil layers surrounding the tunnel, causing ground deformation. This poses a significant risk to nearby buildings and infrastructure that are highly sensitive to ground movement. Therefore, during shield tunnel construction, it is necessary to use synchronous grouting to fill the gap between the shield shell and the tunnel segments to reduce ground disturbance.

[0004] In existing technologies, grout properties are generally studied through experimental methods. However, these methods have the following drawbacks: First, existing experiments focus on the liquid properties of the grout, while less research is conducted on the phase transition process from liquid to solid and the solid-state properties of the grout. Solid-state properties are crucial for long-term subsurface deformation control and tunnel stability. This results in poor accuracy of experimental results and makes it difficult to fully simulate actual engineering environments. Second, existing experimental cycles typically last for several months, resulting in long cycles and high testing costs. Therefore, an experimental method with higher accuracy, shorter cycles, and lower testing costs is urgently needed. Summary of the Invention

[0005] In order to overcome the technical defects of existing slurry performance experimental methods, such as poor accuracy, long cycle and high testing cost, the present invention provides a method for constructing an intelligent prediction model for synchronous grouting slurry performance.

[0006] The method for constructing an intelligent prediction model for synchronous grouting slurry performance provided by the present invention comprises the following steps:

[0007] S1. Obtain liquid performance data under various slurry composition ratios, and solid performance data under various formation pressures and soil permeability coefficients as the initial data set;

[0008] S2. Using a generative adversarial network to expand the initial dataset to generate a target dataset;

[0009] S3. Build an initial model using four algorithms: artificial neural network, random forest, extreme gradient boosting, and support vector regression.

[0010] S4. Using a Bayesian optimization surrogate model and acquisition function to explore the hyperparameter space, and using K-fold cross-validation to divide the target dataset into a validation set and (K-1) training sets, training the initial model to obtain a target model, wherein the target model is used to predict the performance of the grouting slurry;

[0011] S5. Evaluate the performance of the target model by coefficient of determination, root mean square error, and mean absolute error, and explain the prediction of the grouting slurry properties by the target model through SHAP.

[0012] Optionally, in step S1, the slurry composition ratio includes a water-binder ratio, a mortar-sand ratio, a water expansion ratio, and a cement-fly ash ratio.

[0013] Alternatively, the amount of the slurry components is determined by the following formula:

[0014] ;

[0015] in, Indicates the quality of cement, Indicates the mass of fly ash, Indicates the quality of bentonite, Indicates the quality of sand, Indicates the quality of water, represents the water-binder ratio, Indicates the mortar-sand ratio, represents the water expansion ratio, represents the cement-fly ash ratio.

[0016] Optionally, the liquid performance data includes density, water exudation rate, fluidity, consistency and stone rate.

[0017] Optionally, a four-factor three-level orthogonal experiment is used to measure the liquid performance data, where the four factors are water-binder ratio, mortar-sand ratio, water expansion ratio, and cement-fly ash ratio.

[0018] Optionally, the solid-state performance data includes compression deformation, 3-day unconfined compressive strength and 28-day unconfined compressive strength.

[0019] Optionally, the generator of the generative adversarial network is used to learn the distribution characteristics of experimental data and synthesize new data; the discriminator of the generative adversarial network is used to distinguish the authenticity of the data and improve the quality of the data through adversarial training.

[0020] Optional:

[0021] The calculation formula of the determination coefficient is:

[0022] ;

[0023] The calculation formula of the root mean square error is:

[0024] ;

[0025] The calculation formula of the mean absolute error is:

[0026] ;

[0027] in, represents the coefficient of determination, represents the root mean square error, represents the mean absolute error, represents the total number of target datasets, Indicates the measured slurry performance parameters, represents the predicted slurry performance parameters, It represents the average value of the measured slurry performance parameters.

[0028] Optionally, the expression for SHAP is as follows:

[0029] ;

[0030] in, Representation characteristics SHAP value, Indicates that it does not contain features The feature subset of represents the total set of features, represents the number of features, Representing feature subsets The predicted value after inputting the model, Representing feature subsets Add features The predicted value after entering the model.

