A priori knowledge fusion-based asphalt mixture dynamic modulus prediction method and system

Abstract

The application provides an asphalt mixture dynamic modulus prediction method and system based on prior knowledge fusion; the method comprises the following steps: S1, acquiring the dynamic modulus of the asphalt mixture and characteristic variables thereof; S2, optimizing the characteristic variables based on a prior knowledge model, acquiring key characteristic variables, and constructing a dynamic modulus dataset according to the key characteristic variables and the dynamic modulus of the asphalt mixture; S3, constructing an asphalt mixture dynamic modulus prediction model by using a machine learning method based on prior knowledge fusion, and performing model parameter searching training; S4, evaluating the prediction accuracy of the asphalt mixture dynamic modulus prediction model after the parameter searching training, and optimizing the asphalt mixture dynamic modulus prediction model with the best fitting effect; and S5, inputting the key characteristic variables of the asphalt mixture to be predicted into the optimized asphalt mixture dynamic modulus prediction model, and acquiring the dynamic modulus of the asphalt mixture to be predicted. The application solves the technical problem of lacking comprehensive performance of a traditional model.

Description

Technical Field

[0001] This invention relates to the field of predicting the mechanical properties of asphalt mixtures, specifically to a method and system for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion. Background Technology

[0002] In high-temperature areas, asphalt pavements are prone to rutting and other defects under repeated heavy loads. In order to obtain asphalt mixtures that meet the service requirements of roads in high-temperature areas, it is necessary to control the high-temperature performance of asphalt mixtures.

[0003] According to the MEPDG mechanical-empirical design method in the United States, dynamic modulus is an important indicator for evaluating the high-temperature performance of asphalt mixtures. However, the dynamic modulus test specimens for asphalt mixtures are complex to form and the testing time and cost are huge. Therefore, empirical formulas or data-driven methods are now mostly used to predict the dynamic modulus of asphalt mixtures. However, although empirical formula methods are applicable to various asphalt mixtures, the prediction accuracy of empirical formula models is often limited; while data-driven models, although they can have higher accuracy, have poor generalization ability and are difficult to promote and apply. Summary of the Invention

[0004] The purpose of this invention is to propose a method and system for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion, which solves the technical problem of insufficient comprehensive performance of traditional models.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] This invention proposes a method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion, comprising the following steps:

[0007] S1. Obtain the dynamic modulus and characteristic variables of asphalt mixture;

[0008] S2. Based on the prior knowledge model, the feature variables are optimized to obtain key feature variables. The key feature variables are used as inputs and the dynamic modulus of asphalt mixture is used as output to construct a dynamic modulus dataset.

[0009] S3. A machine learning method based on prior knowledge fusion is used to construct a dynamic modulus prediction model for asphalt mixtures, and the model is trained by parameter search.

[0010] S4. The prediction accuracy of the asphalt mixture dynamic modulus prediction model after parameter search training is evaluated using data statistical analysis indicators, and the asphalt mixture dynamic modulus prediction model based on prior knowledge fusion with the best fitting effect is selected.

[0011] S5. Input the key characteristic variables of the asphalt mixture to be predicted into the preferred dynamic modulus prediction model of the asphalt mixture to obtain the dynamic modulus of the asphalt mixture to be predicted.

[0012] Preferably, the dynamic modulus of the asphalt mixture and its characteristic variables are obtained from public databases, public literature, or laboratory measured data.

[0013] Preferably, the characteristic variables include the cumulative sieve residue percentage (%) of 19mm sieve, the cumulative sieve residue percentage (%) of 9.5mm sieve, the cumulative sieve residue percentage (%) of 4.75mm sieve, the cumulative sieve residue percentage (%) of 0.075mm sieve, the porosity (%), the effective asphalt content (%), the aggregate void ratio, the asphalt saturation (%), the logarithmic value of asphalt viscosity (cps), the dynamic modulus loading frequency (Hz), the temperature (°C), the asphalt viscosity parameter A, the asphalt viscosity parameter VTS, the logarithmic value of the asphalt dynamic modulus (psi), and the logarithmic value of the asphalt phase angle (°).

