Cement-based material performance optimization method and system based on machine learning

Abstract

The present invention discloses a method and system for optimizing cement-based material performance based on machine learning, the method comprising: obtaining historical cement-based material data, and preprocessing the historical cement-based material data to obtain a training set and a test set; obtaining multiple preset machine learning models, and performing model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model; performing model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model; obtaining current cement-based material data, and inputting the current cement-based material data into the performance optimization prediction model, and outputting cement-based material performance optimization results. The present invention can accurately predict and optimize the performance of cement-based materials by constructing a performance optimization prediction model, and performing interpretable analysis and multi-objective optimization processing, thereby achieving guidance on the proportioning of cement-based materials.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a cement-based material performance optimization method, system, terminal and computer-readable storage medium based on machine learning. Background Art

[0002] Cement-based materials are core materials in the construction industry and play an important role in various structural applications. Cement-based materials are developing rapidly. Before solidification, cement-based materials have good plasticity, and after hardening, they have high compressive strength and good durability. They can be cast into load-bearing components or structures of various shapes and sizes and used for a long time. As a key cementitious material, it is widely used in civil construction, water conservancy engineering, national defense engineering and other fields.

[0003] However, they face huge challenges in terms of durability and sustainability. In the context of the development of modern building structures towards super-high-rise and large-span designs, two main aspects must be emphasized: first, the inherent shortcomings of traditional cement-based materials, including poor toughness, compressive strength but not tensile strength, and easy cracking; second, concrete components in coastal environments are exposed to seawater for a long time, which will degrade due to factors such as freeze-thaw damage, sulfate corrosion, chloride erosion and mechanical wear, thereby endangering safety and durability.

[0004] In order to solve the above problems, the prior art proposes to improve the performance of nano cement-based materials, but the existing research on the performance improvement of nano cement-based materials mostly focuses on the use of control variable method, trying to avoid the influence of control variables on the research results. The performance improvement of nano cement-based materials is related to multiple factors, and the mutual influence between these factors is often nonlinear, which makes it impossible to predict and optimize the performance of cement-based materials.

[0005] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0006] The main purpose of the present invention is to provide a method, system, terminal and computer-readable storage medium for optimizing the performance of cement-based materials based on machine learning, aiming to solve the problem that the performance improvement of nano cement-based materials in the prior art is related to multiple factors, and the mutual influence between these factors is often nonlinear, which makes it impossible to predict and optimize the performance of cement-based materials.

[0007] To achieve the above object, the present invention provides a cement-based material performance optimization method based on machine learning, and the cement-based material performance optimization method based on machine learning comprises the following steps:

[0008] Acquire historical cement-based material data, and preprocess the historical cement-based material data to obtain a training set and a test set;

[0009] Acquire multiple preset machine learning models, and perform model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model;

[0010] Performing model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model;

[0011] The current cement-based material data is obtained, and the current cement-based material data is input into the performance optimization prediction model, and the cement-based material performance optimization result is output.

[0012] Optionally, in the cement-based material performance optimization method based on machine learning, the historical cement-based material data includes historical cement-based material experimental data and cement-based material literature data; the preprocessing includes data partitioning processing, one-hot code processing and null value interpolation processing;

[0013] The acquiring of historical cement-based material data and preprocessing of the historical cement-based material data to obtain a training set and a test set specifically includes:

[0014] Acquire historical cement-based material experimental data and cement-based material literature data, and perform data segmentation processing on the historical cement-based material experimental data and the cement-based material literature data according to the material properties of the cement-based material to obtain multiple material property data sets;

[0015] Acquire multiple feature classification variables from multiple material property data sets, and perform one-hot coding on the multiple feature classification variables to obtain multiple feature classification vectors;

[0016] Acquire feature null values ​​in a plurality of the feature classification vectors, perform null value interpolation processing on the feature null values ​​using a KNN algorithm to obtain an initial data set, and divide the initial data set into a training set and a test set.

[0017] Optionally, the method for optimizing cement-based material performance based on machine learning, wherein the step of obtaining a plurality of preset machine learning models, and performing model training and model screening on the plurality of preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model, specifically includes:

[0018] Acquire multiple preset machine learning models, wherein the preset machine learning models include an ANN model, a KNN model, a tree model, and an AdaBoost model;

[0019] Performing model training on the plurality of preset machine learning models according to the training set using a k-fold cross-validation method to obtain a plurality of first performance test results, and performing model testing on the plurality of preset machine learning models according to the test set to obtain a plurality of second performance test results;

[0020] The similarities between the multiple first performance test results and the multiple second performance test results are calculated respectively, and the preset machine learning model corresponding to the first performance test result with the greatest similarity is selected from the multiple preset machine learning models to obtain the initial performance optimization prediction model.

[0021] Optionally, in the cement-based material performance optimization method based on machine learning, the model optimization process includes Bayesian optimization process, interpretable analysis process and multi-objective optimization process;

[0022] The performing model optimization processing on the initial performance optimization prediction model to obtain the performance optimization prediction model specifically includes:

[0023] Using a Bayesian optimization method to perform hyperparameter tuning processing on the initial performance optimization prediction model to obtain a first performance optimization prediction model;

[0024] Performing interpretability analysis on the first performance optimization prediction model to obtain a second performance optimization prediction model;

[0025] The second performance optimization prediction model is subjected to multi-objective optimization processing to obtain a performance optimization prediction model.

[0026] Optionally, the cement-based material performance optimization method based on machine learning, wherein the Bayesian optimization method is used to perform hyperparameter tuning processing on the initial performance optimization prediction model to obtain a first performance optimization prediction model, specifically comprising:

[0027] Using a Bayesian optimization method to obtain multiple model parameters in the initial performance optimization prediction model, and combining the multiple model parameters to obtain a multidimensional parameter space;

[0028] Performing Gaussian process processing on the multidimensional parameter space to obtain a probability distribution result of the model parameters;

[0029] Obtaining a preset acquisition function, and performing parameter point sampling on the model parameter probability distribution result according to the preset acquisition function to obtain a parameter performance metric;

[0030] The parameters of the initial performance optimization prediction model are updated according to the parameter performance metric to obtain the first performance optimization prediction model.

