AI-enhanced large-mine-doped hydraulic concrete material performance prediction and mix proportion design integrated system

Abstract

The invention provides an AI-enhanced large mine water-doped concrete material performance prediction and mix proportion design integrated system, aiming at the key problem of large mine water-doped concrete material temperature control, the importance of different influence factors such as raw material-structure-environment and the like is comprehensively and quantitatively evaluated; aiming at a complex raw material system of the large mine water-doped industrial concrete, main features of a performance prediction model such as feature dimension reduction and weight calculation are optimized, and meanwhile, a machine learning algorithm combined with a self-attention mechanism is constructed so as to reduce interference of features weakly related to target performance, and the performance regulation and control capability of the large mine water-doped industrial concrete is enhanced. Furthermore, a multi / super multi-target optimization algorithm containing a constraint system is developed and integrated aiming at targets such as material key performance, carbon emission and cost in the preparation process, and a comprehensive and flexible data-driven solution is provided for temperature control, influence factor evaluation, performance regulation and control and proportion optimization of the large-mine water-doped concrete material. Therefore, the overall decision-making and design process is optimized.

Description

Technical Field

[0001] The present invention relates to an integrated system technology for predicting the properties and designing the mix proportion of AI-enhanced large-mineral admixture hydraulic concrete materials, belonging to the technical fields of artificial intelligence and materials science and technology. Background Art

[0002] Large-mineral admixture hydraulic concrete materials refer to concrete materials that add a large amount of mineral admixtures such as slag powder and fly ash to replace part of the cement on the basis of the bulk raw materials of traditional concrete, and achieve better workability, durability and other properties by optimizing parameters such as the water-binder ratio. Among them, mineral admixtures are mostly industrial waste. By partially replacing the key raw material cement with high carbon emissions due to calcination production, etc., the carbon emissions of concrete materials per unit volume can be significantly reduced, while reducing resource waste and promoting the development of circular economy. Due to the unique properties and environmental protection advantages of large-mineral admixture hydraulic concrete materials, they are widely used in the construction of water conservancy and hydropower projects such as water retaining, power generation, flood discharge, water conveyance, and sediment discharge, as well as underground and underwater projects such as coastal sea dikes and deep-buried tunnel linings. Their properties are directly related to the stability and environmental adaptability of construction projects.

[0003] The diverse and complex raw material composition poses challenges to the performance regulation, mix proportion design, and stable control of the material properties after mixing and forming. For example, due to the differences in the chemical compositions of mineral admixtures from different sources, the rate of hydration reaction and the properties of the final products are affected; due to the differences in particle size distribution, shape, and density among complex components, uneven mixing may occur, affecting the workability and density of concrete materials; due to the addition of a large amount of mineral admixtures, the bleeding property of concrete materials will increase to a certain extent, resulting in an increase in the surface water-binder ratio and prone to the formation of through-capillary pores. At the same time, the use of mineral admixtures in large-mineral admixture hydraulic concrete materials changes the thermal properties of concrete materials to a certain extent, making it more susceptible to the influence of the internal and external temperature difference during pouring and forming in large-volume projects. The overall temperature control directly affects the strength development, shrinkage performance of concrete materials themselves, and the durability of engineering structures. In the application scenarios with long cycles and complex environments, the temperature change of concrete may lead to crack generation and performance deterioration, thus threatening the safety of engineering structures. Therefore, the temperature rise control of large-mineral admixture hydraulic concrete materials is particularly crucial.

[0004] However, the traditional performance regulation and mix design methods for large-mine admixture hydraulic concrete materials mainly rely on a large number of tests such as the empirical knowledge, material performance inspection, and structural model inspection in SL 191-2008 "Code for Design of Hydraulic Concrete Structures" and JGJ 55-2011 "Code for Mix Proportion Design of Ordinary Concrete". Empirical knowledge is difficult to accurately analyze material performance and customize scientific mix ratios when facing unique or different engineering environments. Although laboratory trial mixing can provide important information on the performance of concrete materials, these tests usually consume a large amount of time and resources, especially when long-term environmental impacts need to be simulated. The temperature rise control of large-mine admixture hydraulic concrete materials during engineering construction is also usually based on experimental tests and engineering experience. However, the heat release characteristics of large-mine admixture hydraulic concrete materials in the application scenarios of large-volume projects are complex, and traditional empirical methods are often insufficient to meet the temperature rise control requirements of multi-factor coupling. At the same time, due to the large amount of incorporation of complex raw materials such as mineral admixtures, it is difficult to coordinate the mix design of large-mine admixture hydraulic concrete materials to meet multiple objective requirements such as service performance, raw material cost, and unit carbon emissions from raw material production to finished product preparation. Therefore, the traditional method mainly based on tests combined with empirical guidance has obvious limitations in terms of economy, efficiency, and accuracy under the background of increasingly strict and diverse performance requirements for large-mine admixture hydraulic concrete materials. This limitation prompts the engineering community to seek artificial intelligence methods based on data and machine learning, hoping to more scientifically, intelligently, efficiently, and accurately achieve the performance regulation, temperature rise control, and mix design of large-mine admixture hydraulic concrete materials under various complex working conditions. Summary of the Invention

[0005] Object of the Invention: In order to overcome the limitations of traditional methods and more scientifically, intelligently, efficiently, and accurately achieve the temperature control, evaluation of performance influencing factors, performance regulation, and mix ratio optimization of large-mine admixture hydraulic concrete materials, the present invention provides an AI-enhanced integrated system for performance prediction and mix ratio design of large-mine admixture hydraulic concrete materials.

