Concrete durability prediction method based on data expansion and machine learning

Abstract

The invention discloses a concrete durability prediction method based on data expansion and machine learning, and the method comprises the steps: 1, collecting concrete mix proportion data and environmental parameters in an engineering environment, and building a standard concrete durability database; 2, performing data expansion on the standard concrete durability database based on a Markov Monte Carlo algorithm; 3, constructing an artificial neural network model, taking the expanded database as training sample input, and optimizing and selecting model hyper-parameters; and 4, selecting the artificial neural network model of the hyper-parameter combination with the optimal performance for training, and realizing accurate prediction of the concrete durability under different mix proportions and environmental influences. According to the method, the Markov Monte Carlo algorithm is adopted for data expansion, the loss of the parameter relationship is avoided by adopting an internal preset model method, the defect that huge data is needed to support the establishment of an accurate machine learning model is overcome, and accurate and efficient prediction of the concrete durability chloride ion diffusion coefficient in an engineering environment is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of concrete performance evaluation, and in particular relates to a concrete durability prediction method based on data expansion and machine learning. Background Art

[0002] Concrete is the most widely used building material in the world today because of its abundant raw materials and low price. Durability, as an important factor affecting the life of concrete, has become a research hotspot in the concrete field in recent years. The chloride ion diffusion coefficient is an important parameter for measuring the durability of concrete. However, traditional testing methods are time-consuming, costly, and unable to quickly and accurately predict the remaining life of concrete; empirical models and simulation calculations are often limited by environmental factors and cannot be universally used.

[0003] As an emerging scientific field, machine learning has been widely used in the design, discovery and characterization of engineering materials. It does not require the establishment of equations as models in advance, but can directly establish the relationship between parameters through data, which is time-saving and accurate. However, in order to ensure the accuracy of the model, it often requires huge data sets to support it, which are difficult to obtain.

[0004] In summary, there is an urgent need to propose a machine learning algorithm that can accurately and efficiently predict the durability performance of concrete with limited data support. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method for predicting concrete durability based on data expansion and machine learning. The method can achieve fast, efficient and accurate prediction of machine learning models under the premise of small data samples to solve the problems raised in the background technology.

[0006] Technical solution: A method for predicting concrete durability based on data expansion and machine learning of the present invention comprises the following steps:

[0007] Step 1: Collect concrete mix ratio data and environmental parameters under engineering environment and establish a standard concrete durability database;

[0008] Step 2: Expand the data of standard concrete durability database based on Markov Monte Carlo algorithm;

[0009] Step 3: Build an artificial neural network model, use the expanded database as training sample input, and optimize the model hyperparameters;

[0010] Step 4: Select the artificial neural network model with the best performance hyperparameter combination for training, measure the accuracy of the artificial neural network, and achieve accurate prediction of concrete durability under different mix proportions and environmental influences.

[0011] Furthermore, in step 1, the concrete mix ratio data includes cement dosage, fly ash dosage, silica fume dosage, blast furnace slag dosage, coarse aggregate dosage, fine aggregate dosage, water dosage and water-cement ratio.

[0012] Furthermore, in step 1, the environmental parameters include exposure time, ambient temperature and chloride ion concentration in seawater.

[0013] Furthermore, in step 1, the durability refers to the chloride ion diffusion coefficient of concrete.

[0014] Furthermore, step 2 specifically includes the following steps:

[0015] Step 2.1, normalize the standard concrete durability database and use normal distribution to describe the distribution of each parameter;

[0016] Step 2.2, training a random forest model based on a standard durability database, and pre-establishing the relationship between mix ratio data and environmental parameters and durability;

[0017] Step 2.3, based on the Markov Monte Carlo algorithm, normal sampling is used for the mix ratio data and environmental parameters in the standard durability database to obtain extended data that conforms to the distribution of the original data;

[0018] Step 2.4, according to the mix ratio data and environmental parameters obtained by data expansion, the pre-established random forest model is used to obtain the expanded durability data;

[0019] Step 2.5: Combine the expanded mix ratio data, environmental parameters and durability data, and use reverse normalization to obtain the expanded standard durability database.

