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

Abstract

The invention discloses a grouting parameter type selection optimization method and system based on deep learning, and belongs to the technical field of tunnel grouting, and the method comprises the steps: carrying out the feature extraction of preprocessed geological condition and construction condition data, carrying out the dimension reduction of the extracted features, and obtaining the low-dimensional feature data; performing preliminary importance evaluation on the low-dimensional feature data to obtain a feature importance score; the features with the feature importance scores higher than a set threshold value are input into a pre-trained global optimization model, and grouting effect scores are obtained; and inputting the grouting effect score and the feature data corresponding to the grouting effect into a pre-trained local fine tuning model to obtain a fine tuning value of the grouting parameter. A dual structure of a global optimization model and a local fine tuning model is adopted, global optimization and local fine tuning of grouting parameters are designed respectively, the hierarchical model design enables the system to better adapt to changes under complex construction conditions, and the stability and reliability of the overall grouting effect are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel grouting, and particularly relates to a method and system for optimizing the selection of grouting parameters based on deep learning. Background Technique

[0002] The statements in this part only provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the rapid development of underground engineering, tunnel construction faces increasingly complex geological conditions. Especially in soft surrounding rocks and water-rich strata, the disaster risks such as water and soil influx and collapse increase significantly. As an important engineering means, grouting technology is widely used in fields such as strengthening surrounding rocks, controlling groundwater, and preventing formation deformation. However, the grouting effect is affected by various factors, including geological conditions, grouting material properties, grouting parameters, etc. Its complexity makes it difficult to determine grouting parameters, which in turn affects the construction effect and cost.

[0004] Traditional grouting parameter design mostly relies on the experience of construction personnel and on-site tests, lacking scientificity and systematicness. Due to the uncertainty and complexity of geological conditions, it is difficult to standardize grouting parameters in different engineering environments and often requires continuous adjustment, increasing the construction period and cost. In addition, geological conditions may change at any time during the construction process, such as the instability of the surrounding rock structure, the fluctuation of the underground water flow velocity and flow rate, etc. These factors will directly affect the grouting effect. Therefore, how to achieve real-time optimization and intelligent control of grouting parameters has become an important research direction in current tunnel construction technology.

[0005] In recent years, with the development of big data and artificial intelligence technologies, intelligent optimization methods based on deep learning have provided new solutions for tunnel grouting technology. By constructing and training deep learning models, key features and laws can be automatically extracted based on a large amount of historical data, effectively improving the scientificity of grouting parameter design. Although such intelligent optimization methods have provided new opportunities for tunnel grouting technology, there are still some problems. For example, when the model is applied in a resource-constrained on-site environment, real-time performance and computational efficiency also face certain challenges and may be difficult to meet the rapid response construction requirements; the existing models lag in responding to changes in the construction environment and are difficult to optimize grouting parameters in real time, resulting in insufficient adaptability of the plan and still requiring frequent manual intervention, etc. Summary of the Invention

[0006] To overcome the deficiencies of the above-mentioned existing technologies, the present invention provides a method and system for optimizing the selection of grouting parameters based on deep learning, which adopts a dual structure of "global optimization model + local fine-tuning model", and designs for the global optimization and local fine-tuning of grouting parameters respectively. This hierarchical model design enables the system to better adapt to changes under complex construction conditions, and improves the stability and reliability of the overall grouting effect.

[0007] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0008] In the first aspect, the present invention provides a method for optimizing the selection of grouting parameters based on deep learning, including:

[0009] Obtain the current geological conditions and construction condition data and perform preprocessing;

[0010] Extract features from the preprocessed geological conditions and construction condition data, and perform dimensionality reduction on the extracted features to obtain low-dimensional feature data; perform a preliminary importance assessment on the low-dimensional feature data to obtain a feature importance score;

[0011] Input the features with a feature importance score higher than a set threshold into a pre-trained global optimization model to obtain a grouting effect score;

[0012] Input the grouting effect score and the feature data corresponding to the grouting effect into a pre-trained local fine-tuning model to obtain a fine-tuning value of the grouting parameters.

[0013] In a further technical solution, the construction condition data includes grouting material characteristics, grouting parameters, and grouting effect.

[0014] In a further technical solution, obtaining the low-dimensional feature data specifically includes:

[0015] Extract relevant features from the preprocessed geological conditions and construction condition data and perform normalization processing;

[0016] Calculate the correlation between each feature and the grouting effect index using the Pearson correlation coefficient, and screen the features related to the grouting effect index;

[0017] Apply improved linear discriminant analysis to perform dimensionality reduction processing on the screened features to obtain low-dimensional feature data.

[0018] In a further technical solution, the improved linear discriminant analysis introduces a feature weight matrix and a class contribution weight matrix on the basis of linear discriminant analysis, and is specifically expressed as:

[0019]

[0020] Among them, ζ(w) represents the weighted objective function, and w T is the feature projection direction vector, W c is the class weight matrix, S b is the between-class scatter matrix, W f is the feature weight matrix, w is the corresponding feature vector, S w is the within-class scatter matrix.

