A prediction method for the influence of maintenance operations on the long-term performance of pavement

Abstract

The present invention relates to a prediction method for the influence of maintenance operations on the long-term performance of a road surface, including: selecting the points for maintenance and normal points within the test area, and recording the maintenance operation information; taking pictures of the road surface within a certain range of the selected points to obtain road surface images, and dividing the periods; using an identification algorithm to identify the types and quantities of road surface damages in each road surface image; conducting classified statistics and averaging processing on the identification results to obtain the expected classified damage data of the points in different periods; calculating the change in the road surface performance of the points; using the maintenance operation information of the points and the change in the road surface performance to construct a causal network skeleton; identifying the direction of the causal network through a separation algorithm; constructing and training a prediction model for the change in the road surface performance based on a deep neural network; predicting the change in the road surface performance in different future periods based on the prediction model for the change in the road surface performance. Compared with the prior art, the present invention has the advantages of comprehensive time dimension information, accurate prediction, etc.

Description

Technical Field

[0001] The present invention relates to the field of road pavement performance evaluation, and particularly to a prediction method for the influence of maintenance operations on the subsequent performance of the pavement. Background Art

[0002] With the increasing contradiction between the growth of China's road mileage and the low efficiency of road operation, the intelligentization of road operation is the development trend. Using data-driven informatization means to evaluate road performance and methods for analyzing factors affecting road performance changes and extending road service life based on perceived data are urgent needs to improve road operation efficiency.

[0003] There are various methods and indicators for evaluating road pavement performance. In China's current standards, pavement condition index PCI, ride quality index RQI or international roughness index IRI, etc. can all be used as pavement performance evaluation index methods. Among them, using pavement damage to evaluate pavement performance is a common method. Different types of pavement damage can reflect different stages of pavement performance decay, and the type and quantity of pavement damage on a section can also reflect the current usage state of the road pavement to a certain extent. The image recognition and detection of road pavement damage have been popularized and applied at home and abroad.

[0004] Currently, the evaluation of road maintenance quality mainly relies on the immediate road roughness, elevation difference, skid resistance performance and pavement condition, etc. after the completion of road maintenance operations. These index methods can evaluate whether the performance of the road pavement meets the standards in the short term after the implementation of maintenance operations, but do not consider the impact of the quality of the implemented maintenance operations on the service performance of the road in the subsequent life cycle. The decay of road performance is a long cycle, and the medium- and long-term performance of the road pavement is a key factor in road service life and road operation cost control. A complete set of methods is needed to evaluate the changes in the performance of the road pavement in the subsequent life cycle after maintenance operations. Summary of the Invention

[0005] The purpose of the present invention is to provide a prediction method for the influence of maintenance operations on the subsequent performance of the pavement, accurately predict the changes in pavement performance after the implementation of maintenance operations, and improve the control of the quality of maintenance operations.

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

[0007] A prediction method for the influence of maintenance operations on the subsequent performance of the pavement, comprising the following steps:

[0008] S1: Select the points for maintenance and the normal points without maintenance within the test area, and record the maintenance operation information;

[0009] S2: Track and photograph the pavement within the pre-configured range of the points selected in S1 to obtain pavement images, and divide them into multiple cycles;

[0010] S3: Identify the types and quantities of pavement damages in each pavement image using an identification algorithm, and record the types and quantities of pavement damages at corresponding points at different acquisition times.

[0011] S4: Classify and statistically analyze the identification results of S3, and perform averaging processing on the data of the same cycle to obtain the expected classification damage data of the points at different cycles.

[0012] S5: Calculate the pavement performance of the points based on the expected classification damage data, and calculate the change in pavement performance between cycles.

[0013] S6: Utilize the maintenance operation information of the points and the change in pavement performance to construct the causal network skeleton for each cycle based on the causal inference algorithm.

[0014] S7: Based on the causal network skeleton, identify the direction of the causal network through the separation algorithm to obtain a complete causal relationship network.

[0015] S8: Construct and train a pavement performance change prediction model based on a deep neural network. The input of the model is the maintenance operation factors directly affecting the change in pavement performance selected from the causal relationship network, and the output is the change in pavement performance at different cycles.

[0016] S9: Predict the change in pavement performance at different future cycles based on the trained pavement performance change prediction model.

[0017] The specific steps of S1 are as follows:

[0018] Select the points under maintenance and the adjacent normal non-maintained points within the test area o, and denote the set of points as A. Among them, the maintenance points use the maintenance operation time as the time starting point, and the non-maintained points use the maintenance operation time of the nearest maintenance point in the same section as the time starting point.