[0020] Furthermore, in step 2.2, the specific hyperparameters of the random forest model are 300 tree regressors, 2 minimum samples required for internal node repartitioning, 1 minimum sample number for leaf nodes, and 20 maximum regressor depth.

[0021] Furthermore, step 2.3 is as follows: a Markov chain is constructed by the Markov Monte Carlo algorithm. The steady-state distribution of the chain is the distribution state of the original data. Through a large number of sampling, samples consistent with the target distribution can be obtained. Through standardization, all parameters are made to obey the standard normal distribution. Therefore, for each parameter X, its probability density function is:

[0022]

[0023] in, represents the parameter variance, exp() represents the natural exponential function, X i Represents the input parameter, μ i represents the parameter mean;

[0024] Use normal distribution as the proposed distribution q(x t+1 |x t ), whose mean current state is x t , the standard deviation is a small constant σ q Represents the sampling step size, then the proposed distribution is in the form of:

[0025]

[0026] Based on the mean and standard deviation of the original data, sample from the normal distribution to generate candidate states and calculate the acceptance probability α(x t , x t+1 ), and decide whether to accept the candidate state based on this probability, and repeat this step until enough data is obtained.

[0027] Furthermore, in step 3, the model hyperparameters include the number of hidden layers, the number of neurons in each layer, the learning rate, the number of training rounds, and the batch size.

[0028] Furthermore, in step 3, the optimization selection model hyperparameters is performed by a grid search method, an Adam optimization algorithm is used to construct a training optimizer, and a loss function is constructed based on the mean squared error loss.

[0029] Furthermore, in step 4, the accuracy of the artificial neural network is measured by the mean absolute error MAE and the coefficient of determination R 2 measure.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention establishes a standard concrete durability database based on an engineering environment, supplements the current status of missing engineering data on chloride ion diffusion coefficient of concrete, and establishes a database standardization process for outlier detection, normalization processing, and correlation analysis; the Markov Monte Carlo algorithm is used for data expansion, the original data is described and expanded by normal distribution, and the internal preset model method is used to avoid the loss of input and output parameter correlation, which overcomes the disadvantage that the previous concrete performance prediction requires huge data to support the establishment of an accurate machine learning model; the optimal machine model prediction result is obtained by optimizing the extended durability database and the machine model hyperparameters. Compared with the traditional machine learning model, the prediction mean absolute error (MAE) is reduced by 10%, and the determination coefficient (R 2 ) was improved by 15%, achieving accurate and efficient prediction of chloride ion diffusion coefficient of concrete durability in engineering environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Flow chart of the method of the present invention.

[0032] Figure 2 This is a diagram showing the principal component analysis results of the extended data and original data in the present invention.

[0033] Figure 3 This is a graph showing the top ten hyperparameter combinations optimized using the grid search method used in the present invention.

[0034] Figure 4 This is a graph showing the prediction results of chloride ion diffusion coefficient of concrete obtained using extended data training.

[0035] Figure 5 This is the prediction result of chloride ion diffusion coefficient of concrete obtained by training using original data. DETAILED DESCRIPTION

[0036] The above and other technical features and advantages of the present invention are described in more detail below in conjunction with the embodiments and drawings.

[0037] Reference Figure 1 As shown in the figure, a concrete durability prediction method based on data expansion and machine learning is proposed. The specific steps are as follows:

[0038] S1. Collect concrete mix data and environmental parameters under engineering environment, and establish a standard concrete durability database. Mix data include cement content, fly ash content, silica fume content, blast furnace slag content, coarse aggregate content, fine aggregate content, water content, and water-cement ratio; environmental parameters include exposure time, ambient temperature, and chloride ion concentration in seawater; the prediction parameter is the durability parameter chloride ion diffusion coefficient. The standard concrete durability database includes 846 sets of data, any data pair includes mix data, environmental parameters, and prediction parameters. The basic information is shown in Table 1 below.