[0031] The technical solution provided by the present invention has the following advantages compared with the prior art:

[0032] 1) The method for constructing an intelligent prediction model for the properties of synchronous grouting slurries provided by this invention provides an initial dataset that fully characterizes the slurry properties throughout the entire phase transition process, from initial injection to final solidification. This initial dataset is further enriched using a generative adversarial network to form a target dataset, laying a solid foundation for the development of an interpretable intelligent system capable of describing the complete phase transition process of synchronous grouting slurries.

[0033] 2) The method for constructing an intelligent prediction model for the properties of synchronous grouting slurry provided by the present invention uses four algorithms: artificial neural network, random forest, extreme gradient boosting, and support vector regression to establish an initial model. The artificial neural network demonstrates excellent predictive capabilities for liquid properties, random forest performs best in predicting compression deformation, while extreme gradient boosting and support vector regression excel in predicting three-day and 28-day unconfined compressive strength, respectively. This allows the model constructed using this method to meet the prediction needs of different properties with high prediction accuracy.

[0034] 3) The method for constructing an intelligent prediction model for simultaneous grouting slurry properties provided by this invention uses SHAP analysis to clearly quantify the impact of various factors on liquid properties, clarifying the influence of parameters such as water-binder ratio, mortar-sand ratio, and water expansion ratio on density, water bleeding rate, and fluidity. Furthermore, the analysis identifies different impact patterns for solid properties, clarifying that confining pressure and water-binder ratio dominate compression deformation, and that cement fly ash plays a key role in strength development, with a stronger impact on three-day confined compressive strength than on 28-day unconfined compressive strength. Using SHAP, an advanced interpretability tool, comprehensive insights into the decision-making process of this model can be provided without extensive manual intervention.

[0035] 4) The method for constructing an intelligent prediction model for the properties of synchronous grouting slurry provided by the present invention has a high correlation between the predicted values ​​and the measured values, accurately predicts key parameters, and has stable prediction accuracy for different performance parameters. It has strong applicability and reliability under different conditions and material properties.

[0036] 5) Compared with existing experiments, the method for constructing an intelligent prediction model for the performance of synchronous grouting slurry provided by the present invention can complete comprehensive performance testing in just a few seconds, significantly shortening the testing cycle and reducing experimental costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0039] Figure 1 A schematic diagram showing a flow chart of a model building method according to an embodiment of the present invention;

[0040] Figure 2 SHAP analysis result diagram according to an embodiment of the present invention;

[0041] Figure 3 A graph showing the response of performance to water-binder ratio in an embodiment of the present invention;

[0042] Figure 4 A graph showing the performance response to cement-fly ash ratio in an embodiment of the present invention;

[0043] Figure 5 A diagram showing a comparison of measured and predicted grouting performance in an embodiment of the present invention;

[0044] Figure 6 A MAPE range diagram showing grouting performance according to an embodiment of the present invention;

[0045] Figure 7 It represents the schematic diagram of grouting engineering;

[0046] Figure 8 Schematic diagram showing the performance comparison of two slurries in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to more clearly understand the above-mentioned objectives, features and advantages of the present invention, the scheme of the present invention will be further described below. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features therein can be combined with each other.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present invention, rather than all the embodiments.

[0049] The following combination Figures 1 to 8 Specific embodiments of the present invention are described in detail.

[0050] This embodiment provides a method for constructing an intelligent prediction model for synchronous grouting slurry performance, including steps S1 to S5.

[0051] S1. Obtain liquid performance data under various slurry composition ratios, and obtain solid performance data under various formation pressures and soil permeability coefficients as the initial data set.

[0052] Specifically, the slurry composition ratio includes water-binder ratio, mortar-sand ratio, water expansion ratio and cement-fly ash ratio.

[0053] More specifically, the amounts of the slurry components are determined by the following formula:

[0054] ;

[0055] in, Indicates the quality of cement, Indicates the mass of fly ash, Indicates the quality of bentonite, Indicates the quality of sand, Indicates the quality of water, represents the water-binder ratio, Indicates the mortar-sand ratio, represents the water expansion ratio, represents the cement-fly ash ratio.

[0056] Specifically, liquid performance data include density, water exudation rate, fluidity, consistency and stone rate.

