Asphalt mixture high-temperature performance prediction and design optimization method based on data driving

Abstract

The invention discloses an asphalt mixture high-temperature performance prediction and design optimization method based on data driving. The method comprises the steps of sorting and preprocessing asphalt mixture high-temperature performance data, adopting a grey correlation analysis method, screening a data set based on a grey analysis result, and determining input and output of a high-temperature performance prediction model. Building three machine learning models, namely ANN, SVM and GPR, predicting the high-temperature performance of the asphalt mixture, and carrying out visualization processing on the models; based on the high-temperature performance optimization prediction model, a Bayesian optimization algorithm is utilized to perform targeted design optimization on the mix proportion of the asphalt mixture, and application requirements under different actual working conditions are simulated by limiting part of parameters in the mix proportion. According to the method, reliable prediction information is provided, the test cost of the asphalt mixture in high-temperature performance test and mix proportion design is reduced, the material design and research and development period is shortened, and the digital and intelligent development of asphalt mixture performance prediction and optimization is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of asphalt mixtures, and in particular to a data-driven method for predicting and optimizing the high temperature performance of asphalt mixtures. Background Art

[0002] Asphalt mixture is a temperature-sensitive viscoelastic-plastic material, and its strength and stiffness are significantly affected by temperature. When asphalt pavement is subjected to long-term static loads or repeated dynamic loads in a high-temperature environment, it is easy to produce defects such as rutting, waves, shifting, bumping, and washboard, which not only affect driving comfort but also pose a hidden danger to driving safety. Therefore, in order to make the pavement provide stable and durable services to vehicles, it is particularly important to accurately predict the high-temperature stability of asphalt mixtures and make targeted design optimizations. At present, the high-temperature performance test and design methods of asphalt mixtures are mainly based on the "JTGE20-2011 Test Procedures for Asphalt and Asphalt Mixtures in Highway Engineering", and the design is evaluated through indoor tests such as rutting tests and Marshall tests. However, traditional indoor tests often consume a lot of time, materials and manpower, which to a certain extent limits the design and development of high-performance asphalt mixtures.

[0003] Thanks to the development of artificial intelligence, data-driven intelligent methods have shown broad application prospects in asphalt mixture performance prediction and design optimization. By taking advantage of the machine learning algorithm to effectively handle complex functional relationships, the high-temperature performance of asphalt mixtures can be accurately predicted. This can achieve targeted and efficient design optimization of asphalt mixtures, a multi-phase, multi-scale complex material, under high temperature conditions, reduce the manpower and material consumption caused by traditional testing and design methods, shorten the R&D cycle, improve R&D quality, and reduce R&D costs, and design asphalt mixtures with good high-temperature stability.

[0004] The existing Chinese invention patent with publication number CN118016190A discloses a method for predicting asphalt mixture performance based on random forest. This method uses random forest machine learning method to predict asphalt mixture performance, avoiding the cumbersome experimental testing of traditional methods. However, this method still has shortcomings such as a single prediction model and failure to achieve reverse optimization design of asphalt mixture ratio. Therefore, it is necessary to propose a method for comparing multiple machine learning models and realizing asphalt mixture ratio design optimization method for high temperature performance. Summary of the invention

[0005] Purpose of the invention: The purpose of the invention is to provide a data-driven method for predicting the high temperature performance of asphalt mixture and optimizing the design.

[0006] Technical solution: The present invention comprises the following steps:

[0007] S1: Organize and preprocess the high temperature performance data of asphalt mixtures, and use the gray correlation analysis method to analyze the influence of characteristic index parameters such as asphalt penetration, asphalt softening point, oil-stone ratio, mineral aggregate grading information, mixture void ratio, mineral aggregate gap ratio and effective asphalt saturation on the high temperature performance index parameters of asphalt mixtures. Based on the gray analysis results, screen the data set and determine the input and output of the high temperature performance prediction model;

[0008] S2: Build three machine learning models, namely ANN, SVM and GPR, to predict the high temperature performance of asphalt mixture, visualize the models, use CV cross validation to verify the prediction performance of each machine learning regression model, and select the best prediction model;

[0009] S3: Based on the high-temperature performance optimization prediction model, the Bayesian optimization algorithm is used to carry out targeted design optimization of the mix ratio of asphalt mixture, and some parameters in the mix ratio are limited to simulate the application requirements under different actual working conditions.

