Method for optimizing mixture ratio of recycled aggregate concrete, storage medium and equipment

Abstract

The application discloses a recycled aggregate concrete mix proportion optimization method, a storage medium and equipment, and relates to a concrete mix proportion optimization method, a storage medium and equipment. In order to solve the problems that the current recycled aggregate concrete mix proportion optimization method does not consider cost, carbon emission and the like, and the problems that the current recycled aggregate concrete mix proportion optimization method cannot perform multiple target optimization under the premise of the optimal mix proportion, the application takes the maximum compressive strength of the recycled aggregate concrete predicted by a machine learning model as a first target function, takes the minimum RAC cost, the minimum RAC carbon emission and the minimum RAC implicit energy consumption as second to fourth target functions of the recycled aggregate concrete based on the composition of the recycled aggregate concrete, and then performs multiple target optimization based on actual working conditions by using a CMOPSO algorithm.

Description

Technical Field

[0001] This invention relates to a method, storage medium, and equipment for optimizing concrete mix proportions. Background Technology

[0002] As the most common building material, concrete has seen a surge in demand for raw materials such as natural sand and gravel due to urbanization. Furthermore, rapid urban development generates a large amount of construction waste (CDW). Utilizing CDW to produce recycled aggregate (RA) to replace natural aggregate (NA) can reduce environmental pollution and resource shortages caused by NA extraction, thus achieving green and sustainable development in the construction industry.

[0003] Compared to natural aggregate (NA), recycled aggregate (RA) has a higher water absorption rate and lower apparent density, resulting in generally weaker mechanical properties. To promote the application of recycled aggregate concrete (RAC) in engineering, recycled aggregate concrete refers to new concrete made by crushing, washing, and grading waste concrete blocks, mixing them with aggregates in a certain proportion, partially or completely replacing natural aggregates such as sand and gravel (mainly coarse aggregates), and then adding cement and water. Researchers have proposed many empirical models to predict the mechanical properties of RAC. Although these empirical models have high interpretability, their accuracy is still significantly insufficient due to the highly nonlinear nature of strength prediction. To address the shortcomings of empirical models, artificial intelligence (AI) algorithms that can effectively solve highly nonlinear regression problems are receiving increasing attention from researchers. Studies have shown that machine learning algorithms such as artificial neural networks, Gaussian process regression, and decision trees are far more accurate than traditional empirical models in predicting the mechanical properties of RAC.

[0004] As the "world's factory" and a major manufacturing nation, China should prioritize green and sustainable development while accelerating economic growth. When designing sustainable concrete mixes, simply considering the mechanical properties (compressive strength) and cost of concrete is insufficient; environmental goals such as carbon emissions and implicit energy consumption must also be taken into account. However, these factors often conflict; as mechanical properties improve, costs, carbon emissions, and implicit energy consumption also increase, making it extremely difficult to find the optimal solution. Traditional mix design methods primarily rely on the relationship between water-cement ratio and strength (Paul Mee's formula) to design the compressive strength of concrete. If other objectives exist, such as workability, cost, and carbon emissions, optimization through trial-and-error experiments is necessary, which is both time-consuming and labor-intensive. Furthermore, when considering multiple objectives simultaneously, the mix proportions obtained through traditional trial-and-error experiments only meet the stated requirements and are not necessarily the optimal mix proportions. Summary of the Invention

[0005] This invention aims to address the problems of current methods for optimizing the mix proportion of recycled aggregate concrete, which do not consider issues such as cost and carbon emissions, and cannot perform optimization of multiple objectives under the premise of optimal mix proportion.

[0006] A method for optimizing the mix proportion of recycled aggregate concrete includes the following steps:

[0007] Step 1: Using the water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age of the recycled aggregate concrete as input, a machine learning model is used to predict the compressive strength of the recycled aggregate concrete, i.e., the RAC compressive strength.

[0008] The first objective function is to maximize the compressive strength of recycled aggregate concrete predicted by the machine learning model.