[0014] Preferably, the prior knowledge model includes the Witczak model, the improved Witczak model, the modified Bari-Witczak model, the Hirsch model, the modified Hirsch model, the AI-Khateeb model, the Global model, and the simplified Global model.

[0015] Preferably, the step of optimizing feature variables based on prior knowledge models to obtain key feature variables specifically includes: selecting prior knowledge models according to the types of feature variables obtained in step S1, and setting the input parameters of the selected prior knowledge models as key feature variables to reduce the workload of feature selection in feature engineering.

[0016] Preferably, step S3 specifically includes the following steps:

[0017] S31. Based on the selected prior knowledge model, calculate the prior predicted value of the dynamic modulus of asphalt mixture for all samples in the dynamic modulus dataset according to the key feature variable values.

[0018] S32. Subtract the true dynamic modulus values ​​of all samples in the dynamic modulus dataset from the prior predicted values ​​of the dynamic modulus of asphalt mixture to obtain the dynamic modulus difference of asphalt mixture for all samples; use the key feature variable values ​​as the actual input of the machine learning model and the dynamic modulus difference of asphalt mixture as the output of the machine learning model to construct a dataset based on the key feature variable values ​​and the corresponding dynamic modulus difference of asphalt mixture.

[0019] S33. Divide the dataset constructed in S32 into a training set and a test set, and normalize the training set and the test set respectively; wherein, the normalized training set is used for parameter finding training of the machine learning model, and the normalized test set is used for machine learning model evaluation.

[0020] S34. Use one or more machine learning algorithms to participate in the parameter search and training of the machine learning model: Use genetic algorithms to search for parameters of the machine learning model, and train the machine learning model based on the optimal model hyperparameters obtained from the parameter search.

[0021] S35. Combining the machine learning model trained by parameter seeking and the prior knowledge model, construct a dynamic modulus prediction model for asphalt mixture based on prior knowledge fusion; the output of the constructed dynamic modulus prediction model for asphalt mixture is the sum of the prediction results of the machine learning model and the prior knowledge model.

[0022] Preferably, the optimization objective of the genetic algorithm is to minimize the coefficient of determination R of the k-fold cross-validation of the training set. 2 score.

[0023] Preferably, the machine learning algorithm includes support vector machine regression, multilayer perceptron, random forest, and gradient boosting tree.

[0024] Preferably, the statistical analysis indicators include the coefficient of determination R², mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The closer R² is to 1, the smaller the MSE, MAE, and MAPE, the better the model fit.

[0025] The present invention also proposes an asphalt mixture dynamic modulus prediction system based on prior knowledge fusion, including a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs any of the steps in the above-mentioned asphalt mixture dynamic modulus prediction method.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention enables the rapid selection of key feature variables that play a crucial role in improving the performance of dynamic modulus prediction models based on prior knowledge from empirical formula methods, thereby reducing the workload of feature engineering in machine learning methods. Furthermore, by introducing the predicted values ​​of the empirical formula prior knowledge model into the prediction task of the machine learning model, the empirical formula method and the data-driven method are organically integrated, further improving the prediction accuracy of dynamic modulus. Because the dynamic modulus prediction model relies on the prior knowledge model, it has better generalization and a wider range of application scenarios. At the same time, its data-driven core ensures the prediction accuracy of the model. Attached Figure Description

[0028] Figure 1 This is a flowchart of the dynamic modulus prediction method for asphalt mixtures based on prior knowledge fusion according to the present invention.

[0029] Figure 2 This is a schematic diagram of the structure and calculation process of the dynamic modulus prediction model for asphalt mixtures based on prior knowledge fusion according to the present invention.