[0031] Optionally, the cement-based material performance optimization method based on machine learning, wherein the interpretability analysis processing of the first performance optimization prediction model to obtain the second performance optimization prediction model specifically includes:

[0032] Acquire multiple features in the first performance optimization prediction model, calculate multiple quantized prediction errors corresponding to the multiple features, and perform weighted processing on the multiple quantized prediction errors to obtain multiple contribution values ​​corresponding to the multiple features;

[0033] Sort the multiple features in descending order according to contribution value, select the feature with the largest contribution value, and obtain the best contribution feature;

[0034] The first performance optimization prediction model is subjected to feature selection processing and model optimization processing according to the best contribution feature to obtain the second performance optimization prediction model.

[0035] Optionally, the cement-based material performance optimization method based on machine learning, wherein the multi-objective optimization processing is performed on the second performance optimization prediction model to obtain the performance optimization prediction model, specifically comprising:

[0036] Acquire multiple evolutionary algorithms, and perform multi-objective optimization processing on the second performance optimization prediction model according to the multiple evolutionary algorithms to obtain a trade-off relationship between different performance indicators in the second performance optimization prediction model;

[0037] Calculating a Pareto optimal solution according to the trade-off relationship between different performance indicators in the second performance optimization prediction model, and obtaining an optimization parameter selection range according to the Pareto optimal solution;

[0038] The second performance optimization prediction model is optimized according to the optimization parameter selection range to obtain the performance optimization prediction model.

[0039] In addition, to achieve the above-mentioned purpose, the present invention also provides a cement-based material performance optimization system based on machine learning, wherein the cement-based material performance optimization system based on machine learning includes:

[0040] A data preprocessing module, used to obtain historical cement-based material data and preprocess the historical cement-based material data to obtain a training set and a test set;

[0041] An initial prediction model building module is used to obtain multiple preset machine learning models, and perform model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model;

[0042] A model performance optimization module, used to perform model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model;

[0043] The performance optimization result output module is used to obtain current cement-based material data, input the current cement-based material data into the performance optimization prediction model, and output the cement-based material performance optimization result.

[0044] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a cement-based material performance optimization program based on machine learning stored in the memory and executable on the processor, wherein the cement-based material performance optimization program based on machine learning, when executed by the processor, implements the steps of the cement-based material performance optimization method based on machine learning as described above.

[0045] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a cement-based material performance optimization program based on machine learning, and when the cement-based material performance optimization program based on machine learning is executed by a processor, the steps of the cement-based material performance optimization method based on machine learning as described above are implemented.

[0046] In the present invention, historical cement-based material data is obtained, and the historical cement-based material data is preprocessed to obtain a training set and a test set; multiple preset machine learning models are obtained, and model training and model screening are performed on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model; the initial performance optimization prediction model is subjected to model optimization processing to obtain a performance optimization prediction model; current cement-based material data is obtained, and the current cement-based material data is input into the performance optimization prediction model, and the cement-based material performance optimization result is output. The present invention collects existing experimental data and literature data, and preprocesses these data to construct a training set and a validation set, and trains multiple large machine learning models through these training sets and test sets, and determines the performance optimization prediction model according to the performance test results. Afterwards, the model is optimized through Bayesian optimization, interpretability analysis, and multi-objective optimization processing, and the nonlinear relationship that affects the performance parameters of cement-based materials can be automatically found, so as to achieve accurate prediction and optimization of the optimal performance of cement-based materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of a preferred embodiment of the cement-based material performance optimization method based on machine learning of the present invention;

[0048] Figure 2It is a schematic diagram of the overall architecture implementation process of a preferred embodiment of the cement-based material performance optimization method based on machine learning of the present invention;

[0049] Figure 3 It is a schematic diagram of the programming of multi-objective optimization of GNAs modified cement-based materials in a preferred embodiment of the cement-based material performance optimization method based on machine learning of the present invention;

[0050] Figure 4 It is a structural diagram of a preferred embodiment of the cement-based material performance optimization system based on machine learning of the present invention;

[0051] Figure 5 It is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] Cement-based materials are core materials in the construction industry and play an important role in various structural applications. Cement-based materials are developing rapidly. Before solidification, cement-based materials have good plasticity, and after hardening, they have high compressive strength and good durability. They can be cast into load-bearing components or structures of various shapes and sizes and used for a long time. As a key cementitious material, it is widely used in civil construction, water conservancy engineering, national defense engineering and other fields.

[0054] However, they face huge challenges in terms of durability and sustainability. In the context of the development of modern building structures towards super-high-rise and large-span designs, two main aspects must be emphasized: first, the inherent shortcomings of traditional cement-based materials, including poor toughness, compressive strength but not tensile strength, and easy cracking; second, concrete components in coastal environments are exposed to seawater for a long time, which will degrade due to factors such as freeze-thaw damage, sulfate corrosion, chloride erosion and mechanical wear, thereby endangering the safety and durability of the building.

[0055] In recent years, the incorporation of nanomaterials into cementitious materials has emerged as a key innovation. Due to their small size and large surface area, nanomaterials can interact uniquely with the cement matrix, resulting in improved strength and durability, and potentially improved environmental sustainability. The advent of nanotechnology in cementitious materials has opened up new avenues to address traditional limitations. Studies have shown that nanomaterials such as nanosilica, carbon nanotubes, and graphene can be used to improve the mechanical properties and durability of cementitious composites. These enhancements include increased tensile strength, reduced permeability, and improved durability to environmental factors such as corrosion and freeze-thaw cycles. In addition, the interaction of these nanomaterials with the cement hydration process can produce a denser and finer microstructure, further improving the material's performance. Among the various nanomaterials, GNAs (graphene nanofluid additives) have recently attracted much attention due to their extraordinary mechanical properties and potential to revolutionize cementitious materials.

[0056] However, the prior art on the performance improvement of nano cement-based materials mostly focuses on the use of control variable method, trying to avoid the influence of control variables on the research results. In the only design mix ratio, only one or two interactions such as nanomaterials and chemical admixture dosage can be studied. In fact, nano-modified cement-based materials are also related to multiple factors such as the size of nanomaterials, water-cement ratio, and curing time. The mutual influence between these factors is often unavoidable and nonlinear. Therefore, evaluating the combined impact of these factors is a challenging task. Machine learning models can automatically find nonlinear relationships and better fit data. Therefore, the present invention aims to construct a predictive model for GNAs admixture design through machine learning, and attempt to perform interpretability analysis on the model. Finally, the optimal performance of GNAs modification is searched in the model.