[0009] Step 2: Based on the composition of recycled aggregate concrete, obtain the second to fourth objective functions of recycled aggregate concrete. The second to fourth objective functions are respectively minimizing RAC cost, minimizing RAC carbon emissions, and minimizing RAC implicit energy consumption.

[0010] RAC cost target:

[0011]

[0012] Where n is the quantity of raw materials, m i′ It is the mass of the i′th raw material, C i′ It is the price per kilogram of the i′th raw material;

[0013] RAC carbon emission targets (EC):

[0014] EC = EC R +EC T +EC M

[0015] EC R =Q C EF C +Q S EF S +Q A (1-λ RA )EF A +Q A λ RA EF RA +Q F EF F +Q W EF W

[0016] EC T =(Q C DC +Q S D S +Q A D A +Q F D F )·EF T

[0017] EC M =Q E EF E

[0018] Among them, EC R EC T EC M These represent carbon emissions during production, transportation, and blending processes, respectively; Q C Q S Q A Q F and Q W These represent the contents of cement, sand, coarse aggregate, fly ash, and water, respectively; λ RA Indicates the rate of replacement of recycled aggregates; EF C EF S EF A EF RA EF F EF W These represent the carbon emission factors for cement, sand, coarse aggregate, fly ash, and water, respectively; D C D S D A and D F EF represents the conveying distance for cement, sand, coarse aggregate, and fly ash, respectively; T For transport carbon emission factors; Q E For the power consumption of the RAC mixing process, EF E Carbon emission factor for electricity consumption;

[0019] (c) Implicit energy consumption of RAC:

[0020] The carbon emission factors of each component in the RAC carbon emission target EC are replaced with the corresponding implicit energy consumption factors to obtain the objective function of implicit energy consumption.

[0021] Step 3: Set 3 operating conditions:

[0022] A. Considering only the dual objectives of RAC compressive strength and RAC cost;

[0023] B. Consider the three objectives of RAC compressive strength, RAC cost, and RAC carbon emissions;

[0024] C. Consider the four objectives of RAC: compressive strength, cost, carbon emissions, and implicit energy consumption.

[0025] For working conditions A, B, or C, a multi-objective optimization algorithm is used for multi-objective optimization.

[0026] Furthermore, the machine learning model described in step one is XGboost.

[0027] Furthermore, the XGboost used in step one to predict the compressive strength of recycled aggregate concrete is obtained through the following steps:

[0028] Step 101: Collect test data on the compressive strength of recycled aggregate concrete;

[0029] The compressive strength test data of recycled aggregate concrete includes the following independent variables: water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age; the dependent variable is compressive strength.

[0030] Step 102: Use the XGboost algorithm to establish a compressive strength prediction model. The objective function of the machine learning model corresponding to the XGboost algorithm is:

[0031]

[0032]

[0033] Where L is the loss function, Ω is the regularization term; K is the leaf node tree, ω is the percentage of leaf nodes, γ and λ are overfitting prevention coefficients; i represents the i-th decision tree, N is the total number of decision trees; y i For the actual value of sample i, f represents the predicted values ​​of the first t-1 decision trees for sample i; t (x i Let t be the prediction value of the t-th decision tree for sample i, where t represents the t-th decision tree and T is the total number of decision trees;

[0034] Step 103: Based on the sample dataset, the Bayesian optimization algorithm is selected as the optimization algorithm, and tenfold cross-validation is used to prevent overfitting. The optimization is carried out with the goal of minimizing the RMSE value of the machine learning model, and the hyperparameters of XGboost are obtained.

[0035] Step 104: Randomly obtain a training set from the sample dataset; use the training set to construct XGboost, the hyperparameters of XGboost are the hyperparameters obtained in step 103, and thus obtain XGboost for predicting the compressive strength of recycled aggregate concrete.

