Method and system for predicting effect of additive on grouting material under influence of mineralization degree

Abstract

The invention provides a method and system for predicting the effect of an additive on a grouting material under the influence of mineralization degree, and belongs to the technical field of grouting material performance prediction.The method comprises the steps that data are obtained, and the data comprise mineralization degree parameters, the type and corresponding characteristics of the additive and key performance indexes of the grouting material; the obtained data is preprocessed; screening core characteristic variables from the preprocessed data, and analyzing the indirect influence of the mineralization degree on the reactivity of the additive through a partial correlation analysis method; constructing two single-level sub-network models based on the core feature variables; to-be-predicted mineralization degree parameters, admixture types and grouting material proportion information are input into the two single-level sub-network models to generate prediction output, and the influences, including strength and fluidity key performance index data, of the admixtures on the performance of the cement-based grouting material under different mineralization degree conditions are obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction of the performance of grouting materials, and particularly relates to a method and system for predicting the effect of admixtures on grouting materials under the influence of salinity. Background Technique

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

[0003] In the construction of tunnels and underground projects, grouting technology is widely used in aspects such as waterproofing, anti-seepage, reinforcement, and formation stability. Especially in high-salinity environments with complex and variable geology, the performance of grouting materials has an important impact on the construction effect and safety. However, due to the change of salinity parameters in the on-site environment, the effect of admixtures on the performance of grouting materials may be significantly affected, resulting in an unsatisfactory or unpredictable construction effect. Therefore, how to scientifically and accurately predict the influence of admixtures on the performance of cement-based grouting materials under different salinity conditions, and then optimize the selection and dosage of admixtures, has become a key technical challenge in the field of grouting construction.

[0004] In the prior art, the performance prediction of grouting materials and the selection of admixtures are usually carried out through a large number of on-site tests or simple models based on experience. However, this method has high costs, a long cycle, and low prediction accuracy, and it is difficult to fully capture the complex relationship between salinity, admixture characteristics, and grouting material performance. In addition, due to the multiple effects of salinity changes on admixture activity and grouting material performance, a single test or empirical method is difficult to meet diverse construction needs. The existing feature selection and model optimization methods perform poorly when dealing with complex systems with multiple objectives and parameters, and they cannot update the model in real time or dynamically feedback the application effect, resulting in poor adaptability under complex salinity conditions and prone to construction errors or cost waste. Summary of the Invention

[0005] To overcome the deficiencies of the above prior art, the present invention provides a method for predicting the effect of admixtures on grouting materials under the influence of salinity. This method can capture the complex dependence relationship between variables, significantly improve the accuracy of predicting the material performance in a salinity environment, and effectively avoid the problem that traditional single models ignore the indirect relationship between variables.

[0006] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0007] In the first aspect, a method for predicting the effect of admixtures on grouting materials under the influence of salinity is disclosed, including:

[0008] Obtain data, including salinity parameters, types of admixtures and corresponding characteristics, and key performance indicators of grouting materials;

[0009] Preprocess the acquired data;

[0010] Screen the core feature variables from the preprocessed data, and analyze the indirect influence of salinity on the reactivity of additives through partial correlation analysis;

[0011] Construct two single - level sub - network models based on the core feature variables, including: salinity - additive sub - network model and additive - grouting material property sub - network model;

[0012] Among them, the salinity - additive sub - network model is used to learn the key characteristics of the influence of salinity changes on the reactivity and activity of additives, so as to obtain the law of the influence of salinity on additives under different conditions; the additive - grouting material property sub - network model takes the physical and chemical characteristics of additives as input, learns the direct influence of additive characteristics on the properties of grouting materials, and establishes a model of the effect of additives on grouting materials;

[0013] Input the salinity parameter to be predicted, additive type, and grouting material ratio information into the two single - level sub - network models to generate prediction outputs, and obtain the influence of additives on the properties of cement - based grouting materials under different salinity conditions, including key performance index data such as strength and fluidity.

[0014] As a further technical solution, screen the core feature variables from the preprocessed data, specifically including:

[0015] Adopt the recursive feature elimination method, process the preprocessed data by repeatedly constructing models, and remove the least important features in each iteration until the preset number of features or other stopping conditions are reached.

[0016] As a further technical solution, in the constructed model, select the support vector machine as the basic model, set the initial feature set as all features, denoted as F = {1, 2, …, n}, where n is the number of original features. At the same time, when setting the number of final features to be retained, feature selection stops, use the feature set F to train the model, and evaluate the feature importance using the coefficient vector ω output by the model.