[0057] More specifically, a four-factor, three-level orthogonal experiment was used to measure liquid performance data. The four factors were water-binder ratio, mortar-sand ratio, water expansion ratio, and cement-fly ash ratio. The data from the orthogonal experiment in this example are shown in the following table:

[0058]

[0059] The above table shows the factor levels of grouting materials;

[0060]

[0061] The above table shows the orthogonal experimental design data statistics.

[0062] Following the above steps, orthogonal experimental results of the slurry liquid properties were obtained. In order to enhance the robustness of the dataset beyond the limited orthogonal experimental data, supplementary experimental results from previous studies were integrated.

[0063] Specifically, the solid-state performance data include compression set, 3-day unconfined compressive strength, and 28-day unconfined compressive strength.

[0064] More specifically, this example conducted solid-state performance tests on nine slurry mixtures in three soil types: sand, silt, and clay at 100 kPa, 200 kPa, and 300 kPa.

[0065] S2. Use a generative adversarial network to expand the initial dataset and generate the target dataset.

[0066] Specifically, the generator of the generative adversarial network is used to learn the distribution characteristics of experimental data and synthesize new data; the discriminator of the generative adversarial network is used to distinguish the authenticity of the data and improve the quality of the data through adversarial training.

[0067] It is important to note that generative adversarial networks, an advanced machine learning technique, employ a two-component system consisting of a generator and a discriminator. Through their interaction, the generator and discriminator increase sample diversity while preserving data distribution characteristics. The generated data successfully complements the concentrated areas of the original dataset, providing more comprehensive coverage of the parameter space. This similarity in distribution patterns verifies that the generated data effectively represents the inherent characteristics of the experimental measurements and demonstrates the reliability of the data augmentation process. The target datasets, based on both liquid and solid properties, provide comprehensive data support for the intelligent prediction model, improving accuracy and generalization across a wide range of soil conditions and pressures.

[0068] It is easy to understand that data enhancement is performed using a generative adversarial network. Through this process, the liquid performance dataset and the solid performance dataset are optimized and expanded and integrated into the target dataset.

[0069] S3. Four algorithms, namely artificial neural network, random forest, extreme gradient boosting and support vector regression, were used to build the initial model.

[0070] It is easy to understand that the four algorithms represent different analytical methods: artificial neural networks for pattern recognition, random forests for tree-structured decision making, extreme gradient boosting for gradient boosting optimization, and support vector regression for kernel-based regression.

[0071] S4. A Bayesian optimization surrogate model and acquisition function are used to explore the hyperparameter space. K-fold cross-validation is used to divide the target dataset into a validation set and (K-1) training sets. The initial model is trained to obtain the target model, which is used to predict the performance of the grouting slurry.

[0072] Bayesian optimization (BO) systematically explores the hyperparameter space through surrogate models and acquisition functions to optimize the application of machine learning models to slurry property prediction. Simultaneously, the probabilistic framework intelligently samples hyperparameter combinations based on previous evaluation results. The hyperparameter tuning process is implemented through an iterative procedure, during which the algorithm continuously updates its probabilistic model of the objective function landscape. For all four algorithms, key hyperparameters are optimized within predefined ranges: learning rate, tree depth for tree-based models, kernel parameters for support vector regression, and network architecture parameters for artificial neural networks. The optimization objective combines multiple performance metrics to ensure balanced performance across different slurry properties.

[0073] Specifically, this embodiment adopts a five-fold cross-validation strategy, i.e., K=5, which divides the entire dataset into five equal parts, allowing five model training and validation iterations.

[0074] S5. The performance of the target model was evaluated by the coefficient of determination, root mean square error, and mean absolute error, and the prediction of the grouting slurry properties by the target model was explained by SHAP.

[0075] It should be noted that the coefficient of determination is used to quantify the proportion of variance explained by the model, the root mean square error is used to measure the standard deviation of the prediction residuals, and the mean absolute error is used to calculate the absolute difference between the predicted value and the actual value.

