Asphalt material anti-rutting performance prediction and design optimization method based on data driving

Abstract

The invention discloses an asphalt material anti-rutting performance prediction and design optimization method based on data driving, which comprises the following steps: sorting and preprocessing data obtained by a dynamic shear rheometer test of a rubber powder / SBS composite modified asphalt material with various contents in different aging states, and calculating the rutting resistance of the asphalt material by adopting a grey correlation analysis method; screening the data set based on a gray analysis result, and determining input and output of a prediction model; three machine learning models of GPR, LSBboost integration and ANN are established, the deformation resistance of the asphalt material is predicted, a better model is obtained through multi-index comparison, and the prediction result of the optimal model is explained based on SHAP; and based on the optimal prediction model, performing targeted design optimization on the formula of the asphalt material by using a Bayesian optimization algorithm, and simulating application requirements under different actual working conditions by limiting part of parameters in the formula. According to the invention, the test cost of the asphalt material in deformation resistance test and formula design is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of asphalt materials, and particularly relates to a data-driven and interpretable crumb rubber / SBS composite modified asphalt anti-rutting factor prediction and design optimization method. BACKGROUND

[0002] At present, the anti-deformation ability test and prediction method of rubber / SBS composite modified asphalt in the industry mainly focuses on dynamic shear rheological test (DSR), multi-stress creep recovery test (MSCR), Hamburg wheel tracking test (HWT) and other core methods, and realizes multi-dimensional evaluation combined with finite element simulation and mathematical model. Compared with existing research, this research focuses on different working conditions and chemical reagent effects, and simulates different aging conditions to obtain the anti-rutting ability prediction of crumb rubber / SBS composite modified asphalt material under the influence of multiple factors in the whole life. Compared with the traditional multi-index evaluation which needs a large number of tests, the rheological experiment is simple and efficient. The introduction of anti-rutting characteristic index is of great significance to the design and research of asphalt materials. The rutting factor index in DSR test is derived from the Strategic Highway Research Program (SHRP) in the United States, which is the core index of Superpav asphalt performance specification and is closely related to the high-temperature performance in PG classification, and is used to measure the high-temperature anti-permanent deformation ability of asphalt.

[0003] At present, the anti-rutting performance test and design method of asphalt material is mainly based on "JTGE20-2011 Highway Engineering Asphalt and Asphalt Mixture Test Regulations", and is evaluated and designed through traditional tests such as Marshall test. However, the traditional indoor test is time-consuming and consumes a lot of manpower and material resources, which to some extent limits the design and research of high-performance asphalt materials. The reverse design method of crumb rubber / SBS composite modified asphalt anti-rutting performance under different working conditions is proposed, which provides intelligent support for the development and research of asphalt materials.

[0004] Machine learning method deeply analyzes the structure of known data, adjusts the model structure according to the law of training data, so that it can predict new data. Compared with the traditional regression method relying on statistics and laws, machine learning method can process large-scale data, explore complex function relationship and weaken the interference of external factors. The asphalt material performance prediction and design optimization model based on machine learning method uses artificial intelligence to process complex functions, realizes the processing of a large amount of data, obtains abstract function relationship, and realizes the accurate prediction of the anti-rutting deformation ability of crumb rubber / SBS composite modified asphalt at different aging stages. Compared with the traditional design method of crumb rubber / SBS composite modified asphalt, the machine learning reverse design explores the formula optimization more accurately, and can fully consider the influence of different working conditions such as aging degree and chemical additives, which helps the accurate prediction and rapid design of asphalt materials.

[0005] The existing Chinese invention patent with the publication number CN113533133A discloses a method for predicting the dynamic viscoelastic properties of asphalt based on machine learning. This method uses a vector machine machine learning method to predict the viscoelasticity of asphalt, avoiding the tedious test testing of traditional methods. However, this method still has the disadvantages of single prediction model and inability to realize reverse optimization design of asphalt material formula. Therefore, it is necessary to propose a method for comparing multiple machine learning models and optimizing the design of asphalt material formula for dynamic mechanical properties. SUMMARY

[0006] The purpose of the present application is to overcome the shortcomings of the prior art and provide a data-driven asphalt material anti-rutting performance prediction and design optimization method.