[0021] For a further technical solution, the specific feature importance score is obtained as follows:

[0022] Use the gradient boosting decision tree algorithm to perform a preliminary importance assessment on the low-dimensional feature data, construct and train an XGBoost model to evaluate the importance of features, and output the feature importance score;

[0023] Introduce a weighted feature importance evaluation mechanism, combine engineering experience and target-oriented weight adjustment to make the feature importance score targeted and interpretable; specifically including:

[0024] First, introduce a set of engineering experience weights to adjust the feature importance score, which is expressed as:

[0025] I i ′ = I i ·W e,i

[0026] Among them, I i is the feature importance score of the i-th feature; W e,i is the engineering experience weight; I i ′ is the corrected feature importance score;

[0027] Then, introduce target correlation correction, combine the optimization target to perform a secondary correction on the feature importance, obtain the feature importance score after the secondary correction, and normalize it to obtain the final feature importance score.

[0028] For a further technical solution, the global optimization model is constructed using a multi-layer perceptron, including an input layer, several hidden layers, a Dropout layer, and an output layer, and a Dropout layer is added after each hidden layer of the multi-layer perceptron.

[0029] For a further technical solution, the local fine-tuning model is constructed using a lightweight convolutional neural network, receives the grouting effect score and the feature data corresponding to the grouting effect, sets an optimization target to maximize the grouting effect score, and the objective function is expressed as:

[0030]

[0031] Among them, N is the number of training samples, S local is the local score obtained after the model is optimized, Sglobal is the global evaluation score of grouting effect;

[0032] Then, the local fine-tuning model adjusts the input features, passes through the hidden layer of the multi-layer fully connected network, and finally outputs the fine-tuning values of each grouting parameter, expressed as:

[0033] ΔP inj = W out,1 ·h (L) + b out,1 , ΔT inj = W out,2 ·h (L) + b out,2

[0034] Wherein, ΔP inj and ΔT inj respectively represent the fine-tuning values of grouting pressure and grouting time, W out,i represents the output layer weight matrix, h (L) represents the final output of the hidden layer, and b out,i is the bias of the output layer.

[0035] In a second aspect, the present invention provides a grouting parameter selection and optimization system based on deep learning, including:

[0036] A data acquisition module, which is configured to: acquire current geological condition and construction condition data and perform preprocessing;

[0037] A feature screening module, which is configured to: extract features from the preprocessed geological condition and construction condition data, perform dimensionality reduction on the extracted features to obtain low-dimensional feature data; perform a preliminary importance assessment on the low-dimensional feature data to obtain a feature importance score;

[0038] A global optimization module, which is configured to: input the features with a feature importance score higher than a set threshold into a pre-trained global optimization model to obtain a grouting effect score;

[0039] A local adjustment module, which is configured to: input the grouting effect score and the feature data corresponding to the grouting effect into a pre-trained local fine-tuning model to obtain the fine-tuning values of the grouting parameters.

[0040] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in a grouting parameter selection and optimization method based on deep learning as described in the first aspect.

[0041] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in a grouting parameter selection and optimization method based on deep learning as described in the first aspect are implemented.

[0042] The above one or more technical solutions have the following beneficial effects:

[0043] The present invention adopts a dual structure of "global optimization model + local fine-tuning model", which is designed for the global optimization and local fine-tuning of grouting parameters respectively. The global optimization model mainly focuses on the grouting parameters with greater influence and realizes the global optimal solution through continuous changes; the local fine-tuning model, based on the global optimization model, identifies the potential deficiencies of the current grouting scheme and further optimizes the parameters output by the global optimization model. Through the combined action of the global optimization model and the local fine-tuning model, a set of grouting parameters that can achieve the best grouting effect under the current geological and construction conditions is selected, including grouting pressure, grouting speed, etc. This hierarchical model design enables the system to better adapt to changes under complex construction conditions and improves the stability and reliability of the overall grouting effect.

[0044] The present invention emphasizes the coupling effect between different grouting parameters and realizes the comprehensive prediction of grouting effect through multi-parameter coupling analysis. Different from the traditional linear superposition model, this system can capture the complex non-linear relationships between parameters, thus improving the accuracy of grouting effect prediction. This innovative application enables engineers to more comprehensively consider the mutual influence between various parameters when designing the grouting process, optimize the grouting scheme, and improve construction safety and efficiency.