[0010] Furthermore, the grading information of the mineral material in step S1 includes the pass rates of the key sieve holes of 19 mm, 16 mm, 13.2 mm, 9.5 mm, 4.75 mm, 2.36 mm, 1.18 mm, 0.6 mm, 0.3 mm, 0.15 mm and 0.075 mm, and the three parameters of the Bailey method CA ratio, FA C Ratio, FA f The ratio and grading curve of Weibull distribution have two parameters: shape parameter k and scale parameter λ.

[0011] Furthermore, in step S2, three machine learning models, ANN, SVM and GPR, are constructed based on the Matlab framework, wherein the ANN prediction model includes three hidden layers, the kernel functions of the SVM prediction model and the GPR prediction model are the quadratic kernel function and the Matern 5 / 2 kernel function, respectively, and other optimal hyperparameters are obtained by Bayesian optimization.

[0012] Furthermore, the asphalt mixture performance prediction visualization in step S2 is presented in the form of an APP, and provides switching of three prediction models, Bailey method parameter auxiliary calculation, Weibull distribution auxiliary calculation and high temperature performance prediction functions.

[0013] Furthermore, the objective function of the Bayesian optimization algorithm in step S3 is expressed as:

[0014]

[0015] Among them, obj is the objective function value, prediction is the predicted value obtained after bringing the current optimization formula into the optimization prediction model, targetY is the optimization target value, and gradation is the mineral gradation.

[0016] Furthermore, some parameters defined in step S3 include asphalt index, gradation information and mixture parameters VV and VFA. The asphalt index refers to the gradation, oil usage and compaction degree under target performance for a specific asphalt; the gradation information refers to the use of a certain gradation to obtain the type of asphalt, oil usage and compaction degree under target performance; the mixture parameters VV and VFA refer to the type of asphalt, gradation information and oil usage under specified compaction to obtain the target performance.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention preprocesses the high-temperature data of asphalt mixture and performs secondary screening using the grayscale analysis method, selects high-correlation feature indicators as the input of the machine learning model, so that the calculation efficiency and prediction accuracy of the prediction model are improved, and the three prediction models are visualized to achieve friendly and intuitive human-computer communication; and the prediction effect of each algorithm model is characterized by CV cross-validation, and the three machine learning algorithm models are compared and analyzed, so that the robustness and generalization ability of the preferred model are greatly improved; a Bayesian optimization algorithm is built based on the preferred prediction model to achieve the optimization of asphalt mixture mix design under different working conditions, which is in line with the actual project; reliable prediction information is provided for the design of asphalt pavement materials, and by combining machine learning technology with materials science, the experimental cost of asphalt mixture in high-temperature performance testing and mix design is reduced, the material design and R&D cycle is shortened, and the digital and intelligent development of asphalt mixture performance prediction and optimization is promoted. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the present invention;

[0019] Figure 2 It is the influence degree diagram of different influencing factors on dynamic stability DS;

[0020] Figure 3 It is the visualization interface of the asphalt mixture high temperature performance predictor APP;

[0021] Figure 4 This is a schematic diagram of CV cross validation with 7 folds;

[0022] Figure 5 This is the comparison chart of ANN model predicted value and actual value after CV cross validation;

[0023] Figure 6 This is the residual graph of the ANN model prediction after CV cross validation;

[0024] Figure 7 This is the comparison chart between the predicted value and the actual value of the SVM model after CV cross-validation;

[0025] Figure 8 This is the residual graph of the SVM model prediction after CV cross validation;

[0026] Fig. 9 This is a comparison chart of the predicted value and actual value of the GPR model after CV cross-validation;

[0027] Fig.10 This is the GPR model prediction residual graph after CV cross validation;

[0028] Fig.11 A Bayesian optimization process for asphalt mix proportions targeting high temperature performance;

[0029] Fig.12 Example of Bayesian optimization results for high temperature performance. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0031] like Figure 1 As shown, the present invention comprises the following steps:

[0032] 1. The characteristic index parameters of asphalt mixture and high temperature performance index parameters are sorted out to form an initial data set, and the initial data set is preprocessed by missing value processing, outlier processing, data standardization and other preprocessing measures using KNN filling, 3σ rule, Z-score standard score method, etc.