[0007] The technical solution adopted by the present application is: a data-driven asphalt material anti-rutting performance prediction and design optimization method, comprising the following steps:

[0008] S1: collate and analyze the anti-rutting performance test data of rubber powder / SBS composite modified asphalt material, use the grey correlation analysis method to analyze the influence degree of the characteristic index parameters such as the penetration of base asphalt, the softening point of base asphalt, the ductility of base asphalt, the rubber content, the SBS content, the test temperature, the load frequency, the aging condition and the addition of chemical reagents on the anti-deformation index parameters of the asphalt material, select the output value index at the same time, filter the data set based on the grey analysis result, and determine the input and output of the anti-rutting performance prediction model;

[0009] S2: build GPR, LSBoost and ANN three kinds of machine learning models to realize the prediction of the anti-rutting performance of the asphalt material, use CV cross validation to verify the prediction accuracy and generalization ability of each machine learning regression model, and optimize the best prediction model;

[0010] S3: use the SHapley Additive exPlanations (SHAP) analysis method to compare the contribution of each characteristic index in the optimized model to the prediction result, obtain the action form and weight of different characteristic indexes on the response value in the machine learning process, and verify the action mode of the characteristic obtained by SHAP analysis through the physical properties of the characteristic, prove the accuracy of the machine learning result, and enhance the interpretability of the result;

[0011] S4: based on the anti-rutting performance optimized prediction model, use the Bayesian optimization algorithm to design and optimize the asphalt material formula, and limit part of the parameters in the formula to simulate the application requirements under different actual working conditions.

[0012] Further, the effect of the chemical reagent acting on the asphalt material in step S1: H2O2 or NaClO is expressed by the O element content on the surface of the asphalt material, and the oxidation of the O element on the asphalt material by chemical reaction shows the effect of the chemical reagent.

[0013] Further, the aging condition of the asphalt in step S1 is divided into four categories: no aging, PAV aging after TFOT aging (simulating short-term thermal aging in the construction stage and slow oxidation and environmental aging in the long-term service of the pavement in the use stage), RTFOT aging (simulating short-term thermal oxidation aging in the mixing, paving and compaction in the construction stage), and PAV aging after RTFOT aging (simulating short-term thermal oxidation aging in the mixing, paving and compaction in the construction stage and slow oxidation and environmental aging in the long-term service of the pavement in the use stage), and the aging condition is represented by the occurrence probability in the data set by means of one-hot algorithm.

[0014] Further, in step S2, three machine learning models based on Matlab framework GPR, LSBoost and ANN are built, wherein the kernel function of the GPR prediction model is Matern 5 / 2 kernel function, the weak learner of the LSBoost prediction model is decision tree, and the ANN prediction model contains 3 hidden layers, and other optimal hyperparameters are obtained by Bayesian optimization.

[0015] Further, in step S3, the Shapley additive analysis method is obtained by Matlab software analysis of the optimal model.

[0016] Further, in step S4, the final result of the Bayesian optimization algorithm is obtained by predicting the trained machine learning model after the formula obtained by Bayesian optimization:

[0017] Further, in step S4, the limited parameters include rubber content, SBS content and working condition, the optimized index is the matrix asphalt parameter combination; the matrix asphalt parameter and the working condition, the optimal rubber content or the rubber content and SBS content formula under the target performance are obtained; the matrix asphalt parameter, the rubber content and the SBS content can obtain the environmental conditions suitable for the working of the asphalt material under the target performance.

[0018] Beneficial effects: The present application serves the design of rubber powder / SBS composite modified asphalt, selects the rutting resistance of asphalt material as the analysis object, studies the relationship between the performance of rubber powder / SBS composite modified asphalt material and the content of rubber powder or SBS, the nature of base asphalt and the environmental working condition. The present application preprocesses the rutting resistance data of asphalt material and carries out secondary screening by using gray scale analysis method, selects the indexes with higher correlation as the input of machine learning model, so that the selection of characteristic values is more reasonable, and the prediction effect of each algorithm model is characterized by CV cross validation, comparative analysis of three machine learning algorithm models, so that the accuracy and generalization ability of the optimized model are greatly improved; based on the optimized prediction model, the Bayesian optimization algorithm is built, the formula design optimization of asphalt material under different aging conditions is realized, which is in line with the engineering practice, and provides reliable prediction information for the design of asphalt pavement material. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flowchart of the present application;

