Method for predicting opening degree of cement grouting pressure regulating valve based on sequence model

Through the prediction method based on the sequential model, the inconsistency and low efficiency of pressure regulating valve opening control in cement grouting technology are solved, and automated control is realized, adapting to complex geological conditions, reducing resource waste, and improving grouting quality and efficiency.

CN118938669APending Publication Date: 2024-11-12GEZHOUBA GROUP FOUND ENG +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410990200.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

In the existing cement grouting technology, the pressure control of the grouting process depends on the experience of engineers, resulting in inconsistent opening control of the pressure regulator valve, low efficiency, difficult to adapt to complex geological conditions, and lack of data accumulation and analysis, resulting in fluctuations in grouting quality and waste of resources. .

Method used

Based on the sequential model, the method of predicting the opening of the cement grouting pressure regulating valve is achieved by collecting grouting data from the construction area, constructing a sequential model, training parameters that affect the pressure of the construction pipeline, predicting the opening of the pressure regulating valve, and realizing automated control.

Benefits of technology

It improves the consistency and efficiency of pressure regulating valve opening control, adapts to complex construction conditions, accumulates and analyzes grouting data, reduces resource waste, and improves grouting quality and production efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938669A_ABST
    Figure CN118938669A_ABST
Patent Text Reader

Abstract

The invention proposes a method for predicting the opening degree of a cement grouting pressure regulating valve based on a sequential model, and relates to the technical field of cement grouting, and the method comprises the following steps: collecting a first grouting data set of a constructed region near a to-be-constructed region, carrying out the optimal filtering processing of the first grouting data set, and obtaining a second grouting data set, performing quantitative characterization and standardization processing on the second grouting data set, and dividing the second grouting data set into a training set and a test set; constructing an initial sequence model, and training the initial sequence model through the training set to obtain an optimized prediction model; performing model evaluation on the optimization prediction model through the test set, predicting the opening degree of the pressure regulating valve according to the optimization prediction model when the evaluation is passed, and performing model training again when the evaluation is not passed; and deploying the optimization prediction model into an embedded system, and predicting the opening degree of the cement grouting pressure regulating valve in a to-be-constructed area through the embedded system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of cement grouting, and in particular to a method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model. Background Art

[0002] Due to the effectiveness and economy of cement grouting technology, grouting technology is an indispensable process in the reinforcement, anti-seepage and plugging of water conservancy dam foundations. Due to the mutual influence of stratum changes and grouting parameters, the grouting process presents great uncertainty, time-varying and nonlinearity, and the grouting construction process fails to truly automatically control the grouting pressure. The current large-circulation monitoring system is heavily dependent on the operating experience and professional quality of engineers, and is prone to major accidents.

[0003] The Chinese patent with publication number CN102629107A discloses a pressure closed-loop control method for a large-cycle grouting monitoring system, which controls the grouting pressure in real time by constructing a grouting pressure model. However, this method cannot predict the pressure parameters of the construction pipeline, thereby predicting the pressure control pressure of the pressure regulating valve opening, and cannot achieve the pressure regulating valve opening control requirements for grouting construction. Summary of the invention

[0004] In view of this, the present invention proposes a method for predicting the opening of a cement grouting pressure-regulating valve based on a sequential model. The model is trained by similar grouting geological conditions within a certain range, and the grouting data that has been constructed within the irrigation area is used to guide subsequent grouting construction. The parameters that affect the construction pipeline pressure are trained using the sequential model to establish a model, thereby predicting the opening control pressure of the pressure-regulating valve, which solves the problem that the prior art cannot predict the construction pipeline pressure.

[0005] The technical solution of the present invention is implemented as follows: The present invention provides a method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model, comprising the following steps:

[0006] S1, collecting a first grouting data set of a constructed area near a to-be-constructed area, performing optimal filtering processing on the first grouting data set to obtain a second grouting data set, performing quantitative characterization and standardization processing on the second grouting data set, and dividing it into a training set and a test set;

[0007] S2, constructing an initial sequence model, and training the initial sequence model using the training set to obtain an optimized prediction model;

[0008] S3, performing a model evaluation on the optimization prediction model using the test set, and if the evaluation passes, predicting the opening of the pressure regulating valve according to the optimization prediction model, and if the evaluation fails, returning to step S2 to retrain the model;

[0009] S4, deploying the optimization prediction model to the embedded system, and predicting the opening of the cement grouting pressure regulating valve in the construction area through the embedded system.

[0010] Preferably, step S1 comprises:

[0011] The first grouting data set includes grouting pipeline pressure, grouting flow rate, grouting return flow rate, grouting density, and grouting section depth.

[0012] Preferably, step S1 further includes:

[0013] Performing optimal filtering processing on the first grouting data set, the optimal filtering processing comprising:

[0014] The grouting pipeline pressure and injection rate of the first grouting data set are fitted to obtain a pressure injection rate curve, and the grouting power calculation formula of the first grouting data set is obtained according to the pressure injection rate curve:

[0015] P(n)=F(n)V(n);

[0016] Wherein, P(n) is the grouting power of the first grouting data set, F(n) is the grouting pipeline pressure of the first grouting data set, and V(n) is the injection rate of the first grouting data set;

[0017] Derivative the grouting power to obtain the maximum grouting power P when the derivative value is 0 m , the first grouting data set is screened to remove the grouting power (0.9P m , 1.05P m ) to obtain the second grouting data set.

[0018] Preferably, step S1 further includes:

[0019] Feature extraction is performed on the second grouting dataset to obtain numerical features and type features of the second grouting dataset, the type features are quantified and represented by category vectors, the original values ​​of the numerical data of the second grouting dataset are retained to obtain feature data of the second grouting dataset.

[0020] Preferably, step S1 further includes:

[0021] The characteristic data of the second grouting data set is standardized, and the calculation formula is:

[0022]

[0023] Among them, X BC is the characteristic data after standardization, X is the characteristic data of the second grouting data set, and λ is the characteristic transformation parameter;

[0024] The standardized second grouting dataset is divided into a training set and a test set in a ratio of 7:3.

[0025] Preferably, step S2 comprises:

[0026] Constructing an initial sequential model, wherein the initial sequential model includes an input layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer;

[0027] The output layer inputs a training set, the first fully connected layer includes 64 neurons, the second fully connected layer includes 32 neurons, and the third fully connected layer includes 16 neurons;

[0028] The activation function tanh(x) of the first fully connected layer is:

[0029]

[0030] Among them, x is the input of the first fully connected layer;

[0031] The activation function ReLU(y) of the second fully connected layer is:

[0032] ReLU(y)=max(0,y);

[0033] Among them, y is the input of the second fully connected layer;

[0034] The activation function ELU(z) of the third fully connected layer is:

[0035]

[0036] Among them, z is the input of the third fully connected layer, and α is the hyperparameter of the third fully connected layer;

[0037] The activation function f(s) of the output layer is:

[0038]

[0039] Among them, s is the input of the output layer, and exp is the natural exponential function.