[0033] The characteristic index parameters of asphalt mixture are sorted out. According to the mineral gradation information in the characteristic index of asphalt mixture, three parameters of Bailey method are introduced: CA ratio, FA C Ratio, FA f ratio; the main parameters of Weibull distribution: shape parameter k, scale parameter λ. To further characterize the effect of gradation on the performance of the mixture. The five parameters are obtained from the gradation information of each data. The three parameters of Bailey method and the two parameters of Weibull distribution can be calculated by the following formula:

[0034]

[0035] Among them, P (NMPS / 2) It is the passing rate of the sieve hole corresponding to 0.5 times the maximum nominal size, P PCS is the pass rate of the key sieve hole, It is the passing rate of the sieve hole corresponding to 0.22 times of the PCS point. Is FA CThe point is 0.22 times the corresponding sieve hole's pass rate.

[0036]

[0037] Here, k is the shape parameter and λ is the scale parameter.

[0038] Second, the gray correlation analysis method is used to analyze the influence of each characteristic index of the preprocessed data set on the test index. The data set is screened based on the gray analysis results, and the characteristic index input with a gray correlation degree above 0.7 is selected to determine the input and output of the high temperature performance prediction model.

[0039] The calculation formula of the grey relational degree of each characteristic index is as follows:

[0040]

[0041] Δ i (t)=|x i (t)-y(t)|

[0042]

[0043] Among them, x is the original data of the characteristic index, y is the original data of the response column, and x norm is the data after the characteristic index is standardized, y norm is the data after the response column is standardized, σ is the sample standard deviation, Δ i (t) is the comparison sequence x at time point t i The difference between y(t) and the reference sequence y(t), ξ i (t) is the grey correlation coefficient between the reference sequence y(t) and the ith comparison sequence at time point t, ρ is the resolution coefficient, Δ min is the minimum value among all differences, Δ max is the maximum value among all differences, γ i is the comparison sequence x i is the grey relational degree, n is the number of observation points.

[0044] 3. Build three machine learning models: ANN, SVM, and GPR to predict the high temperature performance of asphalt mixture and visualize the models. Use CV cross validation to verify the prediction performance of each machine learning regression model and select the best prediction model;

[0045] Three machine learning models, ANN, SVM and GPR, were built based on the Matlab framework. The ANN prediction model contained three hidden layers, the kernel functions of the SVM prediction model and the GPR prediction model were quadratic kernel function and Matern 5 / 2 kernel function respectively, and other optimal hyperparameters were obtained by Bayesian optimization.

[0046] The kernel function of the support vector machine (SVM) prediction model is a quadratic kernel function, and its formula is as follows:

[0047] K(x i , x j )=(x i ·x j +c) 2

[0048] Among them, x i , x j is the sample feature vector in the input space, c is an adjustable parameter used to scale the inner product of the input feature vector, and its default value is 1.

[0049] The kernel function of the Gaussian process regression (GPR) prediction model is the Matern 5 / 2 kernel function, and its expression is as follows:

[0050]

[0051] Among them, k Matern 5 / 2 (x, x′) is the Matern 5 / 2 kernel function, x, x′ are two eigenvectors in the input space, l is the length scale parameter, and r is the Euclidean distance between the vectors.

[0052] CV cross-validation is used to consider the generalization ability of the model. When selecting the number of folds, it should be ensured that the amount of data in each fold after folding is more than 10.

[0053] Introducing the coefficient of determination R 2 The three parameters of RMSE, NRMSD and root mean square error characterize the accuracy of each model after CV verification, and their expressions are as follows:

[0054]

[0055] Among them, y i is the actual value, is the predicted value, is the average of the actual values, y max is the maximum actual value, y min is the minimum of the actual values.

[0056] The asphalt mixture performance prediction model is visualized and presented in the form of an APP. The APP provides switching between three prediction models, Bailey method parameter auxiliary calculation, Weibull distribution auxiliary calculation and high temperature performance prediction functions.

[0057] 4. Based on the high-temperature performance optimization prediction model, the Bayesian optimization algorithm is used to carry out targeted design optimization of the mix ratio of asphalt mixture, and some parameters in the mix ratio are limited to simulate the application requirements under different actual working conditions.

