Asphalt pavement permeability coefficient prediction method based on multi-source data

Abstract

The invention provides an asphalt pavement permeability coefficient prediction method based on multi-source data, which is suitable for the field of roadbed and pavement engineering and comprises the steps of data acquisition, data arrangement and input and output coding, prediction model construction, prediction model training and optimization and prediction model application. According to the method, the relationship between the permeability coefficient test result and a plurality of related parameters is comprehensively considered, and the prediction model established by using the BP neural network can effectively capture the complex nonlinear relationship influencing the asphalt pavement permeability coefficient, so that the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to a prediction method for the permeability coefficient of asphalt pavement based on multi-source data, which is applicable to the field of subgrade and pavement engineering. Background Technique

[0002] Due to its good anti-fatigue performance, excellent anti-slip characteristics and wide applicability, asphalt pavement has been widely used in modern transportation infrastructure. With the rapid development of transportation, the quality problems of asphalt pavement have become increasingly prominent. Among them, the permeability coefficient, as an important index to measure the water penetration ability of asphalt pavement, has an important impact on aspects such as the service life, anti-stripping property and driving safety of the pavement. Therefore, accurately predicting the permeability coefficient of asphalt pavement has important practical significance for pavement design, construction and maintenance management.

[0003] Traditional methods for measuring the permeability coefficient of asphalt pavement usually rely on on-site tests and laboratory experiments. These methods are not only time-consuming and laborious, but also affected by factors such as the external environment, test technical means and the experience of operators, resulting in the difficulty of guaranteeing the accuracy and reliability of their results. In addition, due to factors such as the aging of asphalt pavement during use, frequent traffic loads and climate change, the permeability coefficient may change over time. Therefore, relying solely on the results of one-time measurements often cannot fully reflect the actual state of the pavement.

[0004] In recent years, with the rapid development of information technology, the analysis method based on multi-source data has gradually become an effective way to study the permeability coefficient of asphalt pavement. By integrating various types of data, including the physical properties of asphalt pavement, traffic loads, meteorological factors and maintenance history, etc., various factors affecting the permeability coefficient of asphalt pavement can be understood more comprehensively. However, how to effectively utilize these multi-source data and establish an efficient and accurate prediction model is still a major challenge currently faced. Summary of the Invention

[0005] The purpose of the present invention is to propose a prediction method for the permeability coefficient of asphalt pavement based on multi-source data in view of the current urgent need for a method that can effectively utilize multi-source data to establish an efficient and accurate prediction method for the permeability coefficient of asphalt pavement.

[0006] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0007] A prediction method for the permeability coefficient of asphalt pavement based on multi-source data comprises the following steps:

[0008] S101, acquisition of data;

[0009] The acquisition of the data includes obtaining the test results and related parameters of the permeability coefficient of n measuring points on the asphalt pavement, and the test results of the permeability coefficient of the asphalt pavement are marked as Qi where \(i = 1\sim n\), and the corresponding relevant parameters include the measurement result \(N\) of the asphalt pavement void ratio i , the classification result \(M\) of the asphalt pavement by structural type i , the thickness \(D\) of the asphalt pavement i , the cross slope \(L\) of the asphalt pavement i and the maximum nominal particle size \(R\) i ;

[0010] S102, data sorting and input / output coding;

[0011] The data sorting and input / output coding includes coding the classification result \(M\) i according to the classification result code to obtain the coded result \(M'\) i ;

[0012] The data sorting and input / output coding also includes constructing an input data group and an output data group. The input data group is \(C\) i =\(\{N\) i , \(M'\) i , \(D\) i , \(L\) i , \(R\) i \}\), and the corresponding output data group is \(D\) i =\(\{Q\) i \}\). The input data group \(C\) i and its corresponding output data group \(D\) i are subjected to data standardization processing, and then divided into training data and test data according to a certain ratio;

[0013] S103, constructing a prediction model;

[0014] The constructing of the prediction model includes establishing a prediction model based on a BP neural network;

[0015] S104, training and optimizing the prediction model;

[0016] The training and testing of the prediction model include using the training data to carry out model training on the prediction model, finally obtaining the trained model, using the input data group of the test data to input into the trained model to obtain a prediction result, comparing and analyzing the prediction result with the output data group of the test data, and finally obtaining an applicable model according to the comparison and analysis result;

[0017] S105, application of the prediction model;

[0018] The application of the prediction model includes using the applicable model in the prediction work of the asphalt pavement permeability coefficient.

[0019] Further, in the above step S101, the value of n among the n measuring points on the asphalt pavement needs to satisfy the condition that n≥100.

[0020] Further, in the above step S101, the classification result M of the asphalt pavement by structural type i is a dense-graded asphalt concrete pavement, or a semi-open-graded asphalt concrete pavement, or an open-graded asphalt concrete pavement, or others.