[0040] Preferably, step S2 further includes:

[0041] The initial sequential model is trained using the training set, and the optimizer and hyperparameters are configured until an iteration stop condition is reached, and the model training is terminated to obtain an optimized prediction model;

[0042] The hyperparameters include the total number of iterations, the number of samples, and the learning rate of the optimizer;

[0043] The iteration stopping condition is that the total number of iterations is reached.

[0044] Preferably, step S3 comprises:

[0045] The test set is input into the optimized prediction model, and the model is evaluated using the accuracy rate. When the accuracy rate is not less than the preset threshold, the evaluation passes, and the opening of the pressure regulating valve is predicted according to the optimized prediction model; when the accuracy rate is less than the preset threshold, the evaluation fails, and the process returns to step S2, adjusts the optimizer and hyperparameters, and retrains the model.

[0046] Preferably, step S4 comprises:

[0047] An embedded system is selected, a development environment of the embedded system is prepared, the optimization prediction model is converted into a format file corresponding to the development environment of the embedded system, and the converted format file is copied into the embedded system.

[0048] Preferably, step S4 further includes:

[0049] The embedded system is installed in the area to be constructed, and the third grouting data set of the area to be constructed is collected. The opening of the cement grouting pressure regulating valve in the area to be constructed is predicted according to the third grouting data set. The performance of the embedded system is monitored regularly to perform maintenance and further optimize it.

[0050] The method for predicting the opening of cement grouting pressure regulating valve based on the sequential model of the present invention has the following beneficial effects compared with the prior art:

[0051] (1) The model is trained by similar grouting geological conditions within a certain range, and the grouting data that has been constructed within the irrigation area is used to guide the subsequent grouting construction. The parameters that affect the pressure of the construction pipeline are trained using a sequential model to establish a model, thereby predicting the opening control pressure of the pressure regulating valve;

[0052] (2) By collecting the first grouting data set containing multiple key parameters, a rich and comprehensive data foundation is provided for the training of the sequential model, which significantly improves the accuracy and adaptability of the cement grouting pressure regulating valve opening prediction, realizes real-time monitoring and optimization of the grouting process, and provides data support for the accumulation and inheritance of cement grouting knowledge and experience;

[0053] (3) By fitting the pressure injection rate curve, calculating the grouting power, and screening the data under the optimal working conditions, outliers are effectively removed, high-quality data subsets are extracted, data redundancy is reduced, data processing efficiency is improved, the generalization ability and prediction accuracy of the sequential model are improved, and model convergence is accelerated, laying a good data foundation for subsequent model training and application;

[0054] (4) By combining the feature extraction and abstraction capabilities of a multi-layer fully connected structure with the nonlinear transformation and specific advantages of different activation functions, the initial sequential model can effectively learn and represent the complex nonlinear relationships in the cement grouting process, thereby improving the prediction accuracy and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of the method for predicting the opening of the cement grouting pressure regulating valve based on the sequential model of the present invention;

[0057] Figure 2 A schematic diagram of a modeling process of a Sequential model based on a deep neural network according to a second embodiment of the present invention;

[0058] Figure 3 FIG. 4 is a schematic diagram of the model evaluation result of the second embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Embodiment 1

[0061] A method for predicting the opening of cement grouting pressure regulating valve based on sequential model is provided. Figure 1 As shown, the following steps are included:

[0062] S1, collecting a first grouting data set of a constructed area near a to-be-constructed area, performing optimal filtering processing on the first grouting data set to obtain a second grouting data set, performing quantitative characterization and standardization processing on the second grouting data set, and dividing it into a training set and a test set;

[0063] S2, constructing an initial sequence model, and training the initial sequence model using the training set to obtain an optimized prediction model;

[0064] S3, performing a model evaluation on the optimization prediction model using the test set, and if the evaluation passes, predicting the opening of the pressure regulating valve according to the optimization prediction model, and if the evaluation fails, returning to step S2 to retrain the model;

[0065] S4, deploying the optimization prediction model to the embedded system, and predicting the opening of the cement grouting pressure regulating valve in the construction area through the embedded system.

[0066] It should be noted that:

[0067] At present, the control of the opening of the cement grouting pressure regulating valve mainly depends on the experience and judgment of the on-site engineers. Engineers manually adjust the opening of the pressure regulating valve according to the geological conditions of the construction area, construction requirements, grouting pressure and other factors. Although this method can complete the cement grouting task, it has the following disadvantages:

[0068] Inconsistency in manual control: Different engineers have different experiences and judgments, which leads to poor consistency in the opening control of the pressure regulating valve. This will cause fluctuations in grouting quality and make it difficult to ensure the stability and repeatability of construction.

[0069] Low efficiency: Manually adjusting the opening of the pressure regulating valve is a time-consuming process that requires engineers to constantly monitor and adjust. This reduces the efficiency of cement grouting and prolongs the construction period.

[0070] Difficult to adapt to complex construction conditions: For complex geological conditions or special construction requirements, engineers’ experience is not enough to make the best decision on the opening of the pressure regulating valve. This will lead to a decrease in grouting quality or an increase in construction risks.

[0071] Lack of data accumulation and analysis: Traditional manual control methods make it difficult to effectively collect and analyze data. This limits the in-depth understanding and optimization of the cement grouting process, and makes it difficult to learn and improve from past construction experience.

[0072] Waste of resources: Non-optimal pressure regulating valve opening control will lead to excessive use or loss of cement slurry, increasing material costs. At the same time, frequent manual intervention also increases labor costs.

[0073] In view of the shortcomings of the prior art, this embodiment provides a method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model. By collecting grouting data of the constructed area, a sequential model is constructed to predict the optimal opening of the pressure regulating valve.

[0074] Based on the principle that the geological conditions for grouting within a certain range are similar, the grouting data that have been constructed within the irrigation area can guide the subsequent grouting construction. Based on this principle, the sequential model is innovatively used to train the parameters that affect the construction pipeline pressure to establish a model and predict the opening control pressure of the pressure regulating valve.

[0075] The method for predicting the opening of cement grouting pressure regulating valve based on the sequential model in this embodiment has the following beneficial effects:

[0076] Improve control consistency: Through the prediction of the sequential model, the standardization and consistency of the pressure regulating valve opening control can be achieved, the influence of human factors can be reduced, and the stability of grouting quality can be improved.