[0017] As a further technical solution, for the analysis of the indirect influence of salinity on the reactivity of additives through partial correlation analysis, the specific operation is: assume that the additive characteristics are control variables, namely Z 1 , Z 2 , …, Z k , the salinity parameter is the variable X, and the grouting material property is Y, and perform multiple regression;

[0018] Taking X as the dependent variable, Z 1 , Z 2 , …, Zk Perform regression with the independent variable to obtain the regression equation

[0019] X = β 0 + β 1 Z 1 + β 2 Z 2 + … β k Z k + τ 1 , Subtract the predicted value from the actual value to calculate the residual e X

[0020] Similarly, with Y as the dependent variable and Z 1 , Z 2 , …, Z k as the independent variables for regression to obtain the regression equation Y = α 0 + α 1 Z 1 + α 2 Z 2 + … α k Z k + τ 2 , Subtract the predicted value from the actual value to calculate the residual e Y

[0021] Finally, calculate the residual e X and e Y The Pearson correlation coefficient of is the partial correlation coefficient of X and Y under the influence of the control variables Z 1 , Z 2 , …, Z k . The specific formula is:

[0022]

[0023] where e Xi and e Yi are the i-th values of the residual sequences of X and Y respectively, and are the means of the residual sequences of X and Y respectively;

[0024] Finally, calculate the partial correlation coefficient r between the salinity parameter X and the grouting material property Y XY·Z .

[0025] As a further technical solution, the salinity - admixture sub - network model is constructed using a deep neural network, and this network model consists of an input layer, a hidden layer, and an output layer;

[0026] Among them, the input layer contains several nodes, and each node represents a salinity parameter; Set m hidden layers according to requirements, and each hidden layer uses the ReLU activation function to capture the non - linear relationship between the salinity parameter and the admixture characteristics;

[0027] The output layer is the predicted value of the admixture characteristics, and the number of nodes in this layer is the same as the number of admixture characteristic indexes.

[0028] As a further technical solution, the admixture-grouting material performance sub-network model is constructed by using a deep neural network, and this network is composed of an input layer, a hidden layer and an output layer;

[0029] Among them, the input layer is the output layer of the salinity-admixture sub-network model, and the number of nodes remains the same; n hidden layers are set according to requirements, and each hidden layer uses the ReLU activation function to capture the complex relationship between the admixture characteristics and the influence on the grouting material performance;

[0030] The output layer is the performance index of the grouting material, and linear activation is used to generate the predicted value of the grouting material performance.

[0031] Second, a prediction system for the effect of admixtures on grouting materials under the influence of salinity is disclosed, including:

[0032] A data collection and preprocessing module, configured to: obtain data, including salinity parameters, admixture types and corresponding characteristics, and key performance indexes of grouting materials; preprocess the obtained data;

[0033] A feature selection and partial correlation analysis module, configured to: screen core feature variables from the preprocessed data, and analyze the indirect influence of salinity on the reactivity of admixtures through partial correlation analysis;

[0034] A multi-level modeling and optimization module is configured to: construct two single-level sub-network models based on the core feature variables, including: a salinity-admixture sub-network model and an admixture-grouting material performance sub-network model;

[0035] Among them, the salinity-admixture sub-network model is used to learn the key characteristics of the influence of salinity changes on the reactivity and activity of admixtures, so as to obtain the law of the influence of salinity on admixtures under different conditions; the admixture-grouting material performance sub-network model takes the physical and chemical characteristics of the admixture as input, learns the direct influence of the admixture characteristics on the grouting material performance, and establishes a model of the influence of the admixture on the grouting material;

[0036] A performance prediction and application feedback module, configured to: input the salinity parameters, admixture types, and grouting material ratio information to be predicted into the two single-level sub-network models to generate predicted outputs, and obtain the influence of admixtures on the performance of cement-based grouting materials under different salinity conditions, including key performance index data of strength and fluidity.

[0037] The above one or more technical solutions have the following beneficial effects:

[0038] In the technical solution of the present invention, two single - level sub - network models are constructed based on core feature variables, including: the salinity - admixture sub - network model and the admixture - grouting material property sub - network model. Through a multi - level model structure, the influence of salinity on admixtures and the influence of admixtures on the properties of grouting materials are respectively modeled to achieve progressive and refined analysis. This method can capture the complex dependence relationships between variables, significantly improve the accuracy of predicting material properties in a salinity environment, and effectively avoid the problem that traditional single models ignore the indirect relationships between variables.

[0039] In the feature selection of the technical solution of the present invention, partial correlation analysis is introduced to accurately identify the way in which salinity indirectly affects the properties of grouting materials through admixtures, ensuring that the selection of feature variables conforms to the actual physical and chemical relationships. This method not only reduces the interference of redundant data but also enhances the interpretability of the model, making the prediction results more reliable and valuable for reference.

[0040] In the loss function of the technical solution of the present invention, a weight assignment strategy is introduced to optimize various properties of materials for different application requirements. Multi - objective optimization can not only achieve a balance in performance requirements but also enhance the practicality of the model, enabling it to provide more accurate material property predictions and parameter suggestions that meet engineering requirements in specific construction projects.

[0041] In the feature selection of the technical solution of the present invention, the idea of causal analysis is combined. By mining the key influencing factors between salinity and admixtures, the adaptability of the model in practical applications is effectively improved. This method is different from traditional correlation analysis and can more accurately identify the core factors affecting performance, providing more valuable input data for subsequent modeling and significantly enhancing the interpretability and scientific nature of the model.

[0042] Advantages of additional aspects of the present invention will be partially given in the following description, partially become apparent from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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 of the present invention.