[0021] It should be noted that "others" here means that the classification result M of the asphalt pavement by structural type i does not belong to a dense-graded asphalt concrete pavement, a semi-open-graded asphalt concrete pavement, and an open-graded asphalt concrete pavement.

[0022] Further, in the above step S102, the classification result M i is coded according to the classification result to obtain the coded result M'. i The steps are as follows:

[0023] a) Obtain the classification result M i ;

[0024] b) When the classification result M i is a dense-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 1, otherwise proceed to the next step;

[0025] c) When the classification result M i is a semi-open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 2, otherwise proceed to the next step;

[0026] d) When the classification result M i is an open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 3, otherwise proceed to the next step;

[0027] e) When the classification result M i does not belong to any of the three classification results of dense-graded asphalt concrete pavement, semi-open-graded asphalt concrete pavement, and open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 4.

[0028] Further, in the above step S102, the data standardization process uses the maximum-minimum method, and the calculation formula of the maximum-minimum method is shown in Equation (1).

[0029]

[0030] In the formula, X' is the value after standardization processing, X is the original feature value, Xmin and X max are the minimum and maximum values of the original eigenvalues, and a and b are the lower and upper limits of the scaling interval, where a takes the value of -1 / 2 and b takes the value of 1 / 2.

[0031] Furthermore, in the above step S103, the construction steps of the prediction model based on the BP neural network are as follows:

[0032] a) Establish a BP neural network structure, including an input layer, a hidden layer, and an output layer. The input layer contains 5 nodes, the output layer contains one node, the number of layers of the hidden layer is 1, and the number of nodes in the hidden layer is initially determined within a range using the empirical formula method and then determined by the trial-and-error method. The empirical formula is Equation (2),

[0033]

[0034] where L is the number of nodes in the hidden layer, m is the number of nodes in the output layer, n is the number of nodes in the input layer, and a is a constant between 0 and 10;

[0035] b) Set the activation function used in the hidden layer to introduce non-linearity. The activation function uses a piecewise activation function, and the formula of the piecewise activation function is Equation (3),

[0036]

[0037] where u and l are hyperparameters.

[0038] Furthermore, in the above step S104, the steps of comparing and analyzing the prediction result with the output data set of the test data and finally obtaining an applicable model are as follows:

[0039] a) Obtain the prediction result marked as P j , and the permeability coefficient test result Q of the corresponding output data set of the test data j ,

[0040] b) Calculate the average relative error ε. If the average relative error ε is less than or equal to 15%, it meets the requirements, and the model of the obtained prediction value is considered to be an applicable model at this time. Otherwise, parameter adjustment should be performed until the requirements are met. The calculation formula of the average absolute error ε is Equation (4),

[0041]

[0042] where n is the number of data in the test set.

[0043] The present invention has the following beneficial effects: (1) Greatly improving the prediction accuracy: By comprehensively considering the permeability coefficient test results and related parameters of multiple measuring points, and using the prediction model established by the BP neural network, the complex non-linear relationship affecting the permeability coefficient of asphalt pavement can be effectively captured, thus improving the prediction accuracy. Compared with traditional measurement methods, more factors are considered in data processing and model establishment in this method, and more accurate results can be provided for practical applications.

[0044] (2) Data-driven decision support: This method provides a scientific basis for pavement management and maintenance through the acquisition and analysis of multi-source data. Relevant departments can formulate reasonable maintenance plans based on the prediction results, optimize resource allocation, effectively extend the service life of asphalt pavement, and reduce the later maintenance cost.

[0045] (3) Having time and economic benefits: Compared with the methods of manual measurement and laboratory testing, the implementation of this method has higher efficiency; through the automated data collection and analysis process, the manpower and material resources required for on-site testing are reduced, the testing time is greatly shortened, and a more cost-effective solution can be obtained. Description of the Drawings

[0046] Figure 1 It is a flowchart of a method for predicting the permeability coefficient of asphalt pavement based on multi-source data according to the present invention;

[0047] Figure 2 It is a topological structure diagram of the BP neural network of the embodiment of the present invention. Detailed Embodiment

[0048] The following details the specific embodiments of the present invention with reference to the drawings; it should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0049] The following is a specific embodiment of a method for predicting the permeability coefficient of asphalt pavement based on multi-source data.

[0050] S101, Acquisition of data;

[0051] The acquisition of the data includes obtaining the permeability coefficient test results and related parameters of n measuring points of the asphalt pavement, and the permeability coefficient test results of the asphalt pavement are marked as Q i , i = 1 to n, and the corresponding related parameters include the measured result N of the asphalt pavement void ratio i , the classification result M of the asphalt pavement according to the structural type i , the thickness D of the asphalt pavement i , the cross slope L of the asphalt pavement i and the maximum nominal particle size R i ;

[0052] In the further step S101 above, the value of n among the n measuring points of the asphalt pavement needs to satisfy the condition that n ≥ 100.