[0036] Alternatively, the XGboost used in step one to predict the compressive strength of recycled aggregate concrete can be obtained through the following steps:

[0037] Step 101: Collect test data on the compressive strength of recycled aggregate concrete;

[0038] The compressive strength test data of recycled aggregate concrete includes the following independent variables: water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age; the dependent variable is compressive strength.

[0039] Step 102: Use the XGboost algorithm to establish a compressive strength prediction model. The objective function of the machine learning model corresponding to the XGboost algorithm is:

[0040]

[0041]

[0042] Where L is the loss function, Ω is the regularization term; K is the leaf node tree, ω is the percentage of leaf nodes, γ and λ are overfitting prevention coefficients; i represents the i-th decision tree, N is the total number of decision trees; y i For the actual value of sample i, f represents the predicted values ​​of the first t-1 decision trees for sample i; t (x i Let t be the prediction value of the t-th decision tree for sample i, where t represents the t-th decision tree and T is the total number of decision trees;

[0043] Step 103: Randomly obtain a training set from the sample dataset; use the training set to construct XGboost. The hyperparameters of XGboost are: learning rate of 0.413 and maximum depth of 3, thus obtaining XGboost for predicting the compressive strength of recycled aggregate concrete.

[0044] Furthermore, for the test data of compressive strength of recycled aggregate concrete in S101, it is necessary to use the isolated forest algorithm to detect outliers in the collected data, remove the outliers isolated by the algorithm, and use the test data of compressive strength of recycled aggregate concrete after removing outliers as the sample dataset.

[0045] Furthermore, the constraints on the first to fourth objective functions include range constraints, scaling constraints, and volume constraints:

[0046] (1) The range constraint is the range interval of the independent variable, and the interval is taken from the range of each independent variable in the dataset;

[0047] (2) The proportional constraints take into account the water-cement ratio and the sand-gravel ratio, as shown in equation (11);

[0048] (3) The volume of the RAC is limited to 1m³. 3 The volume constraints are shown in equations (12) and (13);

[0049] 0.3≤m W / (m W +m FA )≤1.0 (11)

[0050] 0.3≤m S / (m S +m CA )≤0.6 (12)

[0051]

[0052] In the formula, m represents mass, r represents RCA substitution rate, A and D are the amount of material used and their corresponding densities; the subscripts c, w, s, NCA, RCA and FA represent cement, sand, natural coarse aggregate, recycled coarse aggregate and fly ash respectively.

[0053] Furthermore, the multi-objective optimization algorithm employs the CMOPSO algorithm.

[0054] A computer storage medium storing a computer program, the computer program being loaded and executed by a processor to implement the aforementioned method for optimizing the mix proportion of recycled aggregate concrete.

[0055] A recycled aggregate concrete mix design optimization device includes a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement the recycled aggregate concrete mix design optimization method.

[0056] Beneficial effects:

[0057] 1. Existing technologies have low calculation accuracy, while the calculation accuracy of this invention is far higher than that of existing technologies.

[0058] 2. Existing technologies can only design based on strength. This invention designs based on strength, cost, carbon emissions, and implicit energy consumption. It can design and optimize any number of objectives simultaneously as needed, and obtain the optimal solution through a multi-objective optimization algorithm. This invention can ensure that multiple objectives are optimized under the premise of optimal mix ratio. Attached Figure Description

[0059] Figure 1 This is the biobjective Pareto solution set.

[0060] Figure 2 For a three-objective Pareto solution set.

[0061] Figure 3 For a four-objective Pareto solution set. Detailed Implementation

[0062] To address the limitations of traditional design methods, this invention first uses the XGboost model to establish a predictive model for the compressive strength of recycled aggregate concrete. Then, it employs the NSGA-III optimization algorithm to simultaneously optimize multiple objectives, including strength, cost, and environmental factors, obtaining a Pareto optimal solution set. Finally, TOPSIS is used for sorting, reducing trial-and-error costs while achieving the optimal mix proportion. The specific implementation details are described below. Specific implementation method one:

[0064] This embodiment is a method for optimizing the mix proportion of recycled aggregate concrete, including the following steps:

[0065] Step 1: Collect test data on the compressive strength of recycled aggregate concrete.

