Method and equipment for predicting compressive strength of recycled coarse aggregate concrete based on optimized machine learning algorithm, and medium

Abstract

The invention relates to a method, equipment and medium for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm, and the method comprises the steps: collecting factors which affect the performance of the recycled coarse aggregate concrete from literature data, taking the compressive strength as an output variable, and preprocessing the data; and combining at least one machine learning model and at least one optimization algorithm into a plurality of prediction model combinations. The preprocessed data is divided into a training set and a test set, and an optimal performance model is determined through comprehensive evaluation after the model is trained. And performing reverse prediction on the key factors by using a prediction model with optimal performance to obtain the influence trend of the change of the key factors on the compressive strength of the recycled coarse aggregate concrete. Compared with the prior art, the recycled coarse aggregate concrete compressive strength prediction model based on the optimized machine learning algorithm solves the problems of insufficient precision and poor generalization ability in a traditional prediction method.

Description

Technical Field

[0001] The present invention relates to the field of material science and technology, and in particular to a method, device and medium for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm. Background Art

[0002] Recycled concrete technology effectively solves the problem of construction waste by recycling waste concrete, reducing resource consumption and environmental impact. This technology involves crushing, cleaning, grading and mixing waste concrete with cement to make new concrete, saving resources and reducing environmental damage. Researchers are working hard to improve the performance of recycled concrete to make it close to that of original concrete. Recycled concrete technology plays an important role in promoting the sustainable development of building materials, realizing green buildings and circular economy, and is expected to be more widely used in the future.

[0003] The widespread application of recycled concrete is limited by several challenges. First, the old mortar attached to the surface of recycled aggregate increases the complexity of the interface transition zone, involving four different interface combinations, which directly affects the mechanical properties of concrete and leads to performance instability, making it difficult to accurately evaluate the reliability of structural performance. Second, recycled concrete is affected by many factors, its microstructure is complex, and has a large variability. When exploring the role of mineral additives in improving the mechanical properties of recycled concrete, mineral additives such as fly ash and silica fume show significant benefits. Fly ash, as a volcanic ash material, can effectively improve the workability of concrete mixtures, including fluidity, cohesion and water retention. Its volcanic ash activity reaction not only enhances the density and interface structure of concrete, but also reduces the content of easily corrosive calcium hydroxide through secondary reactions, thereby improving the impermeability and sulfate corrosion resistance of concrete. Silica fume significantly enhances the compression, flexural, impermeability, corrosion, impact and wear resistance of concrete by filling the gaps between cement particles, reacting with hydration products to form gel, and reacting with alkaline magnesium oxide to form gel. Adding an appropriate amount of silica fume can significantly improve the comprehensive mechanical properties of concrete.

[0004] With the development of digital technology, the rapid progress of artificial intelligence technology has enabled the construction industry to explore the application of artificial intelligence technology in the field of civil engineering materials. Compared with traditional methods that rely on subjective experience, machine learning, as a data-driven analysis tool, can deeply explore the complex nonlinear relationship between independent variables and dependent variables. Machine learning technology saves valuable time and labor resources and has been widely used in material performance prediction. For example, in the construction field, machine learning technology is used to enhance the safety of the construction process, improve production efficiency, improve building quality, and optimize maintenance work.

[0005] The above studies show that researchers have conducted in-depth research on the performance of machine learning and achieved certain research results. Current research mainly focuses on the prediction of concrete performance, but there are few studies on the prediction model of the compressive strength of recycled coarse aggregate concrete with optimized machine learning algorithms. Therefore, it is urgent to carry out targeted research and construct an optimized machine learning model suitable for the prediction of the compressive strength of recycled coarse aggregate concrete to fill this research gap and provide strong technical support for the further development of recycled concrete technology. Summary of the invention

[0006] The purpose of the present invention is to provide a method, device, and medium for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm in order to overcome the defects of the above-mentioned prior art. The compressive strength prediction model of recycled coarse aggregate concrete based on an optimized machine learning algorithm is intended to solve the problems of insufficient accuracy and poor generalization ability in traditional prediction methods. By introducing advanced data preprocessing technology and feature selection mechanism, the machine learning model can more accurately capture the relationship between the influencing characteristics and mechanical properties of recycled coarse aggregate concrete, thereby improving the accuracy and reliability of the prediction. The present invention not only provides a new technical approach for the performance evaluation of recycled coarse aggregate concrete, but also has important significance for promoting the resource utilization and sustainable development of construction waste.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] The first aspect of the present invention provides a method, device, and medium for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm. The prediction method comprises the following steps:

[0009] S1. Data collection and processing: Obtain multiple factors that affect the performance of recycled coarse aggregate concrete from literature as input variable data, determine the compressive strength of recycled coarse aggregate concrete as the output variable, and pre-process the collected data;

[0010] S2. Model construction and combination: Use at least one machine learning model and at least one optimization algorithm to combine the selected machine learning model and optimization algorithm to form a variety of different prediction model combinations;

[0011] S3, model training and evaluation: divide the preprocessed data in S1 into training sets and test sets, use the training sets to train the various prediction models constructed in S2, so that the models learn the potential relationship between input variables and output variables, and determine the prediction model with the best performance through comprehensive evaluation;

[0012] S4. Key factor analysis: for the optimal prediction model determined in S3, analyze the importance of input variables to output variables, and determine the key factors affecting the performance of recycled coarse aggregate concrete, wherein the key factors include input variables with higher importance scores in the model, i.e., input variables with a relatively greater influence on the compressive strength of recycled coarse aggregate concrete;

[0013] S5. Reverse prediction and application: Use the prediction model with optimal performance to reversely predict the key factors and obtain the influence trend of the changes in the key factors on the compressive strength of recycled coarse aggregate concrete.