[0045] The present invention uses the gradient boosting decision tree algorithm to evaluate the feature importance and introduces a weighted feature importance evaluation mechanism on this basis, which can effectively identify the features that have the greatest impact on the grouting effect. This process not only helps to optimize the model feature input, improve the training efficiency and prediction performance of the model, but also avoids the computational complexity caused by feature redundancy. The feature dimension reduction process further improves the computational efficiency of the model, enabling the system to provide high-quality prediction results in a shorter time and improving the real-time response of the global optimization model and the local fine-tuning model. Description of the Drawings

[0046] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0047] Figure 1 It is a flowchart of the grouting parameter selection and optimization method in the embodiment of the present invention;

[0048] Figure 2 It is the structural diagram of the global optimization model in the embodiment of the present invention. Specific implementation manners

[0049] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0050] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0051] Without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0052] Embodiment 1

[0053] As Figure 1 shown, this embodiment discloses a method for optimizing the selection of grouting parameters based on deep learning. The method includes the following steps:

[0054] S1: Obtain the current geological conditions and construction condition data and perform preprocessing.

[0055] In this embodiment, various historical data related to tunnel grouting are collected through construction logs, geological reports, and reference documents, including geological condition data and construction condition data of different projects. The construction condition data includes data such as the characteristics of grouting materials, grouting parameters, and grouting effects. The grouting effect is quantified by grouting effect indicators, and a comprehensive database is established.

[0056] Clean and preprocess the collected original data, that is, geological condition and construction condition data: For outliers, use statistical methods or identification methods for processing. For example, for data with significantly deviated compressive strength from the reasonable range, analyze whether it is an abnormal value caused by test errors or other reasons, and decide whether to eliminate or correct it; for missing values, use interpolation methods or methods of deleting missing data to process the null values in the data.

[0057] After the data preprocessing is completed, the data is classified and stored according to its characteristics: multi-dimensional classification is carried out according to the geological conditions of different projects (including rock layer distribution, lithology, groundwater conditions, etc.), the characteristics of grouting materials (including cement slurry ratio, viscosity, setting time, etc.), construction parameters (including grouting pressure, grouting speed, pipeline layout, etc.), and grouting effects (including permeability improvement rate, water resistance, construction stability, etc.). After classification, all data with similar characteristics will be aggregated into their respective independent modules.

[0058] For structured data (including geological parameters, numerical characteristics of grouting materials, etc.), it is stored through a relational database (SQL database) to ensure data integrity, standardization, and retrievability; while for unstructured data (including project logs, construction records, on-site images or videos, etc.), a non-relational database (MongoDB) is used for management to facilitate the storage of information in different formats.

[0059] S2: Extract features from the preprocessed geological conditions and construction conditions data, and perform dimensionality reduction on the extracted features to obtain low-dimensional feature data.

[0060] The specific process of obtaining low-dimensional feature data is as follows: extract relevant features from the preprocessed geological conditions and construction conditions data and perform normalization; calculate the correlation between each feature and the grouting effect index using the Pearson correlation coefficient, and screen the features related to the grouting effect index; apply improved linear discriminant analysis to perform dimensionality reduction on the selected features to obtain low-dimensional feature data. Specifically:

[0061] S201: Extract relevant features from the preprocessed geological conditions and construction conditions data. These features cover key factors affecting the grouting effect, such as grouting parameters, geological environment, grouting material characteristics, and construction conditions. The extracted features are divided into a training set and a test set according to a ratio of 70% and 30% to ensure the generalization ability of the model.

[0062] The features closely related to the grouting effect mainly include: geological condition features: lithology category, rock layer thickness, rock layer fracture rate, groundwater level, geological pressure, etc.; grouting material characteristics: slurry ratio, slurry viscosity, setting time, compressive strength, etc.; grouting parameters: grouting pressure, grouting speed, grouting diffusion radius, construction time, etc.; grouting effect indicators: improvement rate, water resistance index, construction stability, etc.

[0063] Statistical methods are used for feature extraction. Statistical analysis is performed on the known data distribution, the Pearson correlation coefficient of the features is calculated, and parameters with significant influence are screened. The specific steps are as follows:

[0064] First, normalize the data of different dimensions after preprocessing, scale the eigenvalue to the interval [0, 1], and eliminate the dimension difference. Then, use the Pearson correlation coefficient r to calculate the correlation between each feature and the grouting effect index, and screen the features that are significantly correlated with the grouting effect index. In this embodiment, when |r| > 0.5, the feature is considered to be significantly correlated.

[0065] The above Pearson correlation coefficient formula is specifically:

[0066]

[0067] Among them, x i is the eigenvalue of the i-th feature, y i is the grouting effect index, i represents the i-th feature, and n represents the total number of eigenvalues. represents the average value of the feature x, represents the average value of the grouting effect index y.

[0068] The above data normalization processing formula is specifically as follows:

[0069]

[0070] Among them, x is the original eigenvalue, x' is the normalized eigenvalue, and min(x) and max(x) are the minimum and maximum values of this feature respectively.

[0071] S202: Apply the improved linear discriminant analysis (LDA) technology for dimensionality reduction. The specific operation is as follows: Calculate the mean and covariance of the feature data of each category, construct the between-class and within-class scatter matrices, and obtain the optimal projection direction by solving the generalized eigenvalue problem. After projecting the original high-dimensional data onto this direction, low-dimensional feature data is generated and used as the main input for subsequent model training.