[0053] Further, in the above step S101, the classification result M of the asphalt pavement by structural type i is a dense-graded asphalt concrete pavement, or a semi-open-graded asphalt concrete pavement, or an open-graded asphalt concrete pavement, or others.

[0054] In this embodiment, n is taken as 100, that is, in this embodiment, the permeability coefficient results of 100 measuring points and their corresponding relevant parameter results are collected, and the results are shown in Table 1;

[0055] Table 1 Data fragment obtained

[0056]

[0057]

[0058] S102, data sorting and input / output coding;

[0059] The data sorting and input / output coding includes coding the classification result M i according to the classification result code therein to obtain a coded result M'; i ;

[0060] The data sorting and input / output coding also includes constructing an input data group and an output data group. The input data group is C i ={N i , M' i , D i , L i , R i}, and the corresponding output data group is D i ={Q i}. The input data group C i and its corresponding output data group D i are subjected to data standardization processing, and then divided into training data and test data according to a certain ratio;

[0061] Further, in the above step S102, the step of coding the classification result M i according to the classification result code therein to obtain a coded result M' i is as follows:

[0062] a) Obtain the classification result M i ;

[0063] b) When the classification result M iWhen it is a dense-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 1, otherwise proceed to the next step;

[0064] c) When the classification result M i is a semi-open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 2, otherwise proceed to the next step;

[0065] d) When the classification result M i is an open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 3, otherwise proceed to the next step;

[0066] e) When the classification result M i does not belong to any of the three classification results of dense-graded asphalt concrete pavement, semi-open-graded asphalt concrete pavement, and open-graded asphalt concrete pavement, the corresponding coded result M' i takes the value of 4.

[0067] Furthermore, in the above step S102, the data standardization process uses the maximum-minimum method, and the calculation formula of the maximum-minimum method is shown in Equation (1).

[0068]

[0069] In the formula, X' is the value after standardization processing, X is the original feature value, X min and X max are the minimum and maximum values of the original feature value, and a and b are the lower and upper limits of the scaling interval, where a takes the value of -1 / 2 and b takes the value of 1 / 2.

[0070] S103, construct a prediction model;

[0071] The construction of the prediction model includes establishing a prediction model based on a BP neural network;

[0072] Furthermore, in the above step S103, the construction steps of establishing the prediction model based on a BP neural network are as follows:

[0073] a) Establish a BP neural network structure, including an input layer, a hidden layer, and an output layer. The input layer contains 5 nodes, the output layer contains one node, the number of layers of the hidden layer is 1, and the number of nodes in the hidden layer is initially determined within a range using the empirical formula method and then determined by the trial-and-error method. The empirical formula is Equation (2).

[0074]

[0075] Wherein, L is the number of hidden layer nodes, m is the number of output layer nodes, n is the number of input layer nodes, and a is a constant between 0 and 10;

[0076] b) Set the activation function used in the hidden layer to introduce non-linearity. The activation function uses a piecewise activation function, and the formula of the piecewise activation function is Formula (3).

[0077]

[0078] Wherein, u and l are hyperparameters.

[0079] In this embodiment, the number of hidden layer nodes l is calculated according to Formula 2 (where a takes the value of 10) as: The trial-and-error method is adopted to finally obtain the number of hidden layer nodes l = 8. The topological structure of the BP neural network is as Figure 2 shown.

[0080] S104, Training and optimization of the prediction model;

[0081] The training and testing of the prediction model include using the training data to carry out model training on the prediction model, finally obtaining the trained model, inputting the input data group of the test data into the trained model to obtain the prediction result, comparing and analyzing the prediction result with the output data group of the test data, and finally obtaining the applicable model according to the comparison and analysis result;

[0082] Further, in the above step S104, the step of comparing and analyzing the prediction result with the output data group of the test data and finally obtaining the applicable model is:

[0083] a) Obtain the prediction result marked as P j , and the permeability coefficient test result Q of the corresponding output data group of the test data j ,

[0084] b) Calculate the average relative error ε. If the average relative error ε is less than or equal to 15%, it meets the requirements, and the model of the obtained predicted value is considered as the applicable model. Otherwise, parameter adjustment should be carried out until it meets the requirements. The calculation formula of the average absolute error ε is Formula (4).

[0085]

[0086] Wherein, n is the number of data in the test set.

[0087] S105, Application of the prediction model;

[0088] The application of the prediction model includes applying the applicable model to the prediction work of the asphalt pavement permeability coefficient.