[0014] Furthermore, in S1, the preprocessing includes data cleaning to remove outliers, data standardization or normalization to make data features more suitable for subsequent analysis, and correlation analysis to preliminarily explore the relationship between input variables and output variables.

[0015] Further, in S1, the input variable data includes multiple factors affecting the performance of recycled coarse aggregate concrete as input variables, including:

[0016] Data on variables related to the properties of recycled coarse aggregate, including recycled coarse aggregate replacement rate, physical properties of recycled coarse aggregate;

[0017] Variable data related to cementitious materials, including water-cement ratio, mineral admixture replacement rate, etc.;

[0018] Variable data related to aggregates, including water absorption and apparent density of coarse aggregates;

[0019] Data on factors related to maintenance conditions.

[0020] Furthermore, in S3, the input variable data includes 11 input variables, namely, curing age (X-1), water-cement ratio (X-2), sand ratio (X-3), RCA substitution rate (X-4), FA substitution rate (X-5), SF substitution rate (X-6), M substitution rate (X-7), FGS substitution rate (X-8), FGBFS substitution rate (X-9), water absorption rate of CA (X-10), and apparent density of CA (X-11).

[0021] Furthermore, in S2, the machine learning model includes a neural network model, a support vector machine model, a tree-based model, and a kernel function-based model.

[0022] Furthermore, in S2, the optimization algorithm includes an algorithm based on evolutionary theory and an algorithm based on swarm intelligence.

[0023] Furthermore, in S2, a variety of performance evaluation indicators are used to evaluate the performance of the trained model on the test set. The performance evaluation indicators include indicators reflecting the goodness of model fit, indicators measuring prediction errors, and indicators reflecting model stability and reliability.

[0024] Furthermore, in S2, four machine learning models are specifically introduced, namely, the BP model, the SVM model, the RF model, and the RBF model, and two optimization algorithms, namely, the GA optimization algorithm and the PSO optimization algorithm, are combined to obtain 12 different machine learning models.

[0025] A second aspect of the present invention provides an electronic device, comprising a memory and a processor, wherein the processor is used to execute a program in the memory, so as to implement the above-mentioned method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm.

[0026] A third aspect of the present invention provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, is used to execute the above-mentioned method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1) By optimizing the machine learning algorithm, the compressive strength of recycled coarse aggregate concrete can be accurately predicted, providing a new way to evaluate its performance, which helps to deeply understand the influence of various factors on strength. In terms of resource utilization, it can effectively guide the optimization of the proportion of recycled coarse aggregate concrete, improve the utilization rate of recycled coarse aggregate, reduce the mining of natural aggregates, reduce resource consumption, and promote the resource utilization of construction waste. From an environmental perspective, it can reduce the damage of resource mining to the ecological environment, reduce the emission of construction waste, promote the sustainable development of the construction industry, and at the same time improve the quality and efficiency of construction projects, providing strong support for material performance analysis and sustainable practices in the field of civil engineering.

[0029] 2) By introducing advanced data preprocessing technology and feature selection mechanism, the machine learning model can more accurately capture the relationship between the influencing characteristics and mechanical properties of recycled coarse aggregate concrete, thereby improving the accuracy and reliability of prediction. The present invention not only provides a new technical approach for the performance evaluation of recycled coarse aggregate concrete, but also has important significance for promoting the resource utilization and sustainable development of construction waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is the research framework of this invention.

[0031] Figure 2 is the marginal histogram of the database.

[0032] Figure 3 The distribution of the data.

[0033] Figure 4 It is the matrix heat map of the data set.

[0034] Figure 5 Schematic diagram of machine learning model and GA and PSO optimization algorithms.

[0035] Figure 6 It is a performance evaluation indicator for machine learning models.

[0036] Figure 7 Rank the scores of machine learning models.

[0037] Figure 8 are the experimental and predicted values ​​of the machine learning model.

[0038] Fig. 9 are the prediction residuals of different machine learning models.

[0039] Fig.10 Data and error distributions for machine learning models.

[0040] Fig.11 is the importance distribution of the input feature variables.

[0041] Fig.12 The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-1.

[0042] Fig.13 The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-2.

[0043] Fig.14 The compressive strength prediction results of recycled coarse aggregate concrete under different machine learning models and different X-3.

[0044] Fig.15 The compressive strength prediction results of recycled coarse aggregate concrete under different machine learning models and different X-4. DETAILED DESCRIPTION

[0045] Based on this, the present invention constructs a data set containing 11 independent variables and 1 dependent variable by comprehensively analyzing existing literature, and performs preprocessing and correlation analysis on it. The present invention adopts 4 machine learning models (BP, SVM, RF, RBF) and two optimization algorithms (GA, PSO), which are combined into 12 different models. In order to comprehensively evaluate the performance of these machine learning models, the present invention adopts 10 performance evaluation indicators (including R 2, MSE, RMSE, etc.). Through these performance evaluation indicators, we identified the key factors for predicting the compressive strength of recycled coarse aggregate concrete, and used the optimal model to conduct a reverse prediction analysis on these key factors to explore their specific effects on the compressive strength of recycled coarse aggregate concrete in order to optimize the mix ratio of concrete.

[0046] Table 1 Reference table of nouns and abbreviations of the present invention.

[0047]

[0048]

[0049] In general, the compressive strength prediction model of recycled coarse aggregate concrete based on the optimized machine learning algorithm in the present invention includes the following steps:

[0050] Step 1: By sorting out existing literature, a data set containing 11 input variables and 1 output variable is constructed, and preprocessing and correlation analysis are performed on it.