[0076] The specific formula is as follows:

[0077] The formula for calculating the coefficient of determination is:

[0078] ;

[0079] The formula for calculating the root mean square error is:

[0080] ;

[0081] The formula for calculating the mean absolute error is:

[0082] ;

[0083] in, represents the coefficient of determination, represents the root mean square error, represents the mean absolute error, represents the total number of target datasets, Indicates the measured slurry performance parameters, represents the predicted slurry performance parameters, It represents the average value of the measured slurry performance parameters.

[0084] Specifically, the expression of SHAP is as follows:

[0085] ;

[0086] in, Representation characteristics SHAP value, Indicates that features are not included The feature subset of represents the total set of features, represents the number of features, Representing feature subsets The predicted value after inputting the model, Representing feature subsets Add features The predicted value after entering the model.

[0087] SHAP provides a unified framework for interpreting machine learning predictions of slurry properties. It is based on game theory principles and calculates the contribution of each input feature to a single detection.

[0088] Definition: density is y1, bleeding rate is y2, fluidity is y3, consistency is y4, solidity is y5, compression deformation is y6, 3-day unconfined compressive strength is y7, 28-day unconfined compressive strength is y8, water-cement ratio is x1, mortar-sand ratio is x2, water expansion ratio is x3, and cement-fly ash ratio is x4.

[0089] Depend on Figure 2 The SHAP analysis results show that the water-binder ratio x1 is the dominant factor in properties y1-y6, while the cement-fly ash ratio x4 is the main influencing factor in properties y7-y8. Based on the SHAP analysis results, we conducted a parameter sensitivity analysis, and the results are as follows:

[0090] It should be noted that to facilitate comparative evaluation, all input and output parameters were normalized relative to the baseline values. Figure 3 and Figure 4 The y0 in the equation represents the performance standard value, where Figure 3 The medium performance standard value is the y value corresponding to the water-cement ratio x1=1. Figure 4 The performance standard value in the figure is the y value corresponding to the cement-fly ash ratio x4=0.2, y i Indicates the y value of various properties.

[0091] Figure 3 The results show that among all parameters responding to changes in the water-binder ratio x1, bleeding rate y2 and compression set y6 are the most sensitive. Bleeding rate y2 consistently shows a strong positive correlation with the water-binder ratio x1, and its sensitivity is the highest among all properties. With increasing water-binder ratio x1, bleeding rate y2 can increase by up to 60%. This is attributed to increased free water content and weakened particle interactions, leading to increased water migration and segregation. Compression set y6 varies particularly significantly in the low water-binder ratio range (-40% to 0%), where even a small increase in water content can cause drastic changes in deformation behavior. This high sensitivity at low water-binder ratios stems from the critical role of water in particle spacing and paste stability: when the water content is insufficient, even small increases or decreases in water content can significantly affect particle packing density and internal structure formation, leading to abrupt changes in deformation properties. Other properties responded more modestly to changes in the water-binder ratio x1: density y1 and stone formation rate y5 were negatively correlated with increasing water-binder ratio x1, reflecting the dilution effect of solid particle concentration; fluidity y3 and consistency y4 were positively correlated with the water-binder ratio x1, but their sensitivity was much lower than that of bleeding rate y2 and compression set y6. These trends are consistent with the physical mechanism—higher water content increases fluidity but weakens structural stability.

[0092] Figure 4The results show that the cement-fly ash ratio x4 primarily affects strength development, with a strong positive correlation observed between the 3-day unconfined compressive strength y7 and the 28-day unconfined compressive strength y8. The response curves indicate that the strength improvement is more significant at lower ratios, but the enhancement gradually weakens when the ratio exceeds 50%. Notably, the 3-day unconfined compressive strength y7 is slightly more sensitive to the cement-fly ash ratio x4 than the 28-day unconfined compressive strength y8, indicating that cement content has a more pronounced influence on early strength development. This phenomenon stems from the difference between the rapid early hydration of cement and the slower pozzolanic reaction of fly ash.