[0020] Figure 2 The influence degree graph of different influence factors rut factor G* / sinδ;

[0021] Figure 3 The CV cross validation diagram with the number of folds being 10;

[0022] Figure 4 The GPR model prediction value-actual value comparison graph after CV cross validation;

[0023] Figure 5 The GPR model prediction residual graph after CV cross validation;

[0024] Figure 6 The LSBoost model prediction value-actual value comparison graph after CV cross validation;

[0025] Figure 7 The LSBoost model prediction residual graph after CV cross validation;

[0026] Figure 8 The ANN model prediction value-actual value comparison graph after CV cross validation;

[0027] Figure 9 The ANN model prediction residual graph after CV cross validation;

[0028] Figure 10 The SHAP mean value of each characteristic value of LSBoost integrated model;

[0029] Figure 11 The Sharpe Additive Summary of LSBoost integrated model;

[0030] Figure 12Bayesian optimization process for asphalt material formulation;

[0031] Figure 13 Example of Bayesian optimization results. Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0033] like Figure 1 As shown, a data-driven method for predicting and optimizing the rutting resistance of asphalt materials includes the following steps:

[0034] First, the characteristic index parameters of asphalt materials and the DSR test index parameters are organized to form an initial dataset. Then, preprocessing measures such as missing value handling, outlier handling, and data standardization are performed on the initial dataset using KNN imputation, 3σ rule, Z-score standard score method, etc.

[0035] Second, by using grey relational analysis, we analyzed the influence of each feature index of the preprocessed dataset on the experimental index. Based on the grey analysis results, we screened the dataset and selected feature indices with a grey relational degree of 0.8 or higher as inputs to determine the input and output of the rutting resistance performance prediction model.

[0036] The formulas for calculating the grey relational degree of each feature index are as follows:

[0037]

[0038] Δi(t)=|x i (t)-y(t)|

[0039]

[0040] Where x is the original data of the feature index, and y is the original data of the response column, x norm It is the standardized data of the feature index, y norm These are the standardized data of the response column, where σ is the sample standard deviation and Δ is the standard deviation. i (t) is the comparison sequence x at time point t. i The difference between ξ(t) and the reference sequence y(t), i y(t) is the grey relational coefficient between the reference sequence y(t) and the i-th comparison sequence at time t, ρ is the resolution coefficient, and Δt is the gray relational coefficient between the reference sequence y(t) and the i-th comparison sequence at time t. min It is the minimum value among all differences, Δ max It is the maximum value among all differences, γ i It is a comparison sequence x i The grey relational degree, where n is the number of observation points.

[0041] III. Three machine learning models—GPR, LSBoost, and ANN—were built to predict the rutting resistance of asphalt materials. Cross-validation (CV) was used to verify the predictive performance of each machine learning regression model, and the best predictive model was selected.

[0042] Three machine learning models, GPR, LSBoost, and ANN, were built using the Matlab framework. The kernel function of the GPR prediction model is the Matern 5 / 2 kernel function, the weak learner of the LSBoost prediction model is a decision tree, and the ANN prediction model has a two-layer structure. Other optimal hyperparameters were obtained by Bayesian optimization.

[0043] The kernel function for the Gaussian process regression (GPR) prediction model is the Matern 5 / 2 kernel function, and its expression is as follows:

[0044]

[0045] Where, k Matern5 / 2 (x,x') is the Matern 5 / 2 kernel function, x and x' are two feature vectors in the input space, l is the length scale parameter, and r is the Euclidean distance between the vectors.

[0046] When using cross-validation (CV) to assess the model's generalization ability, the number of folds should be chosen such that each fold contains at least 100 data points.

[0047] Introducing the coefficient of determination R 2 The three parameters, root mean square error (RMSE), normalized root mean square deviation (NRMSD), and root mean square error (RMSE), characterize the accuracy of each model after CV validation. Their expressions are as follows:

[0048]

[0049]

[0050]

[0051] Among them, y i This is the actual value. It is a predicted value. It is the average of the actual values, y max It is the maximum value of the actual value, y min It is the minimum value of the actual value.