[0030] Figure 3 This is a schematic diagram illustrating the predictive performance of the dynamic modulus prediction model for asphalt mixtures based on prior knowledge fusion, as per the present invention. Detailed Implementation

[0031] The following is in conjunction with the appendix Figure 1-3 The specific implementation methods of the embodiments of the present invention will be described in detail to demonstrate the process of achieving the technical effects of this application and enhance the understanding of the technical solution. It should be noted that the specific implementation methods described herein are only for illustration and explanation of the embodiments of the present invention and are not intended to limit the present invention.

[0032] In high-temperature areas, asphalt pavements are prone to rutting and other defects under repeated heavy loads. In order to obtain asphalt mixtures that meet the service requirements of roads in high-temperature areas, it is necessary to control the high-temperature performance of asphalt mixtures.

[0033] According to the MEPDG mechanical-empirical design method in the United States, dynamic modulus is an important indicator for evaluating the high-temperature performance of asphalt mixtures. Currently, empirical formulas or data-driven methods are mostly used to predict the dynamic modulus of asphalt mixtures. However, although empirical formula methods have good applicability to various asphalt mixtures, the prediction accuracy of empirical formula models is often limited; while data-driven models, although capable of higher accuracy, have poor generalization ability and are difficult to promote and apply. In summary, there is an urgent need to find a method that can combine generalization, accuracy, and efficiency to predict the dynamic modulus of asphalt mixtures.

[0034] To solve the technical problems existing in the prior art, referring to Figure 1 This invention proposes a method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion, comprising the following steps:

[0035] S1. Obtain the dynamic modulus of asphalt mixture and its characteristic variables from public databases, public documents or laboratory measured data;

[0036] S2. Based on the prior knowledge model, the feature variables are optimized to obtain key feature variables. The key feature variables are used as inputs and the dynamic modulus of asphalt mixture is used as output to construct a dynamic modulus dataset.

[0037] S3. A machine learning method based on prior knowledge fusion is used to construct a dynamic modulus prediction model for asphalt mixtures, and the model is trained by parameter search.

[0038] S4. The prediction accuracy of the asphalt mixture dynamic modulus prediction model after parameter search training is evaluated using data statistical analysis indicators, and the asphalt mixture dynamic modulus prediction model based on prior knowledge fusion with the best fitting effect is selected.

[0039] S5. Input the key characteristic variables of the asphalt mixture to be predicted into the dynamic modulus prediction model of the asphalt mixture fused with prior knowledge, and obtain the dynamic modulus of the asphalt mixture to be predicted.

[0040] Reference Figure 2 One specific embodiment is shown:

[0041] S1. Obtain data on the dynamic modulus and related characteristic variables of several asphalt mixtures from the publicly available dataset NCHRP 9-19. The related characteristic variables in the publicly available dataset include the cumulative sieve residue percentage (%) of 19 mm sieve, the cumulative sieve residue percentage (%) of 9.5 mm sieve, the cumulative sieve residue percentage (%) of 4.75 mm sieve, the cumulative sieve residue percentage (%) of 0.075 mm sieve, the porosity (%), the effective asphalt content (%), the logarithmic value of asphalt viscosity (cps), the frequency (Hz), the temperature (°C), the asphalt viscosity parameter A, the asphalt viscosity parameter VTS, the logarithmic value of asphalt dynamic modulus (psi), and the logarithmic value of asphalt phase angle (°).

[0042] S2. Select the improved Witczak model as the prior knowledge model. The formula for this prior knowledge model is shown below:

[0043]

[0044] In the formula: The dynamic modulus of asphalt;

[0045] δ b The asphalt phase angle;

[0046] p 0.075 The cumulative residue percentage (%) is for a 0.075mm sieve.

[0047] p 4.75 The cumulative residue percentage (%) on a 4.75mm sieve;

[0048] p 9.5 The cumulative sieve residue percentage is 9.5%.

[0049] p 19 The cumulative residue percentage on a 19mm sieve (%)

[0050] Va Porosity (%)

[0051] V beff The effective asphalt content (%) of the asphalt mixture.