[0077] Improve efficiency: Automated pressure regulating valve opening prediction and control can significantly reduce manual intervention, shorten pressure regulation decision-making time, and improve cement grouting efficiency.

[0078] Adaptability to complex construction conditions: By learning from the data of constructed areas, the sequential model can capture the optimal pressure regulating valve opening pattern under complex construction conditions and provide more accurate and adaptable predictions for new construction areas.

[0079] Data accumulation and analysis: This method can continuously collect and accumulate cement grouting data, providing a data basis for in-depth analysis and optimization. Through data mining and learning, the prediction model can be continuously improved to improve the grouting effect.

[0080] Saving resources: By optimizing the opening control of the pressure regulating valve, the waste of cement slurry can be reduced and material costs can be saved. At the same time, automatic control can also reduce manpower input and reduce labor costs.

[0081] Step S1 includes:

[0082] The first grouting data set includes grouting pipeline pressure, grouting flow rate, grouting return flow rate, grouting density, and grouting section depth.

[0083] It should be noted that:

[0084] The grouting pipeline pressure reflects the pressure state of cement slurry in the pipeline and is directly related to the opening of the pressure regulating valve and the grouting effect.

[0085] The grouting flow rate refers to the volume of cement slurry injected into the grouting pipeline per unit time, reflecting the grouting speed and slurry supply.

[0086] The return slurry flow rate refers to the volume of cement slurry returned from the grouting pipeline per unit time, reflecting the slurry loss and circulation during the grouting process.

[0087] Grout density indicates the density of cement slurry, which is related to factors such as slurry proportion and water-cement ratio, and affects the pumpability and consolidation effect of grouting.

[0088] The grouting section depth indicates the depth of the well section where grouting is currently being carried out, reflecting the formation conditions and pressure distribution.

[0089] By collecting these key parameters, the state and characteristics of the cement grouting process can be fully characterized, providing rich characteristic information for the sequential model.

[0090] The construction of this multi-parameter dataset has the following beneficial effects:

[0091] Improved prediction accuracy: By integrating multiple parameters related to the grouting process, the sequential model can more comprehensively understand the relationship between the grouting status and the opening of the pressure regulating valve, thereby improving the accuracy of the prediction.

[0092] Realize process monitoring: Real-time monitoring of the grouting process can be achieved by collecting and analyzing grouting parameter data in real time. This helps to promptly detect abnormal situations, such as pipeline blockage, slurry loss, etc., and take corresponding measures to ensure the quality of grouting.

[0093] Optimize grouting process: By analyzing and mining the collected grouting data, the rules and influencing factors in the grouting process can be discovered. This provides data support for optimizing the grouting process, such as adjusting the slurry ratio, optimizing the grouting parameters, etc., to improve the grouting efficiency and quality.

[0094] Accumulate knowledge and experience: The collected grouting data set is not only used for current model training and prediction, but also can be used as a valuable knowledge asset to provide reference and guidance for subsequent construction projects. Through continuous accumulation and learning, a set of scientific and efficient cement grouting control methods can be formed.

[0095] By collecting the first grouting data set containing multiple key parameters, a rich and comprehensive data basis is provided for the training of the sequential model, which significantly improves the accuracy and adaptability of the cement grouting pressure regulating valve opening prediction, realizes real-time monitoring and optimization of the grouting process, and provides data support for the accumulation and inheritance of cement grouting knowledge and experience.

[0096] Step S1 also includes:

[0097] Performing optimal filtering processing on the first grouting data set, the optimal filtering processing comprising:

[0098] The grouting pipeline pressure and injection rate of the first grouting data set are fitted to obtain a pressure injection rate curve, and the grouting power calculation formula of the first grouting data set is obtained according to the pressure injection rate curve:

[0099] P(n)=F(n)V(n);

[0100] Wherein, P(n) is the grouting power of the first grouting data set, F(n) is the grouting pipeline pressure of the first grouting data set, and V(n) is the injection rate of the first grouting data set;

[0101] Derivative the grouting power to obtain the maximum grouting power P when the derivative value is 0 m , the first grouting data set is screened to remove the grouting power (0.9P m , 1.05P m ) to obtain the second grouting data set.

[0102] It should be noted that:

[0103] The first grouting data set is subjected to optimal filtering, with the goal of selecting the best data subset from the original data to provide high-quality input for subsequent machine learning model training. The beneficial effects of optimal filtering are as follows:

[0104] Removing abnormal data: In the actual cement grouting process, due to measurement errors, equipment failures and other factors, the collected data contains abnormal values ​​or noise. Through optimal filtering processing, these abnormal data can be identified and removed to ensure the accuracy and reliability of the data.

[0105] Extracting the optimal working condition: Grouting power is an important indicator for evaluating grouting performance, which indicates the ability of the grouting system to do work per unit time. By fitting the pressure injection rate curve, the grouting power calculation formula P(n) = F(n)V(n) is obtained, which can quantify the energy transfer efficiency of the grouting process. The grouting power is differentiated and the maximum grouting power Pm corresponding to the derivative being 0 is found, indicating that the grouting system has reached the optimal working condition at this time and the energy transfer efficiency is the highest. By screening the data with grouting power in the range of (0.9Pm, 1.05Pm), a high-quality data subset under the optimal working condition can be obtained.

[0106] Reduce data redundancy: The original collected grouting data contains a lot of redundant information, which increases the overhead of data storage and processing. Through optimal filtering processing, the most representative and informative data subsets can be screened out, which can reduce data redundancy and improve data processing efficiency while ensuring data quality.

[0107] Improve model generalization ability: The performance of machine learning models depends largely on the quality of training data. Through optimal filtering processing, a high-quality second grouting data set can be obtained, which can improve the effective information of model learning, reduce the interference of noise and abnormal data, and thus improve the generalization ability of the model, so that it can make accurate predictions under new grouting conditions.

[0108] Accelerate model convergence: Using high-quality data that has been optimally filtered for model training can accelerate the convergence of the model. Since the data is of higher quality and contains more effective information, the model can learn the key features and laws of the grouting process more quickly, reducing the number of iterations and time required for training.

[0109] Improve prediction accuracy: Through optimal filtering, high-quality data under optimal working conditions can be obtained, which can enable the machine learning model to more accurately capture the key influencing factors and laws of the grouting process. In practical applications, the model trained with these high-quality data can give more accurate prediction results of the pressure regulating valve opening, providing reliable decision support for on-site grouting operations.