[0051] Step 2: Use 4 machine learning models (BP, SVM, RF, RBF) and 2 optimization algorithms (GA, PSO) and combine them into 12 different machine learning models.

[0052] Step 3: Use 10 performance evaluation indicators (including R 2 , MSE, RMSE, etc.) to evaluate the performance of machine learning models.

[0053] Step 4: Use the optimal model to predict the key factors affecting the compressive strength of recycled coarse aggregate concrete.

[0054] Step 5: Use the optimal model to reversely predict the key factors affecting the compressive strength of recycled coarse aggregate concrete, study the influence of key factors on the compressive strength of recycled coarse aggregate concrete, and optimize the mix ratio of concrete.

[0055] Figure 1 It is the research framework of this invention. Figure 1It can be seen that the present invention mainly consists of three parts. The first part is "machine learning modeling and evaluation". The first part mainly includes steps such as data preparation, data preprocessing, data partitioning, training model and obtaining the optimal machine learning model. This part studies a total of 4 machine learning models and 2 optimization algorithms. The 4 machine learning models include BP, SVM, RF and RBF. The 2 optimization algorithms include GA and PSO. The 4 machine learning models and 2 optimization algorithms are arranged and combined into 12 machine learning models. The performance evaluation indicators of 10 machine learning models were studied, including R2, MSE, RMSE, RRMSE, MAE, MAPE, WMAPE, RSR, VAF, and AIC. It aims to evaluate the effectiveness of the developed machine learning model. The second part is "importance analysis of machine learning models". The second part is mainly about the importance of input variables. The third part is "reverse prediction of machine learning models". The third part is mainly for reverse prediction based on the optimal machine learning prediction model.

[0056] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms and other features not clearly described in this technical solution are all considered to be common technical features disclosed in the prior art.

[0057] The following is a further detailed description of the compressive strength prediction model of recycled coarse aggregate concrete based on the optimized machine learning algorithm described in the present invention in combination with specific data.

[0058] The mechanical properties of recycled aggregate concrete are mainly affected by the following factors: water-cement ratio, RCA replacement rate, sand rate, mineral admixtures, etc. For concrete buildings, the compressive strength of concrete is a factor that needs to be considered. Based on this, the present invention selects the curing age, water-cement ratio, sand rate, RCA replacement rate, fly ash replacement rate, silica fume replacement rate, metakaolin replacement rate, ground slag replacement rate, ground blast furnace slag replacement rate, coarse aggregate water absorption rate, and coarse aggregate apparent density as input variables, and selects the compressive strength of recycled aggregate concrete as output variables. The present invention attempts to establish a reliable and effective machine learning model between input variables and output variables. The present invention extracts 756 data samples from 20 peer-reviewed scientific literature. After sorting the data, 687 data samples are selected as the database for training the machine learning model. The database is shown in Table 2. The data samples cover various combinations of RCA replacement rate, water-cement ratio, compressive strength, etc. in concrete tests.

[0059] Table 2 Database.

[0060]

[0061]

[0062] In order to develop a prediction model for the compressive strength of recycled aggregate concrete based on an optimized machine learning algorithm, the present invention uses Matlab programming methods to preprocess, visualize and analyze the data sets in Table 2. The analysis results are shown in Figure 2 and Table 3. Figure 2 is the marginal histogram of the database. Figure 2 It can be seen that the distribution of 11 input variables and 1 output variable is a normal distribution. The curing age is mainly distributed in 0-30d. The water-binder ratio is mainly distributed in 0.3-0.6. The sand ratio is mainly distributed in 30%-45%. The RCA substitution rate is mainly distributed in 0%-50%. The FA substitution rate is mainly distributed in 0%-30%. The SF substitution rate is mainly distributed in 0%-10%. The M substitution rate is mainly distributed in 0%. The FGS substitution rate is mainly distributed in 0%. The FGBFS substitution rate is mainly distributed in 0%. The water absorption rate of CA is mainly distributed in 0%-6%. The apparent density of CA is mainly distributed in 2400-2700. The compressive strength is mainly distributed in 10-60MPa. Table 3 is the statistical distribution of the database. From Table 3, it can be seen that the mean, standard deviation, minimum, maximum, first quartile, median, and third quartile of the 11 input variables and 1 output variable. It can be observed from Table 3 that the numerical range of each variable presents a large span. Specifically, there are significant differences in the size and level of each variable.

[0063] Based on this, formula (1) is used to preprocess the data set. The results obtained after the data preprocessing are plotted into an image, such as Figure 3 shown. Figure 3 It shows that the offset of the feature variable has been effectively controlled and the data distribution presents a normal distribution. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two variables. The Pearson correlation coefficient is obtained by calculating the covariance between the two variables and dividing it by the product of the standard deviations of the two variables. The value range of the result is from -1 to 1. Among them, -1 represents a complete negative correlation and 1 represents a complete positive correlation. Formula (2) is used to calculate the Pearson correlation coefficient of the data set. The result obtained after the data is calculated by formula (2) is plotted into an image, as shown Figure 4 As shown. Figure 4 In the figure, the redder the color, the stronger the positive correlation, and the bluer the color, the stronger the negative correlation. Figure 4 The data show that there is an obvious positive correlation between compressive strength and water-cement ratio, there is an obvious positive correlation between sand ratio and FGBFS replacement rate, and there is an obvious positive correlation between RCA replacement rate and CA water absorption rate.