[0052] IV. The basic optimization prediction model uses the Bayesian optimization algorithm to perform targeted design and optimization of the asphalt material formulation, and by fixing some parameters in the formulation, it simulates the application requirements under different actual working conditions.

[0053] The prior, prediction distribution, acquisition function (using expectation improvement in this method), and expressions for the optimized acquisition function of the LSBoost ensemble model based on the Bayesian optimization algorithm are as follows:

[0054]

[0055] p(f(x)|D,x)=N(μ(x),σ 2 (x))

[0056] EI(x)=E[max(0,f(x)-f(x + ))]

[0057] x next =arg max x AcquisitionFunction(x)

[0058] Where f(x) is the objective function, gp is the Gaussian process prior, and LSBoost is the ensemble model (composed of T base learners h). t (x;θ t The weighted summation is composed of β. t Let θ be the weight of the t-th base learner. t Let p(fx)|D,x) be the posterior distribution of the objective function given the data D, μ(x) be the mean of the posterior distribution, and σ be the parameter. 2 (x) is the variance of the posterior distribution, EI(x) is the expected improvement, and x + It is the current optimal point, x next This is the next experimental point.

[0059] Where obj is the objective function, prediction is the predicted value obtained after substituting the current optimized formula into the optimal prediction model, and targetY is the optimization target value. Some optimization parameters are limited to simulate the actual working conditions as shown in Table 1 below.

[0060] Table 1 Comparison of limiting parameters and simulated operating conditions

[0061]

[0062] Example: This example demonstrates a data-driven, interpretable method for predicting and optimizing the rutting resistance factor of rubber powder / SBS composite modified asphalt, implemented according to the following steps:

[0063] I. Existing data on the rutting resistance of asphalt materials, with the target characteristic parameter being the rutting factor G* / sinδ. The characteristic indices include nine parameters: asphalt base penetration, base asphalt softening point, base asphalt ductility, rubber content, SBS content, test temperature, loading frequency, aging condition, and the effect of chemical reagents. Preprocessing measures were applied to the initial dataset, resulting in 1038 data entries.

[0064] Second, the grey relational degree of each feature indicator is calculated to filter the dataset, resulting in 9 feature indicators as input, and the final dataset is determined. Among them, for example... Figure 2 The figure shows the influence of different influencing factors on the rutting factor G* / sinδ. Table 2 shows the input and output of the rutting resistance performance regression model.

[0065] Table 2: Inputs and Outputs of the Rutting Resistance Regression Model

[0066]

[0067] III. Three machine learning models, namely GPR, LSBoost, and ANN, were built to predict the rutting resistance of asphalt materials. Bayesian optimization was used to optimize some hyperparameters of each model. The Shapley additive analysis method was used to interpret the machine learning results and to analyze and explain the machine learning.

[0068] IV. Select 10 folds for CV cross-validation, such as... Figure 3 As shown, this illustrates a cross-validation diagram with a fold count of 10. The predicted-actual value comparison chart and prediction residual plot for the three prediction models are shown below. Figure 4 , Figure 6 , Figure 8 , Figure 5 , Figure 7 as well as Figure 9 As shown in Table 3, the LSBoost model has the highest R-value. The analysis and comparison of the three prediction models are presented in Table 3. The results show that the LSBoost model has the highest R-value. 2 With the lowest RMSE and NRMSD, this model is the optimal model. Figure 10 The SHAP mean of each feature value in the LSBoost ensemble model; Figure 11 This is a summary of the Shapley additive properties of the LSBoost ensemble model.

[0069] Table 3: Performance Comparison Among Different Prediction Models

[0070]

[0071] V. Based on the selected LSBoost ensemble model, a Bayesian optimization algorithm is used to optimize the asphalt material formulation design, focusing on the rutting resistance of asphalt materials. For example... Figure 12As shown, this embodiment illustrates the Bayesian optimization process for asphalt material formulations targeting rutting resistance. Figure 13 As shown, the rubber content and SBS content were set to 0.2% and 0.05%, respectively. The test temperature was 60℃, the aging condition was unaged, the load frequency was 40 rads / s, and no chemical reagents were added, so the oxygen content was 0%. The target rutting factor G* / sinδ was set to 50 kPa. The simulation aimed to find the optimal formulation of the base asphalt under specific working conditions with the given rubber content and SBS content. Bayesian optimization yielded penetration, softening point, and ductility of 68.27, 49.54, and 92.95, respectively. Substituting the parameter outputs into the LSBoost model, the predicted result was 50.26 kPa, with a relative error of 0.52% compared to the target rutting factor G* / sinδ. This formulation showed satisfactory accuracy.

