An interpretable tunnel gushing water grouting material component optimization method and system

Abstract

The present disclosure provides an interpretable tunnel gushing water grouting material component optimization method and system, relates to the technical field of tunnel grouting material optimization, and comprises the following steps: acquiring actual tunnel grouting engineering condition data and preprocessing; inputting the preprocessed engineering condition data into an integrated model, outputting predicted target material performance by extracting the correlation between engineering conditions and material performance; inputting the target material performance into a self-adaptive generative adversarial network model, first generating material components by combining random noise through a generator, then inputting the generated material components and the target material performance into a discriminator to output a probability value of whether the material components meet the target material performance, introducing an interpretability method, taking the output of the discriminator as an explanation target, quantifying the influence of the input material components on the output judgment probability of the discriminator, and reversely predicting the optimal material components by dynamically adjusting the key feature weights of the discriminator.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of tunnel grouting material optimization, in particular to an interpretable tunnel gushing water grouting material component optimization method and system. BACKGROUND

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

[0003] In tunnel engineering construction, gushing water is one of the common geological disasters, especially in areas with complex geological conditions and abundant groundwater. Gushing water not only causes damage to the tunnel structure, but also seriously affects construction safety and project progress. In order to deal with this problem, grouting treatment method is usually used, which can reinforce the stratum and block the water source by injecting specific materials, so as to achieve the purpose of stabilizing the tunnel.

[0004] Traditional grouting design methods usually rely on construction experience, laboratory test results and subjective judgment of engineers. Although the above methods can solve the tunnel gushing water disaster to some extent, in the performance design stage of grouting materials, the existing design methods rely on experience judgment, and cannot systematically and scientifically consider the complex interaction between multiple engineering conditions, and do not fully consider the influence of actual site conditions on grouting effect. In the material research and design stage, a large number of and tedious manual tests are usually required, which not only has long cycle and high cost, but also is difficult to realize intelligent and efficient material component optimization.

[0005] In recent years, with the development of machine learning and artificial intelligence technology, data-driven optimization methods have gradually become an important way to solve this problem. In the performance design stage of grouting materials, machine learning models can deeply mine and analyze historical data to accurately predict the target performance of grouting materials. In the material research and design stage, machine learning models can be used to train laboratory test data to realize intelligent design of material components for target performance.

[0006] Machine learning has great potential in the optimization design of grouting materials. The commonly used machine learning algorithms include regression model, support vector machine, random forest, artificial neural network, etc. However, these algorithms have certain limitations, which leads to their inability to be widely applied in actual engineering design. For example, regression model is difficult to capture the complex nonlinear relationship between cement-based material performance and each component, and is prone to overfitting, underfitting and multicollinearity problems, and is also very sensitive to outliers. Support vector machine has weak multi-objective optimization ability and is difficult to consider multiple material performance indicators at the same time. Random forest model is difficult to capture high-dimensional complex relationships and has poor extrapolation ability, which cannot generate new material design schemes beyond the distribution of training data, and its generalization ability is limited. The training of artificial neural network depends on a large amount of high-quality data, and also has the problem of poor extrapolation ability. SUMMARY

[0007] In order to solve the above problems, the present disclosure provides an interpretable tunnel gushing water grouting material component optimization method and system. The method optimizes the grouting material components in a data-driven manner, uses a generative adversarial network to construct a machine learning inverse model for inversely inferring material components according to material performance, and introduces an interpretability method for optimization, thereby realizing adaptive adjustment of the generative adversarial network, approaching the complex nonlinear mapping relationship between material components and their performance through adversarial learning, effectively solving the problems of poor interpretability and insufficient generalization ability of traditional machine learning methods, and having higher automation degree and reliability.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] An interpretable tunnel gushing water grouting material component optimization method, comprising:

[0010] Obtaining and preprocessing the engineering condition data of actual tunnel grouting;

[0011] Inputting the preprocessed engineering condition data into an integrated model, wherein the integrated model is a weighted average fusion of a statistical-based model and a machine learning-based model using a fusion strategy, and outputs the predicted target material performance by extracting the correlation between the engineering conditions and the material performance;