[0044] Figure 1 It is a flow chart of a method for predicting the effect of admixtures on cement - based grouting materials under the influence of salinity parameters in an embodiment of the present invention;

[0045] Figure 2 It is a structural diagram of a system for predicting the effect of admixtures on cement - based grouting materials under the influence of salinity parameters in an embodiment of the present invention;

[0046] Figure 3This is the double-layer progressive model framework diagram in the embodiments of the present invention Detailed implementation manners

[0047] It should be noted that the following detailed description is exemplary and is intended to provide further illustration 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

[0048] It should be noted that the terms used herein are only for describing the specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention

[0049] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other

[0050] Embodiment 1

[0051] See the appendix Figure 1 As shown, this embodiment discloses a method for predicting the effect of admixtures on grouting materials under the influence of salinity, including:

[0052] Step (1) Data collection:

[0053] First, clarify the collection scope and data requirements. According to the requirements of grouting material performance prediction, determine the main types and index ranges of the collected data, including salinity parameters, types of admixtures and corresponding characteristics, and key performance indicators (compressive strength, fluidity, etc.) of grouting materials. The ranges and specific collection standards of these data types need to be clearly specified before collection to ensure the systematicness and consistency of the data

[0054] The above-mentioned collected data collects various historical data information through construction logs, geological reports, and reference documents, etc

[0055] In this step, for the above three types of data obtained, among them, the salinity parameter is numerical data and is stored in the database table in the form of specific numerical values; the type of admixture is categorical (such as water reducer, early strength agent, retarder, etc.), and the characteristics of the admixture are numerical (such as percentage or absolute value) and are stored in the multi-dimensional table database; the performance indicators of the grouting material are numerical data, representing the specific values of the performance indicators, and are stored in the database table in the form of specific numerical values

[0056] Step (2) Data cleaning and preprocessing

[0057] First, clean and check the integrity of the collected data. Check each item of data to ensure that there are no missing or incomplete records. For a small amount of missing data, use linear interpolation to fill it to ensure the continuity of the data; for the irreparable missing data, eliminate it according to the actual situation to ensure the quality and consistency of the data

[0058] Due to the wide range of data sources, the dimensions of various data may be different. To ensure the training effect and stability of the model, the data is normalized to eliminate the dimensional differences. The specific operation is as follows: The Min-Max normalization preprocessing method is adopted. For each eigenvalue, the maximum and minimum values in the dataset need to be determined and substituted into the following formula. In this way, the numerical ranges of all features are unified into the interval [0, 1], which is convenient for subsequent model training and analysis.

[0059] During normalization, the above salinity parameters, admixture characteristics, and grouting material performance index features are normalized from specific values to facilitate later input into the model.

[0060] Finally, all the cleaned and preprocessed data is stored in a unified format to ensure that the field names, data formats, and units are consistent, which is convenient for subsequent rapid data reference.

[0061] The above cleaning and integrity checking methods are applicable to all three types of data in step one. Among them, the salinity parameter is continuous numerical data, and linear interpolation is usually suitable for filling missing values; the types of admixtures are categorical data, and outliers are generally directly removed; interpolation is generally used for admixture characteristics; the grouting material performance data has high requirements for integrity, and linear interpolation is generally used to fill the data. If there are too many missing values, they are directly removed.

[0062] Step (3) Feature selection and partial correlation analysis:

[0063] This step is used to extract the initial features that potentially affect the salinity, admixture characteristics, and grouting material performance from the preprocessed data, that is, to preliminarily screen out the features related to the target variable; the target variable is the grouting material performance, and in this example, the grouting material performance refers to: setting time, compressive strength, fluidity, and dynamic water retention rate;

[0064] The initial features are the feature values with higher relevance extracted from the above preprocessed features.

[0065] Then, the partial correlation analysis method is applied to calculate the independent influence of the salinity parameter on the admixture effect, and the influence intensity of the admixture on the grouting material performance under different salinity conditions, and to identify the key interaction relationships.

[0066] Then, using the feature selection algorithm (the selection method based on importance scores), the feature set is further optimized to remove redundant and low-correlated features, ensuring that only the features that contribute the most to the model prediction effect are retained, so as to reduce the calculation amount and improve the accuracy and generalization ability of the model.

[0067] In this step of the present embodiment, first, feature selection is performed. The collected data contains multiple feature variables after preprocessing. To improve the efficiency and accuracy of the model, it is necessary to screen out the core variables that have a greater impact on the performance of the grouting material. The present disclosure uses the recursive feature elimination method (RFE). By repeatedly constructing the model, the least important feature is removed in each iteration until the preset number of features or other stopping conditions are reached.

[0068] The specific operation is as follows: Select the support vector machine (SVM) as the base model, set the initial feature set as all features, denoted as feature set F = {x 1 , x 2 , …, x n}, where n is the number of original features. At the same time, when setting the final number of features to be retained as m (m < n), the feature selection stops. Use the feature set F to train the model, and evaluate the feature importance using the coefficient vector ω output by the model. The larger the absolute value of the coefficient corresponding to the feature, the more important the feature usually is.