[0072] The principle of the GPR model built in this embodiment is as follows:

[0073] Equations (1) and (2) respectively show the definition of Gaussian process and the mathematical expression of Matern 5 / 2 kernel function, while equations (3) and (4) respectively show the specific calculation process of GPR model training and prediction.

[0074] f(x)~gp(m(x),k(x i ,x j )) (1)

[0075]

[0076]

[0077]

[0078] In the formula: x is the input vector value, f(x) is a random function, m(x) is the mean function, l is the length scale parameter, and x * It is the newly input point, k * It is the kernel function vector, k(x) i ,x j ) is the covariance function, K is the kernel matrix between training input points, y is the training target value, r is the Euclidean distance between vectors, and σ is the covariance function. 2 It is the signal variance.

[0079] The principle of the LSBoost ensemble model built in this embodiment is as follows:

[0080] The LSBoost ensemble model iteratively fits the residuals, continuously correcting prediction biases. Combined with a decision tree base learner of a certain complexity, it can accurately capture nonlinear patterns in the data and significantly reduce the risk of underfitting. Equation (5) initializes the base learner, Equation (6) calculates the residuals, Equation (7) trains the base learner, Equation (8) updates the ensemble model, and Equation (9) is the final ensemble model.

[0081]

[0082] r im =y i -f m-1 (x i (6)

[0083]

[0084] f m (x)=f m-1 (x)+γ m h m (x) (8)

[0085]

[0086] Where f0(x) is the predicted value of the initial model, n is the number of samples, and y i r is the actual value of the i-th sample. im f is the residual of the i-th sample in the m-th iteration. m-1 (x) is the predicted value of the i-th sample in the model during the (m-1)-th iteration, h m (x) is the weak learner (decision tree) in the m-th iteration; f m (x) is the model prediction value after the m-th iteration update, γ m It is the learning rate, which controls the contribution of each weak learner to the model, f M (x) is the predicted value of the final trained model, and M is the number of iterations.

[0087] The structure and principle of the ANN model built in this embodiment are as follows:

[0088] The ANN model structure consists of an input layer, hidden layer 1, hidden layer 2, hidden layer 3, and an output layer. The three hidden layers contain 100, 60, and 80 neurons respectively. The training process is as follows: the input layer passes the input data to the hidden layers. The neurons in the hidden layers are calculated using weighting, biasing, and activation functions. The output of the hidden layer neurons is obtained by nonlinear transformation through the activation function, as shown in equations (10)-(12). The neurons in the output layer re-weight and bias the output of the hidden layers to obtain the final predicted value, as shown in equation (13). Finally, the weights and biases of the neural network are updated layer by layer by calculating the gradient of the loss function through backpropagation. The weights and biases are continuously adjusted through multiple iterations to minimize the loss function, as shown in equations (14)-(16), thus obtaining the optimal ANN model.

[0089]

[0090] h j =σ(z) j (11)

[0091] σ(z)=max(0,z) (12)

[0092]

[0093]

[0094]

[0095]

[0096] Where: x is the input data x = [x1, x2, ..., x n ],ω ji Is input x i The weights of neuron j in the hidden layer, b j It is the bias of hidden layer neuron j, z j It is the weighted sum of hidden layer neurons j, σ(z) is the ReLU activation function, and ω oj It is the weight from hidden layer neuron j to the output layer neuron, b o It represents the bias of the output layer neurons, MSE is the mean squared error loss function, and y i This is the actual value. Here, N is the predicted value, η is the number of samples, and η is the learning rate. The loss function is the weight ω ji The partial derivatives, The loss function is relative to the bias b. j The partial derivatives of .