[0052] Based on the input of the improved Witczak prior knowledge model, the feature variables are optimized to obtain key feature variables, including the logarithmic value of the asphalt dynamic modulus lg(|G b * |)(psi), logarithmic value of asphalt phase angle lg(σ) a (°), 0.075mm cumulative sieve residue percentage p 0.075 (%), 4.75mm cumulative sieve residue percentage p 4.75 (%), 9.5mm cumulative sieve residue percentage p 9.5 (%), Cumulative residue percentage on 19mm sieve (p) 19 (%), porosity V a (%), effective asphalt content V beff (%), frequency f m (Hz), temperature T (°C). Although frequency and temperature are not explicitly included in the improved Witczak prior knowledge model, they are actually contained in the dynamic modulus and phase angle of asphalt. Therefore, for ease of input, the frequency and temperature corresponding to the dynamic modulus and phase angle of asphalt are explicitly used as key feature variables. Furthermore, using the key feature variables as input and the logarithmic value of the dynamic modulus of asphalt mixture lg|E*| (psi) as output, a dynamic modulus dataset is constructed.

[0053] The data from the collected dynamic modulus dataset, calculated using statistical parameters, are shown in Table 1 below:

[0054] Table 1. Statistical Parameter Calculation Results of Collected Dynamic Modulus Data of Asphalt Mixtures and Key Characteristic Variables

[0055]

[0056]

[0057] This embodiment provides some representative dynamic modulus data of asphalt mixtures through the improved Witczak model. In this embodiment, dynamic modulus and related characteristic variable data can also be extracted from publicly available literature or laboratory test results, and other prior knowledge models can be selected to optimize key characteristic variables as datasets. No limitations are imposed here.

[0058] The inputs of prior knowledge models are usually based on the viscoelastic properties of asphalt, the proportion of aggregates, and the volumetric properties of asphalt mixtures. They can usually fully consider the feature variables that have a significant impact on the dynamic modulus of asphalt mixtures. The purpose of feature engineering is to select feature variables that have a significant impact on the target. Therefore, integrating prior knowledge models can efficiently select feature variables and significantly reduce the workload of engineers or road designers.

[0059] S3. A machine learning method based on prior knowledge fusion is used to construct a dynamic modulus prediction model for asphalt mixtures, and the model is trained by parameter finding. Step S3 specifically includes the following steps:

[0060] S31. Using the improved Witczak prior knowledge model, the prior predicted values ​​of the dynamic modulus of asphalt mixtures for all samples in the dynamic modulus database are calculated based on the values ​​of key feature variables.

[0061] S32. Subtract the true dynamic modulus values ​​of all samples in the dynamic modulus dataset from the prior predicted values ​​of the dynamic modulus of asphalt mixture to obtain the delta value of the dynamic modulus of asphalt mixture for all samples. The actual input of the machine learning model is the key feature variable, and the output is the delta value of the dynamic modulus of asphalt mixture.

[0062] S33. Divide the input and output of the machine learning method into a training set and a test set, where the training set and the test set account for 70-80% and 20-30% of the input and output, respectively. Then, normalize the training set and the test set respectively.

[0063] S34. Support Vector Machine Regression, Multilayer Perceptron, Random Forest, and Gradient Boosting Tree models are selected for parameter finding and training. Only the training set data can be used for parameter finding and training. The genetic algorithm is used for parameter finding, and the optimization objective of the genetic algorithm is to minimize the coefficient of determination R of the k-fold cross-validation on the training set. 2 Scoring; During training, the model hyperparameters are set to the optimal model hyperparameters obtained through parameter search, and the above machine learning model is fitted using the training set data.

[0064] S35. The prediction results of the machine learning model and the improved Witczak prior knowledge model are summed to obtain the output of the asphalt mixture dynamic modulus prediction model, and then the asphalt mixture dynamic modulus prediction model based on prior knowledge fusion is constructed.