[0089] In the above embodiments, the present invention discloses a prediction method for the permeability coefficient of asphalt pavement based on multi-source data, including data acquisition, data sorting and input / output coding, construction of a prediction model, training and optimization of the prediction model, and application of the prediction model; this method comprehensively considers the relationship between the test results of the permeability coefficient and multiple related parameters, and uses the prediction model established by the BP neural network to effectively capture the complex non-linear relationship affecting the permeability coefficient of asphalt pavement, thereby improving the accuracy of prediction.

[0090] The above are the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting asphalt pavement permeability coefficient based on multi-source data, characterized in that: The following steps are involved: S101, data acquisition; S102, data arrangement and input and output coding; S103, constructing a prediction model; S104, training and optimization of prediction models; S105, Application of prediction models; The data acquisition includes obtaining the permeability coefficient test results and related parameters of n measuring points on the asphalt pavement, and the permeability coefficient test results of the asphalt pavement are marked as Q i , i = 1 to n, the corresponding related parameters include the asphalt pavement void ratio measurement result N i , Classification results of asphalt pavement by structure type M i , Asphalt pavement thickness D i , cross slope of asphalt pavement L i And the maximum nominal particle size R i ; The data arrangement and input and output coding include converting the classification result M i According to the classification result, the coding result M' is obtained. i ; The data arrangement and input / output coding also includes constructing an input data set and an output data set, wherein the input data set is C i = {N i , M' i , D i , L i , R i }, the corresponding output data set is D i = {Q i }, input data set C i The corresponding output data set D i Perform data standardization and then divide it into training data and test data according to a certain ratio; The constructing of the prediction model includes establishing a prediction model based on a BP neural network; The training and testing of the prediction model includes using the training data to carry out model training on the prediction model, finally obtaining a model after the training is completed, using the input data group of the test data to input into the trained model to obtain a prediction result, comparing and analyzing the prediction result with the output data group of the test data, and finally obtaining an applicable model according to the comparison and analysis results; The application of the prediction model includes using the applicable model to predict the permeability coefficient of asphalt pavement.

2. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In the step S101, n of the n measuring points of the asphalt pavement needs to satisfy the condition of n≥100.

3. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In the step S101, the classification result M of the asphalt pavement according to the structural type i It is a densely-graded asphalt concrete pavement, a semi-open-graded asphalt concrete pavement, an open-graded asphalt concrete pavement, or others.

4. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In step S102, the classification result M i According to the classification result, the coding result M' is obtained. i The steps are: a) Obtain classification result M i ; b) When the classification result M i For densely distributed asphalt concrete pavement, the corresponding coded result M' i The value is 1, otherwise proceed to the next step; c) When the classification result M i When it is a semi-open graded asphalt concrete pavement, the corresponding coded result M' i The value is 2, otherwise proceed to the next step; d) When the classification result M i For open-graded asphalt concrete pavement, the corresponding coded result M' i The value is 3, otherwise proceed to the next step; e) When the classification result M i When the pavement does not belong to the three classification results of dense asphalt concrete pavement, semi-open asphalt concrete pavement and open asphalt concrete pavement, the corresponding coded result M' i The value is 4.

5. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In step S102, the data standardization process adopts the maximum and minimum value method, and the calculation formula of the maximum and minimum value method is shown in formula (1): In the formula, X' is the standardized value, X is the original eigenvalue, and X min and X max are the minimum and maximum values ​​of the original eigenvalues, a and b are the lower and upper limits of the scaling interval, where a is -1 / 2 and b is 1 / 2.

6. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In step S103, the steps of building a prediction model based on BP neural network are: a) Establishing a BP neural network structure, including an input layer, a hidden layer and an output layer, wherein the input layer includes 5 nodes, the output layer includes 1 node, the number of the hidden layer is 1, and the number of nodes in the hidden layer is preliminarily determined by an empirical formula method and then determined by trial and error method. The empirical formula is formula (2), Where L is the number of hidden layer nodes, m is the number of output layer nodes, n is the number of input layer nodes, and a is a constant between 0 and 10; b) Setting the activation function used by the hidden layer to introduce nonlinear characteristics, the activation function adopts a piecewise activation function, and the formula of the piecewise activation function is formula (3), Where u and l are hyper parameters.

7. The method for predicting asphalt pavement permeability coefficient based on multi-source data according to claim 1 is characterized in that: In step S104, the step of comparing and analyzing the prediction result with the output data set of the test data and finally obtaining an applicable model according to the comparison and analysis result is as follows: a) Obtain the prediction result and mark it as P j , and the corresponding test data output data set permeability test results Q j , b) Calculate the average relative error ε. If the average relative error ε is less than or equal to 15%, it meets the requirements and the model of the predicted value obtained at this time is considered to be an applicable model. Otherwise, the parameters should be adjusted until the requirements are met. The calculation formula of the average absolute error ε is formula (4), In the formula, n is the number of test set data.

Images