[0072] In this application, on the basis of the original LDA, a feature weight matrix W f is introduced. When constructing the between-class and within-class scatter matrices, combined with the importance score of the feature, weights are assigned to each feature. At the same time, a category contribution weight matrix W c is added to adjust the discrimination direction according to the importance of different geological environment categories. The specific formula is:

[0073]

[0074] Among them, ζ(w) represents the weighted objective function, which represents the optimization objective of w, and w T is the feature projection direction vector; W c is the category weight matrix, which is used to adjust the influence of different categories in the between-class scatter matrix, and W c = diag(α 1, α 2 ,..., α c ), α z represents the engineering importance weight of category z; S b is the between-class scatter matrix, which is used to measure the dispersion degree between the means of different categories; W f is the feature weight matrix, which is used to adjust the contributions of the between-class and within-class scatter matrices, W f = diag(w 1 , w 2 ,..., w d ), w i represents the i-th eigenvector; w is the corresponding eigenvector; S w is the within-class scatter matrix, which is used to measure the dispersion degree of data within the same category.

[0075] In this application, by introducing a feature weighting and category contribution evaluation mechanism and combining specific prior knowledge in engineering, the scatter matrix is adjusted to improve LDA. The optimized generalized eigenvalue problem enhances the category discrimination ability, making the model training more accurate and having stronger generalization ability.

[0076] S3: Conduct a preliminary importance evaluation on the low-dimensional feature data to obtain a feature importance score.

[0077] In this embodiment, the gradient boosting decision tree algorithm is used to conduct a preliminary importance evaluation on the extracted features, construct and train an XGBoost model to evaluate the importance of the features, and output a feature importance score.

[0078] 1) Use the partitioned training set to train the XGBoost model. During the training process, the model continuously adjusts the structure and parameters of the decision tree according to the training data to minimize the loss between the predicted value and the true value. The algorithm iterates step by step, and each iteration generates a new decision tree. The goal of the new tree is to fit the residual between the predicted result of the previous round of the model and the true value. Through multiple iterations, the prediction ability of the model is gradually improved.

[0079] 2) Set the hyperparameters of the XGBoost model. For example, learning rate: A smaller learning rate may lead to too long training time, while a larger learning rate may cause the model not to converge to the optimal value. To ensure the convergence speed of the model and prevent the model from skipping the optimal solution during training, set the learning rate to 0.1.

[0080] 3) Use the metrics of the trained XGBoost model to obtain feature importance scores. For example: The frequency metric indicates the number of times a feature is used as a split node in all decision trees. If a feature frequently appears in the splits of decision trees, it means it has a high influence in the model. Through the evaluation of these metrics, the model can quantify the importance of each feature to the final prediction result and output the feature importance scores, providing a preliminary basis for selecting key feature metrics for training in subsequent steps.

[0081] 4) On the basis of the initial feature importance evaluation by XGBoost, introduce a weighted feature importance evaluation mechanism, and combine engineering experience and target-oriented weight adjustment to make the feature importance scores more targeted and interpretable. The specific improvement steps are as follows:

[0082] First, when initially evaluating feature importance, introduce a set of engineering experience weights W e , and adjust the feature importance scores. The weights are derived from expert experience, historical project analysis, or the priority of specific metrics. The specific formula is:

[0083] I i ′ = I i ·W e,i

[0084] Among them, I i is the feature importance score of the i-th feature (the initial importance score obtained by the XGBoost model); W e,i is the engineering experience weight; I i ′ is the corrected feature importance score.

[0085] Then, introduce target correlation correction, and perform secondary correction on feature importance in combination with the optimization target. For example, if the target is to predict the grouting diffusion radius or the plugging effect, features highly correlated with diffusion, fluid properties, etc. need to increase their weights. The specific formula is:

[0086] I i ″ = I i ′·ρ i

[0087] Among them, I i ′ is the feature importance score after the first correction; I i ″ is the feature importance score after the second correction; ρ i is the correlation weight between the feature and the target variable, obtained by calculating the Pearson correlation coefficient corr(X i , Y): where x i is the feature value of the i-th feature, U is the target variable, and o is the total number of features.

[0088] Finally, after normalization, the final feature importance score is obtained, providing a basis for subsequent key feature selection. The specific formula is:

[0089]

[0090] where I final,i is the feature importance score after normalization, facilitating subsequent feature selection.

[0091] Based on the original XGBoost's evaluation of the initial feature importance, this application introduces engineering experience weights and target correlation weights for improvement, making the model more in line with the actual grouting application requirements. The corrected feature scores avoid misjudgments caused by uneven sample distribution or data noise, improving the stability of feature selection.

[0092] S4: Input the features with feature importance scores higher than the set threshold into a pre-trained global optimization model (large model) to obtain the grouting effect score.