[0019] Record the maintenance operation information of the maintenance points, and establish the set of maintenance operation information factors Z i ={z1, z2, …, z n} of point i in A. Among them, the numerical factors in Z i are expressed in numerical form, and the non-numerical factors are factorized and expressed in digital factors.

[0020] The specific steps of S2 are as follows:

[0021] Starting from the time starting point, use in-vehicle GPS data and visible light cameras to collect information on the pavement within the selected point range r for a period of c, and divide c into n cycles, with each cycle having a duration of c / n. Record pavement images, acquisition times, and geographic information data.

[0022] The specific steps of step S3 are as follows:

[0023] Perform damage identification and classification on the collected road surface images, establish a road surface condition tracking and inspection data table for point i in A, and record the maintenance operation situation Z i And the road surface damage types and quantities identified at different collection times within the point range r

[0024] The specific steps of step S4 are as follows:

[0025] Based on the road surface condition tracking and inspection data table, statistically analyze the different types of damage data for point i in A within the period t and range r and perform averaging processing to obtain the set X of damage statistics for i in the period t i,t ={x1, x2, x3…x k …}, where x k represents the expected damage quantity of damage k

[0026] The specific steps of step S5 are as follows:

[0027] Quantitatively evaluate the road surface performance status of point i within the period t and range r. Based on the X i,t set, multiply the damage severity by the weight to characterize the road surface performance y i,t , and the evaluation method is as follows:

[0028] y i,t =w1x1 + w2x2 + w3x3…

[0029] where, x k represents the quantity of damage k, and w k represents the weight of damage k;

[0030] The change in the road surface performance status within the range r of point i from the starting time to the period t is Δy i,t , where the road surface performance at the starting time is y i,0 , then:

[0031] Δy i,t =y i,t -y i,0

[0032] Let ΔY t be the vector composed of the performance changes of all points in A within the period t, then:

[0033] ΔY t =[Δy 1,t ,Δy 2, ,…,Δy i,t ,…]

[0034] The steps of step S6 include the following steps:

[0035] Step S61: Establish a variable dataset Z, where the variables in Z correspond to the maintenance operation situation Z i , and the target node of the causal network structure is the pavement performance change △y in cycle t t , set the candidate set S(z) = {} for the parent or child nodes of △y t ;

[0036] Step S62: Based on the mRMR criterion, find the nodes in Z that have a direct causal relationship with △y t , where the mRMR criterion is:

[0037]

[0038] Use the incremental search method to search for the factor z with the greatest node dependence on △y t , and incorporate z m into S(z); m ;

[0039] Step S63: Adopt conditional independence testing to remove the nodes in S(z) that have no causal relationship with the pavement performance change △y t ;

[0040] Step S64: Repeat S62 and S63, and iterate until the factors in S(z) no longer increase. Establish a causal network skeleton based on △y t and S(z), and the relationship between nodes is an undirected connection.

[0041] The specific steps of step S7 are as follows:

[0042] Use the principle of d-separation to determine the dependence direction of the edges in the graph, identify the causal network direction, expand the skeleton into a directed acyclic graph, and obtain a complete causal relationship network.

[0043] The deep neural network of the pavement performance change prediction model sets hidden layers, uses the ReLU function as the activation function, and sets a random dropout layer; adopts the cross-validation method to randomly extract b% of the road inspection data as the test set, and the remaining as the training set. After setting the initial learning rate and the number of training times, train the pavement performance change prediction model.

[0044] The specific steps of step S9 are as follows:

[0045] Input the optimized selection feature values of the point to be predicted j into the pavement performance change prediction model, and output the pavement performance change △y corresponding to the time stage t p j,t ;

[0046] According to the pavement performance y before the maintenance operation at point j j,0 , obtain the pavement performance at the time of cycle t after the maintenance operation at the point to be predicted

[0047]

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) The present invention integrates more comprehensive information in the time dimension, can accurately predict the impact of maintenance operations on the subsequent performance of the road surface, and improves the control of the quality of maintenance operations.