[0006] Technical Solution: An AI-enhanced integrated system for performance prediction and mix ratio design of large-mine admixture hydraulic concrete materials, comprising:

[0007] A data management and analysis module, configured to provide the upload and preprocessing of the large-mine admixture hydraulic concrete materials including mix ratio information, raw material information, environmental information, structural information, performance information, and temperature control-related information;

[0008] A feature engineering analysis module, configured to analyze the importance degree of influencing factors during the temperature control of large-mine admixture hydraulic concrete materials, and support the feature dimensionality reduction, feature weight calculation, influence ranking analysis, and feature parameter screening and extraction of the uploaded large-mine admixture hydraulic concrete material data;

[0009] A performance prediction and analysis module, which is used to construct multiple types of performance prediction models based on the required ratio information - raw material information - environmental information - structural information feature parameters and key target performances after feature processing according to user needs, support machine learning algorithms combined with multiple self-attention mechanisms, and at the same time provide hyperparameter range tuning and process evaluation of the models to optimize multiple types of performance prediction models;

[0010] A ratio optimization and design module, which is used to solve the optimized solution set of the large-mineral admixture hydraulic concrete material ratio according to multiple design objectives selected by the user, including compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage performance, carbon emissions and cost during the material preparation process, as well as constraint conditions, using a penalty function combined with a multi- / super multi-objective optimization algorithm.

[0011] Furthermore, the data management and analysis module includes:

[0012] A data upload unit, which is used for uploading data in multiple formats and naming data sets, aligning the consistency of the ratio information - raw material information - environmental information - structural information - performance information of the large-mineral admixture hydraulic concrete materials uploaded and the data labels of the module, and can be switched and used in parallel for the uploaded large-mineral admixture hydraulic concrete material data sets;

[0013] A data preprocessing unit, which is used for precise and fuzzy data search and query, data screening, outlier detection, data cleaning, and normalization of ratio information - raw material information - environmental information - structural information feature parameters of the uploaded large-mineral admixture hydraulic concrete material data set. Among them, outlier detection supports the Mahalanobis distance and K-nearest neighbor methods, and data normalization uses Z-score standardization;

[0014] A statistical analysis unit, which is used to perform statistical analysis on the structured data of the uploaded large-mineral admixture hydraulic concrete material data set after preprocessing, including Pearson correlation, median, interquartile range and missing rate, and includes one or more of data statistical matrix diagrams, violin diagrams and key statistical summaries for data visualization.

[0015] Furthermore, the ratio information includes cementitious material content, aggregate content, water content and admixture content, the raw material information includes particle size distribution, chemical composition, mineral composition, activity of mineral admixtures and bulk density, the environmental information includes curing temperature and curing humidity, and the structural information includes component shape and component size; the performance information includes: work performance information including slump and spread, mechanical performance information including compressive strength, flexural strength and elastic modulus, and deformation performance information including drying shrinkage and autogenous shrinkage; the temperature control-related information includes concrete surface temperature and concrete center temperature.

[0016] Furthermore, the feature engineering analysis module includes:

[0017] A temperature control analysis unit is used for accurately retrieving important information to be considered when controlling the temperature of large-mineral admixture hydraulic concrete materials including the surface temperature and the center temperature of concrete after data preprocessing, and accurately matching the full-scale characteristic parameters of ratio information - raw material information - environmental information - structural information, and supporting the comprehensive quantitative evaluation of the importance degrees of various different influencing factors such as raw material dosage, raw material characteristics, structural dimensions, and the environment on temperature control by numerical regression, extreme gradient boosting tree, and SHAP methods;

[0018] A feature dimensionality reduction unit is used for reducing the dimension of the overall characteristic parameters of ratio information - raw material information - environmental information - structural information in the large-mineral admixture hydraulic concrete material data by using the principal component analysis method PCA, and supporting the drawing of a PCA biplot and a variance proportion plot for analyzing the feature dimensionality reduction results;

[0019] A feature weight calculation and sorting unit is used for the user to select the characteristic parameters of the uploaded large-mineral admixture hydraulic concrete materials and the target performance to be evaluated for the influence situation according to the requirements, and at the same time supports calculating the influence weights of the characteristic parameters by multiple methods, including random forest, extreme gradient boosting tree, and SHAP; and supports sorting the influence weight values of the large-mineral admixture hydraulic concrete material characteristic parameters and the visual drawing of a radar chart;

[0020] A feature screening and recombination unit is used for screening, extracting, and recombining the characteristic parameters in the data of the uploaded large-mineral admixture hydraulic concrete materials to construct an optimal feature combination for the performance prediction model of the large-mineral admixture hydraulic concrete materials.

[0021] Furthermore, the performance prediction and analysis module includes:

[0022] A prediction model training unit is used for training the structured data of the large-mineral admixture hydraulic concrete materials after preprocessing and feature optimization by using multiple machine learning algorithms and establishing a preliminary performance prediction model with default hyperparameters, where the input features and output targets of the trained model support the user to select according to the requirements, and the machine learning algorithms include a multi-layer perceptron combined with a self-attention mechanism, a support vector machine combined with a self-attention mechanism, a random forest combined with a self-attention mechanism, and an extreme gradient boosting tree combined with a self-attention mechanism;

[0023] The model parameter tuning and prediction unit is used to optimize the hyperparameters of the preliminary performance prediction model of large - mine - admixed hydraulic concrete materials by means of random search or grid search. The search range of the model hyperparameters provides default settings and also supports users to set them according to their needs. It also supports visualizing the progress and intermediate process of parameter tuning of the performance prediction model in the form of a progress bar and a dynamic graph. For multiple types of performance prediction models of large - mine - admixed hydraulic concrete materials after tuning, it supports the input of selected features to optimize the key target performance. At the same time, the multiple types of performance prediction models after tuning can be set as the objective function of large - mine - admixed hydraulic concrete materials in the proportion optimization design module.

[0024] The model evaluation unit is used to support the performance evaluation of the large - mine - admixed hydraulic concrete material performance prediction model during the training process and parameter tuning process, as well as the overall performance evaluation of the preliminary performance prediction model and the optimized performance prediction model of large - mine - admixed hydraulic concrete materials. The evaluation indicators adopt the coefficient of determination and the mean square error. It also supports plotting one or more visual evaluation indicators including bar charts and line charts for model analysis.

[0025] Furthermore, in the prediction model training unit, by embedding the self - attention layer between hidden layers, the network can dynamically adjust the feature weights to construct a multi - layer perceptron combined with the self - attention mechanism; by embedding the self - attention layer in the feature space construction stage, the attention weight distribution in the feature dimension is realized to construct a support vector machine combined with the self - attention mechanism; by embedding the self - attention layer in the feature splitting stage of each decision tree, the feature splitting direction is dynamically adjusted to construct a random forest combined with the self - attention mechanism; by embedding the sample attention layer before each round of boosting iteration, the suppression of abnormal samples during training is realized to construct an extreme gradient boosting tree combined with the self - attention mechanism.