[0069] Remove the least important feature from the feature set F. Here, assume it is j, that is, F = F - {j}. Use the updated feature set F to retrain the support vector machine model and evaluate the removal of j again. Iterate in a loop until the final number of features is m and then stop. At this time, the remaining feature set is the final feature subset after recursive feature elimination F = {x 1 , x 2 , …, x m}.

[0070] Furthermore, the coefficient vector ω is calculated as follows:

[0071] First, for the SVM objective function The constraint conditions are y i (w T x i + t) ≥ 1 - ξ i and ξ i ≥ 0 (i = 1, 2, … m). Then, construct the Lagrangian function

[0072]

[0073] where ω is the coefficient vector, t is the bias term; ξ i is the slack variable; α and r are the Lagrange multipliers; x i is the input feature vector, that is, the preprocessed feature in the feature set F, y i is the target variable; C is the penalty parameter, obtained from experience; m is the number of samples in the data set; b is the bias amount, used to adjust the position of the hyperplane; α i and r i are the Lagrange multipliers corresponding to the i-th feature vector; ωT Denotes the transpose of the coefficient vector ω.

[0074] Then, take the partial derivatives of the Lagrangian function L with respect to w, t, and ξ respectively i Take the partial derivative with respect to w: From this, we can obtain Take the partial derivative with respect to t: For ξ i Take the partial derivative:

[0075] Next, through the condition that the above partial derivatives are 0, transform the original problem into a dual problem, that is, a maximization problem about the Lagrange multiplier α i : The constraint condition is 0 ≤ α i ≤ C and where α i and α j are the Lagrange multipliers corresponding to the i-th and j-th parameter samples respectively, x j and y j denote the input feature vector and the target variable corresponding to the j-th parameter sample respectively.

[0076] Select the KKT conditions: where

[0077] According to the constraint conditions and 0 ≤ α i , α j ≤ C, calculate the upper and lower bounds P and H of α j . If y i ≠ y j , then If y i = y j , then where and denote the values of the Lagrange multiplier α i and α j in the previous iteration respectively.

[0078] Establish E i = g(x i ) - y i , E j = g(x j ) - y j , and then calculate where If exceeds the upper and lower bounds P and H, then set it equal to the value of P or H. Then calculate where, E i and E jrespectively represent the prediction errors of the sample feature x i and x j ; and respectively represent the new values of α i and α j after one iteration update.

[0079] Update the threshold b according to the update situations of α i and α j . If then If then Get When the change amount of α , the result converges. Where α i (k) is the Lagrange multiplier vector at the k-th iteration, and k is the number of iterations.

[0080] Finally, calculate the coefficient vector w according to , and then evaluate the feature importance.

[0081] Based on the above feature subsets, apply partial correlation analysis to deeply explore the progressive influence relationship between variables. There is an indirect influence relationship between the salinity parameter and the characteristics of the admixture, that is, the salinity first affects the reaction characteristics of the admixture, and the latter further affects the final performance of the grouting material. Through partial correlation analysis, analyze the indirect influence of salinity on the reactivity of the admixture. The specific operation is: assume that the characteristics of the admixture are control variables, that is, Z 1 , Z 2 , …, Z k , the salinity parameter is variable X, and the performance of the grouting material is Y, and perform multiple regression.

[0082] The variables X, Y, Z are all part of the feature subset, that is, F = {X, Y, Z}, and the formula is to select the variables in the feature subset to perform correlation calculation.

[0083] Taking X as the dependent variable and Z 1 , Z 2 , …, Z k as independent variables for regression, obtain the regression equation, X = β 0 + β 1 Z 1 + β 2 Z 2 + … β k Z k + τ 1 , subtract the predicted value from the actual value, and calculate the residual e X . Similarly, taking Y as the dependent variable and Z 1 , Z 2 , …, Zk Perform regression with the independent variable to obtain the regression equation, Y = α 0 + α 1 Z 1 + α 2 Z 2 + … α k Z k + τ 2 , Subtract the predicted value from the actual value to calculate the residual e Y .

[0084] Finally, calculate the Pearson correlation coefficient of the residuals e X and e Y . This coefficient is the partial correlation coefficient of X and Y under the influence of the control variables Z 1 , Z 2 , …, Z k . The specific formula is:

[0085]

[0086] where, and are respectively the i-th values of the residual sequences of the salinity parameter X and the material property Y, and are respectively the means of the residual sequences of the salinity parameter X and the material property Y.

[0087] When r XY·Z ≠ 0, it indicates that the salinity has an indirect effect on the properties of the grouting material through the admixture. At this time, if r XY·Z > 0, it shows that there is a positive indirect influence relationship between the salinity parameter (after affecting the properties of the admixture) and the properties of the grouting material. If r XY·Z < 0, it means there is a negative indirect influence relationship. Further, the larger the absolute value of r XY·Z , the more crucial the corresponding salinity parameter plays in the indirect influence process, and these parameters can be regarded as core features. In this application, when |r XY·Z | > 0.8, it is considered to have a greater influence, and the corresponding salinity parameter X is the core feature variable.