[0064]

[0065] Where Wi is the normalized value. Qi is the experimental value. μ is the mean value. σ is the standard deviation. ρ X,Y is the Pearson correlation coefficient. cov(X,Y) is the covariance of variables X and Y. σ X is the standard deviation of variable X. σ Y is the standard deviation of variable Y.

[0066] Table 3 Statistical distribution of database

[0067]

[0068]

[0069] Note: Qmin and Qmax are the minimum and maximum values ​​respectively. Q1, Q2 and Q3 are the first quartile, median and third quartile respectively.

[0070] Currently, many fields are applying machine learning technology to predict the behavior patterns of substances and deepen the understanding of their characteristics. The optimization algorithm iteratively improves the parameters of the model so that the machine learning model can more accurately predict the behavior patterns of substances. The distribution of 11 input variables and 1 output variable is a normal distribution, and there is a good nonlinear relationship between the 11 input variables and 1 output variable. This invention mainly introduces 4 machine learning models and 2 optimization algorithms. The 4 machine learning models include BP, SVM, RF and RBF. The 2 optimization algorithms include GA and PSO.

[0071] The BP model is a multi-layer feedforward neural network trained based on the error back propagation algorithm. It consists of an input layer, a hidden layer, and an output layer, and is a supervised learning algorithm. The key elements of the BP model are neurons, weights, biases, activation functions, learning rates, and number of iterations. The BP model receives input data, which first enters the input layer. In each layer of neurons, the input data is weighted and summed, then processed by the activation function, and then the processed data is passed to the next layer. This process continues until the data reaches the output layer, and finally generates a predicted output. The key feature of the BP model is that it can learn complex nonlinear relationships between inputs and outputs without having to know the mathematical expression of this relationship in advance.

[0072] The SVM model is a binary regression model built on the principle of structural risk minimization. The SVM model seeks a decision boundary that maximizes the interval between different categories, that is, it tries to find a hyperplane that maximizes the distance between the nearest positive and negative sample points and the hyperplane. The SVM model can use the kernel technique to map data from the original feature space to a higher-dimensional space, so that data that cannot be linearly distinguished in the low-dimensional space can be linearly divided in the high-dimensional space, see Figure 5(b) The SVM model performs well when processing small-scale data sets because its classification decisions are mainly based on key support vectors rather than relying on all data points. Due to its powerful classification ability, the SVM model has been widely used in many fields, such as image recognition, text analysis, bioinformatics, and financial risk assessment.

[0073] The RF model is an ensemble learning method that builds multiple decision trees to perform classification, regression, and other machine learning tasks, and improves the accuracy and robustness of the overall model by combining the prediction results of these trees. In the RF model, each decision tree is trained independently. When the model training is completed, each decision tree will make predictions independently based on the input samples. The final prediction result of the RF model is determined by majority voting or average. By integrating the prediction results of multiple decision trees, the RF model shows stronger generalization ability than a single decision tree and more effectively avoids overfitting. The disadvantages of the RF model include that the model may become complex and difficult to interpret. Figure 5 (c) is the algorithm diagram of the RF model.

[0074] The RBF model is an efficient nonlinear modeling technique. It is widely used in the field of machine learning to perform various tasks such as classification, regression analysis, and function approximation. The key to the RBF model lies in the radial basis function it uses. The RBF model achieves nonlinear decision boundary demarcation by mapping the input data to a higher-dimensional feature space and then building a linear model in this space. The core of the RBF model is the radial basis function. The radial basis function forms a symmetrical distribution with the center point as the axis of symmetry. As the distance between the input data point and the center point decreases, the output value of the radial basis function increases accordingly. Each radial basis function has a width parameter, which controls the rate at which the function value decreases as the distance from the center point increases. The RBF model has applications in pattern recognition, time series prediction, image processing, and other fields. Figure 5 (a) is the algorithm diagram of the RBF model.

[0075] The GA optimization algorithm, as a heuristic search method based on Darwin's natural selection and evolution theory, effectively solves a variety of optimization problems by simulating mechanisms such as inheritance, hybridization and mutation in biological evolution. The basic principle of the GA optimization algorithm is to use random search technology to find the global optimal solution in a wide solution domain. Its generation process covers key steps such as evaluation, selection, crossover and mutation. The advantage of the GA optimization algorithm is that it does not rely on the derivative information of the problem and can handle complex and nonlinear optimization tasks. However, it often takes a long time to find a solution, and the quality of the solution obtained may be affected by the selection of the initial population and algorithm parameters.

[0076] The PSO optimization algorithm is a meta-heuristic evolutionary algorithm based on swarm intelligence that simulates the social behavior of bird flocks. It is used to optimize nonlinear functions and has the advantages of simple implementation, high accuracy, and fast convergence. The PSO algorithm is a swarm intelligence technology used to optimize nonlinear functions. It tracks the position, speed, and fitness value of particles, and regards the solution of each problem as a particle in the search space. These particles form a particle swarm and work together to find the optimal solution. The PSO algorithm quickly finds the global or approximate optimal solution through information sharing and collaboration between particles. Particles dynamically adjust their direction and speed based on individual and group experience to balance global and local searches and avoid local optimality. Figure 5 (d) is the algorithm diagram of the PSO optimization algorithm.

[0077] The database is randomly divided into two parts in a ratio of 4:1, namely the training set and the test set. Many similar studies have also adopted this specific data division ratio. 80% of the data in the database constitutes the training set, which is used in the training and optimization of the machine learning model. 20% of the data in the database constitutes the test set, which is used to evaluate the generalization ability of the machine learning model on unseen data. The verification results of the test set are only used for the final evaluation of the performance of the machine learning model.

