Method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning

Abstract

The invention relates to the technical field of concrete performance prediction, in particular to a method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning, which comprises the following steps: establishing a data set, establishing an optimal prediction model, dividing the data set and substituting the data set into the optimal prediction model, and performing parameter optimization on the optimal prediction model. And predicting the residual compressive strength. By comparing a plurality of different machine learning algorithms, the optimal machine learning algorithm suitable for predicting the residual compressive strength of the ultra-high performance concrete is found, the optimal prediction model of the residual compressive strength of the ultra-high performance concrete is obtained, and the residual compressive strength of the UHPC under the high-temperature condition can be predicted more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete performance prediction, and in particular to a method for predicting the residual compressive strength of ultra-high performance concrete after high temperature based on machine learning. Background Art

[0002] Ultra-high performance concrete is a new type of concrete, mainly composed of cement, admixtures, fine aggregate, fiber, etc., with ultra-high mechanical properties and toughness as well as excellent durability. UHPC is widely used in bridge construction, high-rise buildings, bay construction, tunnel engineering and other fields, and has become an indispensable building material for contemporary architecture.

[0003] According to statistics from China Fire Department, 450,000 fires were reported nationwide from January to May 2024, of which building fires accounted for the highest proportion, with a total of 138,000. With the development of the times, the proportion of high-rise buildings in my country has increased year by year. When a fire occurs, it will seriously affect the overall stability of high-rise buildings and even cause the structure to collapse continuously. This will cause huge economic losses and serious casualties, so it is urgent to effectively predict the fire resistance of building materials.

[0004] Ultra-high performance concrete is particularly susceptible to temperature because as the strength of concrete increases, the density of the concrete will become better and the porosity will become lower. It is easy to generate a large pore pressure at high temperatures, and the risk of high-temperature bursting is relatively higher. In addition, as the density of the material increases and the permeability is lower, the pore pressure will increase at extremely high temperatures, making it more prone to brittle failure. At the same time, fire resistance tests require a lot of time and money, and it is difficult to effectively evaluate the fire resistance of each project. In order to improve the fire resistance of the material and make it have better bearing capacity after a fire, it is very important to predict the residual compressive strength of ultra-high performance concrete after a fire.

[0005] In the past few decades, with the rapid development of artificial intelligence technology, machine learning models have been widely used in various fields. Many civil engineering problems that are difficult to solve by complex experiments have been effectively solved with the assistance of machine learning models. Machine learning can be used to predict the residual compressive strength of ultra-high performance concrete after high temperature.

[0006] The reaction of ultra-high performance concrete under fire is relatively complex. Not only will complex chemical reactions occur, but it will also be affected by changes in the physical characteristics of ultra-high performance concrete materials. Therefore, it is difficult to derive a more accurate formula when predicting the residual compressive strength of ultra-high performance concrete. Existing technologies mainly use heat conduction models to predict and analyze the fire resistance of ultra-high performance concrete. The heat conduction model considers the thermal conductivity, specific heat capacity, density and other thermal physical properties of ultra-high performance concrete, as well as the temperature distribution and duration of the fire. The temperature changes inside the ultra-high performance concrete can be simulated through ABAQUS software. These temperature data can be used to evaluate the degree of damage and fire resistance of ultra-high performance concrete. The model prediction process is relatively complex and the accuracy is not high. Summary of the invention

[0007] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for predicting the residual compressive strength of ultra-high performance concrete after high temperature based on machine learning. The present invention utilizes machine learning and only needs to collect relevant fire resistance data and effectively train the model. It can predict the residual compressive strength of ultra-high performance concrete with high accuracy. At the same time, it can also effectively observe the influence of various materials on the residual compressive strength of ultra-high performance concrete, thereby promoting the research and development of the fire resistance of ultra-high performance concrete.

[0008] To achieve the above object, the present invention is implemented by the following technical solutions:

[0009] A method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning, comprising the following steps:

[0010] Step 1: Establish a data set: By collecting domestic and foreign publicly published literature, and based on the data in the literature and the principle of fire resistance, select the input parameters that have a greater impact on the residual compressive strength of ultra-high performance concrete, and then establish a data set of residual compressive strength of ultra-high performance concrete;

[0011] Step 2: Establishing the optimal prediction model: Based on the data set, a plurality of different machine learning algorithms are used for training, and the machine learning algorithm with the best training effect is selected as the optimal prediction model;

[0012] Step 3: Divide the data set into a training set, a validation set, and a test set, and bring them into the optimal prediction model;

[0013] Step 4: Optimize the parameters of the optimal prediction model according to the input parameters and the parameter characteristics of the optimal prediction model;

[0014] Step 5: Prediction of residual compressive strength: Input the parameters of the ultra-high performance concrete to be predicted into the optimal prediction model to obtain the predicted value of the residual compressive strength.