[0026] Furthermore, the proportion optimization design module includes:

[0027] The objective function selection unit is used for users to select design objectives including compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage performance, carbon emissions and cost during the material preparation process according to the design requirements of large - mine - admixed hydraulic concrete materials, forming a combination of design objectives, and supporting the determination of the relationship between the design objectives of large - mine - admixed hydraulic concrete materials. Among them, the optimized multiple types of performance prediction models of large - mine - admixed hydraulic concrete materials constructed by users in the performance prediction analysis module are encapsulated as the objective function of material performance, and the standard carbon emission calculation model of large - mine - admixed hydraulic concrete materials from raw material production to the pre - service stage is encapsulated as the objective function of carbon emissions;

[0028] A constraint setting unit is used for users to set the constraints on the target performance, raw material performance, raw material ratio, total amount of raw materials, and specification requirements of large-mine admixture hydraulic concrete materials according to the parameters in the selected objective function. At the same time, for range constraints, it provides default constraint selections based on the numerical ranges of each data label after preprocessing the uploaded large-mine admixture hydraulic concrete material data, including maximum-minimum constraints, one or more preset upper-bound ratio probability distribution quantile - lower-bound ratio probability distribution quantile constraints.

[0029] An objective optimization unit is used to support the algorithm selection and parameter setting of penalty function combined with multi-objective optimization and ultra-multi-objective optimization. The embedding of the penalty function transforms the constrained ratio optimization of large-mine admixture hydraulic concrete materials according to the objective into an unconstrained solution of the ratio solution set, where the penalty function comes from the specific constraints set by the user. At the same time, it supports automatically matching the algorithm type according to the number of design objectives selected by the user.

[0030] A ratio decision optimization unit is used to perform decision-making ranking on the optimized ratio solution set of large-mine admixture hydraulic concrete materials obtained by using the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), and at the same time supports the reverse normalization of the optimized ratio solution set.

[0031] Furthermore, the objective optimization unit automatically matches the algorithm type according to the number of design objectives selected by the user. Specifically, when the number of design objectives is less than or equal to 3, it matches the multi-objective optimization type algorithm, and when the number of design objectives is greater than 3, it matches the ultra-multi-objective optimization algorithm. Among them, multi-objective optimization includes one or more of non-dominated sorting genetic algorithm, decomposition-based multi-objective evolutionary algorithm, multi-objective particle swarm optimization algorithm, multi-objective sand cat optimization algorithm, multi-objective black-winged kite optimization algorithm, etc. Ultra-multi-objective optimization includes one or more of constrained ultra-multi-objective optimization evolutionary algorithm with enhanced mating and environmental selection, two-stage evolutionary ultra-multi-objective optimization algorithm, etc.

[0032] Furthermore, when using the penalty function combined with multi- / ultra-multi-objective optimization algorithm to solve the optimal ratio solution set of large-mine admixture hydraulic concrete materials, through the penalty function, the violation degrees of inequalities and equalities of the target performance, raw material performance, raw material ratio, total amount of raw materials, specification requirements, etc. of large-mine admixture hydraulic concrete materials are incorporated into the objective function to construct a modified objective function vector F ′ (x) = [f 1 (x) + P(x), f 2 (x) + P(x),..., f m (x) + P(x)], where P(x) is the penalty term, quantifying the violation degree of all constraint conditions, and f i (x) is the i-th objective function, i = 1, 2,..., m, and m is the number of design objectives to be optimized; P(x) = Σ j α j ·max(0, g j(x)) ∧ β + Σ k γ k ·∣h k (x)∣ ∧ β, where α j is the penalty weight coefficient of the j-th inequality constraint g j (x); γ k is the penalty weight coefficient of the k-th equality constraint h k (x); β is the penalty exponent; for hierarchical penalties of constraints, higher penalty weight coefficients are assigned to critical constraints.

[0033] Furthermore, the system further includes a project collaboration management module for multiple users to collaborate on the same work project, integrating version control and change tracking functions.

[0034] Advantageous effects: By adopting the above technical solutions, the present invention has the following advantageous effects:

[0035] In the integrated system for predicting the properties and designing the mix proportion of large-ore admixture hydraulic concrete enhanced by AI provided by the present invention, data management and analysis module, feature engineering analysis module, performance prediction analysis module, mix proportion optimization design module, etc. are comprehensively utilized. In terms of data engineering, it realizes the uploading, preprocessing, and statistical analysis of data such as heterogeneous raw material information, mix proportion information, environmental information, structural information, and performance information of large-ore admixture hydraulic concrete materials, simplifying the complexity of data collection, preprocessing, and statistical analysis of complex concrete material systems; in terms of feature engineering, it supports the analysis of the importance degree of factors affecting the temperature control of large-ore admixture hydraulic concrete materials, and at the same time supports feature dimensionality reduction, feature weight calculation of the uploaded data, mining the influence of various features of large-ore admixture hydraulic concrete materials on the target performance, and selecting the main features of the optimal performance prediction model; in terms of performance prediction, a machine learning algorithm combined with a self-attention mechanism is constructed to reduce the interference of problems such as high feature dimensionality and strong nonlinearity of large-ore admixture hydraulic concrete materials, supporting users to build prediction models according to their needs using the required features and targets. After hyperparameter optimization and process evaluation, the optimized model can efficiently predict the key target performances of large-ore admixture hydraulic concrete materials in terms of mechanics, workability, deformation, etc.; in terms of mix proportion design, it supports users to select design targets such as compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage, etc. and carbon emissions and costs during the material preparation process according to their needs, set the constraint ranges of target performances, raw material ratios, specification requirements, etc., and solve the optimized mix proportion of large-ore admixture hydraulic concrete materials through the constructed algorithms of penalty function combined with multi-objective optimization and ultra-multi-objective optimization. The mix proportion solving process includes dynamic adjustment of penalty weights and hierarchical penalty of constraints, providing an effective decision-making reference for mix proportion design. The present invention provides a comprehensive and flexible data-driven solution for the temperature control, performance influence factor evaluation, performance regulation, and mix proportion optimization of large-ore admixture hydraulic concrete materials, optimizing the overall decision-making and design process. Description of the Drawings

[0036] Figure 1 It is the overall architecture diagram of the embodiment of the present invention.