[0078] The present invention proposes 12 machine learning models to predict the compressive strength of recycled aggregate concrete. The trained machine learning model is applied to the test data set to evaluate the prediction performance of the model. In order to evaluate the generalization ability of the model, the present invention uses 10 performance evaluation indicators of the machine learning model, including R2, MSE, RMSE, RRMSE, MAE, MAPE, WMAPE, RSR, VAF, and AIC. The expression of the performance evaluation index of the machine learning model is shown in Table 4. R2 reflects the ability of the machine learning model to explain data changes. VAF measures the ratio of the variability explained by the model to the total variability. The closer R2 and VAF are to 1, the more data changes the machine learning model can explain, showing a better fitting effect. MSE reflects the average of the squared difference between the predicted value and the true value. RMSE reflects the standard deviation of the difference between the predicted value and the true value. RRMSE reflects the relative size of the prediction error of the machine learning model and the data scale. MAE reflects the average of the absolute value of the difference between the predicted value and the true value. MAPE reflects the average of the absolute value of the difference between the predicted value and the true value as a percentage of the true value. WMAPE is a weighted form of MAPE, which takes into account the different importance of each sample. RSR reflects the ratio of the standard deviation of the prediction error of the machine learning model to the standard deviation of the true value. AIC measures the balance between the goodness of fit and complexity of the model. The smaller the MSE, RMSE, RRMSE, MAE, MAPE, WMAPE, RSR, and AIC, the smaller the prediction error of the machine learning model and the stronger the prediction accuracy of the machine learning model.

[0079] Table 4 Performance evaluation indicators of machine learning models.

[0080]

[0081]

[0082] Note: is the predicted value. i is the actual value. n is the number of data samples. is the mean. K is the number of parameters in the model. L is the likelihood function value of the model.

[0083] In order to explain the importance of different machine learning models, the prediction capabilities of the 12 machine learning models are ranked based on the calculated performance evaluation indicators, and the optimal prediction model is determined. Formula (3) is the calculation formula of the ranking method.

[0084]

[0085] Among them, Score is the average evaluation result of the prediction performance of the machine learning model. The better the performance of the machine learning model, the smaller the value of Score. The number of performance evaluation indicators of the machine learning model is m. Rank i It is a single evaluation result based on performance indicators.

[0086] Twelve machine learning models were tested 10 times independently, and after screening, the average of these tests was taken as the final evaluation metric for model training. Figure 6 Table 5 lists the calculation results of 10 model evaluation indicators in the training and testing stages of 12 machine learning models. Figure 6 As can be seen from Table 5, the 12 machine learning models have good generalization ability for predicting compressive strength. Figure 6 In (a), R2 is mainly distributed in the range of 0.7486-0.9899, ​​indicating that there is a good nonlinear correlation between the input variables and the predicted values. Formula (2) is used to rank the prediction capabilities of the 12 machine learning models and determine the optimal prediction model. Figure 7 is the ranking result of the predictive ability of 12 machine learning models. Figure 7 In the , the prediction ability of the 12 machine learning models is ranked as follows:

[0087] GA-RF>PSO-BP=RF>GA-SVM>GA-BP>BP>PSO-SVM>SVM>PSO-RF>PSO-RBF>RBF>GA-RBF. Among them, the GA-RF model has the strongest prediction ability, the PSO-BP model and RF are tied for the second place, and the GA-RBF model has the weakest prediction ability.

[0088] In machine learning models, the performance of nonlinear models is affected by many factors. The complexity of machine learning models varies, which determines their ability to fit the data. A machine learning model that is too simple may ignore the key complexities in the data. A machine learning model that is too complex may overfit the data, resulting in overfitting. Overall, the GA-RF model, PSO-BP model, and RF model show better prediction capabilities than other machine learning models. Therefore, the trained GA-RF model, PSO-BP model, and RF model can better predict the compressive strength of recycled coarse aggregate concrete.

[0089] Table 5 Performance evaluation indicators of machine learning models.

[0090]

[0091]

[0092] Figure 8It is the distribution graph of experimental values ​​and predicted values ​​of the machine learning model. Figure 8 It is shown that, except for a few outliers, the data points of almost all models are distributed within the ±20% error boundary range. The data of the training set and the test set are linearly fitted. The range of the degree of fit of the training set is 0.8375-0.9852. The range of the degree of fit of the test set is 0.7562-0.9361. The GA-RF model, the RF model and the PSO-BP model have a higher degree of fit, indicating that the data points of the GA-RF model and the RF model are distributed more closely. The RBF model, the GA-RBF model and the PSO-RBF model have a lower degree of fit, indicating that the data points of the RBF model, the GA-RBF model and the PSO-RBF model are the most dispersed. The fitting effects of other models are roughly the same, and the data points are relatively dispersed, but the differences between them are not significant. This shows that the GA-RF model, the RF model and the PSO-BP model in the present invention have better generalization ability.

[0093] In summary, the high fit of the GA-RF model, RF model, and PSO-BP model on the training set and test set indicates that they perform well in capturing data patterns and making predictions. These models can not only effectively learn data features during training, but also show strong generalization ability on the test set, that is, they can maintain high prediction accuracy on new data. In contrast, the RBF model, GA-RBF model, and PSO-RBF model have low fits and may have problems of overfitting or underfitting, resulting in relatively weak generalization ability. They are unstable when facing new and different data. This means that these models are more adaptable to specific problems, but may require further adjustment and optimization in a wider range of application scenarios.