[0050] (2) The present invention can adapt to various road conditions, relies on easily obtainable road image data to predict the impact on road surface performance, and has low requirements for hardware facilities, low computing power requirements, and low evaluation costs;

[0051] (3) The present invention can track the short-term, medium-term, and long-term performance of the road after maintenance operations, supplementing the prediction dimension of the quality of road surface maintenance operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flowchart of the method of the present invention;

[0053] Figure 2 is a distribution map of maintenance points selected within the test area of the embodiment of the present invention;

[0054] Figure 3 is a distribution map of non-maintained points selected within the test area of the embodiment of the present invention;

[0055] Figure 4 is an undirected connected causal skeleton diagram composed of factors related to road surface performance changes in the embodiment of the present invention;

[0056] Figure 5 is a causal relationship network composed of factors related to road surface performance changes in the embodiment of the present invention;

[0057] Figure 6 is a deep neural network structure diagram for establishing a road surface performance change prediction model in the embodiment of the present invention;

[0058] Figure 7 is a fitting result diagram of the predicted value and the true value at time stage t of the test set points in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives detailed implementation manners and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0060] This embodiment provides a method for predicting the impact of maintenance operations on the subsequent performance of the road surface, as Figure 1As shown in the figure, it includes the following steps:

[0061] S1: Select the points for maintenance and the normal points without maintenance within the test area, and record the maintenance operation information.

[0062] Specifically, select the points for maintenance within the test area o and the adjacent normal points without maintenance, and record the set of points as A. Among them, the maintenance points use the maintenance operation time as the time starting point, and the non-maintenance points use the maintenance operation time of the nearest maintenance point on the same road section as the time starting point.

[0063] Record the maintenance operation information of the maintenance points, and establish the factor set Z of the maintenance operation information of point i in A i ={z1, z2, …, z n}, where the numerical factors in Z i are expressed in numerical form, and the non-numerical factors are factorized and expressed in digital factors.

[0064] In this embodiment, 273 points for maintenance within the test area o are selected, and the point distribution is as Figure 2 shown, and 117 adjacent normal points without maintenance, and the point distribution is as Figure 3 shown, and the set of points is recorded as A. The maintenance operation information of the maintenance points includes maintenance damage type, damage size, reporting time, operation time, treatment method, immediate treatment effect, geographical information, etc.

[0065] In this embodiment,

[0066] Z i

[0067] +{maintenance damage type, damage size, response time, operation time, treatment method, treatment effect, maintenance season}

[0068] Z i In it, the damage size, response time, and operation time are expressed in numerical values, and the maintenance damage type, treatment method, treatment effect, and maintenance season are factorized and expressed by factors 0, 1, 2…

[0069] S2: Track and photograph the road surface within the pre-configured range of the points selected in S1 to obtain road surface images, and divide them into multiple cycles.

[0070] Specifically, starting from the time starting point, use in-vehicle GPS data and visible light cameras to collect information on the road surface within 3 meters of the selected point range for 1 year, and divide 1 year into 4 cycles, each cycle lasting 3 months, and record the road surface images, collection time, and geographical information data.

[0071] S3: Use the recognition algorithm to identify the types and quantities of road surface damages in each road surface image, and record the types and quantities of road surface damages at different collection times for the corresponding points.

[0072] Specifically, perform damage recognition and classification on the collected road surface images, establish a road surface condition tracking and inspection data table for point i in A, and record the maintenance operation situation Z i and the types and quantities of road surface damages identified at different collection times within a 3-meter range of the point. The process of identifying road surface losses specifically belongs to the conventional settings in this field. To avoid obscuring the purpose of this application, it will not be elaborated here.

[0073] S4: Classify and statistically analyze the recognition results of S3, and perform averaging processing on the data in the same cycle to obtain the expected classification damage data for the point at different cycles.

[0074] Specifically, based on the road surface condition tracking and inspection data table, statistically analyze and perform averaging processing on different types of damage data for point i in A within the range of 3 meters in cycle t (t = 1 / 2 / 3 / 4) to obtain the set X of damage statistics for i in cycle t i,t ={x1, x2, x3…x k …}, where x k represents the expected damage quantity of damage k.

[0075] S5: Calculate the road surface performance of the point based on the expected classification damage data, and calculate the change in road surface performance between cycles.

[0076] Quantitatively evaluate the road surface performance status of point i within the range of 3 meters in cycle t. Based on the set X i,t and according to the severity of the damage multiplied by the weight to characterize the road surface performance y i,t , the evaluation method is as follows:

[0077] y i,t = w1x1 + w2x2 + w3x3…

[0078] where x k represents the quantity of damage k, and w k represents the weight of damage k;

[0079] The change in the road surface performance status within the range r of point i from the starting time to cycle t is Δy i,t , where the road surface performance at the starting time is y i,0 , then:

[0080] Δy i,t = y i,t - y i,0

[0081] Let ΔY tLet ΔY be the vector composed of the performance changes of all points in A at cycle t, then:

[0082] ΔY t = [Δy 1,t , Δy 2, , …, Δy i,t , …]

[0083] S6: Based on the point maintenance operation information and pavement performance changes, construct a causal network skeleton for each cycle using the causal inference algorithm.