[0065] S4. The prediction accuracy of the asphalt mixture dynamic modulus prediction model after parameter training is evaluated using statistical data analysis indicators. These indicators include the coefficient of determination (R²), mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The calculation formulas for each statistical data analysis indicator are shown below:

[0066]

[0067] In the above formula: n is the number of samples, i is the number of input variables, and lg(|E*|) predicted Let lg(|E*|) be the predicted value of the logarithmic dynamic modulus. measured This is the predicted value of the logarithmic value of the dynamic modulus.

[0068] In this embodiment, the statistical analysis results of the data on the predictive performance of the dynamic modulus of asphalt mixtures using different machine learning methods based on prior knowledge fusion are shown in Table 2 below. Furthermore, Figure 3 This further reflects the predictive performance of the above models. Most of the data points of different models are located or distributed near the contour lines, which indicates that there is a good correlation between the predicted values ​​and the measured values ​​of the asphalt mixture dynamic modulus prediction model based on prior knowledge fusion. Furthermore, the asphalt mixture dynamic modulus prediction model composed of the prior knowledge model and the gradient boosting tree model has the best performance.

[0069] Table 2. Statistical Analysis Indicators of Dynamic Modulus Prediction Performance

[0070] Support Vector Machine 0.02914 0.13157 0.02321 0.93786 Multilayer perceptron 0.01474 0.09361 0.01651 0.96857 Random Forest 0.00833 0.06350 0.01153 0.98223 Gradient boosting tree 0.00566 0.05420 0.00963 0.98793

[0071] This embodiment establishes a prediction model for the dynamic modulus of asphalt mixtures by using a machine learning method based on prior knowledge fusion and data from a publicly available dataset. This overcomes the shortcomings of traditional empirical formulas having limited prediction accuracy and data-driven models having poor generalization ability, and finds a method that can predict the dynamic modulus of asphalt mixtures with both generalization and accuracy and high efficiency.

[0072] S5. Input the key characteristic variables of the asphalt mixture to be predicted into the asphalt mixture dynamic modulus prediction model fused with prior knowledge to obtain the dynamic modulus of the asphalt mixture to be predicted. The obtained dynamic modulus of the asphalt mixture is based on the input frequency and temperature, corresponding to the load frequency and service environment temperature of the actual asphalt road. By controlling the dynamic modulus value of the asphalt mixture, a mixture design scheme that can meet the high-temperature service requirements of the actual asphalt road can be quickly obtained, thereby controlling the high-temperature performance of the asphalt mixture.

[0073] In summary, given that dynamic modulus is a key indicator characterizing the high-temperature performance of asphalt mixtures, and addressing the shortcomings of existing technologies, this paper proposes a method based on empirical formulas to quickly select key feature variables that play a crucial role in improving the performance of dynamic modulus prediction models, thereby reducing the workload of feature engineering in machine learning methods. Furthermore, the predicted values ​​from the empirical formula prior knowledge model are introduced into the machine learning model prediction task, organically integrating empirical formula methods and data-driven methods to further improve the prediction accuracy of dynamic modulus. Because the dynamic modulus prediction model relies on prior knowledge models, it has better generalization capabilities and a wider range of application scenarios. Simultaneously, its data-driven core ensures the model's prediction accuracy, safeguarding the high-temperature performance of asphalt mixtures and providing a solid foundation for road design.

[0074] The above embodiments provide a detailed description of the present invention and illustrate the principles and implementation methods of the present invention using specific solutions. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, within the scope of the technical concept of the embodiments of the present invention, any simple combination of the technical solutions of the present invention can be made, but as long as it does not violate the idea of ​​the embodiments of the present invention, it should also be considered to be within the protection scope of the present invention.