[0093] S401: Construct a dataset: Divide the low-dimensional feature data into a training set, a validation set, and a test set. In this embodiment, the ratio of dataset division is set to 70:15:15 to ensure that the training set provides sufficient data samples, the validation set is used for model tuning, and the test set is used for final performance evaluation. The input of the global optimization model includes grouting parameter features with greater influence, that is, features with feature importance scores higher than the set threshold, such as grouting pressure, grouting speed, slurry type, etc., and the output is the grouting effect. In the training set and the validation set, screen out the data features closely related to the grouting effect, that is, the features with feature importance scores higher than the set threshold, and standardize them to improve the efficiency and stability of model training.

[0094] The above-mentioned grouting parameter features with greater influence are screened based on the feature importance score, and the final importance score I final,i of each feature is obtained using the improved XGBoost. Screen out the features with scores higher than the set threshold T as the input values of the global optimization model. Specifically:

[0095] I final,i > T

[0096] where T is the set screening threshold, and in this embodiment, T = 0.5 is taken.

[0097] S402: Construct a global optimization model. As Figure 2As shown in the figure, the global optimization model is constructed using a multi-layer perceptron (MLP), which includes an input layer, several hidden layers, a Dropout layer, and an output layer. A Dropout layer is added after each hidden layer of the multi-layer perceptron. In the specific network configuration, the ReLU activation function is used in the hidden layer to enhance the non-linear modeling ability of the model and prevent the problem of gradient vanishing.

[0098] Among them, the ReLU function is defined as:

[0099] R(x R ) = max(0, x R )

[0100] For negative inputs, the output is 0; for positive inputs, the output is equal to the input.

[0101] In view of the risk of overfitting that the MLP model may face when there are many input features, in this application, a neural network architecture with a Dropout layer is added after each hidden layer. By randomly discarding some neuron features, the dependence of the network during training is reduced, and the robustness of the model is enhanced, thus preventing overfitting. The Dropout formula is specifically as follows:

[0102]

[0103] Among them, is the output after passing through the Dropout layer, representing the output features after discarding; h g is the original output feature of the neuron; z g ∈{0, 1} is a Bernoulli random variable, indicating whether the neuron feature g is retained. This value is determined by the retention probability p. P(z g = 1) = p, that is, the output of this neuron feature is retained with probability p and discarded with probability 1 - p. In this embodiment, p = 0.5 is set, that is, in each training iteration, there is a 50% probability of discarding a neuron feature.

[0104] In this application, for the global optimization model MLP, a neural network architecture with a Dropout layer is added to the hidden layer to prevent simulation overfitting by randomly discarding some neuron features in each iteration, and at the same time improve the training efficiency; for the local fine-tuning model TinyCNN, the diversification of the convolution kernel size and residual connections are introduced to ensure that the convolutional layer can fully capture features of different scales and avoid the problem of gradient vanishing. Both dual models improve the performance on the basis of the original model framework, enhancing the efficiency and accuracy of model training.

[0105] S403: Train the global optimization model. Use the training set data to train the constructed global optimization model. Adopt Stochastic Gradient Descent (SGD) to optimize the network parameters to minimize the prediction error. During the training process, adopt the k-fold cross-validation method to evaluate the generalization ability of the model to ensure good performance of the model on unseen data. At the same time, select the Mean Squared Error (MSE) as the evaluation metric. The mean squared error is the average of the squared differences between the predicted values and the actual values, which can measure the average degree of the prediction error and ensure the reliability of the prediction results.

[0106] The calculation formula of the mean squared error is:

[0107]

[0108] where, G i is the actual value, is the predicted value, and D is the total number of data.

[0109] To further improve the prediction ability of the model, by analyzing the interaction relationships between different parameters, a coupling relationship model between parameters is established as follows:

[0110] First, quantify the importance of parameter interactions based on SHAP values, evaluate the importance of interaction features, and obtain the influence degree of the interaction between features on the model output.

[0111]

[0112] where, x i is the specific value of the feature i to be evaluated; S ∈ F / i is any subset of the feature set after removing feature i; |S| is the number of features included in the feature subset S; |F| is the number of all features; v(S ∪ {i}) is the predicted value of the model when using the feature subset S and feature i; v(S) is the predicted value of the model when only using the feature subset S.

[0113] Then, based on the analysis results, select the coupling relationship model as follows:

[0114]

[0115] where, Y is the grouting effect index; β 0 is the bias constant; X i is the i-th parameter; X i X j is the second-order interaction feature; β ij is the influence of the interaction between variables X i and X j on Y; ε is the random error. β i represents the regression coefficient of the i-th parameter, m represents the total number of parameters, X jIt is represented as the j-th parameter.

[0116] Finally, an output coupling relationship model is presented to reveal the interaction effects among different parameters, identify the parameters that contribute the most or have the strongest influence on the grouting effect, so as to provide a reference for the optimization direction of local fine-tuning of the model and improve the model prediction ability and the optimization effect of grouting parameters. The coupling model refers to the coupling between parameters. For example, the grouting pressure and grouting rate parameters will affect each other and jointly act on the grouting effect. Based on the results output by the coupling relationship model, identify which parameters contribute the most or have the strongest influence on the grouting effect, so as to guide the local fine-tuning of the model. That is, in actual engineering, the real-time monitoring data can be input into the coupling relationship model, and the parameters with strong interaction and the optimized values can be output through model update to guide the dynamic adjustment of construction parameters.