[0066] This implementation method collected 1305 test data on the compressive strength of recycled aggregate concrete from different literatures. The independent variables of these test data included: water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age. The dependent variable was compressive strength, which was used to train the machine learning model later.

[0067] The isolated forest algorithm is used to detect outliers in the collected data, remove the outliers isolated by the algorithm, and obtain a new dataset for training machine learning models.

[0068] The Isolation Forest algorithm is a model for detecting outliers in unsupervised learning. It determines whether a sample is abnormal based on the number of splits required to separate the samples.

[0069] Step 2: Use the new dataset as the sample dataset;

[0070] Step 3: Use the XGboost algorithm to build a compressive strength prediction model. The XGboost algorithm combines individual decision trees to construct a boosting tree and performs parallel computation.

[0071] The XGboost algorithm performs a second-order Taylor expansion on the objective function, enhances the regularization term, reduces the possibility of overfitting, and shortens the training time.

[0072] The objective function of a machine learning model can be expressed as:

[0073]

[0074]

[0075] Where L is the loss function, Ω is the regularization term; K is the leaf node tree, ω is the percentage of leaf nodes, and γ and λ are overfitting prevention coefficients. i represents the i-th decision tree, N is the total number of decision trees; y i For the actual value of sample i, f represents the predicted values ​​of the first t-1 decision trees for sample i; t (x i Let be the prediction value of the t-th decision tree for sample i, where t represents the t-th decision tree and T is the total number of decision trees.

[0076] In order to verify the effectiveness of the XGboost algorithm and compare its effectiveness with other machine learning models (Artificial Neural Network (ANN), Support Vector Machine (GPR), Random Forest (RF), this embodiment also establishes a compressive strength prediction model based on other machine learning models.

[0077] Step 4: Select evaluation metrics to assess the performance of the machine learning model (i.e., the magnitude of the error between the predicted and actual values). For models using machine learning, the evaluation metrics include: RMSE (Root Mean Square Error, the square root of the ratio of the square of the deviation between the predicted and actual values ​​to the number of observations n), MAPE (Mean Absolute Percentage Error, a relative error measure that uses absolute values ​​to avoid positive and negative errors canceling each other out), R... 2 (The coefficient of determination, a numerical characteristic representing the relationship between a random variable and multiple random variables, is a statistical indicator used to reflect the reliability of a regression model in explaining changes in the dependent variable.) Its expression is as follows, where RMSE and MAPE are better the smaller they are, and R... 2 The closer to 1, the better:

[0078]

[0079] Where k is the total number of samples, e i and p i These are the experimental value and the predicted value, respectively. and e i and p i The average value.

[0080] Step 5: Different hyperparameters have a significant impact on the training results of machine learning. Therefore, it is necessary to optimize the hyperparameters of the machine learning model corresponding to the stress intensity prediction model. The optimized hyperparameters are the maximum depth and the learning rate.

[0081] The optimization algorithm selected is Bayesian optimization, and tenfold cross-validation is used to prevent overfitting. The optimization process aims to minimize the RMSE value of the machine learning model trained under a certain hyperparameter. It includes the following steps:

[0082] (1) Tenfold cross-validation: The sample dataset is divided into 10 subsets, of which 9 are training subsets and 1 is test subset. The 10 subsets are used as test subsets in turn, so there are 10 possible allocations.

[0083] (2) Use the Bayesian optimization algorithm, set the number of iterations of the algorithm, and generate a set of hyperparameters for the machine learning model to train in each iteration.

[0084] (3) The training subset is trained using the hyperparameter, and the resulting machine learning model is tested using the test subset. The RMSE index obtained from the test is the Bayesian optimization objective. Since there are ten allocation results, each iteration will train and test 10 times, and the best result among them will be used as the Bayesian optimization result of this iteration.