[0012] Inputting the target material performance into an adaptive generative adversarial network model. First, the generator of the adaptive generative adversarial network model is used to generate material components in combination with random noise. Then, the generated material components and the target material performance are input into the discriminator of the adaptive generative adversarial network model, and the probability value of whether the material components meet the target material performance is output. An interpretability method is introduced, and the output of the discriminator is taken as the explanation target. The influence of the input material components on the output judgment probability of the discriminator is quantified. The input material components are optimized by dynamically adjusting the key feature weights of the discriminator, i.e., inversely predicting the optimal material components.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] An interpretable tunnel gushing water grouting material component optimization system, comprising:

[0015] A data acquisition module for obtaining and preprocessing the engineering condition data of actual tunnel grouting;

[0016] The prediction target acquisition module is configured to input the preprocessed engineering condition data into an integrated model, the integrated model is an integrated model obtained by fusing a weighted average of a statistical-based model and a machine learning-based model by using a fusion strategy, and the integrated model is configured to output a predicted target material performance by extracting a correlation between engineering conditions and material performance;

[0017] The material component reverse prediction module is configured to input the target material performance into the adaptive generative adversarial network model, first generate a material component by combining random noise through a generator of the adaptive generative adversarial network model, then input the generated material component and the target material performance into a discriminator of the adaptive generative adversarial network model, and output a probability value of whether the material component meets the target material performance, introduce an explainability method, take an output of the discriminator as an explanation target, quantify an influence of the input material component on a judgment probability of the output of the discriminator, and optimize the input material component by dynamically adjusting key feature weights of the discriminator, that is, reversely predict an optimal material component.

[0018] According to some embodiments, the present disclosure adopts the technical scheme as follows:

[0019] A computer program product comprising a computer program, which, when executed by a processor, implements the tunnel gushing water grouting material component optimization method with explainability.

[0020] According to some embodiments, the present disclosure adopts the technical scheme as follows:

[0021] A non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the tunnel gushing water grouting material component optimization method with explainability.

[0022] According to some embodiments, the present disclosure adopts the technical scheme as follows:

[0023] An electronic device comprising a processor, a memory, and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the tunnel gushing water grouting material component optimization method with explainability.

[0024] Compared with the prior art, the present disclosure has the following beneficial effects:

[0025] The tunnel gushing water grouting material component optimization method of the present disclosure is capable of constructing a tunnel gushing water grouting treatment material database by using natural language processing technology, fully utilizing historical engineering data, fusing a statistical model and a machine learning model to construct an integrated model, and capturing complex nonlinear relationships by using the respective advantages of the two models, thereby reducing the possible deviation and variance problems of a single model and improving the model generalization ability and stability.

[0026] The tunnel gushing water grouting material component optimization method of the present disclosure is capable of optimizing the grouting material components in a data-driven manner, constructing a machine learning inverse model for inversely inferring material components according to material performance by using a generative adversarial network, and realizing adaptive adjustment of the discriminator by means of a SHAP interpreter, thereby learning and approximating the complex nonlinear mapping relationship between the material components and their performance in an adversarial training manner, solving the difficulty that may not be effectively described by a traditional linear model or a simple regression method, and having higher automation degree and reliability.

[0027] The tunnel gushing water grouting material component optimization method of the present disclosure is capable of realizing the whole process from the tunnel gushing water treatment engineering conditions to the target performance design of the grouting material, and then to the optimization design and explanation of the grouting material components. By means of intelligent and automated methods, the targeted design of the grouting treatment material is realized, the design efficiency and accuracy are greatly improved, an innovative solution is provided for the material design of tunnel engineering grouting treatment, and the intelligent development of engineering design is promoted. BRIEF DESCRIPTION OF DRAWINGS

[0028] The accompanying drawings, which form a part of the present disclosure, are used to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are used to explain the present disclosure, and do not constitute improper limitations on the present disclosure.