[0094] Calculate the prediction residuals of different models and draw graphs such as Fig. 9 As shown. Fig. 9It can be seen that the maximum residuals of the BP model, GA-BP model, PSO-BP model, SVM model, GA-SVM model, PSO-SVM model, RF model, GA-RF model, PSO-RF model, RBF model, GA-RBF model and PSO-RBF model are 31.2911MPa, 23.5315MPa, -28.882MPa, 24.3489MPa, 22.3476MPa, -31.8936MPa, 24.4515MPa, 17.9913MPa, 28.0174MPa, 23.8187MPa, 23.6752MPa and 26.7015MPa, respectively. Among them, the GA-RF model has the smallest absolute residual, which is reduced by 73.92%, 30.79%, 60.53%, 35.34%, 24.21%, 77.27%, 35.91%, 55.73%, 32.39%, 31.59% and 48.41% respectively compared with other models.

[0095] These data show that the GA-RF model has a clear advantage in prediction residuals. Using the GA algorithm to optimize the RF model can significantly reduce the prediction error and improve the prediction accuracy. Compared with the basic RF model, the maximum residual of the GA-RF model is reduced by 55.73%, indicating that by selecting appropriate optimization algorithms and models, the reliability of the prediction results can be significantly improved. Fig. 9 It can also be observed that although the GA-RF model has the smallest absolute value of the maximum residual, there are significant differences in the residual distribution of different models. For example, the residual distribution of the BP model and the PSO-BP model is more dispersed, showing greater volatility, while the residual distribution of the GA-BP model and the PSO-SVM model is relatively concentrated and less volatile. In addition, the residual distribution of the RF model, the GA-RF model, and the PSO-RF model also shows a certain concentration trend.

[0096] Fig.10 (a) is the violin plot of the machine learning model. Fig.10 The violin plot in (a) clearly depicts the density distribution of the data, and can intuitively identify the central tendency, dispersion, and distribution differences of the data. Fig.10 (a) It can be seen that the data distribution of the GA-BP model and the PSO-BP model is closer to the experimental value. This shows that the GA-BP model and the PSO-BP model have better performance in terms of prediction accuracy and can better capture the inherent laws and characteristics of the data. The data distribution of the basic BP model is relatively scattered and has a low degree of consistency with the experimental value.

[0097] Fig.10 (b) and Fig.10 (c) are the residual distribution curve and residual distribution diagram of the machine learning model. Fig.10 (b) and Fig.10 (c) It can be seen that the error distribution of the 12 machine learning models is normally distributed, the data is mainly distributed in the range of ±10%, and the average residual is close to zero. Compared with other machine learning models, the SVM model has the least data points in the range of ±10%, and the GA-BP model and the PSO-BP model have the most data points in the range of ±10%. It shows that the GA-BP model and the PSO-BP model have good generalization ability. From the perspective of the kurtosis and skewness of the residual distribution curve, the residual distribution of the GA-BP model and the PSO-BP model is more symmetrical and has a lower kurtosis, indicating that the error distribution of its prediction results is more concentrated and the degree of deviation from the normal distribution is smaller. These two models have strong stability and robustness in dealing with this problem.

[0098] It can be seen that the GA-BP model, PSO-BP model, RF model and GA-RF model have strong prediction performance in predicting the compressive strength of recycled aggregate concrete. The GA-BP model, PSO-BP model, RF model and GA-RF model are selected to calculate the importance of input variables. When the importance score of an input variable is higher, it indicates that the variable is more important in the model. Fig.11 The importance of the input feature variables is shown, where the sum of the importance scores of each input feature variable is 100. Fig.11 It can be seen that X-1 has the highest importance score in the GA-BP model, PSO-BP model, RF model and GA-RF model, and X-7 has the lowest importance score. In the GA-BP model, PSO-BP model and RF model, the top five input variables with higher scores are: X-1>X-2>X-3>X-4>X-5. In the GA-RF model, the top five input variables with higher scores are: X-1>X-2>X-3>X-4>X-9. It can be seen that the curing age (X-1) is the most important factor for the GA-BP model, PSO-BP model, RF model and GA-RF model to predict the compressive strength of recycled aggregate concrete. The second is the water-cement ratio (X-2), sand ratio (X-3), and RCA replacement rate (X-4). Finally, the FA replacement rate (X-5) and FGBFS replacement rate (X-9).

[0099] These results show that when predicting the compressive strength of recycled aggregate concrete, the curing age of concrete (X-1), water-cement ratio (X-2), sand ratio (X-3) and RCA substitution ratio (X-4) are the key factors affecting the compressive strength of recycled aggregate concrete. Although the FA substitution ratio (X-5) and FGBFS substitution ratio (X-9) also have a certain influence, their influence is smaller than that of the first four factors. These findings have important guiding significance for optimizing the mix proportion of recycled aggregate concrete and improving its compressive strength. By adjusting these key factors, the performance of concrete can be effectively improved, so that it can be more widely used in the construction industry.

[0100] The GA-BP model, PSO-BP model, RF model and GA-RF model training models were selected, and the curing age (X-1), water-cement ratio (X-2), sand ratio (X-3) and RCA replacement ratio (X-4) were selected as the research objects for the optimization of the compressive strength parameters of recycled coarse aggregate concrete.

[0101] The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-1 are as follows: Fig.12 (a) As shown. The nonlinear regression fitting of the compressive strength data can be obtained as follows: Fig.12 (b) and formula (4). Among them, X-1 is set to 1d, 4d, 7d, 14d, 28d, 56d and 90d respectively. The other input parameters X-2, X-3, X-4, X-5, X-6, X-7, X-8, X-9, X-10 and X-11 are set to 0.55, 38%, 0%, 0%, 0%, 0%, 0%, 0%, 1.11% and 2620kg / m 3 .