[0039] Table 1 Basic information of standard concrete durability database

[0040]

[0041] Table 1

[0042]

[0043] S2. Expand the database based on the Markov Monte Carlo algorithm, describe the data distribution of the original durability database through normal distribution, and use normal sampling to generate extended data. The specific process is as follows:

[0044] S21. The z-Score method is used to standardize the original durability database so that all its parameters obey a normal distribution with a mean of 0 and a standard deviation of 1. The z-Score standardization formula is:

[0045]

[0046] Among them, Z represents the standardized parameter data point, X represents the original parameter data point, μ represents the parameter mean, and σ represents the parameter standard deviation.

[0047] S22. Using the existing 846 sets of data, random forest is used as the basic model to pre-establish the relationship between the mix ratio data and environmental parameters and the predicted parameter chloride ion diffusion coefficient, so as to retain the connection between them during the data expansion process. The structure of the random forest model is set as follows: 300 tree regressors, 2 minimum samples required for internal node re-division, 1 minimum sample number for leaf nodes, and 20 maximum regressor depth. Finally, the preset chloride ion diffusion coefficient prediction model is obtained.

[0048] S23. Based on the Markov Monte Carlo algorithm, normal sampling is used for the mix ratio data and environmental parameters in the standard durability database to obtain extended data that conforms to the distribution of the original data.

[0049] The Markov Monte Carlo algorithm is a powerful tool for sampling from complex probability distributions. It constructs a Markov chain whose steady-state distribution is the distribution of the original data. Through a large number of sampling, samples consistent with the target distribution can be obtained. In this example, we standardize all parameters to obey the standard normal distribution, so for each parameter X, its probability density function is:

[0050]

[0051] in, represents the parameter variance, exp() represents the natural exponential function, X i Represents a parameter (one of the twelve input parameters), μ i represents the parameter mean;

[0052] Use normal distribution as the proposed distribution q(x t+1 |x t ), whose mean current state is x t , the standard deviation is a small constant σ q (represents the sampling step length), the form of the proposed distribution is:

[0053]

[0054] Based on the mean and standard deviation of the original data, sample from the normal distribution to generate candidate states and calculate the acceptance probability α(x t , x t+1 ), and decide whether to accept the candidate state based on this probability. Repeat this process until enough data is obtained. A total of 10,000 sets of new data are obtained through the Markov Monte Carlo process.

[0055] S24. Based on the mix ratio data and environmental parameters obtained through data expansion using the Markov Monte Carlo algorithm, a pre-established random forest prediction model is used to obtain the expanded durability data, retaining the relationship between the parameters.

[0056] S25. Combine the expanded mix ratio data, environmental parameters and durability data, and use the reverse normalization method to obtain the expanded standard durability database. The reverse normalization formula is as follows:

[0057] X=Z×σ+μ

[0058] Among them, X represents the actual parameter data point, Z represents the standardized parameter data point, σ represents the parameter standard deviation, and μ represents the parameter mean.

[0059] The extended data generated by the Markov Monte Carlo algorithm retains the distribution state of the original data, making the generated data as similar as possible to the original data, such as Figure 2 As shown in the figure, through principal component analysis, the distribution of the extended data and the original data in two-dimensional space is basically the same; at the same time, the preset random forest model is used to retain the relationship between the mix ratio data and environmental parameters and the chloride ion diffusion coefficient. Compared with the original data, the extended data has a large amount of data, and the data is continuous and closer to the normal distribution, which is conducive to the training and analysis of the subsequent machine learning model.

[0060] S3. Build an artificial neural network model, use the expanded database as training sample input, and optimize the model hyperparameters. The value ranges of the artificial neural network model hyperparameters are shown in Table 2:

[0061] Table 2. Intervals of hyperparameter values ​​for artificial neural networks

[0062] Training rounds: {10,20,50,150,250} Number of neurons in each layer: {5,10,15,20,50} Learning Rate: {0.01,0.001,0.0001} Batch size: {32,64,128,256,512} Number of hidden layers: {1,2,3}

[0063] The specific process of training artificial neural network hyperparameters is as follows:

[0064] S31. Divide the data set into a training set and a test set, where the ratio of the number of data in the training set to the test set is 8:2.