[0029] Figure 1 The flowchart of the optimization design of the tunnel gushing water grouting material components of the present disclosure;

[0030] Figure 2 The adaptive generative adversarial network structure diagram of the present disclosure;

[0031] Figure 3 The adaptive generative adversarial network optimization process diagram of the present disclosure. DETAILED DESCRIPTION

[0032] The present disclosure will be further described below in combination with the drawings and embodiments.

[0033] It should be pointed out that the following detailed description is all exemplary and is intended to provide further description of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present disclosure belongs.

[0034] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components and / or combinations thereof.

[0035] Embodiment 1

[0036] An explainable tunnel gushing water grouting material component optimization method is provided in an embodiment of the present disclosure, comprising:

[0037] Step one: obtaining actual tunnel grouting engineering condition data and preprocessing;

[0038] Step two: inputting the preprocessed engineering condition data into an integrated model, the integrated model being a weighted average fusion of a statistical based model and a machine learning based model using a fusion strategy, outputting a predicted target material performance by extracting the correlation between engineering conditions and material performance;

[0039] Step three: inputting the target material performance into a self-adaptive generative adversarial network model, first generating a material component by combining random noise through the generator of the self-adaptive generative adversarial network model, then inputting the generated material component and the target material performance into the discriminator of the self-adaptive generative adversarial network model, outputting a probability value of whether the material component meets the target material performance, introducing an explainable method, taking the output of the discriminator as an explanation target, quantifying the influence of the input material component on the judgment probability of the discriminator output, and optimizing the input material component by dynamically adjusting the key feature weights of the discriminator, i.e., reversely predicting the optimal material component.

[0040] As an embodiment, the explainable tunnel gushing water grouting material component optimization method of the present disclosure is used to realize autonomous regulation and control of grouting treatment material performance, clearly define the complex action and mutual influence relationship of each component of the material by means of intelligent algorithm, and provide intelligent means for grouting treatment material research and development. The specific implementation process is as follows:

[0041] Step 1: obtaining historical grouting treatment material data, constructing a tunnel gushing water grouting treatment material database, and preprocessing the data to divide the data set;

[0042] 1) The acquisition of historical grouting treatment material data includes: collecting engineering conditions and corresponding grouting treatment material performance parameters from literature, engineering case data or field data;

[0043] Wherein, the engineering conditions include: stratum type, water inflow and tunnel depth; the grouting material performance parameters include: material type, initial setting time, 3-day compressive strength and viscosity.

[0044] 2) Adopt natural language processing technology for historical materials, including data collection and preprocessing, key information extraction processing, and database establishment after processing.

[0045] Specifically, data collection and preprocessing includes: collecting literature, engineering case reports or field records, which can be in the form of PDF or Word documents, web pages, database records, etc. The collected data is preprocessed, including converting non-text format content (such as PDF, charts) into text format, removing elements without text information, cleaning irrelevant information (such as headers and footers, references, etc.), ensuring data simplicity and accuracy, and facilitating subsequent processing.

[0046] Key information extraction includes named entity recognition (NER) and relationship extraction. The present disclosure uses NLP to extract structured databases for subsequent machine learning training, and does not involve complex relationship analysis, so only named entity recognition is required.

[0047] Further, named entity recognition (NER) refers to the recognition of entities with specific meanings in text.

[0048] The specific steps include: annotated data set, model training, automatic recognition and evaluation optimization, the process is as follows:

[0049] 1. Annotated data set: The engineering conditions and material performance contained in the preprocessed data set are annotated as specific entity categories. Specifically, the engineering conditions: stratum type, water inflow, tunnel depth and grouting material performance parameters: material type, initial setting time, 3-day compressive strength, viscosity are set as categories, and their corresponding values are set as entities.

[0050] When manually annotating, each entity and its corresponding category are annotated to construct an entity annotation data set.

[0051] 2. Model training: Select a suitable NER model, which can include BERT, RoBERTa, SpaCy, etc. Use the entity annotation data set to train the model, so that the model learns the features of the entities in the literature, engineering case materials and field data, so as to identify the same or similar types of entities in unannotated data.