[0102] from Fig.12It can be seen that the compressive strength of recycled coarse aggregate concrete under different machine learning models increases with the increase of curing time. With the extension of curing time, the cement hydration reaction inside the concrete can be fully carried out, which in turn promotes the densification of the cement stone structure and enhances the internal bonding and overall strength of the concrete. The change of concrete strength over time can be divided into three stages: initial hardening (slow strength growth), mid-term strengthening (fast strength growth) and late stabilization (stable strength). The compressive strength of recycled coarse aggregate concrete predicted by different machine learning models increases with the increase of curing time. The compressive strength predicted by the GA-BP model at all curing time points is generally high, especially at 90 days, reaching 57.66MPa. The prediction results of the PSO-BP model are second, with a compressive strength of 57.48MPa at 90 days. The prediction results of the RF model are similar to those of the PSO-BP model in the mid-term (28d-56d), but slightly lower in the early (1d-7d) and late (90d). The prediction results of the GA-RF model were slightly lower than those of other models at all time points, but also reached 53.79 MPa at 90 days. The results show that there are differences in the predictions of concrete compressive strength by different machine learning models, but the overall trend is consistent, that is, the compressive strength of concrete gradually increases with the extension of curing time. This trend is related to the progress of cement hydration reaction inside concrete and structural densification over time, and is a natural law for the improvement of concrete material performance.

[0103]

[0104] Among them, f cu is the compressive strength. x1 is the curing age. S 2 is the coefficient of determination.

[0105] The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-2 are as follows: Fig.13 (a) As shown. The nonlinear regression fitting of the compressive strength data can be obtained as follows: Fig.13 (b) and formula (5). Among them, X-2 is set to 0.3, 0.35, 0.4, 0.45, 0.5 and 0.55 respectively. The other input parameters X-1, X-3, X-4, X-5, X-6, X-7, X-8, X-9, X-10 and X-11 are set to 28d, 38%, 0%, 0%, 0%, 0%, 0%, 0%, 1.11% and 2620kg / m 3 .

[0106] from Fig.13It can be seen that the compressive strength of concrete gradually decreases with the increase of water-cement ratio. A higher water-cement ratio means more water and less cement, which may lead to a decrease in the bonding force of the cement paste, thus affecting the compressive strength of the concrete. The GA-BP model predicts a higher compressive strength at a lower water-cement ratio (0.3-0.4), but its predicted value decreases rapidly with the increase of the water-cement ratio. The prediction results of the PSO-BP model are relatively stable in the entire water-cement ratio range and are close to the actual measured values, showing good prediction accuracy. The prediction results of the RF model are similar to those of the PSO-BP model at a medium water-cement ratio (0.35-0.45), but are slightly lower than those of the PSO-BP model at a higher water-cement ratio. The prediction results of the GA-RF model are slightly lower than those of the other models at all water-cement ratios, especially when the water-cement ratio is 0.55, where the predicted compressive strength is the lowest.

[0107]

[0108] Where x2 is the water-to-cement ratio.

[0109] The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-3 are as follows: Fig.14 (a) As shown. The nonlinear regression fitting of the compressive strength data can be obtained as follows: Fig.14 (b) and formula (6). Among them, X-3 is set to 32%, 34%, 36%, 38%, 40% and 42% in sequence. The other input parameters X-1, X-2, X-4, X-5, X-6, X-7, X-8, X-9, X-10 and X-11 are set to 28d, 0.55, 0%, 0%, 0%, 0%, 0%, 0%, 1.11% and 2620kg / m 3 .

[0110] from Fig.14 It can be seen that with the increase of sand ratio, the compressive strength of recycled coarse aggregate concrete shows a trend of first increasing and then decreasing. When the sand ratio is 38%, the compressive strength of concrete reaches its maximum value. The increase in sand ratio improves the workability of concrete, making the bond between aggregate and cement paste more uniform, thereby improving the compressive strength of concrete to a certain extent. When the sand ratio is too high, the specific surface area of ​​aggregate increases, resulting in more cement paste to wrap the aggregate, thereby reducing the volume of cement paste and then reducing the compressive strength of concrete. Fig.14 The fitting curves in (b) further confirm this trend and show the predictive ability of different machine learning models for compressive strength changes. The fitting curves of the GA-BP model and the PSO-BP model are consistent with the actual data points, especially the PSO-BP model, whose determination coefficient S 2It reached 0.9976, indicating that the model has extremely high prediction accuracy. The fitting effect of the RF model is relatively poor, and the determination coefficient S 2 The fitting effect of GA-RF model is between GA-BP and RF, and the determination coefficient S 2 It is 0.8326.

[0111]

[0112] Among them, x3 is the sand ratio.

[0113] The prediction results of compressive strength of recycled coarse aggregate concrete under different machine learning models and different X-4 are as follows: Fig.15 As shown. The nonlinear regression fitting of the compressive strength data can be obtained as follows: Fig.15 (b) and formula (7). Among them, X-4 is set to 0%, 20%, 40%, 60%, 80% and 100% in sequence. The other input parameters X-1, X-2, X-3, X-5, X-6, X-7, X-8, X-9, X-10 and X-11 are set to 28d, 0.55, 38%, 0%, 0%, 0%, 0%, 0%, 1.11% and 2620kg / m 3 .