[0057] The workflow of the present invention is mainly divided into 8 steps: 1. Combining the existing experimental data and the data in the literature, construct the compressive strength test data set, the flexural strength test data set, and the chloride ion diffusion coefficient data set for machine learning model training. 2. The statistical characteristics of each data set are preliminarily explored, the classification characteristics are edited by unique hot code, and the data set is divided into a training set and a test set, and the verification method adopts k-fold cross validation. 3. Using the KNN data interpolation technology, only the input values ​​in the data set are interpolated to obtain a complete and consistent data set. 4. Use multiple machine learning models to generate prediction models for each data set, and compare the prediction performance to preliminarily select models with better performance. 5. Use Bayesian optimization to further adjust the model parameters and optimize the prediction effect of the model. 6. Use an expanded data set composed of experimental data and synthetic data to retrain the prediction model of the chloride ion diffusion coefficient, and compare the effects of unsupervised learning and semi-supervised learning. 7. Use the SHapley additive interpretation method to perform interpretability analysis on the robust model previously obtained through multiple technical comparisons. 8. Multiple evolutionary multi-objective optimization algorithms are used to solve the performance optimization problem of GNAs modified cement-based materials, obtain multiple Pareto optimal solutions when the model performs best, and fit them into the Pareto frontier.

[0058] The method for optimizing cement-based material performance based on machine learning described in the preferred embodiment of the present invention is as follows: Figure 1 and Figure 2 As shown, the cement-based material performance optimization method based on machine learning includes the following steps:

[0059] Step S10, obtaining historical cement-based material data, and preprocessing the historical cement-based material data to obtain a training set and a test set. The historical cement-based material data includes historical cement-based material experimental data and cement-based material literature data; the preprocessing includes data partitioning, one-hot code processing, and null value interpolation processing.

[0060] Strength and durability are key factors in the design and analysis of cement-based composite materials. The present invention can establish an accurate and reliable prediction model based on input variables and data points.

[0061] Specifically, historical cement-based material experimental data and cement-based material literature data are obtained, and data segmentation processing is performed on the historical cement-based material experimental data and the cement-based material literature data according to material properties of the cement-based material to obtain a plurality of material property data sets.

[0062] The present invention constructs three data sets by combining previous experimental data and data in the literature. The prediction targets (i.e., the predicted optimized values ​​output by the model) are compressive strength, flexural strength, and chloride ion diffusion coefficient (i.e., the cement-based material performance optimization results in the present invention), respectively. The data points in the literature come from 18 published experimental studies. There are 300 data points in the three data sets. Among them, the output variables include compressive strength, flexural strength, and chloride ion diffusion coefficient. The number of test data for these three attributes is 117, 103, and 80, respectively, and the number of data set features is 12, 12, and 14, respectively. The input variables include the model value of the strength of the cement ash used "CG", sand ratio "S / C", water-cement ratio "W / C", surfactant "Surfactant", surfactant dosage "SF(%)", defoamer tributyl phosphate dosage "TBP(%)", water reducer dosage "SP(%)", graphene dosage "G(%)", linear size of nanomaterials "D(μm)", thickness of nanomaterials "T(nm)", curing age "CA(d)", compressive strength "CS(MPa)", flexural strength "FS(MPa)", dosage of nanomaterials "Amount(%)", nanomaterials "Nanomaterials", average pore diameter "AD(nm)", porosity "Porosity(%)", specific surface area of ​​nanomaterials "SA(m2 / g)", and chloride ion diffusion coefficient "Dapp(10-12m2 / s)". In order to focus the research as much as possible on the laws of mechanical modification of GO (Graphite Oxide) materials, the data set only covers the experimental data of the mixed system of neat paste and mortar, and the type of cement used is always Portland cement.

[0063] Regarding the input and output variables, there are four points that need to be explained: 1. For all the features that represent the dosage, the percentages all represent the percentage of cement mass. And when the dosage is 0, its related attributes are all set to 0. For example, when the graphene dosage is 0, the linear size and thickness of the graphene in the data set are both set to 0. 2. Many feature values ​​are a range, and for clear characterization, the average value is used as the input value. For example, the thickness of graphene is 2-3nm, so the thickness of graphene in the data set is 2.5 nm. 3. For compression and folding specimens, the input variables are shared. For the specimens of the chloride ion diffusion test, only some features are shared, because this dataset uses the transfer learning method. For example, the dosage of nanomaterials "Amount(%)" in the chloride ion diffusion coefficient dataset is equal to the graphene dosage "G(%)" in the strength dataset, because graphene is also a nanomaterial, but nanomaterials also cover a wider range. 4. For categorical variables such as surfactants "Surfactant", one-hot coding is used to represent them in the form of vectors for learner training.

[0064] A plurality of feature classification variables in a plurality of the material property data sets are obtained, and a plurality of the feature classification variables are processed by one-hot coding to obtain a plurality of feature classification vectors.

[0065] The categorical variables involved in the present invention include types of surfactants and nanomaterials. In order to meet the needs of machine learning model training, the categorical variables are processed by one-hot coding to become a vector.

[0066] The present invention classifies the elemental composition of nanoparticles to make up for the lack of data on GNAs-modified cement-based materials in the literature, and adopts a transfer learning method for the training of chloride ion diffusion coefficients. For example, the nanoparticle TiO2-graphene in the literature can be represented as [1, 0, 0, 1, 1]. This encoding form not only reflects the difference between each nanoparticle, but also facilitates transfer learning.

[0067] Acquire feature null values ​​in a plurality of the feature classification vectors, perform null value interpolation processing on the feature null values ​​using a KNN algorithm to obtain an initial data set, and divide the initial data set into a training set and a test set.

[0068] Among them, cross-validation is one of the commonly used methods to evaluate the performance of machine learning models. It divides the sample data into different training sets and test sets by reusing data, uses the training set to train the model, and uses the test set to evaluate the model's predictive ability. In this way, multiple sets of different training sets and test sets can be obtained, so that the model can be evaluated on different data sets, which increases the credibility and stability of the evaluation results.

[0069] After using the KNN (k=5) algorithm to interpolate the null values ​​of the input features, each data set is divided into a training set and a test set. The default validation method uses k-fold cross validation to fully verify the accuracy of the model and reduce the risk of overfitting and underfitting. Here, the k value is 5, that is, in each fold, 80% of the data is randomly divided into the training set and 20% of the data is divided into the test set. The final evaluation index is the average value of the performance index calculated for each of the 5 folds.

[0070] Step S20: obtain multiple preset machine learning models, and perform model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model.