[0088] Step (4) Multilevel modeling:

[0089] Construct a basic model based on the selected features. Select a deep neural network (DNN) and perform single-level modeling for the influence of salinity parameters on the effect of additives and the effect of additives on the properties of grouting materials respectively, and construct a sub-network of salinity-additive and a sub-network of additive-grouting material properties; then, establish a multi-level model structure, take the influence of salinity on additives as the input, and further predict the influence of additives on the properties of cement-based grouting materials under different salinity conditions to achieve progressive modeling.

[0090] First, the core variables screened by the feature selection and partial correlation analysis module are the core features, which are the input values of the model. Use a deep neural network (DNN) for preliminary modeling. Combining Figure 3 , in this basic model, respectively construct two single-level sub-network models for the influence of salinity parameters on the effect of additives and the effect of additives on the properties of grouting materials: a salinity-additive sub-network and an additive-grouting material property sub-network.

[0091] Among them, the salinity-additive sub-network model learns how salinity changes affect key characteristics such as the reactivity and activity of additives to obtain the law of the influence of salinity on additives under different conditions; the additive-grouting material property sub-network model takes the output parameters of the salinity-additive sub-network model as the input. The output of the salinity-additive sub-network model is the key physicochemical characteristic parameters of additives under different salinity conditions, such as activity, solubility, ion exchange capacity, etc. It focuses on learning the direct influence of additive characteristics on the properties of grouting materials (such as compressive strength, fluidity, etc.), so as to establish a model of the effect of additives on grouting materials.

[0092] The specific construction details are as follows: The salinity-additive sub-network model is constructed using a deep neural network (DNN). This network model consists of an input layer, hidden layers, and an output layer. The input of the salinity-additive sub-network is: salinity-related parameters, that is, the important salinity factors that affect additive characteristics. Specifically, it includes: salt concentration, conductivity, pH value, etc. Among them, the input layer contains several nodes, and each node represents a salinity parameter (salt concentration, conductivity, etc.); Set m hidden layers according to requirements, and each hidden layer uses the ReLU activation function to capture the non-linear relationship between salinity parameters and additive characteristics; The output layer is the predicted value of additive characteristics under different salinity conditions. The output layer of the salinity-additive sub-network is the input layer of the additive-grouting material property sub-network model. The number of nodes in this layer is the same as the number of additive characteristic indexes, providing basic data for the subsequent sub-network.

[0093] The admixture-grouting material performance sub-network model is constructed using a deep neural network (DNN), which consists of an input layer, a hidden layer, and an output layer. The input layer is the output layer of the mineralization-admixture sub-network, and the number of nodes remains the same; n hidden layers are set according to demand, and each hidden layer uses the ReLU activation function to capture the complex relationship between the admixture characteristics and the grouting material performance. The number of nodes in each layer is also designed in a decreasing manner; the output layer is the grouting material performance index, which uses linear activation to generate the predicted value of the grouting material performance, including the material's setting time, compressive strength, fluidity, and dynamic water retention rate.

[0094] The output of the mineralization-admixture sub-network is used as the input of the next network layer, so that the influence of mineralization on admixtures is passed layer by layer to the prediction of grouting material performance. Specifically, the progressive model expresses the influence of mineralization parameters through the change of admixture characteristics, and then further affects the performance of grouting materials based on the change of admixture characteristics. In this way, through the progressive modeling process, the complex correlation between mineralization and the indirect influence of admixture properties on grouting material performance is fully simulated.

[0095] After modeling, model training and multi-objective optimization are performed:

[0096] The preprocessed and feature-selected data is divided into training sets, validation sets, and test sets for model construction and optimization; then, based on the deep neural network (DNN), training is performed, and the network weights are adjusted through the back-propagation algorithm to gradually optimize the model performance. During the training process, k-fold cross-validation and hyperparameter tuning techniques (Bayesian optimization) are used to adjust the model's parameters such as the number of layers, number of neurons, and learning rate to improve the accuracy and stability of the model.

[0097] Specifically, first, the dataset after preprocessing and feature selection is divided into training set, validation set and test set in a ratio of 8:1:1, where the training set is used for model training, the validation set is used for tuning and preventing overfitting, and the test set is used to evaluate the actual prediction effect of the model.

[0098] The two-layer progressive model constructed based on the feature selection data is trained through the back propagation algorithm to adjust the weights and biases between neurons in each layer to optimize the prediction performance. The specific operations are:

[0099] Assuming that the network model in this disclosure has L layers, first input the sample data, and after calculations at each layer, get the output prediction value For each layer of neurons, there are:

[0100] The input signal is the output of the previous layer, set to x (l-1) , the weight matrix is ​​W (l) , the bias vector is t(l) , output y (l) is y (l) = W (l) x (l-1) + t (l) , and y (l) is passed through the ReLU activation function to obtain x (l) = f(y (l) ), and finally the predicted value of the output layer is obtained

[0101] where y (l) is the output of the linear transformation of the neurons in the l-th layer, representing the value obtained after the input signals of the neurons in the current layer are weighted and summed; W (l) is the weight connected to each neuron, randomly generated during initialization; t (l) is the bias vector of the l-th layer, initialized to zero and updated through training later; f(y (l) ) is the ReLU activation function, introducing non-linearity and enhancing the model's ability to fit complex data