[0114] from Fig.15 It can be seen that with the increase of recycled coarse aggregate replacement rate, the compressive strength of recycled coarse aggregate concrete shows a downward trend. The increase of recycled coarse aggregate has a negative impact on the mechanical properties of concrete in many aspects, including interface bonding, internal structure, water absorption, aggregate grading, old mortar influence, aggregate strength and environmental factors. Choosing the appropriate recycled coarse aggregate replacement rate is crucial to optimizing the performance of concrete. However, there are differences in the prediction results of different machine learning models on this trend. Fig.15 (b) shows the prediction ability of different machine learning models for compressive strength changes. The fitting curves of the GA-BP model and the PSO-BP model are consistent with the actual data points, especially the PSO-BP model, whose determination coefficient S 2 It reached 0.9663, indicating that the model has a high prediction accuracy. The fitting effect of the RF model is relatively poor, and the determination coefficient S 2 The GA-RF model has the best fitting effect, and the determination coefficient S 2 It is 0.9893.

[0115]

[0116] Where x4 is the RCA substitution rate.

[0117] The present embodiment also provides an electronic device, including a memory and a processor, wherein the processor is used to execute a program in the memory, so as to implement the above-mentioned method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm. The electronic device can accurately implement the method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm. With the advanced machine learning algorithm optimization strategy, the compressive strength of recycled coarse aggregate concrete is predicted, which effectively improves the accuracy and reliability of the prediction, and has extremely critical application value in related fields such as building material performance evaluation.

[0118] This embodiment also provides a storage medium containing computer executable instructions, which, when executed by a computer processor, is used to execute the above-mentioned method for predicting the compressive strength of recycled coarse aggregate concrete based on the optimized machine learning algorithm. This storage medium provides a convenient way to implement the prediction method. Whether it is in the in-depth exploration of the performance of recycled coarse aggregate concrete by scientific research institutions or in the control of material quality by construction companies, this advanced prediction method can be easily applied with the help of the storage medium, thereby laying a solid foundation for the rational application and promotion of recycled coarse aggregate concrete and promoting the continuous progress and development of the construction industry in the use of sustainable materials.

[0119] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. A method, device, and medium for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm, wherein the prediction method is characterized in that: The following steps are involved: S1. Data collection and processing: Obtain multiple factors that affect the performance of recycled coarse aggregate concrete from literature as input variable data, determine the compressive strength of recycled coarse aggregate concrete as the output variable, and pre-process the collected data; S2. Model construction and combination: Use at least one machine learning model and at least one optimization algorithm to combine the selected machine learning model and optimization algorithm to form a variety of different prediction model combinations; S3, model training and evaluation: divide the preprocessed data in S1 into training sets and test sets, use the training sets to train the various prediction models constructed in S2, so that the models learn the potential relationship between input variables and output variables, and determine the prediction model with the best performance through comprehensive evaluation; S4. Key factor analysis: for the optimal prediction model determined in S3, analyze the importance of input variables to output variables, and determine the key factors affecting the performance of recycled coarse aggregate concrete, wherein the key factors include input variables with higher importance scores in the model, i.e., input variables with a relatively greater influence on the compressive strength of recycled coarse aggregate concrete; S5. Reverse prediction and application: Use the prediction model with optimal performance to reversely predict the key factors and obtain the influence trend of the changes in the key factors on the compressive strength of recycled coarse aggregate concrete.

2. According to claim 1, a method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm, characterized in that: In S1, the preprocessing includes data cleaning to remove outliers, data standardization or normalization to make data features more suitable for subsequent analysis, and correlation analysis to preliminarily explore the relationship between input variables and output variables.

3. According to claim 1, a method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm, characterized in that: In S1, the input variable data includes multiple factors affecting the performance of recycled coarse aggregate concrete as input variables, including: Data on variables related to the properties of recycled coarse aggregate, including recycled coarse aggregate replacement rate, physical properties of recycled coarse aggregate; Variable data related to cementitious materials, including water-cement ratio, mineral admixture replacement rate, etc.; Variable data related to aggregates, including water absorption and apparent density of coarse aggregates; Data on factors related to maintenance conditions.

4. The method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm according to claim 1, characterized in that: In S3, the input variable data includes 11 input variables, namely, curing age (X-1), water-cement ratio (X-2), sand ratio (X-3), RCA substitution rate (X-4), FA substitution rate (X-5), SF substitution rate (X-6), M substitution rate (X-7), FGS substitution rate (X-8), FGBFS substitution rate (X-9), water absorption rate of CA (X-10), and apparent density of CA (X-11).

5. The method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm according to claim 1, characterized in that: In S2, the machine learning model includes a neural network model, a support vector machine model, a tree-based model, and a kernel function-based model.

6. The method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm according to claim 1, characterized in that: In S2, the optimization algorithm includes an algorithm based on evolutionary theory and an algorithm based on swarm intelligence.

7. The method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm according to claim 1, characterized in that: In S2, a variety of performance evaluation indicators are used to evaluate the performance of the trained model on the test set. The performance evaluation indicators include indicators reflecting the goodness of model fit, indicators measuring prediction errors, and indicators reflecting the stability and reliability of the model.

8. The method, device, medium, and prediction method for compressive strength prediction of recycled coarse aggregate concrete based on optimized machine learning algorithm according to claim 1, characterized in that: In S2, four machine learning models are specifically introduced, namely, BP model, SVM model, RF model, and RBF model, and two optimization algorithms, namely, GA optimization algorithm and PSO optimization algorithm, are combined to obtain 12 different machine learning models.

9. An electronic device, comprising a memory and a processor, characterized in that: The processor is used to execute the program in the memory, so as to implement the method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm as described in any one of claims 1 to 8.

10. A storage medium containing computer executable instructions, characterized in that: The storage medium of the computer executable instructions, when executed by a computer processor, is used to execute the method for predicting the compressive strength of recycled coarse aggregate concrete based on an optimized machine learning algorithm as described in any one of claims 1 to 8.

Images