[0052] 3. Automatic recognition: The trained NER model can automatically identify different entities in new text data.

[0053] 4. Evaluation and optimization: Evaluate the performance of the NER model using metrics such as accuracy, recall, F1 score, etc. Adjust the model parameters or increase the training data to improve the accuracy of the model. Also, introduce a domain vocabulary to optimize the recognition ability of the model.

[0054] 3) Establish a database and structure the identified information in the form of a table to associate different engineering conditions and material properties.

[0055] Further, clean and standardize the data to remove noise and errors, such as irrelevant information or missing values. Standardize the units and process various attribute values to ensure data consistency, and finally obtain the tunnel gushing water grouting treatment material database.

[0056] 4) Preprocess the tunnel gushing water grouting treatment material database, including discrete feature numericalization, normalization, and class balancing.

[0057] Among them, discrete feature numericalization can choose one-hot encoding, label encoding, and embedding encoding for conversion. Normalization methods can choose Min-Max normalization, Z-score normalization, and logarithmic normalization. Class balancing can choose undersampling and oversampling methods for processing.

[0058] 5) Data set division can be divided into two parts: randomize the data, select 70% as the training set for model training, and 30% as the test set for model evaluation.

[0059] Step 2: Construct a statistical base model and a machine learning base model, apply a fusion strategy to construct an integrated model, and establish a more accurate relationship between engineering conditions and material properties, including: input the engineering conditions in the tunnel gushing water grouting treatment material database into the integrated model for analysis and feature extraction, and output the target material performance under the engineering conditions.

[0060] Specifically, the integrated model is integrated and fused by the statistical base model and the machine learning base model, wherein the statistical base model can realize linear and nonlinear relationship analysis of engineering conditions and material properties, and the machine learning base model can realize feature extraction of complex nonlinear relationship. Combining traditional engineering experience with modern machine learning technology improves the generalization ability and robustness of the model.

[0061] After training and testing the integrated model, input the current engineering conditions into the integrated model to obtain the required more accurate target material performance under the conditions.

[0062] Further, the training process of the integrated model of the present disclosure is as follows:

[0063] 1. Constructing statistical base model, including exploratory data analysis, model selection, parameter fitting and testing and evaluation.

[0064] Wherein, the exploratory data analysis includes:

[0065] 1) Checking data characteristics, checking whether there is a strong linear or nonlinear between the independent variable engineering conditions and the dependent variable material performance;

[0066] 2) Variable correlation analysis, using scatter plot, correlation coefficient (such as Pearson, Spearman, etc.) to analyze the correlation between independent variables and dependent variables;

[0067] 3) Variable distribution check: check the distribution of data to see if it meets the normality assumption;

[0068] 4) Multiple collinearity detection: use methods such as variance inflation factor (VIF) to detect collinearity problems. If the VIF value is too high, the variables with high correlation need to be deleted, or regularized methods such as ridge regression, Lasso regression, etc. are used.

[0069] Further, the model selection includes: selecting the model according to the data characteristics. Linear relationship can choose ordinary linear regression, generalized linear regression, etc. Nonlinear relationship can choose polynomial regression, and multiple collinearity can choose ridge regression, Lasso regression, etc. Regularization method.

[0070] Further, the parameter fitting includes: dividing the data into training set and test set; using least squares method, maximum likelihood estimation method, etc. to fit the model and get the model parameters; evaluating the goodness of fit (R2) and residual distribution of the model to ensure that the assumptions of the model are met by the data characteristics.

[0071] Further, the testing and evaluation includes: checking the independence and normality of the residuals. If the residuals show patterns, the model may have missing nonlinearity; using t-test or F-test to judge the significance of independent variables on dependent variables; through cross-validation or resampling, the stability of the model under different data distribution is tested.

[0072] Finally, the statistical base model is obtained.

[0073] 2. Constructing machine learning base model, the selectable models include: random forest, support vector machine, KNN and neural network.

[0074] Specifically, for the independent variable engineering conditions and the dependent variable material performance, the above models are tested, and the best model is selected through cross-validation.