[0071] The present invention has considered 12 prediction models based on different data sets: 1. Linear regression method, Ridge Regression Model (RR). 2. Deep neural network, namely artificial neural network (ANN). 3. Non-parametric learner, namely K-nearest neighbor (KNN). 4. Learning methods of many tree types, namely decision tree (DT, Delaunay Triangulation), random forest (RF, Random Forest), ensemble learning classification algorithm (AB, Ada boost), gradient boosting algorithm (GB, Gradient boost), a machine learning algorithm (CB, Catboost), gradient boosting decision tree model (LGBM, LightGBM), supervised model (XGB, XG boost), extreme tree (ET, Extra Trees) and ensemble algorithm (BR, Bagging regressor).

[0072] Specifically, multiple preset machine learning models are obtained, wherein the preset machine learning models include an ANN model, a KNN model, a tree model, and an AdaBoost model.

[0073] Among these models, there are artificial neural networks (ANN), K nearest neighbors (KNN), tree models (such as decision trees and ensemble trees), and AdaBoost (AB). For the ANN model, two hidden layers were set, and the optimal number of neurons was obtained through grid search. The batch size (the number of data samples captured for one training) was selected as 128. Such a large value helps to train multiple samples at the same time and improves training efficiency. The optimizer selected the stochastic gradient descent method, and the activation function was set to the hyperbolic tangent function. This setting makes the model converge faster on small data sets and achieves better results. In the KNN model, the distance metric uses the Euclidean distance, and the K value is optimally obtained through grid search. In the AdaBoost model, the base learner uses the decision tree (DT), which is more appropriate than the default value, indicating that the hyperparameter adjustment of the base learner has an important impact on the performance of the overall model.

[0074] The k-fold cross-validation method is used to perform model training on the multiple preset machine learning models according to the training set to obtain multiple first performance test results, and the multiple preset machine learning models are tested on the model according to the test set to obtain multiple second performance test results; the similarities between the multiple first performance test results and the multiple second performance test results are calculated respectively, and the preset machine learning model corresponding to the first performance test result with the greatest similarity is screened out from the multiple preset machine learning models to obtain the initial performance optimization prediction model.

[0075] The present invention also uses a Taylor Diagram, which is a tool for graphically displaying the similarity between model simulation results and observed data. It comprehensively displays three statistics in the same figure: standard deviation, correlation coefficient, and root mean square error (RMSE). The closer the standard deviation of the model prediction value is to the standard deviation of the true value of the test set, the closer the variability of the model is to the real data. The points of the model are placed according to their standard deviation (radius) and correlation coefficient (angle). The smaller the angle, the closer the correlation coefficient is to 1, indicating that the linear relationship between the predicted value and the true value is stronger.

[0076] In summary, the selection of hyperparameters for machine learning models in the present invention needs to consider multiple factors, including model type and data set characteristics. Appropriate hyperparameter settings can improve the performance and generalization ability of the model, which is critical for model selection and optimization.

[0077] Step S30: Perform model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model. The model optimization processing includes Bayesian optimization processing, interpretability analysis processing and multi-objective optimization processing.

[0078] Specifically, a Bayesian optimization method is used to obtain multiple model parameters in the initial performance optimization prediction model, and the multiple model parameters are combined to obtain a multidimensional parameter space; Gaussian process processing is performed on the multidimensional parameter space to obtain a model parameter probability distribution result; a preset acquisition function is obtained, and parameter point sampling is performed on the model parameter probability distribution result according to the preset acquisition function to obtain a parameter performance metric; the parameters of the initial performance optimization prediction model are updated according to the parameter performance metric to obtain the first performance optimization prediction model.

[0079] After selecting the machine learning model for compressive and flexural strength prediction, the hyperparameters of the selected model are tuned to obtain better model performance.

[0080] The principle of Bayesian optimization for Gradient Boosting learner parameter adjustment is to search the parameter space through an efficient global optimization strategy to find the best parameter combination on the validation set. The specific principles and steps are as follows:

[0081] First, define a performance metric (for example, the accuracy of the 5-fold cross validation is R2), and the parameters of the Gradient Boosting model that need to be optimized. Each possible value of the parameter constitutes a multidimensional parameter space. Bayesian optimization starts with an a priori assumption about the performance of the parameter space, usually assuming that any combination of parameters is possible and that no parameter combination is better than another in the absence of data. Bayesian optimization uses a surrogate model (Gaussian process) to obtain an approximation of the true performance function. The Gaussian process can predict the probability distribution of the model performance under a given parameter combination, including the mean and uncertainty (variance) of the prediction. Then the acquisition function is started. The acquisition function is the basis for selecting the next parameter point in the optimization process. It is based on the prediction and uncertainty of the surrogate model to determine the parameter point that is most likely to improve performance. Finally, based on the acquisition function, Bayesian optimization selects a parameter point to sample, runs the model, and obtains a performance metric. Then, this new data point is used to update the surrogate model, which makes the model more accurately reflect which parameter points are likely to lead to better performance.

[0082] This process will be iterated many times, and each iteration will select the best parameter point for evaluation through the acquisition function, and then use this new data to continuously update the proxy model. As the iterations proceed, the predictions of the proxy model will become more and more accurate, and the optimization algorithm will gradually focus on those parameter areas that are most likely to provide the best performance. When the predetermined number of iterations is reached, or the performance improvement is no longer significant, the optimization process stops. At this point, the best parameter point predicted by the proxy model is considered to be the best parameter setting for the GradientBoosting model.

[0083] After using the parameters optimized by Bayesian, the prediction error MSE value of the stress resistance learner on the test set decreased by 12.94% and the RMSE value decreased by 5.12%. It should be noted that the MSE and RMSE values ​​here are the average results of the 5-fold cross validation, so the MSE value is not directly equal to the RMSE value. The prediction results of the 5-fold cross validation after Bayesian optimization of the stress resistance and fold strength learner show that the predicted values ​​are consistent with the true values, which shows that the model has excellent prediction performance.

[0084] Acquire multiple features in the first performance optimization prediction model, calculate multiple quantitative prediction errors corresponding to the multiple features, and perform weighted processing on the multiple quantitative prediction errors to obtain multiple contribution values ​​corresponding to the multiple features; sort the multiple features in order from large to small according to the contribution value, select the feature with the largest contribution value to obtain the best contribution feature; perform feature selection processing and model optimization processing on the first performance optimization prediction model according to the best contribution feature to obtain the second performance optimization prediction model.