[0093] based on Figure 3 and Figure 4 Based on the parameter analysis results, the following optimization recommendations are made for grouting mix ratios in tunnel projects: Maintaining a water-binder ratio of 0.95-1.05 ensures fluidity while controlling water seepage; when prioritizing deformation resistance and strength, the water-binder ratio should be reduced to below 0.9; when improving early and late strength, the cement-fly ash ratio can be appropriately increased; under multi-objective balanced conditions, it is recommended to select the optimal equilibrium point for each performance determined by the analysis; when controlling ground settlement is critical, a low water-binder ratio should be prioritized to significantly improve deformation resistance at the expense of workability; and for projects with tight deadlines requiring high early strength, a medium water-binder ratio of 0.85-0.95 should be maintained while increasing the cement content. These guidelines can help engineers adjust the mix ratio based on project requirements, optimizing grout performance under specific geological conditions and construction scenarios while reducing the number of test iterations.

[0094] To verify the effectiveness and applicability of the proposed intelligent prediction model for simultaneous grouting slurry properties, experimental data from previous studies were used for validation. Six key indicators were selected for verification: four liquid performance indicators: density, water bleeding, consistency, and solidification rate; and two solid performance indicators: compression deformation and 28-day unconfined compressive strength.

[0095] Figure 5 The results show a high degree of agreement between the measured and predicted values, with correlation coefficients (R²) ranging from 0.91 to 0.97. The predictions of compression set (R²=0.93) and 28-day unconfined compressive strength (R²=0.97) were particularly accurate, demonstrating the model's ability to capture both liquid and solid state properties under different experimental conditions.

[0096] Figure 6Density prediction showed the highest accuracy (MAPE <2%), while bleeding rate predictions showed relatively large fluctuations (median approximately 15%). The main reason for the variation in bleeding rate is the varying fineness of the bentonite used in different studies—the water absorption of bentonite significantly affects the bleeding behavior of the slurry. Notably, the MAPEs for two solid-state properties (compression set and 28-day unconfined compressive strength) remained stable within the 5-15% range, confirming the robustness of the model. Comprehensive validation results demonstrate that this intelligent prediction model can accurately predict slurry properties under varying conditions and material properties, demonstrating its universal applicability in engineering practice.

[0097] The following is a specific example to verify this model:

[0098] A certain project uses earth pressure balance shield tunnel construction technology. The shield cutter head diameter is 6.28 meters, the tunnel lining outer diameter is 6.00 meters, the tunnel lining inner diameter is 5.40 meters, and it is assembled from annular segments with a width of 1.20 meters. Figure 7 As shown, the tunnel was excavated in silty clay, which is easily disturbed during excavation. Surface settlement analysis was conducted at a typical cross-section directly above the tunnel at a depth of 13.3 meters. Settlement monitoring points were located every 5 meters along this cross-section.

[0099] By implementing simultaneous grouting, ground deformation can be precisely adjusted. The original slurry (Slurry) mix ratio used in construction is shown in the table. Based on the intelligent prediction model for simultaneous grouting slurry performance in this embodiment, through reverse inference of performance prediction, the slurry mix ratio with the best performance was selected. The optimized slurry mix ratio (Slurry II) is also listed in the table below.

[0100]

[0101] Based on the intelligent prediction model of synchronous grouting slurry performance of this embodiment, the performance characteristics of two different slurry mixtures were evaluated, such as Figure 8 As shown in the figure, Grout I had a grouting volume of 4.24 cubic meters per ring, and its performance included a fluidity of 245.3 mm, a water bleeding rate of 1.96%, and a compression set of 2.35%. In comparison, the optimized grout (Slurry II) showed improved performance, with a reduced grouting volume of 3.46 cubic meters per ring, an increase in fluidity to 277.5 mm, a water bleeding rate of 3.57%, and a compression set of 1.21%.

[0102] The formation losses associated with these slurry properties were calculated and substituted into the Loganathan settlement prediction formula to calculate the theoretical settlement curve. Comparison of the predicted settlement data with the measured settlement data showed that the prediction results of the original slurry were very accurate. The maximum settlement of the original slurry was about 6 mm, while the optimized slurry reduced the maximum settlement to about 3.5 mm, a decrease of 42%, which was a significant improvement in settlement control. The high degree of consistency between the predicted settlement and the measured settlement verified the reliability of the interpretable intelligent system. In addition, the results showed that optimizing the slurry performance through the intelligent prediction model of the synchronous grouting slurry performance of this embodiment can effectively reduce surface settlement, providing a powerful tool for settlement control in urban tunnel projects.