[0085] (4) Repeat steps (2) and (3). The number of repetitions is the number of iterations set by the user, and finally the optimized hyperparameters are obtained.

[0086] In this implementation, the optimized hyperparameters of XGboost are as follows: learning rate of 0.413 and maximum depth of 3.

[0087] Step 6: Randomly divide the sample dataset into a training set and a test set, with the training set accounting for 70% of the total data and the test set accounting for 30% of the total data.

[0088] XGBoost is built using the training set, and the hyperparameters of XGBoost are the same as those in step 5. After XGBoost is built, the accuracy of XGBoost in the face of new data is tested using the test set.

[0089] Similarly, other machine learning models (Artificial Neural Networks, Support Vector Machines, Random Forests) can be trained using the training set, and then the accuracy of these models on new data can be tested using the test set.

[0090] The evaluation metrics of various machine learning models are compared, and the results are shown in Table 1. As can be seen from Table 1, XGB (i.e. XGboost) outperforms other machine learning models on this dataset.

[0091] Table 1 Model Performance

[0092]

[0093]

[0094] Step 7: Define the multi-objective optimization objective functions: The XGboost prediction model is used as one of the objective functions, namely, maximizing the compressive strength of the RAC. The other objective functions are minimizing the cost of the RAC, minimizing the carbon emissions of the RAC, and minimizing the implicit energy consumption of the RAC. The calculation methods for each objective function are as follows:

[0095] (1) RAC cost target:

[0096]

[0097] Where n is the quantity of raw materials, m i′ It is the mass of the i′th raw material, C i′ It is the price per kilogram of the i′th raw material.

[0098] Table 2 shows the current prices of various raw materials.

[0099] Table 2. Prices and densities of various raw materials

[0100]

[0101] (2) RAC carbon emission targets:

[0102] The carbon emissions and energy consumption of RAC (Resource Acquisition, Energy Consumption, and Chemicals) mainly come from the generation, transportation, and blending of raw materials. The RAC carbon emission (EC) objective function can be expressed as:

[0103] EC = EC R +EC T +EC M (7)

[0104] EC R =Q C EF C +Q S EF S +Q A (1-λ RA )EF A +Q A λ RA EF RA +Q F EF F +Q W EF W (8)

[0105] EC T =(Q C D C +Q S D S +Q A D A +Q F D F )·EF T(9)

[0106] EC M =Q E EF E (10)

[0107] Among them, EC R EC T EC M These represent carbon emissions during production, transportation, and blending processes, respectively; Q C Q S Q A Q F and Q W These represent the contents of cement, sand, coarse aggregate, fly ash, and water, respectively; λ RA Indicates the rate of replacement of recycled aggregates; EF C EF S EF A EF RA EF F EF W These represent the carbon emission factors for cement, sand, coarse aggregate, fly ash, and water, respectively; D C D S D A and D F EF represents the conveying distance for cement, sand, coarse aggregate, and fly ash, respectively; T For transport carbon emission factors; Q E For the power consumption of the RAC mixing process, EF E The carbon emission factor is the amount of electricity consumed.

[0108] (3) Implicit energy consumption of RAC:

[0109] By replacing the carbon emission factors of each component in the RAC carbon emission target with the corresponding implicit energy consumption factors, the objective function of implicit energy consumption can be obtained.

[0110] Table 3 shows the implicit energy consumption factor and carbon emission factor for all raw materials. The carbon emission factor for electricity consumption is 0.77CO. 2e / (kW·h). The transportation distance for each raw material is set at 50 km. The transportation condensation energy and mixing condensation energy are set at 2.6 MJ / (ton-km) and 2.7 kWh / m3, respectively. The transportation carbon emission factor is 0.18 kgCO2e / (ton-km).

[0111] Table 3 Implicit Energy Consumption and Carbon Emission Factors of Raw Materials

[0112]

[0113] Step 8: Define the constraints, namely the range constraint, the scale constraint, and the volume constraint.