[0085] The machine learning explainability tool (SHAP, Shapley Additive explanation), Shapley (Shapley index) value originated from game theory and is used to measure the importance of each feature in the model. It explains the predicted results by calculating the contribution of each feature to the prediction. A positive SHAP value indicates that the feature has a positive contribution to the predicted result, a negative contribution indicates a negative contribution, and a zero value indicates that the feature has no effect on the predicted result. In addition to helping understand how the model makes predictions, SHAP can also identify interactions between features in the dataset. SHAP explanations are globally legal, so detailed explanations of the prediction results of each sample can be obtained. SHAP values ​​can explain the predictions of the entire dataset by pairing expected values ​​with the predicted values ​​of each sample. These values ​​can be presented in graphical and numerical reports for more intuitive understanding. SHAP can also identify the most important features by ranking the importance of features. These features can be used to guide feature selection and model optimization.

[0086] Importance of feature k to the output of model f according to SHAP This feature is the output of the model The weighted sum of the contributions of all possible feature combinations. It is expressed as:

[0087] ,

[0088] in, , , is a subset of features, is the number of features in the model, is the total number of features, To multiply by the number of features, Does not contain features CA subset of . In SHAP, feature importance is measured by quantifying the prediction error while perturbing a given feature value. The sensitivity of the prediction error when perturbing the feature value is weighted by the importance of the feature. SHAP is also used to describe the performance of a trained machine learning model. SHAP uses an additive feature attribution approach, where an output model is considered to be the linear sum of the input variables to describe an interpretable model. For example, an input variable The model, in which Varies between 1 and n, where n is the number of input variables. , simplify input Explanation model Expressed as ;in, represents the number of input features, Represents a constant value when no information is provided (or no input is used), that is, a baseline value. Enter the variables for the simplified input variables, Features The Shapley value of Contribution to the model output.

[0089] In the present invention, the application of transfer learning is based on a core assumption: the effects of different nanomaterials on the properties of cement-based materials (including the chloride ion diffusion coefficient), although the specific mechanisms may be different, follow similar physical and chemical laws to a certain extent. Therefore, the model obtained by training on a wider data set of nanomaterials can capture these generally applicable laws and characteristics. Next, this pre-trained model is applied to the specific scenario of GNAs modified cement-based materials, and the model parameters are fine-tuned to better adapt to the characteristics of the target task. This process not only includes subtle adjustments to the model weights, but may also involve moderate modifications to the model structure to more accurately reflect the impact of GNAs on the performance of cement-based materials.

[0090] The transfer learning strategy includes two key steps: 1. Use a wide range of nanomaterials (including graphene, nano-titanium dioxide, etc.) performance data to train the model. At this stage, the model learns the general laws of the impact of nanomaterials on the performance of cement-based materials. 2. Fine-tune the model on limited GNAs-modified cement-based materials data to capture the unique impact characteristics of GNAs.

[0091] In the next stage of the present invention, synthetic data is introduced by unsupervised learning methods to enhance the training data set and improve the model performance. Specifically, the original data set is widened by different synthetic data ratios, with a widening gradient of 20% and a maximum widening of 300%. Four different data generation models are used in the present invention, namely conditional generative adversarial network (CTGAN), Gaussian Copula, variational autoencoder (TVAE) and conjugate generative adversarial network (Copula GAN). These models have their own unique characteristics and working principles, and are used to generate synthetic data to simulate the distribution characteristics of real data. Among them, the Gaussian conjugate model generates data using the characteristics of Gaussian distribution, the conditional generative adversarial network generates qualified data samples through adversarial training, the conjugate generative adversarial network combines the advantages of Copula function (correlation between variables, connection function) and generative adversarial network, and can better simulate complex data distribution, while the variational autoencoder uses encoding and decoding networks to generate data and learn the potential representation of data. The generated data is only used to expand the training set, while the test set remains a subset of the unenlarged real data set, and 5-fold cross validation is used to ensure the authenticity of the evaluation results.

[0092] Acquire multiple evolutionary algorithms, and perform multi-objective optimization processing on the second performance optimization prediction model according to the multiple evolutionary algorithms to obtain the trade-off relationship between different performance indicators in the second performance optimization prediction model; calculate the Pareto optimal solution according to the trade-off relationship between different performance indicators in the second performance optimization prediction model, and obtain the optimization parameter selection range according to the Pareto optimal solution; perform model optimization processing on the second performance optimization prediction model according to the optimization parameter selection range to obtain the performance optimization prediction model.

[0093] The performance-based multi-objective optimization process for GNAs modified cement-based materials is as follows: Pareto optimality theory is an important concept in economics, which describes an ideal state of resource allocation. Under the Pareto optimal state, the allocation of resources can achieve the improvement of at least one target benefit without compromising any target benefit. In other words, the Pareto optimal state refers to the optimization of the benefits of the existing resource allocation, that is, maximizing the overall benefit by rationally allocating resources. The multi-objective optimization algorithm can be used to simulate the resource allocation problem in Pareto theory. The algorithm makes trade-offs between multiple objectives so that resource allocation can meet all objectives as much as possible. The combination of Pareto theory and multi-objective optimization algorithm has unique advantages, which can help decision makers draw more accurate conclusions in specific application scenarios in a more scientific and effective way.

[0094] like Figure 3As shown, the design scheme of the present invention is a three-objective optimization problem for the compressive strength, flexural strength and chloride ion diffusion coefficient of GNAs modified cement-based materials. It aims to simultaneously improve the compressive strength and flexural strength of the material, while reducing the chloride ion diffusion coefficient, so as to achieve the purpose of improving the overall performance of the material. To this end, an evolutionary algorithm is used to find the optimal material composition and process parameters. First, by loading three pre-trained machine learning models, which represent predictors of compressive strength, flexural strength and chloride ion diffusion coefficient respectively. Then a multi-objective optimization problem is defined, which includes 20 optimization variables: 17 shared feature variables and 3 variables specific to the chloride ion model. On this basis, the type and value range of the variables are determined, as well as the mark of whether each goal is maximized or minimized.