[0015] In the above scheme, by using multiple machine learning algorithms for training and selecting the optimal prediction model, the residual compressive strength of UHPC under high temperature conditions can be predicted more accurately, which helps to reduce experimental costs, accelerate the research and development of new materials, and improve the accuracy of predictions.

[0016] As a further optimization scheme of the present invention, in step 1, the input parameters include the following materials: water, cement, silica fume, fly ash, mineral powder, coarse aggregate, fine aggregate, water reducer, steel fiber, synthetic fiber, heating rate, constant temperature time, specimen size, compressive strength at room temperature, and temperature.

[0017] In the above scheme, the specific content of the input parameters is clarified to ensure the comprehensiveness and reliability of the data set. By covering all key variables, it can more accurately reflect the behavior of ultra-high performance concrete under high temperature conditions.

[0018] As a further optimization scheme of the present invention, the units of water, cement, silica fume, fly ash, mineral powder, coarse aggregate, fine aggregate, water reducing agent, steel fiber and synthetic fiber in the input parameters are kg / m 3 ; The unit of heating rate is ℃ / min; the unit of constant temperature time is h; the unit of specimen size is mm; the unit of compressive strength at room temperature is MPa; the unit of temperature is ℃.

[0019] In the above scheme, the units of input parameters are specified to ensure the consistency and standardization of data; this is conducive to improving the comparability and repeatability of the model and ensuring the accuracy of the prediction results.

[0020] As a further optimization scheme of the present invention, in step 2, a five-fold cross-validation technique is used to train multiple different machine learning algorithms to obtain the optimal prediction model.

[0021] As a further optimization scheme of the present invention, the five-fold cross-validation technology includes: randomly dividing the data set in step 1 into five parts, selecting one part of the five parts for training in turn, and using the remaining four parts for testing, training to obtain five determination coefficients of multiple different machine learning algorithms, taking the average value of the determination coefficients, and the machine learning algorithm with the average value closest to 1 is used as the optimal prediction model.

[0022] In the above scheme, the five-fold cross-validation technology is used for model training, which improves the stability and generalization ability of the model; through multiple training and testing, the performance of the model can be evaluated more accurately and the risk of overfitting can be reduced.

[0023] As a further optimization scheme of the present invention, in step 2, the multiple different machine learning algorithms include extreme gradient boosting algorithm (XGBoost), support vector machine algorithm (SVM), random forest algorithm (RF) and boosted decision tree algorithm (LSBoost).

[0024] In the above scheme, a variety of different machine learning algorithms are listed, providing a rich range of choices; by comparing the effects of multiple algorithms, the best model that best suits the current task can be selected.

[0025] As a further optimization scheme of the present invention, in step 3, the training set is used to obtain the optimal prediction model for the residual compressive strength of ultra-high performance concrete, the validation set is used to adjust the hyperparameters of the optimal prediction model, evaluate the generalization ability of the model, and prevent overfitting, and the test set is used to evaluate the final result of the optimal prediction model.

[0026] As a further optimization scheme of the present invention, in step 3, the training set accounts for 70% of the data set, and the validation set and the test set each account for 15% of the data set.

[0027] In the above scheme, the purpose of the training set, validation set and test set is clearly defined, ensuring the effectiveness of model training and the accuracy of evaluation; through reasonable data division, the performance of the model can be effectively evaluated and the occurrence of overfitting can be prevented.

[0028] As a further optimization scheme of the present invention, in step 4, the extreme gradient boosting algorithm is used as the optimal prediction model, and the adjustment of its parameters is divided into general parameters, Booster parameters and learning target parameters. The general parameters are controlled by macro functions, the Booster parameters control the booster of each step, and the learning target parameters control the performance of the training target. The default values ​​are usually used for the general parameters, and the learning target parameters and the Booster parameters need to be appropriately tuned according to the parameter selection of the data set.

[0029] In the above scheme, the parameter adjustment method when the extreme gradient boosting algorithm is used as the optimal prediction model is explained in detail; through refined parameter adjustment, the prediction performance of the model can be improved and the effectiveness of the model in practical applications can be ensured.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) The present invention compares a variety of different machine learning algorithms to find the best machine learning algorithm suitable for predicting the residual compressive strength of ultra-high performance concrete, and obtains the optimal prediction model for the residual compressive strength of ultra-high performance concrete, which can more accurately predict the residual compressive strength of UHPC under high temperature conditions. Model training through five-fold cross-validation technology can not only improve the accuracy of the model, but also effectively evaluate the generalization ability of the model and reduce the risk of overfitting. Through a meticulous parameter adjustment process, especially parameter optimization for the extreme gradient boosting algorithm, the accuracy and stability of the prediction model can be further improved.