[0102] The mean squared error loss function is used to calculate the error between the predicted value and the true value:

[0103]

[0104] where m is the number of samples in the dataset, y i is the true value of the i-th sample of the grouting material performance index, and the specific value is obtained through experiments or on-site measurements; is the predicted value of the i-th sample of the grouting material performance index, obtained from the output result of the above forward propagation

[0105] To achieve multi-objective optimization, considering the effectiveness of additives and the key properties of grouting materials (such as strength and fluidity) under different salinity parameters, weight allocation is introduced in the loss function, and different weight coefficients are set for the error terms of different objectives to find a more appropriate balance among multiple optimization objectives. Finally, ensure that the model obtains stable prediction results on the test set, and gradually improve its reliability in practical applications by continuously updating data and model parameters

[0106] Starting from the output layer, the error is propagated back layer by layer to calculate the gradients of the weights and biases of each layer. Among them, the error gradient δ (L) of the output layer and the error gradient δ (l) of the hidden layer are calculated as follows:

[0107]

[0108] δ (l) = (W (l+1) ) T δ(l+1) ·f′(y (l) )

[0109] where δ (L) represents the gradient value that the error of the output layer propagates to the current layer; is the difference between the predicted value and the true value; δ (l) represents the gradient value that the error of the previous layer passes through the weights of the current layer to the hidden layer; f'(y (L) ) is the derivative of the activation function of the output layer.

[0110] Calculate the weight gradient and bias gradient of each layer as follows:

[0111]

[0112] Update the weights and biases using the gradient descent method, and finally obtain:

[0113]

[0114] where represents the contribution of each weight to the total loss; represents the contribution of the bias to the total loss; η is the learning rate, which is determined by grid search.

[0115] Based on the training of the backpropagation algorithm, to further improve the model accuracy and stability, k-fold cross-validation and the Bayesian optimization algorithm are used to adjust the hyperparameters. Specifically: the data is further divided into k subsets. In each validation, one of the subsets is selected as the validation set, and the remaining parts are used as the training set. Define the Bayesian optimization objective function (in this disclosure, the mean squared error loss function is used as the objective function), randomly sample several parameter combinations from the hyperparameter space, train the model, record the validation set error of each combination, and generate new hyperparameter combinations. Iteratively repeat the above steps until the preset maximum number of iterations is reached. After the Bayesian optimization is completed, select the hyperparameter combination with the lowest validation set error in the cross-validation, and use this combination as the final hyperparameters, and save the model.

[0116] Step (5) Model training and multi-objective optimization

[0117] First, input relevant salinity parameters, admixture types, grouting material ratios, etc. into the pre-trained deep neural network (DNN) model. The system generates a predicted output based on these input data, and calculates the influence of admixtures on the properties of cement-based grouting materials under different salinity conditions, especially key performance indicators such as strength and fluidity. This prediction provides an estimate of the application effect of admixtures in different salinity environments for the project site, facilitating the selection and adjustment of the construction plan before construction.

[0118] After the prediction is completed, the system compares the prediction results with the performance data collected in actual applications. By feeding back the actual application effects into the model, the prediction effects of the model under the current environmental and material conditions can be evaluated. For example, if there are deviations between the predicted strength or fluidity of the model and the actual measured values, the system will record these differences and conduct error analysis to optimize the weight and parameter configuration of the model. At the same time, to ensure the applicability of the model in various actual environments, the system will dynamically update the database, adding data such as salinity, admixture characteristics, and environmental conditions during the construction process to improve the coverage and accuracy of the prediction. By collecting the latest data under different salinity conditions and updating the model parameters, the system gradually improves the prediction accuracy, enabling the model to adapt to different salinity conditions in future predictions and optimizations and achieving high-precision predictions of the performance of admixtures.

[0119] Example Two

[0120] The purpose of this embodiment is to provide 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 of the above method are implemented.

[0121] Example Three

[0122] The purpose of this embodiment is to provide a computer-readable storage medium.

[0123] A computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the above method are executed.

[0124] Example Four

[0125] See the appendix Figure 2 As shown, the purpose of this embodiment is to provide a prediction system for the effect of admixtures on grouting materials under the influence of salinity, specifically including the following processes:

[0126] (1) Data acquisition and preprocessing module, which is used to provide accurate, comprehensive, and high-quality input data for the model to ensure the reliability of the model prediction results. First, this module obtains data through multiple channels, including on-site tests, laboratory tests, and consulting relevant literature, to collect key data such as salinity parameters, types of admixtures and their specific dosages, and performance indicators of grouting materials. After the data acquisition is completed, the system imports the data into the data cleaning process. First, perform data integrity checks, identify missing data, and complete the filling process through interpolation. Secondly, detect and remove outliers to ensure that the data quality is not interfered by extreme values. At the same time, standardize each data through normalization processing to ensure that variables with different dimensions can be uniformly processed by the model. After completing the above steps, the system stores the preprocessed data in a standardized manner to ensure consistent formats for easy model calling and quick reference.