Claims

1. A method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion, characterized in that: Includes the following steps: S1. Obtain the dynamic modulus and characteristic variables of asphalt mixture; S2. Based on the prior knowledge model, the feature variables are optimized to obtain key feature variables. The key feature variables are used as inputs and the dynamic modulus of asphalt mixture is used as output to construct a dynamic modulus dataset. S3. A machine learning method based on prior knowledge fusion is used to construct a dynamic modulus prediction model for asphalt mixtures, and the model is trained by parameter finding. The specific steps include: S31. Based on the selected prior knowledge model, calculate the prior predicted value of the dynamic modulus of asphalt mixture for all samples in the dynamic modulus dataset according to the key feature variable values. S32. Subtract the true dynamic modulus values ​​of all samples in the dynamic modulus dataset from the prior predicted values ​​of the dynamic modulus of asphalt mixture to obtain the dynamic modulus difference of asphalt mixture for all samples; use the key feature variable values ​​as the actual input of the machine learning model and the dynamic modulus difference of asphalt mixture as the output of the machine learning model to construct a dataset based on the key feature variable values ​​and the corresponding dynamic modulus difference of asphalt mixture. S33. Divide the dataset constructed in S32 into a training set and a test set, and normalize the training set and the test set respectively; wherein, the normalized training set is used for parameter finding training of the machine learning model, and the normalized test set is used for machine learning model evaluation. S34. Use one or more machine learning algorithms to participate in the parameter search and training of the machine learning model: Use genetic algorithms to search for parameters of the machine learning model, and train the machine learning model based on the optimal model hyperparameters obtained from the parameter search. S35. Combining the machine learning model trained by parameter seeking and the prior knowledge model, construct a dynamic modulus prediction model for asphalt mixture based on prior knowledge fusion; the output of the constructed dynamic modulus prediction model for asphalt mixture is the sum of the prediction results of the machine learning model and the prior knowledge model. S4. The prediction accuracy of the asphalt mixture dynamic modulus prediction model after parameter search training is evaluated using data statistical analysis indicators, and the asphalt mixture dynamic modulus prediction model based on prior knowledge fusion with the best fitting effect is selected. S5. Input the key characteristic variables of the asphalt mixture to be predicted into the preferred dynamic modulus prediction model of the asphalt mixture to obtain the dynamic modulus of the asphalt mixture to be predicted.

2. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The dynamic modulus of the asphalt mixture and its characteristic variables were obtained from public databases, public literature, or laboratory measured data.

3. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The characteristic variables include the cumulative sieve residue percentage of 19mm sieve, the cumulative sieve residue percentage of 9.5mm sieve, the cumulative sieve residue percentage of 4.75mm sieve, the cumulative sieve residue percentage of 0.075mm sieve, porosity, effective asphalt content, aggregate void ratio, asphalt saturation, logarithmic value of asphalt viscosity, dynamic modulus loading frequency, temperature, asphalt viscosity parameters, logarithmic value of asphalt dynamic modulus, and logarithmic value of asphalt phase angle.

4. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The prior knowledge models include the Witczak model, the improved Witczak model, the modified Bari-Witczak model, the Hirsch model, the modified Hirsch model, the AI-Khateeb model, the Global model, and the simplified Global model.

5. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 4, characterized in that, The step of optimizing feature variables based on prior knowledge models to obtain key feature variables specifically includes: selecting prior knowledge models based on the types of feature variables obtained in step S1, and setting the input parameters of the selected prior knowledge models as key feature variables to reduce the workload of feature selection in feature engineering.

6. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The optimization objective of the genetic algorithm is to minimize the coefficient of determination R0 of the k-fold cross-validation on the training set. 2 score.

7. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The machine learning algorithms include support vector machine regression, multilayer perceptron, random forest, and gradient boosting tree.

8. The method for predicting the dynamic modulus of asphalt mixtures based on prior knowledge fusion according to claim 1, characterized in that, The statistical analysis indicators for the data include the coefficient of determination R², mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).

9. A dynamic modulus prediction system for asphalt mixtures based on prior knowledge fusion, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the asphalt mixture dynamic modulus prediction method as described in any one of claims 1-8.

Images