[0032] (2) The present invention collects publicly published papers and documents at home and abroad as data sets, and the trained model has more universal applicability. The method relies on a large amount of experimental data. By analyzing and modeling these data, more reliable prediction results can be obtained, which has more practical application value than traditional empirical formulas or theoretical calculations. Its input parameters are relatively easy to obtain, which can provide a reference for the fire resistance test of ultra-high performance concrete. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is the schematic diagram of the five-fold cross fold.

[0034] Figure 2 Determination coefficient R for four machine learning models under 5-fold cross validation technique 2 Box plot of .

[0035] Figure 3 This is a comparison chart of the residual compressive strength test value of ultra-high performance concrete and the predicted value of the XGBoost model. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the technical solution of the present invention, the preferred implementation scheme of the present invention is described below in conjunction with specific embodiments, but it should not be understood as limiting the present invention and is only used as an example.

[0037] The experimental methods or test methods described in the following examples are conventional methods unless otherwise specified.

[0038] This embodiment provides a method for predicting residual compressive strength of ultra-high performance concrete based on machine learning, and the detailed steps are described as follows:

[0039] (1) Collect all domestic and foreign literature related to ultra-high performance concrete fire resistance. Summarize the data in the literature in turn and collect them into Excel files. For data points that are not marked with details, use Getdata software to collect data. The collected data needs to be further screened. For a single input parameter, if the amount of data is small, this part of the data should be eliminated. The reason is that the amount of data for this parameter is small, and machine learning requires a large amount of data sets as a training model to have a better effect. The collected data is shown in Table 1.

[0040] Table 1 Dataset parameters (482 points in total)

[0041]

[0042]

[0043] Note: Q25% and Q75% in the table represent the 25th and 75th percentiles respectively.

[0044] (2) Four machine learning models were constructed, namely, extreme gradient boosting (XGBoost), support vector machine (SVM), random forest (RF), and boosted decision tree (LSBoost). These four machine learning models are all from models that are currently widely used in the field of concrete. Among them, XGBoost is a machine learning algorithm based on the gradient boosting framework, which is used to solve regression and classification problems; SVM is a commonly used supervised learning model, mainly used for classification and regression analysis; RF is an ensemble learning algorithm based on decision trees. It improves the stability and generalization ability of the model by constructing multiple decision trees and combining their outputs; LSBoost is an ensemble learning algorithm based on decision trees, which gradually improves the performance of the model by iteratively training decision trees.

[0045] (3) The prediction effects of four machine learning models were evaluated using a five-fold cross-validation technique. Figure 1 As shown, the data set in step (1) is evenly divided into five parts, one of which is selected for training, and the remaining four are used for testing. The five coefficients of determination of the four machine learning test sets are obtained and the average is taken.

[0046] (4) Draw a box plot based on the determination coefficient obtained by the five-fold cross-validation technique in step (3) as shown in Figure 2 As shown. Figure 2It can be seen that, except for XGBoost, other machine learning algorithms have outliers. The larger the average determination coefficient and the closer it is to 1, the better the performance of the model. The XGBoost model has the largest average determination coefficient and has higher accuracy in predicting residual compressive strength. Combining outliers and average determination coefficients, and comparing the four machine learning algorithms, it can be found that XGBoost has the best performance, better stability and higher prediction accuracy. That is, XGBoost is the best model for predicting the residual compressive strength of ultra-high performance concrete.

[0047] (5) The data set in step (1) is randomly divided into three subsets, namely, training set, validation set, and test set. The training set is 70%, i.e., 337 samples, and the validation set and test set are both 15%, i.e., 72 samples. 338 samples are used for XGBoost training, 72 samples are used for validation, and the last 72 samples are used for testing. The training set is used to obtain a prediction model for the residual compressive strength of ultra-high performance concrete, the validation set is used to adjust the model's hyperparameters, evaluate the model's generalization ability, and prevent overfitting, and the test set is used to evaluate the final result of the model.