[0065] S32. Optimize hyperparameters through grid search, use Adam optimization algorithm to build training optimizer, build loss function based on mean square error loss, and train hyperparameters with the goal of minimizing loss function until the best hyperparameter combination is obtained: 250 training rounds, 20 neurons per layer, 0.01 learning rate, 128 batch size, 2 hidden layers, and the top ten hyperparameter combinations are as follows: Figure 3 shown.

[0066] S4. Substitute the best performance hyperparameter combination into the artificial neural network model, use the expanded concrete durability database for training, and predict the chloride ion diffusion coefficient of concrete under different parameters to be predicted. Finally, the following is obtained: Figure 4 , 5 The model performance parameters are mean absolute error (MAE) 1.23, determination coefficient (R 2 )0.89, compared with the performance before data expansion (MAE 2.19, R 2 0.72) have been greatly improved.

Claims

1. A concrete durability prediction method based on data expansion and machine learning, characterized in that: The steps include: Step 1: Collect concrete mix ratio data and environmental parameters under engineering environment and establish a standard concrete durability database; Step 2: Expand the data of standard concrete durability database based on Markov Monte Carlo algorithm; Step 3: Build an artificial neural network model, use the expanded database as training sample input, and optimize the model hyperparameters; Step 4: Select the artificial neural network model with the best performance hyperparameter combination for training, measure the accuracy of the artificial neural network, and achieve accurate prediction of concrete durability under different mix proportions and environmental influences.

2. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 1, the concrete mix ratio data includes cement dosage, fly ash dosage, silica fume dosage, blast furnace slag dosage, coarse aggregate dosage, fine aggregate dosage, water dosage and water-cement ratio.

3. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 1, the environmental parameters include exposure time, ambient temperature and chloride ion concentration in seawater.

4. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 1, the durability refers to the chloride ion diffusion coefficient of concrete.

5. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1, normalize the standard concrete durability database and use normal distribution to describe the distribution of each parameter; Step 2.2, training a random forest model based on a standard durability database, and pre-establishing the relationship between mix ratio data and environmental parameters and durability; Step 2.3, based on the Markov Monte Carlo algorithm, normal sampling is used for the mix ratio data and environmental parameters in the standard durability database to obtain extended data that conforms to the distribution of the original data; Step 2.4, according to the mix ratio data and environmental parameters obtained by data expansion, the pre-established random forest model is used to obtain the expanded durability data; Step 2.5: Combine the expanded mix ratio data, environmental parameters and durability data, and use reverse normalization to obtain the expanded standard durability database.

6. The method for predicting concrete durability based on data expansion and machine learning according to claim 5, characterized in that: In step 2.2, the specific hyperparameters of the random forest model are 300 tree regressors, 2 minimum samples required for internal node repartitioning, 1 minimum sample number for leaf nodes, and 20 maximum regressor depth.

7. The method for predicting concrete durability based on data expansion and machine learning according to claim 5, characterized in that: Step 2.3 is as follows: a Markov chain is constructed by using the Markov Monte Carlo algorithm. The steady-state distribution of the chain is the distribution state of the original data. Through a large number of sampling, samples consistent with the target distribution can be obtained. Through standardization, all parameters are made to obey the standard normal distribution. Therefore, for each parameter X, its probability density function is: in, represents the parameter variance, exp() represents the natural exponential function, X i Represents the input parameter, μ i represents the parameter mean; Use normal distribution as the proposed distribution q(x t+1 |x t ), whose mean current state is x t , the standard deviation is a small constant σ q Represents the sampling step size, then the proposed distribution is in the form of: Based on the mean and standard deviation of the original data, sample from the normal distribution to generate candidate states and calculate the acceptance probability α(x t , x t+1 ), and decide whether to accept the candidate state based on this probability, and repeat this step until enough data is obtained.

8. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 3, the model hyperparameters include the number of hidden layers, the number of neurons in each layer, the learning rate, the number of training rounds, and the batch size.

9. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 3, the optimization selection model hyperparameters is performed by a grid search method, the Adam optimization algorithm is used to construct a training optimizer, and the loss function is constructed based on the mean square error loss.

10. The method for predicting concrete durability based on data expansion and machine learning according to claim 1, characterized in that: In step 4, the accuracy of the artificial neural network is measured by the mean absolute error MAE and the coefficient of determination R 2 measure.

Images