[0127] (2) Feature selection and partial correlation analysis module, which is used to identify the key influencing factors of salinity parameters and admixture characteristics on the performance of grouting materials, so as to provide scientific input data for the model and ensure the model is more targeted and accurate. First, in the feature selection stage, the collected feature variables are screened through an algorithm based on importance scores. This method assigns scores to all features, selects the feature variables significantly related to the performance of grouting materials, eliminates redundant information in the data, retains the core features, reduces complexity and improves the modeling efficiency. Next, this module applies the partial correlation analysis method to reveal the indirect influence between salinity and admixture characteristics and grouting material performance by calculating the correlation coefficient under controlled variables. Specifically, the partial correlation analysis calculates the correlation between salinity and admixture characteristics, and then combines the partial correlation between admixture characteristics and material performance to further clarify the progressive influence path between variables.

[0128] (3) Multi-level modeling and optimization module, which is used to accurately capture the multiple influence relationships between salinity, admixture characteristics and their influence on the performance of grouting materials. This module includes two main sub-models. Through layer-by-layer modeling and progressive fusion, the associations between parameters are gradually established to support the accurate prediction of the overall model.

[0129] First, a sub-model of the influence of salinity parameters on admixture characteristics is constructed. The input of this sub-model is the salinity parameter. By analyzing how the change of salinity affects the key characteristics such as the activity, reactivity, dissolution rate, etc. of the admixture, the function of mapping the salinity parameter to the reaction activity of the admixture is realized. This model includes an input layer, a hidden layer and an output layer: the input layer receives the salinity parameter, the hidden layer uses non-linear activation functions and regularization strategies to model the complex influence relationship of salinity on admixture performance, and the output layer generates the intermediate output data of the influence of salinity on admixture characteristics.

[0130] Next, a sub-model of the influence of admixtures on the performance of grouting materials is constructed. This model takes the output of the previous sub-model as the input to analyze the specific influence of different admixture characteristics on the performance of grouting materials (such as strength, fluidity, etc.) under salinity conditions. This sub-model also includes an input layer, a hidden layer and an output layer, especially focusing on how the admixture activity affects the macroscopic performance indicators of grouting materials under different salinity conditions. The hidden layer combines non-linear mapping and neuron interaction to further capture the diffusion, reaction and strengthening effects of the admixture in the material to achieve performance prediction.

[0131] Through the twin model structure of this hierarchy, the progressive transmission of model information and the multi-dimensional relationship modeling are realized. The hierarchical fusion between the sub-model of the influence of salinity on the additive and the sub-model of the influence of the additive on the properties of the grouting material ensures the dynamic interaction relationship among variables under different salinity conditions. To ensure the accuracy and generalization ability of each hierarchical model, hyperparameter optimization and cross-validation techniques are further introduced. Specifically, the Bayesian optimization method is used to tune the hyperparameters of the model, and the k-fold cross-validation technique is used to evaluate the effects of different parameter combinations on the validation set, so as to select the parameter configuration with the best prediction performance. Finally, through the combination of the multi-level structure and the optimization strategy, this module realizes the accurate prediction and optimization of the properties of the grouting material under different salinity conditions, ensuring the reliability and applicability of the model in practical applications.

[0132] (4) Performance prediction and application feedback module. It is used to apply the prediction results of the multi-level model to practice and continuously optimize the model performance through the feedback mechanism. First, input parameters such as salinity and the characteristics of the additive into the model to predict the key performance impacts of the additive on the grouting material (such as strength and fluidity) under different salinity conditions. Then, compare and analyze the prediction results with the actual application situation to identify deviations and evaluate the accuracy and applicability of the model. According to the actual feedback, dynamically adjust the model parameters and optimization algorithms to gradually improve the prediction accuracy of the model. At the same time, collect new data to enrich the database, so that the model can better adapt to different salinity environments and application requirements, and achieve continuous improvement and accurate prediction.

[0133] Example Five

[0134] The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any one of the above embodiments.

[0135] The steps involved in the devices of the above embodiments correspond to those in Method Embodiment 1, and the specific implementation manners can be referred 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.

[0136] 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 made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to be implemented. The present invention is not limited to any specific combination of hardware and software.

[0137] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made without creative efforts on the basis of the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting the effect of admixtures on grouting materials under the influence of mineralization, characterized in that: include: Obtain data, including mineralization parameters, admixture types and corresponding characteristics, and key performance indicators of grouting materials; Preprocess the acquired data; The core characteristic variables were screened from the pre-processed data, and the indirect effect of mineralization on admixture reactivity was analyzed by partial correlation analysis. Based on the core characteristic variables, two single-level sub-network models are constructed, including: mineralization-admixture sub-network model and admixture-grouting material performance sub-network model; Among them, the mineralization-admixture sub-network model is used to study the key characteristics of the reactivity and activity of admixtures affected by mineralization changes, so as to obtain the law of the influence of mineralization on admixtures under different conditions; the admixture-grouting material performance sub-network model uses the physical and chemical properties of admixtures as input, studies the direct influence of admixture properties on the performance of grouting materials, and establishes a model of the effect of admixtures on grouting materials; The mineralization parameters to be predicted, the type of admixture, and the ratio of grouting materials are input into two single-level sub-network models to generate prediction outputs, and the effects of admixtures on the performance of cement-based grouting materials under different mineralization conditions are obtained, including key performance indicator data of strength and fluidity.