[0103] Compared to traditional laboratory methods, the intelligent prediction model for simultaneous grouting slurry properties in this embodiment achieves revolutionary improvements in efficiency and resource optimization. Taking a typical orthogonal experiment consisting of nine test groups as an example, the traditional testing process requires significant resources: experimental design requires one technician over three days, material preparation requires two technicians over three days at a cost of 471 yuan, liquid performance testing requires two technicians over five days at a cost of 1,087 yuan, and solid performance testing requires three technicians over 10 days at a cost of 3,262 yuan. Furthermore, the unconfined compressive strength (UCS) test requires a 28-day curing period, followed by an additional five days of data analysis by one technician. Overall, completing a set of grouting performance evaluations using traditional methods requires approximately two months, three personnel, and a direct cost of 4,821 yuan, not including labor costs. In contrast, the intelligent prediction model for simultaneous grouting slurry properties significantly simplifies this process, completing a comprehensive performance prediction in just 1.67 seconds.

[0104] The above is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Although detailed descriptions have been made with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents; and such modifications or replacements do not deviate from the essence of the corresponding technical solutions within the scope of the technical solutions of the embodiments, and they should all be covered by the scope of protection of the claims.

Claims

1. A method for constructing an intelligent prediction model for synchronous grouting slurry performance, characterized in that: The steps include: S1. Obtain liquid performance data under various slurry composition ratios, and solid performance data under various formation pressures and soil permeability coefficients as the initial data set; S2. Using a generative adversarial network to expand the initial dataset to generate a target dataset; S3. Build an initial model using four algorithms: artificial neural network, random forest, extreme gradient boosting, and support vector regression. S4. Using a Bayesian optimization surrogate model and acquisition function to explore the hyperparameter space, and using K-fold cross-validation to divide the target dataset into a validation set and (K-1) training sets, training the initial model to obtain a target model, wherein the target model is used to predict the performance of the grouting slurry; S5. Evaluate the performance of the target model by coefficient of determination, root mean square error, and mean absolute error, and explain the prediction of the grouting slurry properties by the target model through SHAP.

2. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 1, characterized in that: In step S1, the slurry composition ratio includes water-binder ratio, mortar-sand ratio, water expansion ratio and cement-fly ash ratio.

3. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 2, characterized in that: The dosage of the slurry components is determined by the following formula: ； in, Indicates the quality of cement, Indicates the mass of fly ash, Indicates the quality of bentonite, Indicates the quality of sand, Indicates the quality of water, represents the water-binder ratio, Indicates the mortar-sand ratio, represents the water expansion ratio, represents the cement-fly ash ratio.

4. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 2 or 3, characterized in that: The liquid performance data include density, water bleeding rate, fluidity, consistency and stone rate.

5. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 4, characterized in that: The liquid performance data were measured using a four-factor three-level orthogonal experiment, where the four factors were water-binder ratio, mortar-sand ratio, water expansion ratio, and cement-fly ash ratio.

6. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 1, characterized in that: The solid state performance data includes compression set, 3-day unconfined compressive strength and 28-day unconfined compressive strength.

7. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 1, characterized in that: The generator of the generative adversarial network is used to learn the distribution characteristics of experimental data and synthesize new data; the discriminator of the generative adversarial network is used to distinguish the authenticity of the data and improve the quality of the data through adversarial training.

8. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 1, characterized in that: The calculation formula of the determination coefficient is: ； The calculation formula of the root mean square error is: ； The calculation formula of the mean absolute error is: ； in, represents the coefficient of determination, represents the root mean square error, represents the mean absolute error, represents the total number of target datasets, Indicates the measured slurry performance parameters, represents the predicted slurry performance parameters, It represents the average value of the measured slurry performance parameters.

9. The method for constructing an intelligent prediction model for synchronous grouting slurry performance according to claim 1, characterized in that: The expression of SHAP is as follows: ； in, Representation characteristics SHAP value, Indicates that it does not contain features The feature subset of represents the total set of features, represents the number of features, Representing feature subsets The predicted value after inputting the model, Representing feature subsets Add features The predicted value after entering the model.

Images