[0084] Specifically, apply the causal inference algorithm to construct the causal network skeleton, and use the maximum relevance - minimum redundancy feature selection method (mRMR) to find the set of main factor nodes that affect the subsequent pavement performance in the maintenance operation. Considering that the maintenance operation may have different impacts on different stages of the road life cycle, establish a causal network for each cycle t:

[0085] Step S61: Establish a variable data set Z, where the variables in Z correspond to the maintenance operation situation Z i , and the target node of the causal network structure is the pavement performance change Δy at cycle t t , set the candidate set of the parent or child nodes of Δy t S(z) = {}.

[0086] Step S62: Find the nodes in Z that have a direct causal relationship with Δy t based on the mRMR criterion:

[0087] The correlation between Z and the target node Δy t is defined by the average value of all mutual information values between each factor node z i and Δy t . The redundancy of all factors in the set Z is the average value of all mutual information values between factor z i and factor z j . The mRMR criterion is a combination of correlation and redundancy, defined as follows:

[0088]

[0089] Use the incremental search method to search for the factor z t with the greatest node dependence on Δy m , and incorporate z m into S(z).

[0090] Step S63: Remove the nodes in S(z) that have no causal relationship with the pavement performance change Δy t using conditional independence testing.

[0091] Step S64: Repeat S62 and S63, iterate until the factors in S(z) no longer increase, and establish a causal network skeleton based on △y t and S(z), where the relationship between nodes is an undirected connection.

[0092] The causal network skeleton obtained in this embodiment is as Figure 4 shown.

[0093] S7: Based on the causal network skeleton, identify the direction of the causal network through the separation algorithm to obtain a complete causal relationship network.

[0094] Specifically, use the principle of d-separation to determine the dependence direction of the edges in the graph, identify the direction of the causal network, expand the skeleton into a directed acyclic graph, and obtain a complete causal relationship network, as Figure 5 shown.

[0095] S8: Construct and train a pavement performance change prediction model based on a deep neural network.

[0096] The input of the pavement performance change prediction model is the maintenance operation factors directly affecting the pavement performance change selected from the causal relationship network, and the output is the pavement performance change in different periods.

[0097] The deep neural network of the pavement performance change prediction model sets hidden layers, uses the ReLU function as the activation function, and sets a random dropout layer. The structure of the deep neural network is as Figure 6 shown.

[0098] In this embodiment, the cross-validation method is used to randomly extract b% of the road inspection data as the test set, and the remaining data is used as the training set. After setting the initial learning rate and the number of training times, the pavement performance change prediction model is pre-trained.

[0099] After pre-training, use sequential feature selection to forward or backward search for the best feature group from the maintenance operation factors affecting the pavement performance change according to the cross-validation score and estimator, that is, start feature selection from zero features and find a group of features that can maximize the cross-validation score when training a machine learning model for a single feature. Use the selected feature group for model training to obtain an optimized pavement performance change prediction model.

[0100] S9: Predict the pavement performance change in different future periods based on the trained pavement performance change prediction model.

[0101] Specifically, input the optimized selected feature values of the point to be predicted j into the pavement performance change prediction model, and output the pavement performance change △y p j,t ;

[0102] According to the road surface performance y before the maintenance operation at point j j,0 , the road surface performance at the predicted point after the maintenance operation at cycle t is obtained

[0103]

[0104] The fitting results of the predicted values and the true values of the road surface performance changes at the points in the test set of this embodiment are as Figure 7 shown

[0105] Specifically, the methods used in steps S6 - S9 can predict the overall road surface performance. If it is necessary to predict the generation and change of a specific type of damage k, y can be taken i,t = x k , and repeat the modeling and prediction process of steps S6 - S9 to predict the number of damages at the predicted point i at time stage t after the maintenance operation

[0106] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should be within the protection scope determined by the claims