[0037] Figure 2 It is the flowchart of the data management and analysis module of the embodiment of the present invention.

[0038] Figure 3 It is the flowchart of the feature engineering analysis module of the embodiment of the present invention.

[0039] Figure 4 It is the flowchart of the performance prediction analysis module of the embodiment of the present invention.

[0040] Figure 5 It is the flowchart of the mix proportion optimization design module of the embodiment of the present invention. Detailed Embodiments

[0041] The present invention will be further illustrated below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent modifications made by those skilled in the art to the present invention fall within the scope defined by the appended claims of this application.

[0042] See Figure 1 , which is the overall architecture diagram of the integrated system for performance prediction and mix ratio design of AI-enhanced large-aggregate admixture hydraulic concrete materials in an embodiment of the present invention, including:

[0043] The interaction layer includes a user interface interaction module, which is mainly used to implement the visual design of the system page, the interface-based point selection operation of the user, and friendly interactive use, etc.

[0044] The data layer includes a data management and analysis module, which is mainly used to upload various data formats of large-aggregate admixture hydraulic concrete materials. The uploaded data includes, but is not limited to, mix ratio information such as cementitious material content, aggregate content, water content, admixture content, etc., raw material information such as particle size distribution, chemical composition, mineral composition, activity of mineral admixtures, bulk density, etc., environmental information such as curing temperature, curing humidity, etc., structural information such as component shape, component size, etc., workability information such as slump, spread, etc., mechanical properties information such as compressive strength, flexural strength, elastic modulus, etc., deformation properties information such as drying shrinkage, autogenous shrinkage, etc., and important information that needs to be considered for temperature control when large-aggregate admixture hydraulic concrete materials are applied to the casting of large-volume projects, such as concrete surface temperature, concrete center temperature, etc. And it provides preprocessing operations such as precise and fuzzy search queries, data screening, outlier detection, and normalization of characteristic parameters such as mix ratio information - raw material information - environmental information - structural information for large-aggregate admixture hydraulic concrete materials. At the same time, based on statistical analysis, the data information of large-aggregate admixture hydraulic concrete materials is visualized in the form of drawing, key statistical summaries, etc.

[0045] The business layer includes a feature engineering analysis module, a performance prediction analysis module, and a mix ratio optimization design module, which are mainly used to analyze the importance degree of influencing factors when large-aggregate admixture hydraulic concrete materials are subjected to temperature control, mine out the influence of each feature on the target performance of large-aggregate admixture hydraulic concrete materials, and select the main features of the optimal performance prediction model; at the same time, it is used to implement the construction of an optimized multi-class performance prediction model for large-aggregate admixture hydraulic concrete materials by using a machine learning algorithm combined with multiple self-attention mechanisms, and efficiently predict and analyze the key target performance of large-aggregate admixture hydraulic concrete materials in terms of mechanics, workability, deformation, etc.; and it is used to implement the mix ratio optimization of large-aggregate admixture hydraulic concrete materials based on the selection of design objectives, the setting of constraint ranges such as target performance - raw material performance - specification requirements, and the selection of an optimization algorithm combined with a penalty function constructed.

[0046] Expansion layer, including project collaboration management module, extensibility and multi-language support module, mainly used to support multiple users to collaborate on the same work project for large mine admixture concrete materials and integrate version control and change tracking functions, and at the same time used to provide a plug-in system and public API and allow users to customize functions and integrate external tools, as well as used to expand support for different languages and provide usage tutorials and practice guides.

[0047] Among them, the interaction layer provides a zero-code, user-friendly operation for the data layer, business layer and expansion layer; the data layer provides a data stream of large mine admixture concrete materials including mix ratio information - raw material information - environmental information - structural information - performance information, etc. for the feature engineering analysis module, performance prediction analysis module, and mix ratio optimization design module in the business layer; the feature engineering analysis module in the business layer can provide feature decisions of large mine admixture concrete materials for the performance prediction analysis module and mix ratio optimization design module, and the performance prediction analysis module provides the objective function after encapsulation of multiple performance prediction models of large mine admixture concrete materials for the mix ratio optimization design module; the project collaboration management module, extensibility and multi-language support module in the expansion layer provide collaboration and expansion support for the data layer and business layer.

[0048] Specifically, in the embodiments of the present invention, the user interface interaction module includes the following units:

[0049] Interface configuration unit: Set the font through matplotlib, configure the initial state of the Streamlit page title, navigation bar, layout method and sidebar, use the container, column and sidebar elements of Streamlit to organize the page content, and display relevant logos and pictures on the page.

[0050] Visual configuration and interaction unit: Use CSS and Markdown to configure the visual style and set multiple elements for user interaction, such as text input boxes, file uploaders, selection boxes and buttons, etc.

[0051] See Figure 2 , which is the flowchart of the data management and analysis module in the embodiments of the present invention, including the following units:

[0052] Data upload unit: Support users to upload data in multiple formats such as csv or xlsx and name the data set, and at the same time support aligning the mix ratio information - raw material information - environmental information - structural information - performance information, etc. of the large mine admixture concrete materials uploaded with the data labels of the module, and can switch and use the uploaded large mine admixture concrete material data sets in parallel;

[0053] Data preprocessing unit: It supports preprocessing operations such as precise and fuzzy data search queries, data screening, outlier detection, data cleaning, and normalization of characteristic parameters such as mix ratio information - raw material information - environmental information - structural information for the uploaded large - mine - admixture hydraulic concrete material dataset. Among them, outlier detection supports two methods: Mahalanobis distance and K - Nearest Neighbor. Z - score standardization is used for data normalization;

[0054] The calculation formula of Mahalanobis distance is: where d is the Mahalanobis distance, x is the feature vector of the large - mine - admixture hydraulic concrete material sample point, μ is the mean vector of the dataset, and Σ is the covariance matrix of the dataset.