2. The method for predicting the effect of admixture on grouting material under the influence of mineralization as claimed in claim 1, characterized in that: The core feature variables are screened for the preprocessed data, including: Recursive feature elimination is used to process the preprocessed data by repeatedly building the model, removing the least important features in each iteration until the preset number of features or other stopping conditions are reached.

3. The method for predicting the effect of admixture on grouting material under the influence of mineralization as claimed in claim 1, characterized in that: The support vector machine is selected as the basic model in the constructed model, and the initial feature set is set to all features, denoted as F = {x1, x2, ..., x n }, where n is the number of original features. At the same time, when the final number of features to be retained is set to , feature selection stops, the feature set F is used to train the model, and the coefficient vector ω output by the model is used to evaluate the feature importance.

4. The method for predicting the effect of admixture on grouting material under the influence of mineralization as claimed in claim 1, characterized in that: The partial correlation analysis method is used to analyze the indirect effect of mineralization on the reactivity of admixtures. The specific operation is as follows: Assuming that the admixture characteristics are the control variables, that is, Z1, Z2, ..., Z k , the mineralization parameter is the variable X, and the grouting material performance is Y, and multiple regression is performed; With X as the dependent variable, Z1, Z2, …, Z k Perform regression for the independent variables and obtain the regression equation; X=β0+β1Z1+β2Z2+…β k Z k +τ1, actual value minus predicted value, calculate the residual e X ; Similarly, with Y as the dependent variable, Z1, Z2, …, Z k Regression is performed on the independent variable, and the regression equation is obtained: Y = α0 + α1Z1 + α2Z2 + ... α k Z k +τ2, actual value minus predicted value, calculate the residual e Y ; Finally, calculate the residual e X and e Y The Pearson correlation coefficient is the correlation coefficient between X and Y in the control variables Z1, Z2, ..., Z k The partial correlation coefficient under the influence of is: in, and The i-th value of the residual sequence of mineralization parameter X and material property Y, respectively, and are the means of the residual series of mineralization parameter X and material property Y respectively; Finally, the partial correlation coefficient r between the mineralization parameter X and the grouting material performance Y is calculated. XY·Z .

5. The method for predicting the effect of admixture on grouting material under the influence of mineralization as claimed in claim 1, characterized in that: The mineralization-admixture sub-network model is constructed using a deep neural network, which consists of an input layer, a hidden layer and an output layer; The input layer contains several nodes, each of which represents a mineralization parameter; m hidden layers are set, and each hidden layer uses the ReLU activation function to capture the nonlinear relationship between the mineralization parameter and the admixture characteristics; The output layer is the predicted value of the admixture characteristics, and the number of nodes in this layer is consistent with the number of admixture characteristic indicators.

6. The method for predicting the effect of admixture on grouting material under the influence of mineralization as claimed in claim 1, characterized in that: The admixture-grouting material performance sub-network model is constructed using a deep neural network, which consists of an input layer, a hidden layer and an output layer; Among them, the input layer is the output layer of the mineralization-admixture sub-network model, and the number of nodes remains consistent; n hidden layers are set according to the needs, and each hidden layer uses the ReLU activation function to capture the complex relationship between the influence of admixture characteristics on the performance of grouting materials; The output layer is the performance index of the grouting material, which uses linear activation to generate the predicted value of the grouting material performance.

7. A prediction system for the effect of admixtures on grouting materials under the influence of mineralization, characterized by: include: The data acquisition and preprocessing module is configured to: acquire data, including mineralization parameters, admixture types and corresponding characteristics, and key performance indicators of grouting materials; preprocess the acquired data; The feature selection and partial correlation analysis module is configured to: screen the core feature variables from the preprocessed data and analyze the indirect effect of mineralization on admixture reactivity through partial correlation analysis; The multi-level modeling and optimization module is configured to: construct two single-level sub-network models based on the core characteristic variables, including: mineralization-admixture sub-network model and admixture-grouting material performance sub-network model; Among them, the mineralization-admixture sub-network model is used to study the key characteristics of the reactivity and activity of admixtures affected by mineralization changes, so as to obtain the law of the influence of mineralization on admixtures under different conditions; the admixture-grouting material performance sub-network model uses the physical and chemical properties of admixtures as input, studies the direct influence of admixture properties on the performance of grouting materials, and establishes a model of the effect of admixtures on grouting materials; The performance prediction and application feedback module is configured to: input the mineralization parameters to be predicted, the type of admixture, and the grouting material ratio information into two single-level sub-network models to generate prediction outputs, and obtain the impact of admixtures on the performance of cement-based grouting materials under different mineralization conditions, including key performance indicator data of strength and fluidity.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. 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 of the method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are performed.

Images