Pavement Risk Early Warning Method Based on Unsupervised Learning of FE-FCM Multidimensional Diseases

Abstract

The present invention provides a pavement risk early warning method based on unsupervised learning of FE-FCM multi-dimensional diseases, comprising the following steps: S1. Obtain pavement disease data; S2. Perform missing value and standardization processing on the pavement disease data, and then perform feature extraction to reduce the data dimension; S3. Based on the FCM model of the sample dispersion probability in the cluster, calculate the pavement performance risk level on each road section; S4. Give early warning to high-risk road sections according to the calculation results. The present invention has the characteristics of easy implementation and high compatibility. Multi-feature data under road diseases can be flexibly selected according to the disease characteristics of different test sections, and the calibrated number of categories can be flexibly selected through the optimal target of the silhouette coefficient. Moreover, the model can be calibrated by continuously improving the later data to improve the accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of risk warning, and particularly to a road surface risk warning method based on unsupervised learning of multi-dimensional diseases by FE-FCM. Background Art

[0002] Since asphalt pavements are sensitive to the damage caused by heavy traffic, under the long-term action of heavy traffic, asphalt pavements on highways and urban roads will show damages of different severities. Along with the continuous attenuation of the performance indicators of asphalt pavements, diseases such as cracks, rutting, water damage, and slippage occur on the road surface, posing potential hazards to road traffic safety. As the degree of road aging increases continuously, the driving comfort of drivers and passengers is severely reduced, which has a serious impact on the running speed and traffic capacity. On this basis, it is easy to lead to traffic accidents, and at the same time, it also brings great pressure to the road operation managers.

[0003] In order to maintain the good functionality of asphalt pavements, it becomes particularly important how to efficiently and plannedly utilize limited resources. By reasonably dividing sections, using statistical methods to quantify the risks of common diseases of asphalt pavements, and objectively evaluating asphalt pavements from the perspectives of road safety and the comfort of road participants, high-risk asphalt pavements under multi-dimensional diseases can be identified, and focused and efficient disease treatment can be achieved, thereby improving road safety and the comfort of road participants.

[0004] Dividing the risk levels of segmented asphalt pavements under multi-dimensional diseases can reasonably study the risk levels under the functional performance of sections and the urgency and severity of disease treatment from a data perspective, so as to extract high-risk sections for early warning, providing technical support for the disease treatment work of highways and urban roads.

[0005] In view of this, it is necessary to design an improved road surface risk warning method based on unsupervised learning of multi-dimensional diseases by FE-FCM to solve the above problems. Summary of the Invention

[0006] In order to overcome the deficiencies of the above-mentioned prior art, the purpose of the present invention is to provide a road surface risk warning method based on unsupervised learning of multi-dimensional diseases by FE-FCM, which has the characteristics of easy implementation and high compatibility. Multi-feature data under road diseases can be flexibly selected according to the disease characteristics of different test sections. At the same time, the calibrated number of categories can be flexibly selected through the optimal objective of the silhouette coefficient, and the model can be calibrated by continuously improving the later data to improve the accuracy of the model.

[0007] To achieve the above object of the invention, the present invention provides a road surface risk warning method based on unsupervised learning of multi-dimensional diseases by FE-FCM, including the following steps:

[0008] S1. Obtain pavement disease data;

[0009] S2. Perform missing value and normalization processing on the pavement disease data;

[0010] S3. Calculate the pavement performance risk level for each road section based on the FCM model of the sample dispersion probability in the cluster;

[0011] S4. Issue a warning for high-risk road sections according to the calculation results;

[0012] Among them, the acquisition method of the pavement disease data includes:

[0013] S11. Assign ID numbers to the road sections participating in data collection on the pavement. Each road section assigned an ID number is used as an experimental road section, and each test section includes all the characteristic dimensions in this road section;

[0014] S12. Collect and count the multi-dimensional characteristic values of the pavement disease data according to the test sections; the pavement disease data includes cracks, rutting, slippage and water damage;

[0015] Step S3 includes: determining the probability relative to the centroid of each cluster by assigning a data point to multiple clusters, so as to obtain the cluster where the data point is located, and realizing the division of the pavement risk level;

[0016] Step S3 includes:

[0017] S31. Specify the number k of cluster centroids, and then randomly assign cluster sample points to each centroid;

[0018] S32. Calculate the centroid μ of each cluster (k) , k ∈ {1,..., n};

[0019] S33. Update the sample points of each cluster;

[0020] S34. Loop steps S32 and S33 until the sample points in the cluster no longer change;

[0021] The centroid μ of the cluster (k) is calculated by the weighted average of all data points:

[0022]

[0023] The similarity within the cluster is measured by distance. When J m is the smallest, the similarity of the internal cluster sample points is the highest. The calculation formula is as follows:

[0024]

[0025] Among them, ω(i,j) Indicates the probability that each sample belongs to each cluster. m is the fuzzifier, x (i) is the cluster sample point, n is the total sample size, and k is the centroid point:

[0026]

[0027] where p is the initial accumulation amount;

[0028] Judge the effect of the risk level classification of pavement diseases under multi-dimensional diseases through the silhouette coefficient s (i) :

[0029]

[0030] where b (i) is the cluster separation degree between the cluster and the nearest cluster, and a (i) is the cluster cohesion.