[0055] The distance calculation in K - Nearest Neighbor uses Euclidean distance, and the calculation formula is: where x i and x j are two characteristic - performance sample points of the large - mine - admixture hydraulic concrete material respectively. Their feature vectors are (x i1 , x i2 , …, x id ) and (x j1 , x j2 , …, x jd ) respectively, and d is the feature dimension of the sample.

[0056] The calculation formula of Z - score standardization is: where Z is the value after normalization of the large - mine - admixture hydraulic concrete material characteristic information, x is the value before normalization of the large - mine - admixture hydraulic concrete material characteristic information, μ is the sample mean, and σ is the sample standard deviation.

[0057] Statistical analysis unit: It supports performing statistical analyses such as Pearson correlation, median, interquartile range, and missing rate on the pre - processed structured data of the uploaded large - mine - admixture hydraulic concrete material dataset, and visualizing the data in ways such as data statistical matrix diagrams, violin diagrams, etc., as well as key statistical summaries;

[0058] Among them, the execution result after the data upload unit in the data management and analysis module is to enable the parallel use of the large - mine - admixture hydraulic concrete material dataset. The execution result after the data preprocessing unit is to construct a structured large - mine - admixture hydraulic concrete material dataset. Then, the execution result after the statistical analysis unit is to achieve the overall analysis and visualization of the statistical results of the large - mine - admixture hydraulic concrete material data.

[0059] See Figure 3 , which is the flowchart of the feature engineering analysis module of this invention embodiment, including the following units:

[0060] Temperature Control Analysis Unit: It supports the precise retrieval of important information such as the surface temperature of concrete and the center temperature of concrete after data preprocessing for temperature control of large-mineral admixture hydraulic concrete materials, as well as the precise matching of all characteristic parameters such as mix ratio information, raw material information, environmental information, and structural information. At the same time, it supports comprehensive quantitative evaluation of the importance degree of different influencing factors such as raw material dosage, raw material characteristics, structural dimensions, and environment on temperature control by methods such as numerical regression, Extreme Gradient Boosting (XGBoost), and SHAP;

[0061] Feature Dimensionality Reduction Unit: It supports dimension reduction of the overall characteristic parameters such as mix ratio information, raw material information, environmental information, and structural information in the data of large-mineral admixture hydraulic concrete materials by using the Principal Component Analysis (PCA), and supports drawing the PCA biplot and variance proportion plot for the analysis of the feature dimensionality reduction results to assist in the decision-making of feature selection;

[0062] Feature Weight Calculation and Ranking Unit: It supports users to select the characteristic parameters of the uploaded large-mineral admixture hydraulic concrete materials and the target performance to be evaluated according to their needs, and at the same time supports multiple methods for calculating the influence weights of the characteristic parameters, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), SHAP, etc.; and supports the ranking and radar chart visualization of the influence weight values of the characteristic parameters of large-mineral admixture hydraulic concrete materials to assist in the analysis and decision-making of feature selection;

[0063] Feature Screening and Recombination Unit: It supports screening, extracting, and recombining the characteristic parameters in the data of the uploaded large-mineral admixture hydraulic concrete materials to construct an optimal feature combination for the performance prediction model of large-mineral admixture hydraulic concrete materials;

[0064] Among them, the execution result after the Temperature Control Analysis Unit in the Feature Engineering Analysis Module is the quantification of the importance degree of influencing factors during the temperature control of large-mineral admixture hydraulic concrete materials. The execution result after the Feature Dimensionality Reduction Unit is the visualization of the feature dimensionality reduction result of large-mineral admixture hydraulic concrete materials. Then, the execution result after the Feature Weight Calculation and Ranking Unit is the visualization of the feature weight values and rankings of large-mineral admixture hydraulic concrete materials. Finally, the execution result after the Feature Screening and Recombination Unit is the optimal feature combination for the performance prediction model of large-mineral admixture hydraulic concrete materials.

[0065] See Figure 4 , which is the flowchart of the performance prediction analysis module of the embodiment of the present invention, including the following units:

[0066] Prediction model training unit: It supports using a variety of developed machine learning algorithms to train the structured data of large-aggregate admixture hydraulic concrete materials after preprocessing and feature optimization, and establish a preliminary performance prediction model with default hyperparameters. The input features and output targets of the trained model support users to select according to their needs. The machine learning algorithms include multi-layer perceptron (MLP) combined with self-attention mechanism, support vector machine (SVM) combined with self-attention mechanism, random forest (RF) combined with self-attention mechanism, and extreme gradient boosting tree (XGBoost) combined with self-attention mechanism to address the problems of high feature dimension and strong non-linearity of large-aggregate admixture hydraulic concrete materials;

[0067] The calculation formula for the attention weight is: where q i is the query vector, k j is the key vector, n is the number of key vectors, and d k is the dimension of the key vector.

[0068] Specifically, the implementation of MLP combined with self-attention mechanism is as follows: By embedding the self-attention layer between hidden layers, the network can dynamically adjust the feature weights to construct MLP combined with self-attention mechanism.

[0069] The implementation of SVM combined with self-attention mechanism is as follows: By embedding the self-attention layer in the feature space construction stage, the attention weight distribution on the feature dimension is realized to construct SVM combined with self-attention mechanism.

[0070] The implementation of RF combined with self-attention mechanism is as follows: By embedding the self-attention layer in the feature splitting stage of each decision tree, the feature splitting direction is dynamically adjusted to construct RF combined with self-attention mechanism.

[0071] The implementation of XGBoost combined with self-attention mechanism is as follows: By embedding the sample attention layer before each round of boosting iteration, the suppression of abnormal samples during training is realized to construct XGBoost combined with self-attention mechanism.