[0075] Where, cross-validation is used five-fold cross-validation or ten-fold cross-validation, record the performance indicators (such as precision, recall, mean square error, etc.) of each fold, in order to compare the pros and cons of the model. For the selected model using grid search or random search, adjust the hyperparameters of the model, check the error distribution, precision, recall and feature importance of the model, etc., and finally get the machine learning base model.

[0076] 3. Apply fusion strategy to integrate statistical base model and machine learning base model, the selected suitable fusion strategy includes weighted average, Boosting, Stacking, etc., combine the prediction results of statistical base model and machine learning base model to improve the overall prediction performance, that is, establish a more accurate relationship between engineering conditions and material performance. Optionally, the present disclosure uses weighted average method for model fusion, the specific process is:

[0077] 1) For each input sample (engineering condition) x i and the corresponding output (material performance) y i , the prediction output of the statistical base model is The prediction output of the machine learning base model is Calculate the error of each model, preferably the mean square error:

[0078]

[0079] Where, N is the number of samples.

[0080] 2) Find the best weight by minimizing the weighted average prediction error, the weighted average prediction error is:

[0081]

[0082] Where, is the prediction output of the statistical base model and the machine learning base model for the i-th sample, N is the number of samples, ω stat , ω ml are the weights of the statistical base model and the machine learning base model, respectively.

[0083] 3) Minimize the weighted average prediction error using the least squares method: take the partial derivative of ω stat , ω ml respectively:

[0084]

[0085] Make the partial derivative zero, and solve the weight:

[0086]

[0087] The final prediction value can be obtained by weighted average of the models:

[0088]

[0089] wherein, is the prediction output of the statistical base model, is the prediction output of the machine learning base model.

[0090] Further, the training of the integrated model should be performed separately on the base models. For the statistical base model, the engineering conditions and material properties are fitted and tested using the selected regression method; for the machine learning base model, similarly, the engineering conditions are taken as input and the material properties are taken as output, and the selected machine learning method is trained and tested.

[0091] Finally, the trained integrated model is obtained.

[0092] Further, by inputting the engineering conditions into the integrated model, the target material properties under the engineering conditions can be output.

[0093] Step 3: By establishing a grouting material component and performance database, a machine learning inverse model for inversely inferring material components according to material properties is constructed, and the target material properties output by the integrated model are input into the inverse model, so that the predicted material components are obtained.

[0094] Step 1) Establish a grouting material component and performance database;

[0095] Specifically, the establishment of the grouting material component and performance database can be performed by means of laboratory test data collection and data augmentation. The material properties of various grouting materials under different components are collected through tests, and specifically, in the treatment of tunnel gushing water grouting, the cement and admixture of cement-based materials have the most direct and important influence on the material properties.

[0096] The material components include water-cement ratio, cement type, water-reducing agent type and addition amount, retarder type and addition amount, and early strength agent type and addition amount; the material properties are consistent with the output of the integrated model, i.e., material type, preliminary setting time, 3-day compressive strength, and viscosity.

[0097] After being augmented by data augmentation means, the data is preprocessed, and finally the grouting material component and performance database is established. The data augmentation means can enhance the number of data samples and enrich the diversity of data in the case of insufficient test data, effectively improving the generalization ability of the model. The optional means include interpolation method, variational encoder, SMOTE, etc.

[0098] Further, the data preprocessing includes one-hot encoding combined with numerical processing of the type and amount of water reducing agent, the type and amount of retarder, and the type and amount of early strength agent, and standardization and normalization processing of the material component data.

[0099] Step 2) The prediction of material components is realized by a machine learning inverse model, and a machine learning inverse model is constructed to inversely infer material components according to material performance and is trained.

[0100] Specifically, the machine learning inverse model adopts a self-adaptive generative adversarial network model, which includes a generative adversarial network model and an explainer. The generative adversarial network model includes a generator and a discriminator. The generator accepts target material performance as a conditional variable input and generates material components in combination with random noise. The generator adopts a convolutional neural network structure. The generation process of the generator is as follows:

[0101]

[0102] Wherein, G is the generator, c is the random noise, y is the conditional variable, i.e. the material performance, is the output of the generator, i.e. the material component.