[0031] As a further improvement of the present invention, in step S2, the normalization processing formula is as follows:

[0032]

[0033] x (i) is the input sample point, μ x is the total sample standard deviation, σ x is the total sample variance, is the result after standardizing a single sample.

[0034] As a further improvement of the present invention, step S2 further includes feature extraction on the data after normalization processing, and the steps are as follows:

[0035] S21. Input the sample set D = {x1, x2,..., x n} and the low-dimensional space dimension d' = 3;

[0036] S22. Assume that the new coordinates after the projection transformation are W = {ω1, ω2,..., ω i ,..., ω d}, where ω i is the standard orthogonal basis vector, and calculate the reconstructed sample point Starting from the entire sample, the distance between the original sample point x i and the reconstructed sample point is:

[0037]

[0038] where n is the total sample size, x i is the sample point, is the sample point x iThe coordinates in the j - dimensional space of the low - dimensional coordinate system, d′ is the dimension of the low - dimensional space, z i =(z i1 , z i2 ,..., z id′ ) is the projection of the sample point x i in the low - dimensional coordinate system, const is a constant, and d is the initial dimension;

[0039] S23. Minimize formula (6) to make the distance from the sample point to the hyperplane the closest, and the optimization objective is:

[0040]

[0041] S24. Adopt the Lagrange multiplier method XX T ω i =λ i ω i , where XX T is the covariance matrix, calculate the eigenvalue λ i , sort the obtained eigenvalues: λ1≥λ2≥...≥λ d , and take the first d′ eigenvectors to form the solution.

[0042] As a further improvement of the present invention, step S4 includes: according to the calculation results of the model, select the first m categories before sorting as high - risk, and export the corresponding road sections to give risk warnings.

[0043] As a further improvement of the present invention, conduct inspections once every half month, import the newly added road surface disease data into the data file, and train and correct the established model.

[0044] The beneficial effects of the present invention are:

[0045] The present invention provides a pavement risk warning method based on FE - FCM multi - dimensional disease unsupervised learning, which provides a basis for the warning and treatment of highway pavements and urban asphalt pavements under multiple diseases, and lays a foundation for the improvement of road use performance and the enhancement of traffic safety. This method has the characteristics of easy implementation and high compatibility. The multi - feature data under road diseases can be flexibly selected according to the disease characteristics of different test sections. At the same time, the number of calibrated categories can be flexibly selected through the optimal objective of the silhouette coefficient, and the model can be calibrated by continuously improving the later - stage data to improve the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is the flow block diagram of the pavement risk warning method based on FE - FCM multi - dimensional disease unsupervised learning of the present invention.

[0047] Figure 2The performance of cracks in the four grades of i, ii, iii, and iv in the embodiments.

[0048] Figure 3 The performance of flatness in the four grades of i, ii, iii, and iv in the embodiments.

[0049] Figure 4 The performance of rutting in the four grades of i, ii, iii, and iv in the embodiments. Detailed implementation manners

[0050] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with specific embodiments.

[0051] Here, it should also be noted that in order to avoid obscuring the present invention due to unnecessary details, only the structures and / or processing steps closely related to the solution of the present invention are shown in the specific embodiments, and other details less related to the present invention are omitted.

[0052] In addition, it should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0053] A pavement risk early warning method based on FE-FCM multi-dimensional disease unsupervised learning provided by the present invention has the following detailed implementation steps:

[0054] 1. Investigate and count all functional diseases (such as cracks, ruts, slips, and water damage, etc.) on the asphalt pavement by section, specifically as follows:

[0055] (11) Assign ID numbers to the sections of the asphalt pavement participating in data collection. Each section assigned an ID number is used as an experimental section, and each test section includes all the characteristic dimensions in that section;

[0056] (12) Collect and count multi-dimensional characteristic values such as the number of cracks, the number of ruts, and the area of slips according to the test sections; the present invention numbers and collects the functional disease data by section, which is convenient for data collection and classification, and also convenient for risk early warning of each section.