[0072] Model parameter tuning and prediction unit: It supports hyperparameter optimization of the preliminary performance prediction model for large-aggregate admixture hydraulic concrete materials by means of random search or grid search. The search range of model hyperparameters provides default settings and also supports users to set according to their needs. It also supports visualizing the progress and intermediate process of parameter tuning of the performance prediction model in the form of a progress bar and a dynamic graph; For the multi-class performance prediction models of large-aggregate admixture hydraulic concrete materials after tuning, it supports the input of selected features to optimize the key target performances in aspects such as material mechanics, workability, and deformation. At the same time, the multi-class performance prediction models after tuning can be set as the objective function of large-aggregate admixture hydraulic concrete materials in the mix proportion optimization design module;

[0073] Model evaluation unit: It supports the performance evaluation of the large-mine admixture hydraulic concrete material property prediction model during the training process and parameter tuning process, as well as the overall performance evaluation of the preliminary performance prediction model and the optimized performance prediction model of the large-mine admixture hydraulic concrete material. The evaluation index adopts the coefficient of determination R 2 and the mean squared error MSE. At the same time, it supports drawing visual evaluation indexes such as bar charts and line charts for model analysis;

[0074] The calculation formula of the coefficient of determination is: where Y i ′ represents the model prediction value, and Y i represents the actual value, represents the average value of the actual values, and n represents the number of data samples for constructing the large-mine admixture hydraulic concrete material property prediction model.

[0075] The calculation formula of the mean squared error is: where Y i represents the model prediction value, Y represents the actual value, and n represents the number of data samples for constructing the large-mine admixture hydraulic concrete material property prediction model.

[0076] Among them, the execution result after the prediction model training unit in the performance prediction analysis module is to train the model and establish a preliminary large-mine admixture hydraulic concrete material property prediction model. The execution result after the model parameter tuning and prediction unit is to establish a multi-class property prediction model of the large-mine admixture hydraulic concrete material after hyperparameter tuning and support the prediction of the selected features and targets such as mechanics, workability, and deformation. Then, the execution result after the model evaluation unit is the performance evaluation during the training process and the overall performance evaluation of the large-mine admixture hydraulic concrete material property prediction model.

[0077] See Figure 5 , which is the flowchart of the ratio optimization design module of the embodiment of the present invention, including the following units:

[0078] Objective function selection unit: It supports users to select performance such as compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage, etc. and design goals such as carbon emissions and cost during the material preparation process according to the design requirements of the large-mine admixture hydraulic concrete material to form a combination of design goals. At the same time, it supports determining the high or low relationship between the numerical requirements of the design goals of the large-mine admixture hydraulic concrete material. Among them, the optimized multi-class property prediction model of the large-mine admixture hydraulic concrete material constructed by the user in the performance prediction analysis module is encapsulated as the objective function of the material properties, and the standard carbon emission calculation model of the large-mine admixture hydraulic concrete material from raw material production to the pre-service stage is encapsulated as the objective function of carbon emissions;

[0079] Constraint Setting Unit: It supports users to set inequality and equality constraints on the target performance, raw material performance, raw material ratio, total amount of raw materials, specification requirements, etc. of large-mine admixture hydraulic concrete materials according to the parameters in the selected objective function. At the same time, for the inequality range constraints, it provides default constraint selections based on the numerical ranges of each data label after the data preprocessing of the uploaded large-mine admixture hydraulic concrete materials, including maximum - minimum constraint, upper boundary 90% probability distribution quantile - lower boundary 10% probability distribution quantile constraint, upper boundary 75% probability distribution quantile - lower boundary 25% probability distribution quantile constraint;

[0080] Objective Optimization Unit: It supports the selection and parameter setting of algorithms that combine multiple constructed penalty functions with multi-objective optimization and ultra-multi-objective optimization to deal with the problem of optimizing the mix proportion of large-mine admixture hydraulic concrete materials with constraints, where the penalty function comes from the specific constraints set by users; at the same time, it supports automatically matching the algorithm type according to the number of design objectives selected by users. When the number of design objectives is less than or equal to 3, it matches the multi-objective optimization type algorithm, and when the number of design objectives is greater than 3, it matches the ultra-multi-objective optimization algorithm. Among them, multi-objective optimization includes non-dominated sorting genetic algorithm (NSGA-II), multi-objective evolutionary algorithm based on decomposition (MOEA / D), multi-objective particle swarm optimization algorithm (MOPSO), multi-objective sand cat optimization algorithm (MOSCSO), multi-objective black-winged kite optimization algorithm (MOBKA), and ultra-multi-objective optimization includes constrained multi / ultra-multi-objective optimization evolutionary algorithm with enhanced mating and environmental selection (CMME), two-stage evolutionary ultra-multi-objective optimization algorithm (MaOEA-IT);

[0081] Specifically, the mathematical expression of the problem of optimizing the mix proportion of large-mine admixture hydraulic concrete materials with constraints is as follows:

[0082] minF(x)=[f 1 (x),f 2 (x),…,f m (x)]

[0083] s.t.g j (x)≤0,j=1,…,p

[0084] h k (x)=0,k=1,…,q

[0085]

[0086] Among them, x is the decision variable vector, representing the mix proportion parameters of large-mine admixture hydraulic concrete materials; F(x) is the objective function vector, including m design objectives to be optimized (such as performance such as compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage, etc. and carbon emissions and cost during the material preparation process); f i(x) is the i-th objective function, where i = 1, 2, …, m; g j (x) is the j-th inequality constraint function; h k (x) is the k-th equality constraint function; p is the total number of inequality constraints; q is the total number of equality constraints; n is the dimension of decision variables.

[0087] By means of penalty functions, the violation degrees of inequalities and equality constraints such as the target performance, raw material performance, raw material ratio, total amount of raw materials, and specification requirements of large-mine blended hydraulic concrete materials are incorporated into the objective function, and a modified objective function vector is constructed: F ′ (x) = [f 1 (x) + P(x), f 2 (x) + P(x),..., f m (x) + P(x)]. Where F ′ (x) is the modified objective function vector; P(x) is the penalty term, which quantifies the violation degrees of all constraint conditions.

[0088] The specific mathematical expression of the penalty term P(x) is defined as: P(x) = Σ j α j · max(0, g j (x)) ∧ β + Σ k γ k · ∣h k (x) ∣ ∧ β, where α j is the penalty weight coefficient of the j-th inequality constraint; γ k is the penalty weight coefficient of the k-th equality constraint; β is the penalty exponent; max(0, g j (x)) is the calculation of the violation degree of the inequality constraint, and the penalty is imposed only when g j (x) > 0, that is, when the constraint is violated; ∣h k (x) ∣ is the calculation of the violation degree of the equality constraint, and the absolute value is taken to measure the deviation.