[0048] (6) Adjust the parameters according to the characteristics of the XGBoost algorithm. The parameters that need to be adjusted are generally the learning target parameters and the Booster parameters. Among them, the Booster parameters can first set all its parameters to the default values, and then adjust the maximum depth of the tree, the largest node on the tree, the weight of the tree and other parameters according to the data characteristics. The learning target parameters are used to control the ideal optimization goal and the measurement method of each step result. Take the adjustment of the eval_metric parameter as an example. For regression problems, the default value is rmse, and for classification problems, the default value is error. This data set belongs to a regression problem, so its default value is rmse. Finally, adjust until the determination coefficient of the XGBoost test set is close to 1.

[0049] (7) The simulation results of the final test set are as follows Figure 3 As shown by Figure 3 It can be seen that the test value and the predicted value are in good agreement. That is, XGBoost has a higher accuracy in predicting the residual compressive strength of ultra-high performance concrete. The trained model can be used to predict the residual compressive strength of ultra-high performance concrete after high temperature.

[0050] In summary, the present invention provides a prediction method based on machine learning, which can accurately predict the residual compressive strength of ultra-high performance concrete under high temperature conditions, which can not only improve the accuracy of prediction, but also reduce costs and speed up the development of new materials. Through strict model training and verification processes, the stability and generalization ability of the model are ensured, and the credibility and practicality of the model are also improved.

[0051] The above are only preferred embodiments of the present invention. It should be noted that the above preferred embodiments should not be regarded as limiting the present invention, and the protection scope of the present invention should be based on the scope defined by the claims. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the spirit and scope of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning, characterized in that: The following steps are involved: Step 1: Establish a data set: By collecting domestic and foreign publicly published literature, select the input parameters that have a greater impact on the residual compressive strength of ultra-high performance concrete, and then establish a data set of residual compressive strength of ultra-high performance concrete; Step 2: Establishing the optimal prediction model: Based on the data set, a plurality of different machine learning algorithms are used for training, and the machine learning algorithm with the best training effect is selected as the optimal prediction model; Step 3: Divide the data set into a training set, a validation set, and a test set, and bring them into the optimal prediction model; Step 4: Optimize the parameters of the optimal prediction model according to the input parameters and the parameter characteristics of the optimal prediction model; Step 5: Prediction of residual compressive strength: Input the parameters of the ultra-high performance concrete to be predicted into the optimal prediction model to obtain the predicted value of the residual compressive strength.

2. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1 is characterized in that: In step 1, the input parameters include the following materials: water, cement, silica fume, fly ash, mineral powder, coarse aggregate, fine aggregate, water reducing agent, steel fiber, synthetic fiber, heating rate, constant temperature time, specimen size, compressive strength at room temperature, and temperature.

3. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 2 is characterized in that: The unit of water, cement, silica fume, fly ash, mineral powder, coarse aggregate, fine aggregate, water reducing agent, steel fiber and synthetic fiber in the input parameters is kg / m 3 ; The unit of heating rate is ℃ / min; the unit of constant temperature time is h; the unit of specimen size is mm; the unit of compressive strength at room temperature is MPa; the unit of temperature is ℃.

4. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1 is characterized in that: In step 2, a five-fold cross-validation technique is used to train a variety of different machine learning algorithms to obtain the optimal prediction model.

5. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 4 is characterized in that: The five-fold cross validation technique includes: The data set in step 1 is randomly divided into five parts, and one part of the five parts is selected for training, and the remaining four parts are used for testing. Five determination coefficients of multiple different machine learning algorithms are obtained through training, and the average value of the determination coefficient is taken. The machine learning algorithm with the average value closest to 1 is taken as the optimal prediction model.

6. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1 or 4, characterized in that: In step 2, the multiple different machine learning algorithms include extreme gradient boosting algorithm, support vector machine algorithm, random forest algorithm and boosted decision tree algorithm.

7. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1 is characterized in that: In step 3, the training set is used to obtain the optimal prediction model for the residual compressive strength of ultra-high performance concrete, the validation set is used to adjust the hyperparameters of the optimal prediction model, evaluate the generalization ability of the model, and prevent overfitting, and the test set is used to evaluate the final result of the optimal prediction model.

8. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1 or 7, characterized in that: In step 3, the training set accounts for 70% of the data set, and the validation set and the test set each account for 15% of the data set.

9. The method for predicting residual compressive strength of ultra-high performance concrete after high temperature based on machine learning according to claim 1, characterized in that: In step 4, the extreme gradient boosting algorithm is used as the optimal prediction model, and the adjustment of its parameters is divided into general parameters, Booster parameters and learning target parameters. The general parameters are controlled by macro functions, the Booster parameters control the booster of each step, and the learning target parameters control the performance of the training target. The default values ​​are usually used for the general parameters, and the learning target parameters and the Booster parameters need to be appropriately tuned according to the parameter selection of the data set.

Images