[0114] (1) The range constraint is the range interval of the independent variable, and the interval is taken from the range of each independent variable in the dataset;

[0115] (2) The proportional constraints take into account the water-cement ratio and the sand-gravel ratio, as shown in equation (11);

[0116] (3) The volume of the RAC is limited to 1m³. 3 The volume constraints are shown in equations (12) and (13);

[0117] 0.3≤m W / (m W +m FA )≤1.0 (11)

[0118] 0.3≤m S / (m S +m CA )≤0.6 (12)

[0119]

[0120] In the formula, m represents mass, r represents RCA substitution rate, A and D are the amount of material used and their corresponding densities; the subscripts c, w, s, NCA, RCA and FA represent cement, sand, natural coarse aggregate, recycled coarse aggregate and fly ash respectively.

[0121] In this invention, RCA refers to recycled coarse aggregate, and RAC refers to recycled aggregate concrete.

[0122] Step 9: Define operating conditions. This invention sets three operating conditions for the user:

[0123] A. Considering only the dual objectives of RAC compressive strength and RAC cost;

[0124] B. Consider the three objectives of RAC compressive strength, RAC cost, and RAC carbon emissions;

[0125] C. Consider the four objectives of RAC: compressive strength, cost, carbon emissions, and implicit energy consumption.

[0126] Step 10: For the three working conditions, perform multi-objective optimization using the CMOPSO algorithm respectively.

[0127] To optimize the calculation example from an engineering practice perspective, several variables related to material properties were determined before optimization: the curing time of RAC, the water absorption rates of NCA and RCA, and the maximum particle size of coarse aggregate were 28 days, 0.5%, 4.0%, and 20 mm, respectively. The optimization results are as follows: Figure 1 , 2 As shown in Figure 3, the blue dots (labeled as experimental results) represent the curing time of RAC, the water absorption rates of NCA and RCA in the original dataset, and the maximum particle size of coarse aggregate at 28 days, 0.5%, 4.0%, and 20 mm, respectively. The others represent the multi-objective optimization results. The colors, from light to dark, represent TOPSIS scores from low to high, with the highest score being the optimal mix design. The optimal mix design is shown in Table 4.

[0128] The TOPSIS method obtains the relative proximity by calculating the distance between each solution and the positive and negative ideal solutions, and then sorts the Pareto sets according to the proximity. The positive and negative ideal solutions can be expressed as:

[0129]

[0130]

[0131] in, and These are the positive and negative ideal solutions for the j-th optimization objective.

[0132] Pareto sets are such that each solution is and The distance between them can be calculated using Euclidean distance:

[0133]

[0134]

[0135] in, Is the i-th solution and The distance between them.

[0136] Finally, the solution to TOPSIS can be determined as follows:

[0137]

[0138] C i The larger the value, the better the solution. Then, according to C... i The values ​​are sorted to obtain the optimal solution.

[0139] Figure 1 , Figure 2 , Figure 3 explain: Figure 1 As a bi-objective Pareto solution set, each star in the two-dimensional graph represents a Pareto solution. The blue dots are test points (i.e., partial true values ​​in the dataset, where the compressive strength is the collected data and the cost is the mix proportion calculated by formula (6) at that strength). Figure 2The dataset is a three-objective Pareto solution set, so each star in the 3D graph represents a Pareto solution, and the blue dots represent some of the true values ​​in the dataset. Figure 3 The graph is a Pareto solution set with four objectives. Therefore, each axis in the four-axis graph represents an objective, and each line represents a Pareto solution. The blue line represents the optimal solution calculated in TOPSIS. For the sake of simplicity, points from the dataset are not included.

[0140] Table 4 Optimal Mix Proportions for Each Working Condition

[0141] Specific Implementation Method Two:

[0143] This embodiment is a computer storage medium that stores a computer program. The computer program is loaded and executed by a processor to implement a method for optimizing the mix proportion of recycled aggregate concrete.