[0095] The following four evolutionary algorithms are used in multi-objective optimization research. 1. NSGA-II (Non-dominated Sorting Genetic Algorithm II) is a widely used multi-objective optimization algorithm that maintains population diversity through non-dominated sorting and crowding distance comparison, so that it can quickly converge to the optimal solution of the target problem. 2. MOEA / D (Multi-objective Evolutionary Algorithm / Decomposition) adopts the strategy of decomposing the multi-objective optimization problem into a series of sub-problems optimized for a single objective, and solves these problems simultaneously. MOEA / D uses the proximity between sub-problems to exchange information, which speeds up the convergence of the algorithm and can effectively approach the Pareto front. 3. RVEA (Reference Vector Guided Evolutionary Algorithm) uses a set of dynamically adjusted reference vectors to guide the search process, ensuring uniform exploration of the Pareto front. This algorithm is suitable for problems with significant conflicting objectives, and can effectively balance exploration and utilization behaviors, thereby obtaining a uniformly distributed Pareto front solution set. 4. Parameter settings for the above algorithms. Encoding='RI': This means that the population encoding is a mixed "real + integer" encoding, that is, some variables are encoded as real values ​​and others as integer values, which corresponds to the mixed nature of continuous and discrete variables in the optimization problem.

[0096] The final result of multi-objective optimization is a set of solutions that meet multiple objectives. These solutions form a set called the Pareto front. The Pareto front is a solution to a multi-objective optimization problem. It refers to a set of solutions that are not inferior to any other solutions in all objectives. The Pareto front is composed of Pareto solutions. Therefore, the Pareto front represents the optimal solution space of the problem, which provides a method to select the best solution from multiple solutions.

[0097] Step S40: obtaining current cement-based material data, and inputting the current cement-based material data into the performance optimization prediction model, and outputting cement-based material performance optimization results.

[0098] The present invention introduces an admixture prediction model constructed by a multi-objective optimization algorithm, the purpose of which is to optimize the mechanical properties and durability of cement-based materials by GNAs. In the workflow of the present invention, the purpose is defined as a multi-objective optimization problem of the performance of GNAs-modified cement-based materials, and then advanced machine learning techniques are used to predict and optimize the mechanical properties and durability of the materials.

[0099] Technical effects:

[0100] 1. For model selection and Bayesian optimization, the present invention compares multiple initial models to draw Taylor diagrams to select a better model, and then adjusts the model hyperparameters through Bayesian optimization, which significantly improves the prediction accuracy of the model. It can make the model show high accuracy and consistency in key performance indicators such as compressive strength.

[0101] 2. Transfer learning, supervised learning and cluster analysis. The present invention uses transfer learning to build a prediction model to make up for the lack of trainable data in this field. Then, more training data is synthesized through unsupervised learning. By clustering features, the model can identify the key factors that affect the performance of cement-based materials. Ultimately, the prediction performance of the model is significantly improved.

[0102] 3. Verification of the performance prediction model. The present invention uses K-fold cross validation to verify the prediction effect of the model, which shows a high degree of accuracy and stability. The performance of mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R²) all confirm the reliability and effectiveness of the prediction model.

[0103] 4. Model interpretability analysis: The present invention adopts the SHAP method to perform model interpretability analysis, quantifies the influence of various characteristics on the prediction of mechanical properties and durability performance, further verifies the importance of factors such as porosity structure, admixtures, and nanomaterial size to material properties, and at the same time, quantifies the influence of surfactant functional groups in the modification.

[0104] 5. Application of multi-objective optimization algorithm: The present invention introduces a variety of evolutionary algorithms (such as NSGA-II (non-dominated sorting genetic algorithm 2), NSGA-III (non-dominated sorting genetic algorithm 3), MOEA / D (multi-objective evolutionary algorithm / decomposition), etc.) for multi-objective optimization, aiming to simultaneously improve the compressive strength and flexural strength of the material and reduce the chloride ion diffusion coefficient. The obtained optimization results reveal the trade-off relationship between different performance indicators and find the Pareto frontier of the optimal solution.

[0105] 6. Analysis of Pareto optimal solution: The present invention analyzes in detail the Pareto optimal solution obtained based on the NSGA-II algorithm, provides a specific parameter selection range for the performance optimization of GNAs modified cement-based materials, and also demonstrates the possibility of achieving the best trade-off among the three objectives of compressive strength, flexural strength and chloride ion diffusion coefficient.

[0106] The research of this invention not only improves the understanding of the factors affecting the performance of GNAs modified cement-based materials, but also provides an effective methodology and tool for the accurate prediction and optimization of material properties.

[0107] Furthermore, if Figure 4 As shown, based on the above-mentioned cement-based material performance optimization method based on machine learning, the present invention also provides a cement-based material performance optimization system based on machine learning, wherein the cement-based material performance optimization system based on machine learning includes:

[0108] A data preprocessing module 51 is used to obtain historical cement-based material data and preprocess the historical cement-based material data to obtain a training set and a test set;

[0109] An initial prediction model building module 52 is used to obtain a plurality of preset machine learning models, and perform model training and model screening on the plurality of preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model;

[0110] A model performance optimization module 53 is used to perform model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model;

[0111] The performance optimization result output module 54 is used to obtain current cement-based material data, input the current cement-based material data into the performance optimization prediction model, and output the cement-based material performance optimization result.

[0112] Furthermore, if Figure 5 As shown, based on the above-mentioned cement-based material performance optimization method and system based on machine learning, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 5 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0113] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a cement-based material performance optimization program 40 based on machine learning is stored on the memory 20, and the cement-based material performance optimization program 40 based on machine learning can be executed by the processor 10, thereby realizing the cement-based material performance optimization method based on machine learning in the present application.

[0114] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the cement-based material performance optimization method based on machine learning.

[0115] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0116] In one embodiment, when the processor 10 executes the cement-based material performance optimization program 40 based on machine learning in the memory 20, the steps of the cement-based material performance optimization method based on machine learning as described above are implemented.

[0117] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a cement-based material performance optimization program based on machine learning, and when the cement-based material performance optimization program based on machine learning is executed by a processor, the steps of the cement-based material performance optimization method based on machine learning as described above are implemented.

[0118] In summary, the present invention provides a method for optimizing cement-based material performance based on machine learning and related equipment, the method comprising: obtaining historical cement-based material data, and preprocessing the historical cement-based material data to obtain a training set and a test set; obtaining multiple preset machine learning models, and performing model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model; performing model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model; obtaining current cement-based material data, and inputting the current cement-based material data into the performance optimization prediction model, and outputting cement-based material performance optimization results. The present invention collects existing experimental data and literature data, and preprocesses these data to construct a training set and a validation set, and trains multiple large machine learning models through these training sets and test sets, and determines a performance optimization prediction model according to performance test results. Afterwards, the model is optimized through Bayesian optimization, interpretability analysis, and multi-objective optimization processing, and the nonlinear relationship affecting the performance parameters of cement-based materials can be automatically found, so as to achieve accurate prediction and optimization of the optimal performance of cement-based materials.