[0057] 2. Perform data preprocessing on the collected disease data, mainly including missing value processing, standardization processing, and feature extraction, specifically as follows:

[0058] (21) Process the missing values of the collected multi-dimensional disease characteristic values;

[0059] (22) Based on the sample data after missing value processing, perform standardization according to formula (1); where x (i) is the input sample point, μ x is the total sample standard deviation, σ x is the total sample variance, is the result after standardizing a single sample:

[0060]

[0061] Transform the feature data into a standard normal distribution with a mean of 0 and a variance of 1, so that the selected rutting, pavement smoothness, and crack data have the same unit, preventing only larger numerical data from being captured in the next feature extraction, resulting in an inaccurate model, and thus enabling the model to obtain the best performance.

[0062] (23) Since the data is multi-dimensional data, in order to exclude as much redundant data information as possible, identify the correlation between samples to obtain the rules between data, and improve the accuracy of the model, it is necessary to perform feature extraction on the multi-dimensional data. The feature extraction method is as follows:

[0063] (231) Input the sample set D = {x1, x2,..., x n} and the low-dimensional space dimension d′ = 3.

[0064] (232) Assume that the new coordinates after projection transformation are W = {ω1, ω2,..., ω i ,..., ω d}, where ω i is the standard orthogonal basis vector, and calculate the reconstructed sample point Starting from the entire sample, the distance between the original sample point x i and the reconstructed sample point is:

[0065]

[0066] Among them, n is the total sample size, x i is the sample point, is the j-dimensional coordinate of the sample point x i in the low-dimensional coordinate system, d′ is the low-dimensional space dimension, z i = (z i1 , z i2 ,..., z id′ ) is the projection of the sample point x i in the low-dimensional coordinate system, and const is a constant.

[0067] (233) Among them, in order to make the distance from the sample point to the hyperplane the closest, formula (2) should be minimized, and the optimization objective is:

[0068]

[0069] (234) The Lagrange multiplier method is adopted for XX T ω i = λ i ω i , where XX T is the covariance matrix, and the eigenvalue λ is calculated i , and the obtained eigenvalues are sorted: λ1 ≥ λ2 ≥... ≥ λ d , and the first d' eigenvectors are taken to form the solution.

[0070] 3. By assigning a data point to multiple clusters to determine the probability relative to the centroid of each cluster, the cluster where the data point is located is obtained, and the division of the risk level of the asphalt pavement is realized. The key steps are as follows:

[0071] (31) Specify the number of centroids k, and then randomly assign cluster sample points to each centroid;

[0072] (32) Calculate the centroid μ of each cluster (k) , k ∈ {1,..., n};

[0073] (33) Update the sample points of each cluster;

[0074] (34) Loop steps 2 and 3 until the sample points in the cluster no longer change.

[0075] Among them, the similarity within the cluster is measured by the distance J m , when J m is the smallest, the similarity of the internal cluster sample points is the highest. The calculation method of the objective function of this algorithm is:

[0076]

[0077] ω (i,j) represents the probability that each sample belongs to each cluster, m is the fuzzifier, controlling the degree of fuzziness, x (i) is the cluster sample point, n is the total sample size, and k is the centroid point:

[0078]

[0079] In the formula, p is the initial accumulation amount.

[0080] The weighted average of all data points is calculated as the centroid of the cluster:

[0081]

[0082] 4. Use the silhouette coefficient s (i)To judge the effect of risk level classification of asphalt pavement diseases under multi-dimensional diseases;

[0083]

[0084] b (i) is the cluster separation degree between the cluster and the nearest cluster, and a (i) is the cluster cohesion;

[0085] 5. According to the calculation results of the model, select the m categories before sorting as high risks, and export the corresponding road sections to give risk warnings; m is a positive integer from 1 to 4.

[0086] 6. Implement inspections once every half month, import the newly added pavement disease data into the data file, train and correct the previously established model, and improve the classification accuracy.

[0087] Embodiment

[0088] Select the disease data on the experimental road sections in the United States and Canada for analysis and verification.

[0089] Preprocess the collected multi-dimensional disease data according to the above steps. First, normalize the data and convert the feature values into the standard normal distribution to ensure the balance of the importance of all feature values. Use the feature extraction method to reduce the dimension of the 14-dimensional disease feature vector, and while excluding the interference of abnormal data, improve the accuracy of the model. Classify the disease risk levels of the asphalt pavement through step (3), and check the classification effect of the model for different numbers of clusters through the silhouette coefficient to find the optimal number of clusters. As can be seen from Table 1, when k is 4, the silhouette coefficient is the largest, so four risk levels are preferably selected.

[0090] Table 1 Silhouette coefficients corresponding to different numbers of clusters

[0091] Number of clusters k 3 4 5 6 7 8 9 10 Silhouette coefficient 0.577 0.604 0.589 0.595 0.597 0.584 0.592 0.592

[0092] According to the classification results of the model, calibrate the level of the functional performance of the asphalt pavement into four categories (i, ii, iii, and iv), where categories iii and iv are selected as severe risk levels, especially for the test sections corresponding to category iv, risk warnings need to be given. In the process of disease statistical analysis, mainly select pavement functional performances such as cracks, rutting, and pavement smoothness to represent the overall road performance of the existing asphalt pavement. The performances of cracks, rutting, and pavement smoothness under the four levels of i, ii, iii, and iv are as Figure 2-3 shown.