[0110] By fitting the pressure injection rate curve, calculating the grouting power, and screening the data under the optimal working conditions, outliers can be effectively removed and high-quality data subsets can be extracted. This can not only reduce data redundancy and improve data processing efficiency, but also improve the generalization ability and prediction accuracy of machine learning models, accelerate model convergence, and lay a good data foundation for subsequent model training and application. Optimal filtering processing is an indispensable part of the data-driven cement grouting pressure regulating valve opening prediction method, which is of great significance for improving prediction performance and field application effects.

[0111] Step S1 also includes:

[0112] Feature extraction is performed on the second grouting dataset to obtain numerical features and type features of the second grouting dataset, the type features are quantified and represented by category vectors, the original values ​​of the numerical data of the second grouting dataset are retained to obtain feature data of the second grouting dataset.

[0113] It should be noted that:

[0114] The purpose of feature extraction on the second grouting data set is to extract the information that best reflects the key features of the grouting process from the original data, providing more targeted and discriminative input for the training of the machine learning model. The beneficial effects are as follows:

[0115] Extracting numerical features: The second grouting dataset contains many numerical variables, such as grouting pressure, injection rate, grouting time, etc. These numerical features directly reflect the quantitative information of the grouting process and are crucial for establishing an accurate prediction model. By retaining the original values ​​of these numerical features, the quantitative information they contain can be retained to the maximum extent, providing rich feature representation for model learning.

[0116] Extracting type features: In addition to numerical features, the second grouting dataset also contains some type features, such as grouting material type, construction environment type, etc. These type features reflect the qualitative information of the grouting process and are of great significance for distinguishing different grouting conditions. In order to incorporate type features into the machine learning model, they need to be quantitatively characterized. Converting type features into numerical form through category vectors (such as one-hot encoding) enables the model to effectively process and learn this qualitative information.

[0117] Improve feature discrimination: Through feature extraction, the subset of features that have the greatest impact on the grouting process and are most discriminative can be screened out from the original data. These features show obvious differences under different grouting conditions and can effectively distinguish different operating conditions and results. Extracting highly discriminative features can enable the machine learning model to more accurately capture the key influencing factors of the grouting process and improve the prediction performance.

[0118] Reduce feature redundancy: The original grouting data contains some redundant or highly correlated features, which have limited contribution to model learning and prediction, but increase the computational complexity. Through feature extraction, these redundant features can be identified and removed, and only the most informative and discriminative feature subsets can be retained. Reducing feature redundancy can not only simplify the model structure, but also improve the training and prediction efficiency.

[0119] Promote model generalization: The goal of feature extraction is to extract key features that are universal and robust to the grouting process. These features can effectively reflect the inherent laws of the grouting process under different grouting conditions. By learning these generalizable features, the machine learning model can better adapt to new grouting data and give reliable prediction results under unknown conditions.

[0120] Improve model interpretability: Through feature extraction, we can select features that are closely related to the grouting process and easy to understand and explain. These features usually have clear physical or engineering meanings, such as grouting pressure, injection rate, etc. The prediction results of the machine learning model established based on these interpretable features are also easier for engineers to understand and accept, which helps to improve the credibility and practicality of the model.

[0121] By extracting numerical features and type features and quantifying the type features, the key information of the grouting process can be retained to the maximum extent, and the distinguishability and generalization ability of the features can be improved. Feature extraction can not only reduce feature redundancy and simplify the model structure, but also improve the interpretability and practicality of the model. The machine learning model based on high-quality features can more accurately predict the opening of the pressure regulating valve and provide reliable decision support for on-site grouting operations.

[0122] Step S1 also includes:

[0123] The characteristic data of the second grouting data set is standardized, and the calculation formula is:

[0124]

[0125] Among them, X BC is the characteristic data after standardization, X is the characteristic data of the second grouting data set, and λ is the characteristic transformation parameter;

[0126] The standardized second grouting dataset is divided into a training set and a test set in a ratio of 7:3.

[0127] It should be noted that:

[0128] The feature data of the second grouting dataset are standardized and divided into datasets, with the aim of further optimizing the feature representation and providing a suitable dataset for the training and evaluation of the machine learning model.

[0129] The beneficial effects are as follows:

[0130] Feature standardization: The BOX-COX standardization method is used to standardize feature data, which can achieve the following purposes:

[0131] Eliminate the impact of dimensions: Different features have different dimensions and value ranges. Directly using the original values ​​will cause some features to dominate the model training, while ignoring the contribution of other features. Through BOX-COX normalization, the impact of dimensions can be eliminated, so that different features can be compared and processed on the same scale.

[0132] Improve model convergence speed: Standardized feature data usually has a smaller value range and a mean close to 0. This data distribution helps speed up the convergence of machine learning models, especially for optimization algorithms based on gradient descent. Standardization makes the contribution of different features to the model more balanced, which helps to reach the optimal solution faster.

[0133] Enhance the generalization ability of the model: Standardization can make the feature data present a more symmetrical and concentrated distribution, reducing the impact of outliers. This helps the model learn the inherent laws of the data instead of overfitting to certain extreme cases. The standardized feature data has better generalization ability, allowing the model to give reliable prediction results on new grouting data.

[0134] Reasonable data set division: The standardized second grouting data set is divided into training set and test set in a ratio of 7:3, which has the following beneficial effects:

[0135] Model training and evaluation: Dividing the data set into a training set and a test set allows for an objective evaluation of the model's performance while the model is being trained. The training set is used to train the machine learning model so that it can learn the key features and laws of the grouting process. The test set is used to evaluate the model's predictive ability on unknown data and verify the model's generalization performance.

[0136] Prevent overfitting: By dividing the data set into a training set and a test set, the model can be effectively prevented from overfitting. Overfitting refers to the phenomenon that the model performs well on the training set but performs poorly on the test set or in actual applications. By evaluating the model performance on the test set, the overfitting problem can be discovered in time and corresponding measures can be taken, such as adjusting the model structure and introducing regularization terms.

[0137] Reasonable partition ratio: A 7:3 partition ratio, that is, 70% of the data is used for training and 30% of the data is used for testing, is a common rule of thumb. This partition ratio ensures sufficient training data while leaving enough test data for performance evaluation. A reasonable partition ratio helps to obtain reliable model performance estimates while avoiding inaccurate performance estimates caused by too little test data.

[0138] Through BOX-COX standardization processing, the dimensional effect can be eliminated, the model convergence speed can be improved, and the generalization ability of the model can be enhanced. Reasonable division of the training set and the test set can be used to evaluate the performance while training the model, prevent overfitting problems, and obtain reliable performance estimates. These processing steps provide a high-quality data foundation for subsequent model training and prediction, help to establish an efficient and accurate cement grouting pressure regulating valve opening prediction model, and provide reliable decision support for engineering applications.