[0119] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0120] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.

[0121] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for optimizing cement-based material performance based on machine learning, characterized in that: The cement-based material performance optimization method based on machine learning includes: Acquire historical cement-based material data, and preprocess the historical cement-based material data to obtain a training set and a test set; The historical cement-based material data includes historical cement-based material experimental data and cement-based material literature data; the preprocessing includes data partitioning processing, one-hot code processing and null value interpolation processing; The acquiring of historical cement-based material data and preprocessing of the historical cement-based material data to obtain a training set and a test set specifically includes: Acquire historical cement-based material experimental data and cement-based material literature data, and perform data segmentation processing on the historical cement-based material experimental data and the cement-based material literature data according to the material properties of the cement-based material to obtain multiple material property data sets; Acquire multiple feature classification variables from multiple material property data sets, and perform one-hot coding on the multiple feature classification variables to obtain multiple feature classification vectors; Acquire feature null values ​​in a plurality of the feature classification vectors, perform null value interpolation processing on the feature null values ​​using a KNN algorithm to obtain an initial data set, and divide the initial data set into a training set and a test set; Acquire multiple preset machine learning models, and perform model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model; The obtaining of multiple preset machine learning models, and performing model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model specifically includes: Acquire multiple preset machine learning models, wherein the preset machine learning models include an ANN model, a KNN model, a tree model, and an AdaBoost model; Performing model training on the plurality of preset machine learning models according to the training set using a K-fold cross-validation method to obtain a plurality of first performance test results, and performing model testing on the plurality of preset machine learning models according to the test set to obtain a plurality of second performance test results; Respectively calculating the similarities between the plurality of the first performance test results and the plurality of the second performance test results, and selecting the preset machine learning model corresponding to the first performance test result with the greatest similarity from the plurality of the preset machine learning models, to obtain the initial performance optimization prediction model; Using a transfer learning strategy to fine-tune the initial performance optimization prediction model; The adopting of the transfer learning strategy to fine-tune the initial performance optimization prediction model specifically includes: The model is trained using the performance data of nanomaterials to obtain the influence of nanomaterials on the performance of cement-based materials, wherein the nanomaterials include graphene and nano-titanium dioxide; Fine-tuning the model on limited data on cementitious materials modified by GNAs to capture the unique impact characteristics of GNAs; Unsupervised learning methods are used to introduce synthetic data to enhance training datasets and improve model performance; The unsupervised learning method is used to introduce synthetic data to enhance the training data set and improve the model performance, including: The original data set is widened by using different synthetic data ratios, with the gradient of each widening being 20% ​​of the original data set, and the data amount in the original data set is widened to a maximum of 300% of the original data set; Four different data generation models, namely conditional generative adversarial network model, Gaussian conjugate model, variational autoencoder model and conjugate generative adversarial network model, are used to generate synthetic data to simulate the distribution characteristics of real data. Among them, the Gaussian conjugate model uses the characteristics of Gaussian distribution to generate data; The conditional generative adversarial network generates qualified data samples through adversarial training; The conjugate generative adversarial network combines the Copula function and the generative adversarial network; The variational autoencoder uses the encoding and decoding networks to generate data and learn the potential representation of the data. The generated data is used to expand the training set, while the test set is kept as a subset of the unenlarged real data set. When using it, 5-fold cross validation is used to ensure the authenticity of the evaluation results. Performing model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model; The model optimization process includes Bayesian optimization process, interpretability analysis process and multi-objective optimization process; The performing model optimization processing on the initial performance optimization prediction model to obtain the performance optimization prediction model specifically includes: Using a Bayesian optimization method to perform hyperparameter tuning processing on the initial performance optimization prediction model to obtain a first performance optimization prediction model; Performing interpretability analysis on the first performance optimization prediction model to obtain a second performance optimization prediction model; Performing multi-objective optimization processing on the second performance optimization prediction model to obtain a performance optimization prediction model; The method of using the Bayesian optimization method to perform hyperparameter tuning on the initial performance optimization prediction model to obtain a first performance optimization prediction model specifically includes: Using a Bayesian optimization method to obtain multiple model parameters in the initial performance optimization prediction model, and combining the multiple model parameters to obtain a multidimensional parameter space; Performing Gaussian process processing on the multidimensional parameter space to obtain a probability distribution result of the model parameters; Obtaining a preset acquisition function, and performing parameter point sampling on the model parameter probability distribution result according to the preset acquisition function to obtain a parameter performance metric; Performing parameter updates on the initial performance optimization prediction model according to the parameter performance metric to obtain the first performance optimization prediction model; The performing interpretability analysis on the first performance optimization prediction model to obtain a second performance optimization prediction model specifically includes: Acquire multiple features in the first performance optimization prediction model, calculate multiple quantized prediction errors corresponding to the multiple features, and perform weighted processing on the multiple quantized prediction errors to obtain multiple contribution values ​​corresponding to the multiple features; The contribution value is expressed as: ； in, is the feature subset value, For feature set A subset of is the feature set, is the feature subset value The absolute value of is the factorial symbol, is the number of features in the model; Sort the multiple features in descending order according to contribution value, select the feature with the largest contribution value, and obtain the best contribution feature; Performing feature selection processing and model optimization processing on the first performance optimization prediction model according to the best contribution feature to obtain the second performance optimization prediction model; The performing multi-objective optimization processing on the second performance optimization prediction model to obtain the performance optimization prediction model specifically includes: Acquire multiple evolutionary algorithms, and perform multi-objective optimization processing on the second performance optimization prediction model according to the multiple evolutionary algorithms to obtain a trade-off relationship between different performance indicators in the second performance optimization prediction model; Calculating a Pareto optimal solution according to the trade-off relationship between different performance indicators in the second performance optimization prediction model, and obtaining an optimization parameter selection range according to the Pareto optimal solution; Performing model optimization processing on the second performance optimization prediction model according to the optimization parameter selection range to obtain the performance optimization prediction model; The current cement-based material data is obtained, and the current cement-based material data is input into the performance optimization prediction model, and the cement-based material performance optimization result is output.