[0144] It should be understood that the computer storage media described in this embodiment include, but are not limited to, magnetic computer storage media and optical computer storage media; the magnetic computer storage media include, but are not limited to, RAM, ROM, and other computer storage media such as hard disks and USB flash drives. Specific implementation method three:

[0146] This embodiment is a recycled aggregate concrete mix proportion optimization device. The device includes a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement a recycled aggregate concrete mix proportion optimization method.

[0147] It should be understood that the device described in this embodiment includes, but is not limited to, devices including processors and memory, and may also include other devices corresponding to units or modules with information acquisition, information interaction, and control functions, such as signal acquisition devices. The device includes, but is not limited to, PCs, workstations, and mobile devices.

[0148] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for optimizing the mixture ratio of recycled aggregate concrete, characterized by, Includes the following steps: Step 1: Using the water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age of the recycled aggregate concrete as input, a machine learning model is used to predict the compressive strength of the recycled aggregate concrete, i.e., the RAC compressive strength. The first objective function is to maximize the compressive strength of recycled aggregate concrete predicted by the machine learning model. Step 2: Based on the composition of recycled aggregate concrete, obtain the second to fourth objective functions of recycled aggregate concrete. The second to fourth objective functions are respectively minimizing RAC cost, minimizing RAC carbon emissions, and minimizing RAC implicit energy consumption. RAC cost target: ； Where n is the quantity of raw materials, It is the first The quality of the raw materials It is the first The price per kilogram of the raw materials; RAC carbon emission targets (EC): ； ； ； ； Among them, EC R EC T EC M These represent carbon emissions during production, transportation, and blending processes, respectively; Q C Q S Q A Q F and Q W These represent the contents of cement, sand, coarse aggregate, fly ash, and water, respectively; λ RA Indicates the rate of replacement of recycled aggregates; EF C EF S EF A EF F EF W These represent the carbon emission factors for cement, sand, coarse aggregate, fly ash, and water, respectively; D C D S D A and D F EF represents the conveying distance for cement, sand, coarse aggregate, and fly ash, respectively; T For transport carbon emission factors; Q E For the power consumption of the RAC mixing process, EF E Carbon emission factor for electricity consumption; RAC implicit energy consumption: The carbon emission factors of each component in the RAC carbon emission target EC are replaced with the corresponding implicit energy consumption factors to obtain the objective function of implicit energy consumption. Step 3: Set 3 operating conditions: A. Considering only the dual objectives of RAC compressive strength and RAC cost; B. Consider the three objectives of RAC compressive strength, RAC cost, and RAC carbon emissions; C. Consider the four objectives of RAC: compressive strength, cost, carbon emissions, and implicit energy consumption. For working conditions A, B, or C, a multi-objective optimization algorithm is used for multi-objective optimization.

2. The method according to claim 1, characterized in that, The machine learning model described in step one is XGboost.

3. The method according to claim 2, wherein, Step 1, using XGboost to predict the compressive strength of recycled aggregate concrete, is obtained through the following steps: Step 101: Collect test data on the compressive strength of recycled aggregate concrete; The compressive strength test data of recycled aggregate concrete includes the following independent variables: water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age; the dependent variable is compressive strength. Step 102: Use the XGboost algorithm to establish a compressive strength prediction model. The objective function of the machine learning model corresponding to the XGboost algorithm is: ； ； Where L is the loss function, Ω is the regularization term; K is the leaf node tree, ω is the percentage of leaf nodes, and γ and λ are coefficients to prevent overfitting; Representing the Decision tree, Total number of decision trees; For the sample The actual value, For the first t-1 decision trees, pair the samples The predicted value; For the t-th decision tree, pair of samples The predicted value, Representing the Decision tree, Total number of decision trees; Step 103: Based on the sample dataset, the Bayesian optimization algorithm is selected as the optimization algorithm, and tenfold cross-validation is used to prevent overfitting. The optimization is carried out with the goal of minimizing the RMSE value of the machine learning model, and the hyperparameters of XGboost are obtained. Step 104: Randomly obtain a training set from the sample dataset; use the training set to construct XGboost, the hyperparameters of XGboost are the hyperparameters obtained in step 103, and thus obtain XGboost for predicting the compressive strength of recycled aggregate concrete.