[0139] Step S2 includes:

[0140] Constructing an initial sequential model, wherein the initial sequential model includes an input layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer;

[0141] The output layer inputs a training set, the first fully connected layer includes 64 neurons, the second fully connected layer includes 32 neurons, and the third fully connected layer includes 16 neurons;

[0142] The activation function tanh(x) of the first fully connected layer is:

[0143]

[0144] Among them, x is the input of the first fully connected layer;

[0145] The activation function ReLU(y) of the second fully connected layer is:

[0146] ReLU(y)=max(0,y);

[0147] Among them, y is the input of the second fully connected layer;

[0148] The activation function ELU(z) of the third fully connected layer is:

[0149]

[0150] Among them, z is the input of the third fully connected layer, and α is the hyperparameter of the third fully connected layer;

[0151] The activation function f(s) of the output layer is:

[0152]

[0153] Among them, s is the input of the output layer, and exp is the natural exponential function.

[0154] It should be noted that:

[0155] By building an initial sequential model, including an input layer, three fully connected layers, and an output layer, and selecting different activation functions for each fully connected layer, this model structure and activation function selection has the following beneficial effects:

[0156] Multi-layer fully connected structure: The initial sequential model uses a three-layer fully connected structure, each containing a different number of neurons (64, 32, and 16). This multi-layer fully connected structure has the following advantages:

[0157] Feature extraction and abstraction: Each fully connected layer can learn different levels of feature representation of the input data. The bottom fully connected layer (such as the first fully connected layer) can extract the combination and transformation of the original features and learn higher-level features. The high-level fully connected layers (such as the second and third fully connected layers) can further extract and abstract these high-level features to capture complex patterns and relationships in the data.

[0158] Enhance the expressiveness of the model: By using multiple fully connected layers, the model can learn the nonlinear relationship between input data and output. Each fully connected layer can perform nonlinear transformation on the input by adjusting weights and biases, so that the model can fit more complex functional relationships. The multi-layer structure makes the model more expressive and can handle various influencing factors and nonlinear relationships in the grouting process.

[0159] Dimensionality reduction and information compression: From the first fully connected layer to the third fully connected layer, the number of neurons gradually decreases (64 to 32 and then to 16). This structure can achieve data dimensionality reduction and information compression. By reducing the number of neurons layer by layer, the model can extract the most critical features and remove redundant and noisy information, thereby improving the generalization ability and computational efficiency of the model.

[0160] Choice of activation function: Each fully connected layer uses a different activation function (Tanh, ReLU, and ELU), and the output layer uses the sigmoid activation function. This choice of activation function has the following benefits:

[0161] Nonlinear transformation: The activation function introduces nonlinear transformation, which enables the model to learn and represent nonlinear relationships. Tanh, ReLU and ELU are all commonly used nonlinear activation functions, which can perform nonlinear mapping on the input and enhance the expressive power of the model.

[0162] Advantages of specific activation functions:

[0163] Tanh activation function: maps the input to the range of [-1, 1], with the characteristic of zero centering. This helps to accelerate the convergence of the model and alleviate the gradient vanishing problem.

[0164] ReLU activation function: maps negative input to 0, and keeps positive input unchanged. ReLU has the advantages of high computational efficiency and fast convergence speed, and can alleviate the gradient vanishing problem.

[0165] ELU activation function: Similar to ReLU, but for negative input, ELU gives a small negative output. This feature allows ELU to maintain the advantages of ReLU while also alleviating the "dead ReLU" problem and improving the robustness of the model.

[0166] Sigmoid output layer: Use the sigmoid activation function to map the output to the range of [0, 1]. This mapping is suitable for binary classification problems or scenarios that require probabilistic output. In the prediction of cement grouting pressure regulating valve opening, the sigmoid function can interpret the model output as the probability of valve opening, which facilitates subsequent decision-making and control.

[0167] Different activation functions have different nonlinear transformation characteristics. By using different activation functions in different fully connected layers, the model can learn and represent various types of nonlinear relationships. Tanh provides bidirectional nonlinear transformation, ReLU provides unidirectional nonlinear transformation, and ELU introduces additional nonlinear transformation in the negative part. This variety of nonlinear transformation can enhance the model's ability to capture complex patterns and relationships.

[0168] A reasonable combination of activation functions can promote the flow of gradients in the network and accelerate the convergence of the model. The Tanh activation function is used in the first fully connected layer to map the input data to a zero-centered range, which helps accelerate convergence. The ReLU activation function is used in the second fully connected layer, which has good gradient flow and computational efficiency and can further accelerate convergence. The ELU activation function is used in the third fully connected layer, which can alleviate the "dead ReLU" problem while maintaining the advantages of ReLU and improve the robustness of the model.

[0169] The Tanh activation function can learn the bidirectional feature representation of the data in the first fully connected layer, the ReLU activation function can learn the unidirectional feature representation of the data in the second fully connected layer, and the ELU activation function can learn the feature representation with additional nonlinear transformation of the negative part in the third fully connected layer. This multi-level and multi-angle feature representation can better characterize the complex characteristics of the cement grouting process.

[0170] By combining the feature extraction and abstraction capabilities of the multi-layer fully connected structure, as well as the nonlinear transformation and specific advantages of different activation functions, the initial sequential model can effectively learn and represent the complex nonlinear relationships in the cement grouting process, improving the accuracy and generalization of predictions. At the same time, through the reasonable selection of activation functions, the model can accelerate convergence, alleviate problems such as gradient vanishing and "dead ReLU", and improve the robustness and stability of the model. This lays a good foundation for subsequent model training and optimization, and helps to obtain a high-performance cement grouting pressure regulating valve opening prediction model.

[0171] Step S2 also includes:

[0172] The initial sequential model is trained using the training set, and the optimizer and hyperparameters are configured until an iteration stop condition is reached, and the model training is terminated to obtain an optimized prediction model;

[0173] The hyperparameters include the total number of iterations, the number of samples, and the learning rate of the optimizer;

[0174] The iteration stopping condition is that the total number of iterations is reached.

[0175] It should be noted that:

[0176] The initial sequential model is trained with the training set, and the beneficial effects are as follows:

[0177] Optimizer selection and configuration: The optimizer plays a key role in model training, which determines the update method and speed of model parameters. Common optimizers include SGD, Adam, RMSprop, etc. By selecting an appropriate optimizer and configuring a suitable learning rate, you can effectively optimize model parameters, accelerate convergence, and improve model performance. The learning rate is an important hyperparameter of the optimizer, which controls the step size of each parameter update. A suitable learning rate can reach the optimal solution in a reasonable time and avoid convergence problems caused by too large or too small learning rates.