2. A cement-based material performance optimization system based on machine learning, characterized in that: The cement-based material performance optimization system based on machine learning includes: A data preprocessing module, used to obtain historical cement-based material data and preprocess the historical cement-based material data to obtain a training set and a test set; The historical cement-based material data includes historical cement-based material experimental data and cement-based material literature data; the preprocessing includes data partitioning processing, one-hot code processing and null value interpolation processing; The acquiring of historical cement-based material data and preprocessing of the historical cement-based material data to obtain a training set and a test set specifically includes: Acquire historical cement-based material experimental data and cement-based material literature data, and perform data segmentation processing on the historical cement-based material experimental data and the cement-based material literature data according to the material properties of the cement-based material to obtain multiple material property data sets; Acquire multiple feature classification variables from multiple material property data sets, and perform one-hot coding on the multiple feature classification variables to obtain multiple feature classification vectors; Acquire feature null values ​​in a plurality of the feature classification vectors, perform null value interpolation processing on the feature null values ​​using a KNN algorithm to obtain an initial data set, and divide the initial data set into a training set and a test set; An initial prediction model building module is used to obtain multiple preset machine learning models, and perform model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model; The obtaining of multiple preset machine learning models, and performing model training and model screening on the multiple preset machine learning models according to the training set and the test set to obtain an initial performance optimization prediction model specifically includes: Acquire multiple preset machine learning models, wherein the preset machine learning models include an ANN model, a KNN model, a tree model, and an AdaBoost model; Performing model training on the plurality of preset machine learning models according to the training set using a K-fold cross-validation method to obtain a plurality of first performance test results, and performing model testing on the plurality of preset machine learning models according to the test set to obtain a plurality of second performance test results; Respectively calculating the similarities between the plurality of the first performance test results and the plurality of the second performance test results, and selecting the preset machine learning model corresponding to the first performance test result with the greatest similarity from the plurality of the preset machine learning models, to obtain the initial performance optimization prediction model; Using a transfer learning strategy to fine-tune the initial performance optimization prediction model; The adopting of the transfer learning strategy to fine-tune the initial performance optimization prediction model specifically includes: The model is trained using the performance data of nanomaterials to obtain the influence of nanomaterials on the performance of cement-based materials, wherein the nanomaterials include graphene and nano-titanium dioxide; Fine-tuning the model on limited data on cementitious materials modified by GNAs to capture the unique impact characteristics of GNAs; Unsupervised learning methods are used to introduce synthetic data to enhance training datasets and improve model performance; The unsupervised learning method is used to introduce synthetic data to enhance the training data set and improve the model performance, including: The original data set is widened by using different synthetic data ratios, with the gradient of each widening being 20% ​​of the original data set, and the data volume in the original data set is widened to a maximum of 300% of the original data set; Four different data generation models, namely conditional generative adversarial network model, Gaussian conjugate model, variational autoencoder model and conjugate generative adversarial network model, are used to generate synthetic data to simulate the distribution characteristics of real data. Among them, the Gaussian conjugate model uses the characteristics of Gaussian distribution to generate data; The conditional generative adversarial network generates qualified data samples through adversarial training; The conjugate generative adversarial network combines the Copula function and the generative adversarial network; The variational autoencoder uses the encoding and decoding networks to generate data and learn the potential representation of the data. The generated data is used to expand the training set, while the test set is kept as a subset of the unenlarged real data set. When using it, 5-fold cross validation is used to ensure the authenticity of the evaluation results. A model performance optimization module, used to perform model optimization processing on the initial performance optimization prediction model to obtain a performance optimization prediction model; The model optimization process includes Bayesian optimization process, interpretability analysis process and multi-objective optimization process; The performing model optimization processing on the initial performance optimization prediction model to obtain the performance optimization prediction model specifically includes: Using a Bayesian optimization method to perform hyperparameter tuning processing on the initial performance optimization prediction model to obtain a first performance optimization prediction model; Performing interpretability analysis on the first performance optimization prediction model to obtain a second performance optimization prediction model; Performing multi-objective optimization processing on the second performance optimization prediction model to obtain a performance optimization prediction model; The method of using the Bayesian optimization method to perform hyperparameter tuning on the initial performance optimization prediction model to obtain a first performance optimization prediction model specifically includes: Using a Bayesian optimization method to obtain multiple model parameters in the initial performance optimization prediction model, and combining the multiple model parameters to obtain a multidimensional parameter space; Performing Gaussian process processing on the multidimensional parameter space to obtain a probability distribution result of the model parameters; Obtaining a preset acquisition function, and performing parameter point sampling on the model parameter probability distribution result according to the preset acquisition function to obtain a parameter performance metric; Performing parameter updates on the initial performance optimization prediction model according to the parameter performance metric to obtain the first performance optimization prediction model; The performing interpretability analysis on the first performance optimization prediction model to obtain a second performance optimization prediction model specifically includes: Acquire multiple features in the first performance optimization prediction model, calculate multiple quantized prediction errors corresponding to the multiple features, and perform weighted processing on the multiple quantized prediction errors to obtain multiple contribution values ​​corresponding to the multiple features; The contribution value is expressed as: ； in, is the feature subset value, For feature set A subset of is the feature set, is the feature subset value The absolute value of is the factorial symbol, is the number of features in the model; Sort the multiple features in descending order according to contribution value, select the feature with the largest contribution value, and obtain the best contribution feature; Performing feature selection processing and model optimization processing on the first performance optimization prediction model according to the best contribution feature to obtain the second performance optimization prediction model; The performing multi-objective optimization processing on the second performance optimization prediction model to obtain the performance optimization prediction model specifically includes: Acquire multiple evolutionary algorithms, and perform multi-objective optimization processing on the second performance optimization prediction model according to the multiple evolutionary algorithms to obtain a trade-off relationship between different performance indicators in the second performance optimization prediction model; Calculating a Pareto optimal solution according to the trade-off relationship between different performance indicators in the second performance optimization prediction model, and obtaining an optimization parameter selection range according to the Pareto optimal solution; Performing model optimization processing on the second performance optimization prediction model according to the optimization parameter selection range to obtain the performance optimization prediction model; The performance optimization result output module is used to obtain current cement-based material data, input the current cement-based material data into the performance optimization prediction model, and output the cement-based material performance optimization result.

3. A terminal, characterized in that: The terminal includes: a memory, a processor, and a cement-based material performance optimization program based on machine learning stored in the memory and executable on the processor. When the cement-based material performance optimization program based on machine learning is executed by the processor, the steps of the cement-based material performance optimization method based on machine learning as described in claim 1 are implemented.

4. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a cement-based material performance optimization program based on machine learning, and when the cement-based material performance optimization program based on machine learning is executed by a processor, the steps of the cement-based material performance optimization method based on machine learning as described in claim 1 are implemented.

Images