[0058] The expressions of the Gaussian process prior, posterior distribution, acquisition function (this method uses expected improvement) and optimized acquisition function of the Bayesian optimization algorithm are as follows:

[0059] f(x)~gp(m(x), k(x, x′))

[0060] p(f(x)|D,x)=N(μ(x,σ 2 (x)

[0061] EI(x) = E[max(0, f(x)-f(x) + ))]

[0062] x next = arg max x AcquisitionFunction(x)

[0063] Where, f(x) is the objective function, gp is the Gaussian process prior, m(x) is the Gaussian process mean function, k(x, x′) is the Gaussian process kernel function, p(f(x)|D, x) is the posterior distribution of the objective function given the data D, μ(x) is the mean of the posterior distribution, σ 2 (x) is the variance of the posterior distribution, EI(x) is the expected improvement, x + is the current optimal point, x next This is the next experimental point.

[0064] The objective function expression of the Bayesian optimization algorithm is:

[0065]

[0066] Wherein, obj is the objective function value (optimization error), prediction is the predicted value obtained by bringing the current optimization formula into the optimal prediction model, targetY is the optimization target value, gradation is the gradation of mineral materials, and some optimization parameters are limited to simulate the actual working conditions as shown in Table 1 below.

[0067] Table 1: Limit some optimization parameters to simulate actual working conditions:

[0068]

[0069] Embodiment: The data-driven high temperature performance prediction and design optimization method of asphalt mixture in this embodiment is implemented according to the following steps:

[0070] 1. Existing high temperature performance data of asphalt mixture, among which the target characteristic parameter is dynamic stability, and the characteristic index parameters are: 25℃ needle penetration, softening point, oil-stone ratio, 19mm, 16mm, 13.2mm, 9.5mm, 4.75mm, 2.36mm, 1.18mm, 0.6mm, 0.3mm, 0.15mm and 0.075mm grade key sieve hole pass rate, void ratio VV, mineral gap ratio VMA, effective asphalt saturation VFA, CA ratio, FA C Ratio, FA f The initial data set was preprocessed and 77 high-temperature data were finally obtained.

[0071] Second, calculate the gray correlation degree of each characteristic index to screen the data set, and finally get 20 characteristic index inputs and determine the final data set. Figure 2 The figure shows the influence degree of different influencing factors on dynamic stability DS. Table 2 shows the input and output of the high temperature performance regression model.

[0072] Table 2: Inputs and outputs of the high temperature performance regression model

[0073]

[0074] 3. Build three machine learning models, ANN, SVM and GPR, to predict the high temperature performance of asphalt mixture, and use Bayesian optimization to optimize the hyperparameters of each model. And develop the "Asphalt Mixture High Temperature Performance Predictor APP" based on Matlab framework programming, and its user interface is as follows: Figure 3 As shown in the figure, after the user selects the prediction model to be used, inputs the needle penetration, softening point, oil-stone ratio, aggregate gradation, VV and VFA values, uses the additional parameter calculator to calculate the Bailey method parameters and Weibul1 distribution parameters, and then clicks the prediction button to predict the dynamic stability DS of the asphalt mixture.

[0075] 4. Select the fold number of CV cross validation as 7, such as Figure 4 As shown in the figure, a schematic diagram of CV cross validation with a fold of 7 is shown. The predicted value-actual value comparison diagram and the predicted residual diagram of the three prediction models are shown in Figure 5 , Figure 7 , Fig. 9 , Figure 6 , Figure 8 as well as Fig.10 The three prediction models are analyzed and compared, and the results are shown in Table 3. The results show that the GPR model has the highest R 2 and the lowest RMSE and NRMSD, so this model is the optimal model.

[0076] Table 3: Performance comparison between different high temperature performance prediction models

[0077]

[0078] 5. Based on the preferred GPR model, the Bayesian optimization algorithm is used to optimize the asphalt mixture mix design for high temperature stability. Fig.11 As shown in FIG. 1 , the Bayesian optimization process of asphalt mixture ratio for high temperature performance in this embodiment is demonstrated. Fig.12 As shown, it is demonstrated that the mix ratio output can be achieved based on the preferred GPR model and the constructed Bayesian optimization algorithm under the condition of limiting some parameters. Among them, the asphalt penetration and softening point are limited to 67.3 (0.1mm, 25℃, 100g) and 50 (℃), respectively, and the grading type is limited to AC-16, and the target dynamic stability DS is set to 3000 (times / mm). It simulates the working condition that an asphalt mixture with a dynamic stability of 3000 (times / mm) needs to be designed under the conditions of having a specific asphalt and requiring a specific grading. The other formula parameters are output, and the preferred GPR model is brought in to predict its DS as 3125.4 (times / mm), with a relative error of 4.18% from the target DS. Considering the excellent performance of the GPR model in step 4 in predicting the high temperature performance of asphalt mixtures, the accuracy of the design optimization formula can be well supported.