[0117] The parameter results output by the global optimization model are used as inputs, and the Bayesian optimization technique is adopted to optimize and adjust the hyperparameters of the global optimization model to maximize the grouting effect, so as to achieve higher grouting stability and economy. The specific steps are as follows:

[0118] 1) Define the hyperparameters to be optimized and their value ranges. For example, the learning rate range is set to [0.01, 0.1] to control the learning step size of the model; the number of network layers ranges from [2, 4] to control the depth of the neural network, etc. And the Gaussian process (GP) is used as the surrogate model, and the objective function is defined as:

[0119] f(θ) = -MSE val (θ)

[0120] where θ represents different combinations of hyperparameters, and MSE val represents the mean squared error evaluated on the validation set.

[0121] 2) Define an acquisition function to select combinations from the hyperparameter space and use the surrogate model to evaluate these combinations. The acquisition function is defined as follows:

[0122] EI(A) = E[max(f(A) - f best )]

[0123] where f best is the current best performance. f(A) represents the performance value predicted by the surrogate model at point A and is a random variable; EI(A) represents the expected improvement value, which represents the expected potential improvement value at point A in the hyperparameter space. E represents the expected value, and the improvement value at point A is weighted and averaged under its probability distribution.

[0124] 3) Update the surrogate model GP by incorporating the new evaluation results to improve the accuracy of the surrogate model.

[0125] 4) Repeat steps 2 and 3 until the preset number of iterations is reached. Output the current optimal hyperparameter combination and apply it to the final training of the model to ensure the best prediction ability.

[0126] After completing the model training and hyperparameter tuning, input the validation set data into the global optimization model for prediction to evaluate its accurate prediction of the grouting effect. Specifically: Generate a scatter plot of the predicted values and the actual values, observe the overall distribution of the data and the fitting effect of the model, and check whether the scatter points are closely distributed near the ideal 45-degree diagonal line. If it is satisfied, the prediction effect meets the standard, fix the model parameters, and save the global optimization model; otherwise, there is a prediction deviation and the training should be iterated again.

[0127] S5: Input the grouting effect score and the feature data corresponding to the grouting effect into a pre-trained local fine-tuning model (small model) to obtain the fine-tuning value of the grouting parameters.

[0128] The grouting effect corresponds one-to-one with the feature data. Different feature data inputs result in different grouting effects. Another way of expressing the feature data corresponding to the grouting effect is: a certain grouting effect caused by a certain part of the feature values, that is, the feature data is a subset of the features with a feature importance score higher than the set threshold, and it is a part of the feature data, because not all feature data needs to be fine-tuned.

[0129] The input of the local fine-tuning model has two parts. One is the grouting effect score output by the global optimization model, and the other is the feature data group corresponding to the grouting effect score, that is, the data group corresponding to the data input into the global optimization model to obtain the corresponding score.

[0130] S501: Construct a data set: Divide the input features into a training set and a validation set.

[0131] S502: Construct a local fine-tuning model: The local fine-tuning model is constructed using a lightweight convolutional neural network (TinyCNN), including 2 - 3 convolutional layers and 1 - 2 fully connected layers. In the convolutional layer, small-sized convolutional kernels are used to capture the local correlations of the input features. The pooling layer uses max pooling operations to further extract features and reduce the dimension of the feature map. Finally, the prediction result is output through the fully connected layer.

[0132] To ensure that the convolutional layer can fully capture features of different scales and avoid the influence of the gradient vanishing problem, this application diversifies the convolutional kernel sizes and introduces residual connections, specifically as follows:

[0133] The first convolutional layer: Use a 3×3 convolutional kernel, a stride of 1, and an output channel number of 32;

[0134] The second convolutional layer: Use a 5×5 convolutional kernel, a stride of 1, and an output channel number of 64;

[0135] Third - layer Convolution: Use a 3×3 convolution kernel, with a stride of 1 and 128 output channels.

[0136] The specific formula for the residual connection is:

[0137]

[0138] Among them, J is the output signal of the convolutional layer; W i is the convolutional kernel parameter; is the output of the residual module, representing the result after a series of convolutional operations; K is the output signal. In this way, the gradient can be directly back - propagated through the skip connection, thus solving the problem of gradient vanishing.

[0139] S503: Train the local fine - tuning model: Use the Adam optimization algorithm, combined with mini - batch gradient descent for training. During the training process, gradually reduce the loss value and improve the prediction accuracy. Adopt k - fold cross - validation to evaluate the performance stability of the local fine - tuning model. Select the mean squared error (MSE) as the loss function to quantify the error of the model under different grouting environments for evaluation. The specific content is as shown in S403 above.