[0178] Setting the total number of iterations: The total number of iterations determines the duration of model training and the degree of data utilization. Setting an appropriate total number of iterations can ensure that the model fully learns the features and patterns in the training data while avoiding overfitting. By setting the total number of iterations as the iteration stop condition, you can control the model training time and computing resource consumption while ensuring that the model reaches a certain performance level.

[0179] Setting the number of samples: The number of samples refers to the number of samples used to train the model in each iteration, also known as the batch size. The appropriate number of samples can balance computational efficiency and memory consumption. A larger number of samples can take advantage of parallel computing and accelerate the training process, but it will also increase memory consumption. A smaller number of samples can reduce memory consumption, but it will affect the speed and convergence of training. By weighing computational efficiency and memory consumption, choosing the appropriate number of samples can optimize the performance and resource utilization of model training.

[0180] Setting of iteration stop condition: The iteration stop condition determines when the model training is terminated. In step S2, the total number of iterations is used as the iteration stop condition. This means that when the model training reaches the preset total number of iterations, the training process will automatically terminate. The advantage of this stop condition is that it is simple and clear, and can control the training time and the consumption of computing resources. However, it also has certain limitations, because the preset total number of iterations cannot guarantee that the model achieves optimal performance. In order to further optimize model training, other stop conditions can be considered, such as performance indicators on the validation set, early stopping method, etc.

[0181] Model optimization and performance improvement: By selecting the appropriate optimizer, configuring hyperparameters, and setting the iteration stop condition, the model parameters can be effectively optimized and the model performance can be improved. During the training process, the optimizer continuously updates the model parameters to adjust them in the direction of minimizing the loss function. Through multiple iterations, the model gradually learns the key features and patterns in the data, improving the accuracy and generalization of predictions. At the same time, reasonable hyperparameter settings and iteration stop conditions can prevent overfitting and ensure the performance of the model on unknown data.

[0182] By selecting a suitable optimizer, configuring the learning rate, setting the total number of iterations and the number of samples, and reasonable iteration stop conditions, the model parameters can be effectively optimized, the performance and generalization ability of the model can be improved, and the model can be ensured to learn key features and patterns from the training data and make accurate predictions of the opening of cement grouting pressure regulating valves in practical applications. At the same time, reasonable hyperparameter settings and iteration stop conditions can also help control the training time and the consumption of computing resources, making the model training process more efficient and controllable.

[0183] Step S3 includes:

[0184] The test set is input into the optimized prediction model, and the model is evaluated using the accuracy rate. When the accuracy rate is not less than the preset threshold, the evaluation passes, and the opening of the pressure regulating valve is predicted according to the optimized prediction model; when the accuracy rate is less than the preset threshold, the evaluation fails, and the process returns to step S2, adjusts the optimizer and hyperparameters, and retrains the model.

[0185] It should be noted that:

[0186] The optimized prediction model is evaluated through the test set, and the beneficial effects are as follows:

[0187] Model performance evaluation: By inputting the test set into the optimized prediction model, the performance of the model on unknown data can be evaluated. Accuracy is a commonly used evaluation metric that measures the consistency between the model's prediction results and the actual labels. By calculating the accuracy of the model on the test set, the model's predictive ability and generalization performance can be objectively evaluated. This step can help determine whether the model has achieved the expected performance requirements and provide a basis for subsequent model applications.

[0188] Setting of preset threshold: The preset threshold is a key parameter that determines whether the model performance reaches an acceptable level. By setting a reasonable preset threshold, it can be ensured that the model has sufficient reliability and accuracy in practical applications. If the accuracy of the model is not less than the preset threshold, it is considered that the performance of the model meets the requirements and can be used to predict the opening of the pressure regulating valve. On the contrary, if the accuracy is less than the preset threshold, it indicates that the performance of the model needs to be improved and further optimization is required.

[0189] Model iterative optimization: When the accuracy of the model does not reach the preset threshold, step S3 provides an iterative optimization mechanism. By returning to step S2, the model can be further optimized and adjusted. This includes adjusting the selection of the optimizer, adjusting the value range of the hyperparameters, etc. Through continuous iteration and optimization, the performance of the model can be gradually improved to meet the requirements of the preset threshold. This iterative optimization process can ensure that the model achieves optimal performance in continuous improvement and perfection.

[0190] Reliability and stability of the model: By setting preset thresholds and iterative optimization mechanisms, the model can be ensured to have high reliability and stability in practical applications. When the accuracy of the model reaches the preset threshold, it means that the model has sufficient prediction capabilities and can accurately predict the opening of the pressure regulating valve in actual scenarios. At the same time, the iterative optimization process can continuously improve the robustness of the model, enabling it to adapt to different data distributions and noise interference, and improve the stability of the model.

[0191] Practical application of the model: Once the model passes the evaluation and reaches the preset threshold requirements, it can be applied to the actual cement grouting pressure regulating valve opening prediction. By inputting relevant feature data, the optimization prediction model can predict the optimal opening of the pressure regulating valve based on historical data and current status. This machine learning-based prediction method can provide more accurate and real-time decision support, optimize the cement grouting process, and improve production efficiency and quality.

[0192] By setting a preset threshold, the prediction ability and generalization performance of the model can be objectively evaluated to ensure that the model meets the requirements of practical applications. When the model does not reach the preset threshold, by returning to step S2 for iterative optimization, the performance of the model can be continuously improved to achieve the ideal accuracy level. At the same time, the iterative optimization process also improves the reliability and stability of the model, enabling it to adapt to different data distributions and noise interference. Ultimately, through a qualified optimization prediction model, accurate and real-time cement grouting pressure regulating valve opening prediction can be achieved, the production process can be optimized, and efficiency and quality can be improved.

[0193] Step S4 includes:

[0194] An embedded system is selected, a development environment of the embedded system is prepared, the optimization prediction model is converted into a format file corresponding to the development environment of the embedded system, and the converted format file is copied into the embedded system.

[0195] It should be noted that:

[0196] Deploy the optimized prediction model to the embedded system to realize the practical application of the model, and the beneficial effects are as follows:

[0197] Selecting a suitable embedded system: An embedded system is a special-purpose computer system with the characteristics of small size, low power consumption, and strong real-time performance. Selecting a suitable embedded system is crucial for the actual deployment and application of the model. It is necessary to consider factors such as the processing power, memory size, and interface type of the embedded system to ensure that it can meet the requirements of the model operation. At the same time, it is also necessary to consider the cost, reliability, scalability, etc. of the embedded system to meet the requirements of actual applications.