4. The method according to claim 3, characterized in that, Step 103 includes the following steps: (1) Tenfold cross-validation: The sample dataset is divided into 10 subsets, of which 9 are training subsets and 1 is test subset. The 10 subsets are used as test subsets in turn, so there are 10 possible allocations. (2) Use the Bayesian optimization algorithm, set the number of iterations of the algorithm, and generate a set of hyperparameters for the machine learning model to train in each iteration; (3) The training subset is trained using the hyperparameter, and the resulting machine learning model is tested using the test subset. The RMSE index obtained from the test is the Bayesian optimization objective. Each iteration will be trained and tested 10 times, and the best result among them will be used as the Bayesian optimization result of this iteration. (4) Repeat steps (2) and (3) to finally obtain the optimized hyperparameters.

5. The method according to claim 2, wherein, Step 1, using XGboost to predict the compressive strength of recycled aggregate concrete, is obtained through the following steps: Step 101: Collect test data on the compressive strength of recycled aggregate concrete; The compressive strength test data of recycled aggregate concrete includes the following independent variables: water content, cement content, cement strength, sand content, coarse aggregate dosage, recycled aggregate replacement rate, recycled aggregate water absorption rate, natural aggregate water absorption rate, maximum coarse aggregate particle size, and curing age; the dependent variable is compressive strength. Step 102: Use the XGboost algorithm to establish a compressive strength prediction model. The objective function of the machine learning model corresponding to the XGboost algorithm is: ； ； Where L is the loss function, Ω is the regularization term; K is the leaf node tree, ω is the percentage of leaf nodes, and γ and λ are coefficients to prevent overfitting; Representing the Decision tree, Total number of decision trees; For the sample The actual value, For the first t-1 decision trees, pair the samples The predicted value; For the t-th decision tree, pair of samples The predicted value, Representing the Decision tree, Total number of decision trees; Step 103: Randomly obtain a training set from the sample dataset; use the training set to construct XGboost. The hyperparameters of XGboost are: learning rate of 0.413 and maximum depth of 3, thus obtaining XGboost for predicting the compressive strength of recycled aggregate concrete.

6. The method for optimizing the mixture proportion of recycled aggregate concrete according to claim 3, 4 or 5, characterized in that, For the compressive strength test data of recycled aggregate concrete in S101, the isolated forest algorithm needs to be used to detect outliers in the collected data, remove the outliers isolated by the algorithm, and use the outlier-removed recycled aggregate concrete compressive strength test data as the sample dataset.

7. The method according to claim 6, wherein the method is characterized by, The constraints on the first to fourth objective functions include range constraints, scaling constraints, and volume constraints: The range constraint is the range interval of the independent variable, which is taken from the range of each independent variable in the dataset. The proportional constraints take into account the water-cement ratio and the sand-gravel ratio, as shown in equation (11); The volume of the RAC is limited to 1 m 3 The volume constraint is given by equations (12), (13) （11） （12） （13） In the formula, m represents mass, r represents RCA substitution rate, A and D are the amount of material used and their corresponding densities; the subscripts c, s, NCA, RCA and FA represent cement, sand, natural coarse aggregate, recycled coarse aggregate and fly ash, respectively.

8. The method according to claim 7, wherein the method is characterized by, The multi-objective optimization algorithm used is the CMOPSO algorithm.

9. A computer storage medium, characterized in that The computer storage medium stores a computer program, which is loaded and executed by a processor to implement the method for optimizing the mix proportion of recycled aggregate concrete as described in any one of claims 1 to 8.

10. A recycled aggregate concrete mix proportion optimization apparatus, characterized by, The device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the method for optimizing the mix proportion of recycled aggregate concrete as described in any one of claims 1 to 8.

Images