[0140] Adopt Bayesian optimization technology to optimize the hyperparameters of the local fine - tuning model. The specific operation is as shown in S4.

[0141] In the actual grouting project, according to the current environmental monitoring data and the output of the global optimization model, that is, the grouting effect score and the corresponding characteristic data group, input them into the local fine - tuning model, and the local fine - tuning model outputs the fine - tuning values of parameters such as grouting pressure and grouting time.

[0142] After receiving the grouting effect score from the global optimization model and the characteristic data corresponding to this effect (such as geological characteristics, construction conditions, etc.), according to the score or effect data, the local fine - tuning model identifies the potential deficiencies of the current grouting plan and sets the optimization goal to maximize the grouting effect score. Among them, the objective function is expressed as:

[0143]

[0144] Among them, N is the number of training samples, S local is the local score obtained after model optimization, and S global is the global grouting effect score.

[0145] Then, the local fine - tuning model adjusts the input features, such as optimizing the grouting pressure, grouting time, etc. Specifically:

[0146] The model maps the input data X input to the fine - tuned grouting parameters, expressed as:

[0147] ΔP inj , ΔT inj = f(X input )

[0148] Among them, ΔP inj is the fine-tuning value of the grouting pressure, and ΔT inj is the fine-tuning value of the grouting time. The hidden layer adopts a multi-layer fully connected network, which is expressed as:

[0149] h (l) = σ(W (l) ·h (l-1) + b (l) )

[0150] Among them, h (l) is the output of the l-th layer of hidden units, W (l) , b (l) are the weight matrix and the bias vector respectively, and σ() is the activation function.

[0151] Finally, the fine-tuning values of each grouting parameter are output, which are expressed as:

[0152] ΔP inj = W out,1 ·h (L) + b out,1 , ΔT inj = W out,2 ·h (L) + b out,2

[0153] Among them, ΔP inj and ΔT inj respectively represent the fine-tuning values of the grouting pressure and the grouting time, W out,i represents the output layer weight matrix, h (L) represents the final output of the hidden layer, and b out,i is the bias of the output layer.

[0154] These fine-tuning parameters are further adjusted on the basis of the global optimization model to cope with different geological and construction conditions. In this way, it is ensured that the grouting effect reaches the optimum under the coupling action of the parameters, realizing real-time dynamic fine-tuning of the parameters to meet the changing requirements of the grouting environment.

[0155] Real-time Application and Dynamic Adjustment of Grouting Parameter Selection Optimization Method: Obtain data on current geological conditions and construction conditions, and after preprocessing, feature extraction, feature dimensionality reduction, and feature importance scoring, input it into the global optimization model to output a grouting effect score; input the grouting effect score and the feature data corresponding to the grouting effect into the local fine-tuning model to adjust the grouting parameters. The global optimization model has been trained with a large amount of historical data in the early stage, covering grouting cases under various different geological and construction scenarios. It can provide a relatively macroscopic optimal parameter combination for the current situation as a basis. On this basis, the local fine-tuning model plays its role of local adjustment and further optimizes the parameters output by the global optimization model. Through the collaborative work of the global optimization model and the local fine-tuning model, the deep learning model can generate the best grouting parameters applicable to the current conditions in real time.

[0156] When the geological conditions change during the construction process (such as encountering changes in different formation hardness, groundwater level changes, etc.) or other factors affect the grouting effect, the present invention can timely obtain new data, re-analyze and calculate, and quickly adjust the grouting parameters to ensure the stability and reliability of the grouting effect. For example, if it is found during the construction process that the actual grouting effect deviates from the expectation, the subsequent grouting parameters can be recalculated and adjusted based on the real-time monitoring data to adapt to the changing situation and ensure the smooth progress of the entire grouting project and achieve the expected effect.

[0157] Embodiment 2

[0158] This embodiment discloses a grouting parameter selection optimization system based on deep learning, including:

[0159] A data acquisition module, which is configured to: acquire data on current geological conditions and construction conditions and perform preprocessing;

[0160] A feature screening module, which is configured to: extract features from the preprocessed geological conditions and construction conditions data, perform dimensionality reduction on the extracted features to obtain low-dimensional feature data; perform a preliminary importance assessment on the low-dimensional feature data to obtain a feature importance score;

[0161] A global optimization module, which is configured to: input the features with a feature importance score higher than a set threshold into a pre-trained global optimization model to obtain a grouting effect score;

[0162] A local adjustment module, which is configured to: input the grouting effect score and the features with a feature importance score higher than a set threshold into a pre-trained local fine-tuning model to obtain a fine-tuning value of the grouting parameters.

[0163] Embodiment 3

[0164] The objective of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method in Embodiment 1 are implemented.

[0165] Embodiment 4

[0166] The objective of this embodiment is to provide a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the method in Embodiment 1 are executed.