[0198] Prepare the development environment for embedded systems: The development environment for embedded systems is usually different from that for ordinary computers. It requires specific tools such as compilers, linkers, debuggers, and libraries and drivers that match the embedded system hardware. Preparing the development environment for embedded systems is a necessary step for model deployment, which ensures that the model can be correctly compiled, linked, and run on the embedded system. By preparing a suitable development environment, the efficiency of model migration can be improved and compatibility issues can be reduced.

[0199] Model format conversion: Different embedded systems and development environments support different model formats. In order to deploy the optimized prediction model into an embedded system, the model needs to be converted into a format file corresponding to the development environment of the embedded system. This format conversion ensures that the model can be loaded and run normally on the embedded system. Common model formats include TensorFlow Lite, ONNX, Caffe, etc. Different embedded systems and development environments support these formats to varying degrees. By converting the model into a suitable format file, the model can be efficiently deployed and executed on the embedded system.

[0200] Copying and deploying model files: Copying the converted model format files to the embedded system is a key step in model deployment. This process needs to consider factors such as the storage space of the embedded system, the file system format, and the file transfer method. By copying the model files to the appropriate location of the embedded system and performing the necessary configuration and initialization, it can be ensured that the model can be loaded and run normally on the embedded system. This step realizes the migration of the model from the development environment to the actual application environment, laying the foundation for the real-time prediction and control of the model.

[0201] Realize real-time prediction and control of the model: After the optimization prediction model is deployed to the embedded system, the real-time prediction and control functions of the model can be realized. The embedded system can collect relevant data of the cement grouting process in real time through sensors and other devices, and input it into the prediction model for real-time prediction. According to the prediction results, the embedded system can adjust the opening of the cement grouting pressure regulating valve in real time to realize automatic control and optimization of the production process. By combining the model with the embedded system, real-time monitoring, prediction and control of the cement grouting process can be realized, improving production efficiency and quality.

[0202] The process of deploying the optimization prediction model to the embedded system realizes the migration of the model from the development environment to the actual application environment. By selecting a suitable embedded system, preparing the development environment, converting the model format and copying the files, the efficient deployment and execution of the model on the embedded system can be ensured. This process lays the foundation for the real-time prediction and control of the model, making the automated control and optimization of the cement grouting process possible. By combining the machine learning model with the embedded system, the efficiency, quality and stability of the cement grouting process can be significantly improved, providing strong support for intelligent cement production.

[0203] Step S4 also includes:

[0204] The embedded system is installed in the area to be constructed, and the third grouting data set of the area to be constructed is collected. The opening of the cement grouting pressure regulating valve in the area to be constructed is predicted according to the third grouting data set. The performance of the embedded system is monitored regularly to perform maintenance and further optimize it.

[0205] It should be noted that:

[0206] The embedded system is installed in the area to be constructed, and data collection, prediction and control, as well as regular monitoring and maintenance are carried out. The beneficial effects are as follows:

[0207] Realize real-time data collection: After the embedded system is installed in the construction area, the relevant data of the cement grouting process can be collected in real time through sensors and other equipment to form a third grouting data set. These data can include parameters such as pressure, flow, density of cement slurry, as well as factors such as ambient temperature and humidity. Through real-time data collection, the real state and dynamic changes of the cement grouting process in the construction area can be obtained, providing data support for subsequent prediction and control. Real-time data collection can improve the timeliness and accuracy of data and ensure the reliability of prediction and control.

[0208] Realize real-time prediction and control: Based on the collected third grouting data set, the embedded system can use the deployed optimization prediction model for real-time prediction. By inputting the real-time collected data into the prediction model, the optimal opening of the cement grouting pressure regulating valve in the construction area can be predicted. According to the prediction results, the embedded system can adjust the opening of the cement grouting pressure regulating valve in real time to achieve automatic control and optimization of the cement grouting process. Real-time prediction and control can reduce manual intervention, improve the efficiency and quality of the cement grouting process, and ensure the stability and reliability of construction.

[0209] Improve the adaptability of the cement grouting process: By installing an embedded system in the construction area and performing real-time data collection and prediction, the opening of the cement grouting pressure regulating valve can be dynamically adjusted according to changes in the actual construction environment. Different construction areas have differences in geological conditions, environmental factors, etc. Using an embedded system for real-time prediction and control can make adaptive adjustments based on actual conditions to ensure the optimal effect of the cement grouting process. This adaptive adjustment can improve the flexibility and robustness of the cement grouting process and adapt to the requirements of different construction environments.

[0210] Realize remote monitoring and management: Through the network communication function of the embedded system, remote monitoring and management of the cement grouting process can be realized. Managers can remotely access the embedded system to view the various parameters and status of the cement grouting process in real time, understand the construction progress and equipment operation. At the same time, managers can also remotely adjust the parameter settings of the embedded system to remotely control and optimize the cement grouting process. Remote monitoring and management can reduce the workload of on-site personnel, improve management efficiency, and promptly discover and deal with potential problems.

[0211] Regular monitoring and maintenance: Embedded systems run in harsh construction environments for a long time and will be affected by various factors, such as temperature, humidity, vibration, etc. Regular monitoring of the performance of embedded systems can timely detect abnormal conditions and potential failures of the system and perform corresponding maintenance and optimization. Regular monitoring and maintenance of embedded systems can ensure the stable operation of the system, extend the service life of the system, and continuously optimize the performance of the system. Regular monitoring and maintenance can reduce the occurrence of system failures and improve the reliability and continuity of the cement grouting process.

[0212] Continuous optimization and improvement: In the actual application process, through the performance monitoring and data analysis of the embedded system, the deficiencies and room for improvement of the system can be discovered. Based on the feedback from the actual application and the results of the data analysis, the embedded system can be continuously optimized and improved. This involves retraining and optimizing the prediction model, upgrading and updating the hardware and software of the embedded system, and adjusting and improving the control strategy. Through continuous optimization and improvement, the performance and efficiency of the embedded system can be continuously improved to adapt to the ever-changing construction requirements and technological developments.