[0097] The Bayesian optimization algorithm used in this example follows the following principle:

[0098] In the specific implementation process, the Bayesian optimization algorithm first selects a suitable surrogate model and, based on the existing data points, gives the probability distribution of the objective function at other points. Its specific steps are mainly divided into establishing the prior of the LSBoost ensemble model, updating it to the posterior distribution based on the existing data, selecting the next experimental point through the acquisition function, performing verification and updating the model, and continuously iterating the above process to gradually approach the global optimal solution of the objective function. Equations (17)-(20) respectively show the expressions for calculating the weight distribution, acquisition function (this method adopts expectation improvement), and optimizing the acquisition function of the LSBoost ensemble model.

[0099]

[0100] p(f(x)|D,x)=N(μ(x),σ 2 (x)) (18)

[0101] EI(x)=E[max(0,f(x)-f(x + (19)

[0102] x next =arg max x AcquisitionFunction(x) (20)

[0103] Where f(x) is the objective function, and the LSBoos ensemble model consists of T base learners h t (x;θ t The weighted summation is composed of β. t Let θ be the weight of the t-th base learner. t Let p(f(x)|D,x) be the posterior distribution of the objective function given the data D, μ(x) be the mean of the posterior distribution, and σ be the parameter. 2 (x) is the variance of the posterior distribution, EI(x) is the expected improvement, and x + It is the current optimal point, x next This is the next experimental point.

Claims

1. A data-driven method for predicting and optimizing the rutting resistance of asphalt materials, characterized in that, The method includes the following steps: S1: Compile and analyze the test data of the rutting resistance performance of rubber powder / SBS composite modified asphalt materials. Use grey relational analysis to analyze the influence of characteristic index parameters such as the penetration of base asphalt, softening point of base asphalt, ductility of base asphalt, rubber content, SBS content, test temperature, loading frequency, aging condition and added chemical reagents on the deformation resistance index parameters of asphalt materials. At the same time, select the output value index. Based on the grey analysis results, filter the dataset to determine the input and output of the rutting resistance performance prediction model. S2: Build three machine learning models, GPR, LSBoost and ANN, to predict the rutting resistance of asphalt materials. Use cross-validation (CV) to verify the prediction accuracy and generalization ability of each machine learning regression model, and select the best prediction model. S3: The Shapley additive analysis method is used to compare the contributions of each feature index in the optimal model to the prediction results, to obtain the form and weight of the effect of different feature indexes on the response value in the machine learning process, and to verify the feature action mode obtained by the SHAP analysis through the physical properties of the features, thus proving the accuracy of the machine learning results and enhancing the interpretability of the results. S4: Based on the rutting resistance performance optimization prediction model, the Bayesian optimization algorithm is used to conduct targeted design optimization of asphalt material formulation, and by limiting some parameters in the formulation, the application requirements under different actual working conditions are simulated.

2. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 1, characterized in that, In step S1, the effect of the chemical reagents H2O2 or NaClO on the asphalt material is characterized by the content of O element on the surface of the asphalt material. The effect of the chemical reagents on the oxidation of the asphalt material by the O element obtained through the chemical reaction is demonstrated.

3. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 2, characterized in that, In step S1, the asphalt aging status is divided into four categories: no aging, PAV aging after TFOT aging, RTFOT aging, and PAV aging after RTFOT aging. The one-hot algorithm is used to represent the aging status feature in the dataset as the probability of occurrence.

4. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 3, characterized in that, In step S2, three machine learning models based on the Matlab framework are built: GPR, LSBoost, and ANN. The kernel function of the GPR prediction model is the Matern 5 / 2 kernel function, the weak learner of the LSBoost prediction model is a decision tree, and the ANN prediction model contains three hidden layers. Other optimal hyperparameters are obtained by Bayesian optimization.

5. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 4, characterized in that, In step S3, the Shapley additive analysis method is obtained by analyzing the optimal model using Matlab software.

6. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 5, characterized in that, The final result of the Bayesian optimization algorithm in step S4 is predicted by a trained machine learning model after the formula is obtained through Bayesian optimization.

7. The data-driven method for predicting and optimizing the rutting resistance of asphalt materials according to claim 6, characterized in that, In step S4, the parameters in the defined formula include rubber content, SBS content, and operating conditions. The optimized index is the combination of base asphalt parameters. The base asphalt parameters and operating conditions yield the optimal rubber content and SBS content formula under the target performance. The base asphalt parameters, rubber content, and SBS content can obtain the environmental conditions suitable for the working of this asphalt material under the target performance.

Images