[0167] The steps involved in the devices in the above Embodiments 3 and 4 correspond to those in Method Embodiment 1. For specific implementation manners, reference may be made to the relevant description part of Embodiment 1. The term "computer-readable storage medium" should be understood to include a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0168] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0169] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0170] Although the specific implementation manners of the present invention are described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that based on the technical solutions of the present invention, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. A grouting parameter selection optimization method based on deep learning, characterized in that: include: Obtain current geological conditions and construction conditions data and perform preprocessing; Performing feature extraction on the preprocessed geological condition and construction condition data, and performing dimensionality reduction on the extracted features to obtain low-dimensional feature data; performing a preliminary importance assessment on the low-dimensional feature data to obtain a feature importance score; The features whose feature importance scores are higher than the set threshold are input into the pre-trained global optimization model to obtain the grouting effect score; The grouting effect score and the characteristic data corresponding to the grouting effect are input into a pre-trained local fine-tuning model to obtain the fine-tuning value of the grouting parameter.

2. A grouting parameter selection optimization method based on deep learning as claimed in claim 1, characterized in that: The construction condition data include grouting material characteristics, grouting parameters and grouting effects.

3. A grouting parameter selection optimization method based on deep learning as claimed in claim 1, characterized in that: The low-dimensional feature data is obtained as follows: Extract relevant features from the preprocessed geological conditions and construction conditions data and perform normalization processing; The Pearson correlation coefficient was used to calculate the correlation between each feature and the grouting effect index, and the features related to the grouting effect index were screened; The improved linear discriminant analysis is applied to reduce the dimension of the screened features to obtain low-dimensional feature data.

4. A grouting parameter selection optimization method based on deep learning as claimed in claim 3, characterized in that: The improved linear discriminant analysis introduces a feature weight matrix and a category contribution weight matrix based on the linear discriminant analysis, which is specifically expressed as follows: Among them, ζ(w) represents the weighted objective function, w T is the feature projection direction vector, W c is the category weight matrix, S b is the inter-class scatter matrix, W f is the feature weight matrix, w is the corresponding feature vector, S w is the within-class scatter matrix.

5. A grouting parameter selection optimization method based on deep learning as claimed in claim 1, characterized in that: The feature importance score is obtained as follows: Using a gradient boosting decision tree algorithm to perform a preliminary importance assessment on the low-dimensional feature data, constructing and training an XGBoost model to assess the importance of features, and outputting feature importance scores; A weighted feature importance evaluation mechanism is introduced, combining engineering experience and goal-oriented weight adjustment to make feature importance scores more targeted and interpretable; specifically: First, a set of engineering experience weights are introduced to adjust the feature importance score, which is expressed as: I i ′=I i ·W e,i Among them, I i Score the feature importance of the i-th feature; W e,i is the engineering experience weight; I i ′ is the corrected feature importance score; Then, the target relevance correction is introduced, and the feature importance is corrected twice in combination with the optimization target to obtain the feature importance score after the second correction, and the final feature importance score is obtained after normalization.

6. A grouting parameter selection optimization method based on deep learning as claimed in claim 1, characterized in that: The global optimization model is constructed by using a multi-layer perceptron, including an input layer, a plurality of hidden layers, a Dropout layer and an output layer, and a Dropout layer is added after each hidden layer of the multi-layer perceptron.

7. A grouting parameter selection optimization method based on deep learning as claimed in claim 1, characterized in that: The local fine-tuning model is constructed using a lightweight convolutional neural network, receives the grouting effect score and the feature data corresponding to the grouting effect, sets the optimization goal, and maximizes the grouting effect score. The objective function is expressed as: Where N is the number of training samples, S local is the local score obtained after model optimization, S global Give a global score for the grouting effect; Then the local fine-tuning model adjusts the input features and passes through the hidden layer of the multi-layer fully connected network, and finally outputs the fine-tuning values ​​of each grouting parameter, which is expressed as: ΔP inj =W out,1 ·h (L) +b out,1 ,ΔT inj =W out,2 ·h (L) +b out,2 Where ΔP inj With ΔT inj Respectively represent the fine-tuning values ​​of grouting pressure and grouting time, W out,i represents the output layer weight matrix, h (L) represents the final output of the hidden layer, b out,i is the bias of the output layer.

8. A grouting parameter selection and optimization system based on deep learning, characterized in that: include: A data acquisition module is configured to: acquire current geological conditions and construction conditions data and perform preprocessing; The feature screening module is configured to: extract features from the preprocessed geological condition and construction condition data, and reduce the dimension of the extracted features to obtain low-dimensional feature data; perform a preliminary importance assessment on the low-dimensional feature data to obtain a feature importance score; A global optimization module is configured to: input features with feature importance scores higher than a set threshold into a pre-trained global optimization model to obtain a grouting effect score; The local adjustment module is configured to: input the grouting effect score and the characteristic data corresponding to the grouting effect into a pre-trained local fine-tuning model to obtain a fine-tuning value of the grouting parameter.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a grouting parameter selection and optimization method based on deep learning as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the grouting parameter selection and optimization method based on deep learning as described in any one of claims 1-7 are implemented.

Images