[0103] The task of the discriminator is to receive the material performance and the generated sample of the generator output as input, and output the probability value of whether the material component meets the target material performance, i.e. to judge whether the generated material component is "real". The discriminator adopts a multi-layer perceptron (MLP) structure. The generation process of the discriminator is as follows:

[0104] D(z) = σ(ω T z+b)

[0105] Wherein, ω and b are the hyperparameters of the discriminator, ω is the feature weight, b is the feature bias, z is the input feature, which should be the generated sample of the generator output, σ is the Sigmoid activation function, and D(z) represents the probability that the data z is real data.

[0106] The generator loss and the discriminator adversarial loss are defined respectively:

[0107] The calculation formula of the generator loss is:

[0108] L G = -log[D(y,G(c,y))]

[0109] Wherein, G(c,y) is the generated sample of the generator output, and D(y,G(c,y)) is the judgment probability of the discriminator on the generated sample.

[0110] The adversarial loss of the discriminator is:

[0111] L D = Lreal +L fake = -log(D(y,z real ))-log(1-D(y,z fake ))

[0112] The adversarial loss of the discriminator is the sum of the real sample and the generated sample loss, where y and z real are the material performance in the grouting material component and performance database and the corresponding material real component, i.e. the real sample; z fake is the material component output by the generator, i.e. the generated sample; D(y,z real ) and D(y,z fake ) are the probability outputs of the real sample and the generated sample by the discriminator.

[0113] Further, the model training is to use the preprocessed data to alternately train the generator and the discriminator, and the specific process is:

[0114] 1) Train the discriminator: randomly extract a batch of real samples y and z real to the discriminator, calculate the loss of the real sample; use the generator to randomly generate a batch of initial material components y fake and the corresponding material performance z fake , as a false sample input to the discriminator, calculate the loss of the generated sample; calculate the gradient by back propagation and update its parameters to minimize the total loss.

[0115] 2) Train the generator: pass random noise and material performance into the generator, and the generator generates the corresponding material component through its neural network, i.e. the generated sample. Evaluate these generated samples using the discriminator, and update the generator parameters by calculating the gradient by back propagation to maximize the generator's comprehensive loss.

[0116] 3) Alternately repeat the process of training the discriminator and the generator. Each training step will optimize the respective parameters through back propagation and gradient update.

[0117] Further, the back prediction is to input the target material performance output by the integrated model and the random noise into the trained generative adversarial network, i.e. the generator output can predict the material component, which can better meet the input target performance under the component allocation.

[0118] Step 3) Introduce an explainer after the generative adversarial network. The explainer applies the SHAP (Shapley Additive exPlanations) explainability method to quantify which generated samples (i.e. material components and corresponding material performance) in the discriminator help generate "more real" samples, so as to optimize the process of predicting material components;

[0119] Specifically, the adaptive generative adversarial network dynamically adjusts the key feature weights of the discriminator with the help of the SHAP interpreter, and then optimizes the reverse prediction output. The specific steps are as follows:

[0120] 1) The output score probability of the discriminator is taken as the explanation target, and SHAP is used to quantify the influence of the input generated sample on the output judgment probability D(y, z) of the discriminator.

[0121] For the Shapley value of the generated sample, the calculation formula is:

[0122]

[0123] Where M is the set of all input features, that is, the set of all features of the generated sample z fake ; M| represents the total number of features; i represents the i-th sample; S is any subset of the i-th sample in the set M, which can be an empty set; |S| represents the size of the subset S; f(S) represents the output of the discriminator when only the features in the subset S are used as input, that is, D(y, z fake ).

[0124] 2) According to the size and sign of the SHAP value, identify which material components contribute most to the authenticity or performance of the discriminator output. The larger the SHAP value, the greater the contribution of the feature to the discriminator output (i.e., the probability of the authenticity of the sample). A positive value indicates that the feature makes the sample more likely to be judged as real, and a negative value indicates that the feature makes the sample more likely to be judged as false. Specifically, a threshold is set, and features exceeding the threshold are considered key features.