[0079] The structure and principle of the ANN model built in this embodiment are as follows:

[0080] The ANN model structure is input layer, hidden layer 1, hidden layer 2, hidden layer 3 and output layer. The three hidden layers contain 280, 63 and 84 neuron nodes respectively. The training process is as follows: the input layer passes the input data to the hidden layer, the neurons in the hidden layer are calculated through weighting, bias and activation functions, and the output of the neurons in the hidden layer is obtained by nonlinear transformation through the activation function, as shown in formula (1)-(3). The neurons in the output layer weight and bias the output of the hidden layer again to obtain the final prediction value. As shown in formula (4). Finally, the gradient of the loss function is calculated through back propagation, and the weights and biases of the neural network are updated layer by layer. The weights and biases are continuously adjusted through multiple iterations to minimize the loss function, as shown in formula (5)-(6), and the optimal ANN model is obtained.

[0081]

[0082] h j =σ(z j ) (2)

[0083] σ(z)=max(0,z)(3)

[0084]

[0085] Where: x is the input data x = [x1, x2, ..., x n ],ω ji is the input x i to the weight of the hidden layer neuron j, b j is the bias of hidden layer neuron j, z j is the weighted sum of neurons j in the hidden layer, σ(z) is the ReLU activation function, ω oj is the weight from hidden layer neuron j to output layer neuron, b o is the bias of the output layer neuron, MSE is the mean square error loss function, y i is the actual value, is the predicted value, N is the number of samples, η is the learning rate; is the loss function with respect to weight ω ji The partial derivative of is the loss function with respect to the bias b j The partial derivative of .

[0086] The principle of the SVM model built in this embodiment is as follows:

[0087] The goal of SVM is to find a function that makes the prediction error within the range of ∈ and minimizes the model complexity. Its objective function and constraints are shown in equations (7) and (8), respectively. By introducing a kernel function (this method uses a quadratic kernel function) to process nonlinear data, as shown in equation (9), the data is mapped to a high-dimensional space. In this space, by constructing a Lagrangian function and solving the dual problem, the weight vector and bias are calculated, as shown in equations (10)-(12). Finally, the prediction of new data is completed through equation (13).

[0088]

[0089] K(x i , x j )=(x i ·x j +c) 2 (9)

[0090]

[0091]

[0092] Where: w is the weight vector, C is the regularization parameter, x i is the input feature vector of the i-th sample, x j is the input feature vector of the jth sample, b is the bias, ξ i , is the slack variable, K is the quadratic kernel function, c is an adjustable parameter used to scale the inner product of the input feature vector, the default value is 1, α i , is the Lagrange multiplier and N is the number of samples.

[0093] The principle of the Matern 5 / 2GPR model constructed in this embodiment is as follows:

[0094] The GPR model is based on Bayesian probability theory and is used to model random functions. The core of GPR is the kernel function (covariance function), which defines the similarity and correlation between any two points, thereby affecting the smoothness and shape of the function. In GPR regression, the predicted mean and variance of the new input point can be calculated using the covariance matrix of the training data and the covariance of the new input point, thereby realizing the prediction of unknown data and quantifying the uncertainty of the prediction. The GPR model of this embodiment selects the Matern5 / 2 kernel function. Formulas (14) and (15) respectively show the Gaussian process definition and the Matern 5 / 2 kernel function expression, and the training and prediction expressions of the GPR model are shown in Formulas (16) and (17), respectively.

[0095] f(x)~gp(m(x),k(x,x′)) (14)

[0096]

[0097] Where: x is the input vector, f(x) is the random function, m(x) is the mean function, l is the length scale parameter, x* is the new input point, k * is the kernel function vector, K is the kernel matrix between the training input points, y is the training target value, k(x, x′) is the covariance function, and r is the Euclidean distance between the vectors.

[0098] The principle of the Bayesian optimization algorithm used in this example is as follows:

[0099] The core idea of ​​the Bayesian optimization algorithm is to use a probability model to approximate the objective function and use the model to guide the selection of new experimental points. The specific steps are mainly divided into establishing a Gaussian process prior, updating it to the posterior distribution based on the existing data, selecting the next experimental point through the acquisition function, updating the model after verification, and continuously iterating the above process to gradually approach the global optimal solution of the objective function. Equations (18)-(21) respectively show the expressions of the Gaussian process prior, posterior distribution, acquisition function (this method uses expected improvement) and optimized acquisition function.