[0093] In summary, the pavement risk early warning method based on FE-FCM multi-dimensional disease unsupervised learning provided by the present invention provides a basis for the early warning and treatment of highway pavements and urban asphalt pavements under multiple diseases, and lays a foundation for the improvement of road use performance and the enhancement of traffic safety. This method has the characteristics of easy implementation and high compatibility. The multi-feature data under road diseases can be flexibly selected according to the disease characteristics of different test sections. At the same time, the calibrated number of categories can be flexibly selected through the optimal objective of the silhouette coefficient, and the model can be calibrated by continuously improving the later data to improve the accuracy of the model.

[0094] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A pavement risk warning method based on FE-FCM multi-dimensional disease unsupervised learning, characterized by: The following steps are involved: S1. Obtain pavement damage data; S2. Process missing values and standardize the pavement damage data; S3. Calculate the road performance risk level for each road segment based on the FCM model of sample dispersion probability in the cluster; S4. Issue early warnings for high-risk road sections based on the calculation results; The method for obtaining the pavement damage data includes: S11. ID the road sections involved in data collection. Each ID-numbered road section serves as an experimental section, and each experimental section includes all characteristic dimensions of the section. S12. Collect and compile statistics on multidimensional characteristic values of pavement damage data according to the test section; the pavement damage data includes cracks, rutting, slippage, and water damage; Step S3 includes: assigning a data point to multiple clusters to determine the probability relative to the centroid of each cluster, thereby obtaining the cluster where the data point is located and achieving the classification of road risk levels; Step S3 includes: S31. Specify the number of cluster centroids k, and then randomly assign cluster sample points to each centroid; S32. Calculate the centroid μ of each cluster (k) , k∈{1,...,n}; S33. Update the sample points of each cluster; S34. Loop steps S32 and S33 until the sample points in the cluster no longer change; The centroid μ of the cluster (k) Calculated by weighted average of all data points: The similarity within the cluster is measured by distance. When J m When it is minimum, the similarity of internal cluster sample points is the highest, and the calculation formula is as follows: Among them, ω (i,j) represents the probability that each sample belongs to each cluster, m is the fuzzifier, x (i) is the cluster sample point, n is the total sample size, and k is the centroid point: Among them, p is the initial amount of accumulation; By the silhouette coefficient s (i) Determine the effectiveness of risk classification of pavement damage under multi-dimensional damage: Among them, b (i) is the cluster separation between the cluster and the nearest cluster, a (i) is the cluster cohesion.

2. The pavement risk early warning method based on FE-FCM multidimensional disease unsupervised learning according to claim 1 is characterized in that: In step S2, the normalization formula is as follows: x (i) is the input sample point, μ x is the total sample standard deviation, σ x is the total sample variance, The results are normalized for a single sample.

3. The pavement risk early warning method based on FE-FCM multidimensional disease unsupervised learning according to claim 2 is characterized in that: Step S2 also includes feature extraction of the normalized data, the steps are as follows: S21. Input sample set D = {x1, x2, ..., x n } and the dimension of the low-dimensional space d′ = 3; S22. Assume that the new coordinates after projection transformation are W = {ω1,ω2,...,ω i ,...,ω d }, where ω i Is the standard orthogonal basis vector, calculate the reconstructed sample points Starting from the entire sample, the original sample point x i And the reconstructed sample points The distance is: Where n is the total sample size, x i is the sample point, is the sample point x i The j-dimensional coordinate in the low-dimensional coordinate system, d′ is the dimension of the low-dimensional space, z i =(z i1 , z i2 ,...,z id ′) is the sample point x i The projection in the low-dimensional coordinate system, const is a constant, d is the initial dimension; S23. Minimize formula (6) to minimize the distance between the sample point and the hyperplane. The optimization objective is: S24. Using Lagrange multiplier method XX T ω i =λ i ω i , where XX T is the covariance matrix, calculate the eigenvalue λ i , sort the desired eigenvalues: λ1≥λ2≥...≥λ d , take out the first d′ eigenvectors to form the solution.

4. The pavement risk early warning method based on FE-FCM multidimensional disease unsupervised learning according to claim 1 is characterized in that: Step S4 includes: selecting the top m categories as high-risk according to the calculation results of the model, and exporting the corresponding road sections to provide risk warnings.

5. The pavement risk early warning method based on FE-FCM multi-dimensional disease unsupervised learning according to claim 1 is characterized in that: Carry out inspections once every two weeks, import new pavement disease data into the data file, and train and correct the established model.

Images