[0125] 3) According to the key features identified by the SHAP value, construct an adaptive discriminator that gives higher weights to key features when judging real and false samples. Specifically, the weights of the identified key features are expanded by 5 times.

[0126] 4) Re-train and reverse predict the inverse model, and re-calculate the SHAP value of the input generated sample to the discriminator, and iteratively update the weights. When the change of the feature weight is less than 1% or the iteration number reaches a predetermined threshold, stop the iteration process, and save the current weight of the discriminator.

[0127] 5) Output the current generator generated sample, that is, the optimized material component with higher quality and interpretability.

[0128] Example 2

[0129] An embodiment of the present disclosure provides an interpretable tunnel gushing water grouting material component optimization system, comprising:

[0130] A data acquisition module is configured to acquire actual tunnel grouting engineering condition data and pre-process the data.

[0131] a prediction target acquisition module configured to input the preprocessed engineering condition data into an integrated model, the integrated model being an integrated model in which a weighted average of a statistical-based model and a machine learning-based model is integrated using a fusion strategy, and configured to output a predicted target material performance by extracting a correlation between an engineering condition and a material performance;

[0132] a material component reverse prediction module configured to input the target material performance into a self-adaptive generative adversarial network model, to generate a material component by combining random noise through a generator of the self-adaptive generative adversarial network model, to input the generated material component and the target material performance into a discriminator of the self-adaptive generative adversarial network model, and to output a probability value of whether the material component meets the target material performance, and configured to introduce an explainability method, to take an output of the discriminator as an explanation target, to quantify an influence of the input material component on a judgment probability of the output of the discriminator, and to optimize the input material component by dynamically adjusting key feature weights of the discriminator, that is, to reversely predict an optimal material component.

[0133] Embodiment 3

[0134] In an embodiment of the present disclosure, a computer program product is provided, which includes a computer program, and the computer program, when executed by a processor, implements the tunnel gushing water grouting material component optimization method with explainability.

[0135] Embodiment 4

[0136] In an embodiment of the present disclosure, a non-transitory computer readable storage medium is provided, which is configured to store computer instructions, and the computer instructions, when executed by a processor, implement the tunnel gushing water grouting material component optimization method with explainability.

[0137] Embodiment 5

[0138] In an embodiment of the present disclosure, an electronic device is provided, which includes a processor, a memory and a computer program, wherein the processor is connected with the memory, and the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device implements the tunnel gushing water grouting material component optimization method with explainability.

[0139] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0140] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0141] Although the present disclosure has been described with reference to specific on embodiments thereof, a person of ordinary skill in the art understands that various modifications or changes can be made on the technical solutions of the present disclosure without paying creative labor, and these modifications or changes shall still fall within the protection scope of the present disclosure.

Claims

1. An explainable tunnel gushing water grouting material component optimization method, characterized in that, The method comprises the following steps: acquiring and preprocessing actual tunnel grouting engineering condition data; inputting the preprocessed engineering condition data into an integrated model, wherein the integrated model is obtained by integrating and fusing a statistical basic model and a machine learning basic model by using a fusion strategy, and the integrated model outputs predicted target material performance by extracting the correlation between engineering conditions and material performance; the integrated model is obtained by integrating and fusing the statistical basic model and the machine learning basic model by using the fusion strategy, wherein the statistical basic model analyzes the linear and nonlinear relationship between engineering conditions and material performance, and the machine learning basic model extracts the complex nonlinear relationship of engineering condition data, the two are weighted and averaged to obtain the final predicted target material performance: wherein, is a prediction output of the statistical base model, is a prediction output of the machine learning base model, are weights of the statistical base model and the machine learning base model, respectively; inputting the target material performance into a self-adaptive generative adversarial network model, first generating material components by combining random noise through the generator of the self-adaptive generative adversarial network model, then inputting the generated material components and the target material performance into the discriminator of the self-adaptive generative adversarial network model, and outputting the probability value of whether the material components meet the target material performance, introducing an explainability method, taking the output of the discriminator as an explanation target, quantifying the influence of the input material components on the output judgment probability of the discriminator, and optimizing the input material components by dynamically adjusting the key feature weights of the discriminator, that is, reversely predicting the optimal material components.