[0089] The developed penalty function combined with multi-objective optimization and ultra-multi-objective optimization algorithms can realize dynamic adjustment of penalty weights, that is, adjust α j and γ k according to the number of iterations during the optimization solution. In the initial stage, the infeasible region of the optimal mix ratio of large-mine blended hydraulic concrete materials is explored with low penalties, and in the later stage, high penalties are used to guide the convergence to the feasible region; at the same time, the algorithm can also realize hierarchical penalties for constraints, that is, higher α j is given to the key constraints of large-mine blended hydraulic concrete materials (such as the constraints on factors with higher importance for controlling the surface temperature and center temperature of concrete obtained by the temperature control analysis unit), and lower α j is given to the secondary constraints, reflecting the engineering priorities.

[0090] Ratio decision optimization unit: It supports using the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) to perform decision ranking on the optimized ratio solution set of the large ore admixture hydraulic concrete materials to be solved, and at the same time supports the reverse normalization of the optimized ratio solution set;

[0091] The basic formula of TOPSIS is: where R i represents the similarity between the i-th solution and the ideal solution, A i represents the sum of the attribute values of the i-th solution, M i represents the minimum value of the sum of the attribute values of all solutions, M max represents the maximum value of the sum of the attribute values of all solutions, M min represents the minimum value of the sum of the attribute values of all solutions.

[0092] Among them, the execution result after the objective function selection unit in the ratio optimization design module is the determination of the relationship between the design objectives of the large ore admixture hydraulic concrete materials and the design objectives. The execution result after the constraint setting unit is the determination of the constraint settings such as target performance, raw material performance, raw material ratio, and specification requirements. Then, the execution result after the objective optimization unit is the matching and determination of the multi-objective optimization and ultra-multi-objective optimization algorithms and the determination of the parameters of the optimization algorithms. Then, the execution result after the ratio decision optimization unit is the optimized ratio solution set of the large ore admixture hydraulic concrete materials after decision ranking.

Claims

1. An AI-enhanced integrated system for predicting the performance of large-scale concrete materials and designing the mix ratio, characterized in that: include: The data management and analysis module is used to upload and preprocess the information related to the concrete materials of the large mine, including the proportion information, raw material information, environmental information, structural information, performance information and temperature control; Feature engineering analysis module, used to analyze the importance of factors affecting temperature control of large-scale concrete materials, supports feature dimension reduction, feature weight calculation, influence ranking analysis, and feature parameter screening and extraction of uploaded large-scale concrete material data; The performance prediction analysis module is used to build multiple performance prediction models based on user needs using the required ratio information, raw material information, environmental information, structural information characteristic parameters and key target performance after feature processing. It supports machine learning algorithms combined with multiple self-attention mechanisms, and provides model hyperparameter range tuning and process evaluation to optimize multiple performance prediction models. The mix optimization design module is used to solve the optimal solution set of the material mix ratio of large-scale water-mixed concrete based on multiple design objectives selected by the user, including compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage performance, carbon emissions in the material preparation process, and cost, as well as constraints, using penalty functions combined with multi / super multi-objective optimization algorithms.

2. According to claim 1, an AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water is characterized in that: The data management and analysis module comprises: The data upload unit is used for uploading data in various formats and naming data sets, as well as aligning the uploaded large-scale concrete material ratio information, raw material information, environmental information, structural information, and performance information with the data labels of the modules, and can switch and use the uploaded large-scale concrete material data sets in parallel; The data preprocessing unit is used for accurate and fuzzy data search and query, data screening, outlier detection, data cleaning, and normalization of the characteristic parameters of the ratio information, raw material information, environmental information, and structural information of the uploaded large-scale water-mixed concrete material data set. The outlier detection supports the Mahalanobis distance and K nearest neighbor methods, and the data normalization adopts Z-score standardization; The statistical analysis unit is used to perform statistical analysis on the pre-processed structured data of the uploaded large-scale concrete material dataset, including Pearson correlation, median, interquartile range and missing rate, and to visualize the data in one or more ways including data statistical matrix diagram, violin diagram and key statistical summary.

3. According to claim 1, an AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water is characterized in that: The mix ratio information includes cementitious material content, aggregate content, water content and admixture content; the raw material information includes particle size grading, chemical composition, mineral composition, mineral admixture activity and bulk density; the environmental information includes curing temperature and curing humidity; the structural information includes component shape and component size; the performance information includes: working performance information including slump and expansion, mechanical property information including compressive strength, flexural strength and elastic modulus, deformation performance information including drying shrinkage and autogenous shrinkage; the temperature control related information includes concrete surface temperature and concrete center temperature.

4. According to claim 1, an AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water is characterized in that: The feature engineering analysis module comprises: The temperature control analysis unit is used to accurately retrieve important information that needs to be considered when controlling the temperature of large-scale concrete materials, including the surface temperature of concrete and the center temperature of concrete after data preprocessing, and accurately match the full range of characteristic parameters of proportion information, raw material information, environmental information and structural information. It supports numerical regression, extreme gradient boosting tree and SHAP methods to comprehensively and quantitatively evaluate the importance of various influencing factors in the environment, including raw material dosage, raw material characteristics, structural dimensions and environment, for temperature control; The feature dimension reduction unit is used to reduce the dimension of the overall feature parameters of the ratio information, raw material information, environmental information and structural information in the large-scale water-mixed concrete material data using the principal component analysis method PCA, and supports the drawing of PCA biplots and variance ratio plots to analyze the feature dimension reduction results; The feature weight calculation and sorting unit is used for users to select the uploaded feature parameters of the large-scale water-mixed concrete materials and the target performance of the impact situation to be evaluated according to their needs. It also supports multiple methods for calculating the impact weights of feature parameters, including random forest, extreme gradient boosting tree and SHAP; and supports the sorting of the impact weight values ​​of the feature parameters of the large-scale water-mixed concrete materials and the visual drawing of radar charts; The feature screening and recombination unit is used to screen, extract and recombine the feature parameters in the uploaded large-scale water-mixed concrete material data, and construct the optimal feature combination of the large-scale water-mixed concrete material performance prediction model.