[0100] f(x)~gp(m(x),k(x,x′)) (18)

[0101] p(f(x)|D,x)=N(μ(x,σ 2(x)) (19)

[0102] EI(x) = E[max(0, f(x)-f(x) + ))] (20)

[0103] x next = arg max x AcquisitionFunction(x) (21)

[0104] Where: f(x) represents the objective function, gp represents the Gaussian process prior, m(x) represents the Gaussian process mean function, k(x, x′) represents the Gaussian process kernel function, p(f(x)|D, x) represents the posterior distribution of the objective function given the data D, μ(x) represents the mean of the posterior distribution, σ 2 (x) represents the variance of the posterior distribution, EI(x) represents the expected improvement, and x + represents the current optimal point, x next Indicates the next experimental point.

Claims

1. A data-driven method for predicting and optimizing the high temperature performance of asphalt mixtures, characterized in that: The method comprises the following steps: S1: Organize and preprocess the high temperature performance data of asphalt mixtures, and use the gray correlation analysis method to analyze the influence of characteristic index parameters such as asphalt penetration, asphalt softening point, oil-stone ratio, mineral aggregate grading information, mixture void ratio, mineral aggregate gap ratio and effective asphalt saturation on the high temperature performance index parameters of asphalt mixtures. Based on the gray analysis results, screen the data set and determine the input and output of the high temperature performance prediction model; S2: Build three machine learning models, namely ANN, SVM and GPR, to predict the high temperature performance of asphalt mixture, visualize the models, use CV cross validation to verify the prediction performance of each machine learning regression model, and select the best prediction model; S3: Based on the high-temperature performance optimization prediction model, the Bayesian optimization algorithm is used to carry out targeted design optimization of the mix ratio of asphalt mixture, and some parameters in the mix ratio are limited to simulate the application requirements under different actual working conditions.

2. The data-driven asphalt mixture high temperature performance prediction and design optimization method according to claim 1 is characterized in that: The grading information of the mineral material in step S1 includes the pass rates of the key sieve holes of 19 mm, 16 mm, 13.2 mm, 9.5 mm, 4.75 mm, 2.36 mm, 1.18 mm, 0.6 mm, 0.3 mm, 0.15 mm and 0.075 mm, and the three parameters of the Bailey method, CA ratio, FA C Ratio, FA f The ratio and grading curve of Weibull distribution have two parameters: shape parameter k and scale parameter λ.

3. The data-driven asphalt mixture high temperature performance prediction and design optimization method according to claim 1 is characterized in that: In the step S2, three machine learning models, ANN, SVM and GPR, are constructed based on the Matlab framework, wherein the ANN prediction model includes three hidden layers, the kernel functions of the SVM prediction model and the GPR prediction model are quadratic kernel function and Matern 5 / 2 kernel function respectively, and other optimal hyperparameters are obtained by Bayesian optimization.

4. The data-driven asphalt mixture high temperature performance prediction and design optimization method according to claim 1 is characterized in that: The asphalt mixture performance prediction visualization in step S2 is presented in the form of an APP, and provides switching of three prediction models, Bailey method parameter auxiliary calculation, Weibull distribution auxiliary calculation and high temperature performance prediction functions.

5. The data-driven high temperature performance prediction and design optimization method for asphalt mixture according to claim 1 is characterized in that: The objective function of the Bayesian optimization algorithm in step S3 is expressed as: Among them, obj is the objective function value, prediction is the predicted value obtained after bringing the current optimization formula into the optimization prediction model, targetY is the optimization target value, and gradation is the mineral gradation.

6. The data-driven high temperature performance prediction and design optimization method for asphalt mixture according to claim 1 is characterized in that: Some parameters defined in step S3 include asphalt index, gradation information and mixture parameters VV and VFA. The asphalt index refers to the gradation, oil usage and compaction degree under target performance for a specific asphalt; the gradation information refers to the use of a certain gradation to obtain the type of asphalt, oil usage and compaction degree under target performance; the mixture parameters VV and VFA refer to the type of asphalt, gradation information and oil usage under specified compaction to obtain the target performance.

Images