2. The method for optimizing components of an interpretable tunnel gushing water grouting material according to claim 1, wherein The engineering condition data includes stratum type, water inflow and tunnel depth, and the material performance of grouting includes material type, initial setting time, compressive strength and viscosity; the preprocessing includes discrete feature numericalization, normalization and category balance.

3. The method of claim 1, wherein the tunnel inrush water grouting material component optimization is interpretable. The material components include water-cement ratio, cement type, water-reducing agent type and addition amount, retarder type and addition amount, and early strength agent type and addition amount. The reverse prediction of the material components is realized by a machine learning reverse model. The machine learning reverse model adopts a self-adaptive generative adversarial network model. The self-adaptive generative adversarial network model comprises a generator, a discriminator and an explainer. The generator accepts the target material performance as a conditional variable input and generates material components by combining random noise.

4. The method for optimizing components of an interpretable tunnel gushing water grouting material according to claim 3, characterized in that, The discriminator receives the target material performance and the generated sample output by the generator as input, and outputs the probability value of whether the generated sample output by the generator meets the target material performance, that is, whether the generated material components are real. The discriminator selects a multi-layer perception structure. The generation process of the discriminator is as follows: where, and b is a hyperparameter of the discriminator, is the input feature, i.e., the generated sample output by the generator, is a Sigmoid activation function, denotes the data z is the probability of the real data.

5. The method for optimizing components of an interpretable tunnel gushing water grouting material according to claim 1, wherein The reverse prediction is to input the target material performance output by the integrated model and the generated random noise into the self-adaptive generative adversarial network, and output the predicted material components by the generator. In the explainer, the key feature weights of the discriminator are dynamically adjusted by introducing the explainability method, and then the output of the reverse prediction is optimized. The Shapley value of the sample generated by the generator is calculated as follows: where, M is the set of all input features, i.e. the set of all features of the generated sample ; denotes the total number of features; i denotes the i-th sample; i S is an arbitrary subset of the set M excluding the i-th sample, which can be the empty set; i denotes the size of the subset S; S denotes the output of the discriminator when only using the features in the subset S as input, i.e. .​​​ 6. An interpretable tunnel inrush grouting material component optimization system, characterized in that, The method comprises the following steps: a data acquisition module for acquiring and preprocessing actual tunnel grouting engineering condition data; The prediction target acquisition module is configured to input the preprocessed engineering condition data into an integrated model, the integrated model is an integrated model obtained by fusing a weighted average of a statistical-based model and a machine learning-based model by using a fusion strategy, and the integrated model is configured to output a predicted target material performance by extracting a correlation between an engineering condition and a material performance; The material component reverse prediction module is configured to input the target material performance into the adaptive generative adversarial network model, first generate a material component by combining random noise through a generator of the adaptive generative adversarial network model, then input the generated material component and the target material performance into a discriminator of the adaptive generative adversarial network model, and output a probability value of whether the material component meets the target material performance, introduce an explainability method, take an output of the discriminator as an explanation target, quantify an influence of the input material component on a judgment probability of the output of the discriminator, optimize the input material component by dynamically adjusting key feature weights of the discriminator, and obtain an optimal material component through reverse prediction.

7. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the interpretable tunnel gushing water grouting material component optimization method of any one of claims 1-5.

8. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions are executed by the processor to implement the interpretable tunnel gushing water grouting material component optimization method of any one of claims 1-5.

9. An electronic device, comprising: The computer program is executed by the processor to implement the interpretable tunnel gushing water grouting material component optimization method of any one of claims 1-5. The computer program is executed by the processor to implement the interpretable tunnel gushing water grouting material component optimization method of any one of claims 1-5.

Images