5. According to claim 1, an AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water is characterized in that: The performance prediction and analysis module comprises: The prediction model training unit is used to use a variety of machine learning algorithms to train the structured data of large-scale concrete materials after preprocessing and feature optimization, and to establish a preliminary performance prediction model with default hyperparameters. The input features and output targets of the trained model support users to select according to their needs. The machine learning algorithms include multi-layer perceptron combined with self-attention mechanism, support vector machine combined with self-attention mechanism, random forest combined with self-attention mechanism, and extreme gradient boosting tree combined with self-attention mechanism; The model parameter tuning and prediction unit is used to optimize the hyperparameters of the preliminary performance prediction model of large-scale water-mixed concrete materials by random search or grid search. The search range of the model hyperparameters provides default settings and also supports users to set them according to their needs. It also supports visualization of the progress and intermediate process of parameter tuning of the performance prediction model in the form of progress bars and dynamic graphs. The multi-category performance prediction model of large-scale water-mixed concrete materials after tuning supports the input of selected features to optimize the key target performance of the prediction. At the same time, the tuned multi-category performance prediction model can be set as the objective function of large-scale water-mixed concrete materials in the mix optimization design module. The model evaluation unit is used to support the performance evaluation of the large-scale water-mixed concrete material performance prediction model in the training process and parameter optimization process, as well as the overall performance evaluation of the large-scale water-mixed concrete material preliminary performance prediction model and the optimized performance prediction model. The evaluation indicators use the determination coefficient and the mean square error, and support the drawing of one or more visual evaluation indicators including bar charts and line charts for model analysis.

6. The AI-enhanced integrated system for predicting material properties and mix design of large-scale concrete mixed with water according to claim 1 is characterized in that: In the prediction model training unit, by embedding the self-attention layer between hidden layers, the network can dynamically adjust the feature weights to construct a multi-layer perceptron combined with a self-attention mechanism; by embedding the self-attention layer in the feature space construction stage, the attention weight distribution on the feature dimension is realized to construct a support vector machine combined with a self-attention mechanism; by embedding the self-attention layer in the feature splitting stage of each decision tree, the feature splitting direction is dynamically adjusted to construct a random forest combined with a self-attention mechanism; by embedding the sample attention layer before each round of boosting iteration process, the suppression of abnormal samples during training is realized to construct an extreme gradient boosting tree combined with a self-attention mechanism.

7. The AI-enhanced integrated system for predicting material properties and mix design of large-scale concrete mixed with water according to claim 1 is characterized in that: The ratio optimization design module includes: The objective function selection unit is used for users to select design objectives including compressive strength, elastic modulus, slump, autogenous shrinkage, drying shrinkage performance, and carbon emissions and costs in the material preparation process according to the design requirements of large-scale water-mixed concrete materials, to form a design objective combination, and to support the determination of the relationship between the design objectives of large-scale water-mixed concrete materials. The optimized multi-category performance prediction model of large-scale water-mixed concrete materials constructed by users in the performance prediction and analysis module is encapsulated as the objective function of material performance, and the standard carbon emission calculation model of large-scale water-mixed concrete materials from raw material production to the service stage is encapsulated as the objective function of carbon emissions; The constraint setting unit is used for users to set the target performance, raw material performance, raw material ratio, total raw material amount and specification requirements of large-scale water-mixed concrete materials according to the parameters in the selected objective function. At the same time, for range constraints, default constraint selection is provided based on the numerical range of each data label after the uploaded large-scale water-mixed concrete material data is preprocessed, including maximum-minimum constraint, one or more preset upper boundary ratio probability distribution quantile-lower boundary ratio probability distribution quantile constraints; The target optimization unit is used to support the algorithm selection and parameter setting of penalty function combined with multi-objective optimization and super-multi-objective optimization. The embedding of penalty function transforms the target-based constrained proportion optimization of large-scale water-mixed concrete materials into an unconstrained solution of the proportion solution set, where the penalty function comes from the specific constraints set by the user; at the same time, it supports automatic matching of algorithm types according to the number of design objectives selected by the user; The proportion decision optimization unit is used to make decision sorting for the optimized proportion solution set of large-scale water-mixed concrete materials by using a sorting method close to the ideal solution, and supports the reverse normalization of the optimized proportion solution set.

8. The AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water according to claim 7 is characterized in that: The target optimization unit automatically matches the algorithm type according to the number of design targets selected by the user, specifically: when the design target is less than or equal to 3, it matches the multi-target optimization type algorithm; when the design target is greater than 3, it matches the super-target optimization algorithm, wherein the multi-target optimization includes one or more of a non-dominated sorting genetic algorithm, a decomposition-based multi-target evolutionary algorithm, a multi-target particle swarm optimization algorithm, a multi-target sand cat optimization algorithm, and a multi-target black kite optimization algorithm; and the super-target optimization includes one or more of a constrained super-target optimization evolutionary algorithm with enhanced mating and environmental selection and a two-stage evolutionary super-target optimization algorithm.

9. The AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water according to claim 1 is characterized in that: The penalty function is combined with multi-objective optimization algorithm to solve the optimization solution of the material ratio of large-scale water-mixed concrete. The penalty function is used to integrate the inequality and equality constraint violation degree of large-scale water-mixed concrete materials including target performance, raw material performance, raw material ratio, total raw material amount and specification requirements into the objective function, and the modified objective function vector F is constructed. ′ (x)=[f1(x)+P(x),f2(x)+P(x),...,f m (x)+P(x)], P(x) is the penalty term, which quantifies the degree of violation of all constraints, where f i (x) is the i-th objective function, i = 1, 2, ..., m, m is the number of design objectives to be optimized; P(x) = Σ j α j max(0,g j (x) ∧ β+Σ k γ k ·∣h k (x)∣ ∧ β, where α j is the jth inequality constraint g j The penalty weight coefficient of (x); γ k is the kth equality constraint h k (x) is the penalty weight coefficient; β is the penalty index; for the hierarchical penalty of constraints, a higher penalty weight coefficient is assigned to the key constraints.

10. The AI-enhanced integrated system for predicting material properties and designing mix ratios of large-scale concrete mixed with water according to claim 1 is characterized in that: Also included is a project collaboration management module for multiple users to collaborate on the same work project, integrating version control and change tracking capabilities.

Images