[0213] Installing the embedded system in the construction area and performing real-time data collection, prediction and control, as well as regular monitoring and maintenance, can significantly improve the automation and intelligence level of the cement grouting process. Through real-time data collection and prediction, the opening of the cement grouting pressure regulating valve can be dynamically adjusted according to the changes in the actual construction environment to achieve adaptive control and optimization of the cement grouting process. At the same time, through remote monitoring and management, management efficiency can be improved, and potential problems can be discovered and handled in a timely manner. Regular monitoring and maintenance can ensure the stable operation of the embedded system, and continuous optimization and improvement can continuously improve the performance and efficiency of the system. The combined effect of these effects can significantly improve the efficiency, quality and reliability of the cement grouting process, providing strong support for intelligent cement construction.

[0214] Embodiment 2

[0215] There are many interference factors in cement grouting pressure regulation, including grouting pipeline pressure P, grouting flow rate Q+, return grouting flow rate Q-, slurry density Md, grouting section depth H and other factors. The input parameters include grouting pipeline pressure P, grouting flow rate Q+, return grouting flow rate Q-, slurry density Md and grouting section depth H, and the output parameter is the regulating valve opening O. The data set D (P, Q+, Q-, Md, H) is obtained by collecting field data and filtering it through the optimal grouting power. After quantitative characterization and standardization, the data set D is randomly divided into training set D1 and test set D2. The model is built using the Sequential model based on deep neural network. The activation function in the model compilation process selects Rule function, the optimization function selects Adam optimizer, and the loss function selects MAE (mean square error loss). The training set D1 is brought into the training, the input layer is D1, and the output layer is the regulating valve opening. After the training is completed, the test set D2 is used for model evaluation. The model that passes the evaluation can be used to guide the grouting pressure regulation opening at the next moment. Figure 2 This is the modeling process of the Sequential model based on a deep neural network in this embodiment. Figure 3 This is the result of model evaluation (comparison between predicted value and true value). According to the error percentage of the output layer (control valve opening) of the model evaluation, the error between the control valve opening of more than 95% of the test samples and the theoretical predicted opening is within ±3%. The very few large errors are related to data cleaning and screening, indicating that this model matches the current construction parameters well and can be used to guide intelligent pressure regulation work.

[0216] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model, characterized in that: The following steps are involved: S1, collecting a first grouting data set of a constructed area near a to-be-constructed area, performing optimal filtering processing on the first grouting data set to obtain a second grouting data set, performing quantitative characterization and standardization processing on the second grouting data set, and dividing it into a training set and a test set; S2, constructing an initial sequence model, and training the initial sequence model using the training set to obtain an optimized prediction model; S3, performing a model evaluation on the optimization prediction model using the test set, and if the evaluation passes, predicting the opening of the pressure regulating valve according to the optimization prediction model, and if the evaluation fails, returning to step S2 to retrain the model; S4, deploying the optimization prediction model to the embedded system, and predicting the opening of the cement grouting pressure regulating valve in the construction area through the embedded system.

2. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 1, characterized in that: Step S1 includes: The first grouting data set includes grouting pipeline pressure, grouting flow rate, grouting return flow rate, grouting density, and grouting section depth.

3. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 2, characterized in that: Step S1 also includes: Performing optimal filtering processing on the first grouting data set, the optimal filtering processing comprising: The grouting pipeline pressure and injection rate of the first grouting data set are fitted to obtain a pressure injection rate curve, and the grouting power calculation formula of the first grouting data set is obtained according to the pressure injection rate curve: P(n)=F(n)V(n); Wherein, P(n) is the grouting power of the first grouting data set, F(n) is the grouting pipeline pressure of the first grouting data set, and V(n) is the injection rate of the first grouting data set; Derivative the grouting power to obtain the maximum grouting power P when the derivative value is 0 m , the first grouting data set is screened to remove the grouting power (0.9P m , 1.05P m ) to obtain the second grouting data set.

4. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 3, characterized in that: Step S1 also includes: Feature extraction is performed on the second grouting dataset to obtain numerical features and type features of the second grouting dataset, the type features are quantified and represented by category vectors, the original values ​​of the numerical data of the second grouting dataset are retained to obtain feature data of the second grouting dataset.

5. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 4, characterized in that: Step S1 also includes: The characteristic data of the second grouting data set is standardized, and the calculation formula is: Among them, X BC is the characteristic data after standardization, X is the characteristic data of the second grouting data set, and λ is the characteristic transformation parameter; The standardized second grouting dataset is divided into a training set and a test set in a ratio of 7:

3.

6. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 5, characterized in that: Step S2 includes: Constructing an initial sequential model, wherein the initial sequential model includes an input layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and an output layer; The output layer inputs a training set, the first fully connected layer includes 64 neurons, the second fully connected layer includes 32 neurons, and the third fully connected layer includes 16 neurons; The activation function tanh(x) of the first fully connected layer is: Among them, x is the input of the first fully connected layer; The activation function ReLU(y) of the second fully connected layer is: ReLU(y)=max(0,y); Among them, y is the input of the second fully connected layer; The activation function ELU(z) of the third fully connected layer is: Among them, z is the input of the third fully connected layer, and α is the hyperparameter of the third fully connected layer; The activation function f(s) of the output layer is: Among them, s is the input of the output layer, and exp is the natural exponential function.

7. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 6, characterized in that: Step S2 also includes: The initial sequential model is trained using the training set, and the optimizer and hyperparameters are configured until an iteration stop condition is reached, and the model training is terminated to obtain an optimized prediction model; The hyperparameters include the total number of iterations, the number of samples, and the learning rate of the optimizer; The iteration stopping condition is that the total number of iterations is reached.

8. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 7, characterized in that: Step S3 includes: The test set is input into the optimized prediction model, and the model is evaluated using the accuracy rate. When the accuracy rate is not less than the preset threshold, the evaluation passes, and the opening of the pressure regulating valve is predicted according to the optimized prediction model; when the accuracy rate is less than the preset threshold, the evaluation fails, and the process returns to step S2, adjusts the optimizer and hyperparameters, and retrains the model.

9. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 8, characterized in that: Step S4 includes: An embedded system is selected, a development environment of the embedded system is prepared, the optimization prediction model is converted into a format file corresponding to the development environment of the embedded system, and the converted format file is copied into the embedded system.

10. The method for predicting the opening of a cement grouting pressure regulating valve based on a sequential model as claimed in claim 9, characterized in that: Step S4 also includes: The embedded system is installed in the area to be constructed, and the third grouting data set of the area to be constructed is collected. The opening of the cement grouting pressure regulating valve in the area to be constructed is predicted according to the third grouting data set. The performance of the embedded system is monitored regularly to perform maintenance and further optimize it.

Citation Information

Cited By

  • Grouting parameter type selection optimization method and system based on deep learning

    CN120068620A

  • A deep learning-based grouting parameter selection optimization method and system

    CN120068620B