Claims

1. A prediction method for the influence of maintenance operations on the subsequent performance of the road surface, characterized in that, It includes the following steps: S1: Select the points for maintenance and the normal points without maintenance within the test area, and record the maintenance operation information; S2: Track and photograph the road surface within the pre-configured range of the points selected in S1 to obtain road surface images, and divide them into multiple cycles; S3: Use an identification algorithm to identify the types and quantities of road surface damages in each road surface image, and record the types and quantities of road surface damages at the corresponding points at different acquisition times; S4: Conduct classification statistics on the identification results of S3, and perform averaging processing on the data in the same cycle to obtain the expected classification damage data of the points in different cycles; S5: Calculate the road surface performance of the points based on the expected classification damage data, and calculate the change in road surface performance between cycles; S6: Utilize the point maintenance operation information and the change in road surface performance to construct a causal network skeleton for each cycle based on the causal inference algorithm; S7: Based on the causal network skeleton, identify the direction of the causal network through a separation algorithm to obtain a complete causal relationship network; S8: Construct and train a road surface performance change prediction model based on a deep neural network. The input of the model is the maintenance operation factors directly affecting the change in road surface performance selected from the causal relationship network, and the output is the change in road surface performance in different cycles; S9: Predict the change in road surface performance in future different cycles based on the trained road surface performance change prediction model; The specific content of step S5 is: For point positions i During the cycle t , range r Quantitatively evaluate the pavement performance within, based on the point positions i During the cycle t Set of damage statistics , characterize the pavement performance by multiplying the damage severity by the weight y i,t , the evaluation method is as follows: Among them, represents k the number of injuries, represents k the weight of the injury; Point location i Range r The change in the pavement performance condition within the range from the starting point in time to the cycle t is △ y i,t , where the pavement performance at the starting point in time is y i,0 , then: Let △ Y t be A the vector composed of the performance changes of all points in the cycle t , then: Step S6 includes the following steps: Step S61: Establish a variable data set Z , Z The variables corresponding to the maintenance operation conditions Z i , The target node of the causal network structure is the cycle t The pavement performance change △ y t , Set △ y t The candidate set of parent nodes or child nodes ; Step S62: Based on the mRMR criterion, find in Z the nodes that have a direct causal relationship with △ y t where the mRMR criterion is as follows: Search for the factor with the greatest node dependence degree using the incremental search method and △ y t the factor with the greatest node dependence degree , incorporate into ; Step S63: Use conditional independence test to remove nodes that have no causal relationship with the change in pavement performance △ from y t ; Step S64: Repeat S62 and S63, and iterate until the factors in y t no longer increase. According to △ and establish a causal network skeleton, and the relationship between nodes is an undirected connection.

2. The prediction method for the influence of maintenance operations on the pavement's later performance according to claim 1, wherein, The specific content of step S1 is: Select the test area o and the points for maintenance within it and the adjacent normal points without maintenance, and denote the set of points as A , where the maintenance points take the maintenance operation time as the starting time, and the non-maintenance points take the maintenance operation time of the nearest maintenance point on the same road section as the starting time; Record the maintenance operation information of the maintenance points and establish A midpoint i maintenance operation information factor set , where the numerical factors in are expressed in numerical form, and the non-numerical factors are factorized and expressed in digital factors.

3. The prediction method for the influence of maintenance operations on the pavement's later performance according to claim 2, characterized in that, The specific content of step S2 is: Starting from the time origin, using in-vehicle GPS data and visible light cameras to collect information on the road surface within the selected point range r for a period of c and dividing c into n cycles, with each cycle lasting c / n , recording road surface images, collection times, and geographic information data.

4. The prediction method for the influence of maintenance operations on the long-term performance of the road surface according to claim 3, characterized in that, The specific content of step S3 is: Perform damage recognition and classification on the collected road surface images, and establish A midpoint i road surface condition tracking and inspection data sheet to record the maintenance operation situation Z i and the point range r The types and quantities of road surface damages identified at different collection times within the range.

5. The prediction method for the influence of maintenance operations on the long-term performance of the road surface according to claim 4, characterized in that, The specific content of step S4 is: Based on the road surface condition tracking inspection data table, for A the midpoint i in the cycle t , range r statistically analyze different types of damage data within and perform averaging processing to obtain i the set of damage statistics in the cycle t , where represents the expected number of damages of the damage k .

6. The prediction method for the influence of maintenance operations on the later-stage performance of the road surface according to claim 1, wherein, The specific content of step S7 is: Use d Determine the dependency direction of the edges in the graph using the principle of separation, identify the causal network direction, expand the skeleton into a directed acyclic graph, and obtain a complete causal relationship network.

7. A prediction method for the influence of maintenance operations on the long-term performance of the road surface according to claim 1, characterized in that, The deep neural network of the road surface performance change prediction model sets hidden layers, uses the ReLU function as the activation function, and sets a random dropout layer; The road inspection data is randomly selected by b% as the test set using the cross-validation method, and the rest is used as the training set. After setting the initial learning rate and the number of training times, the road surface performance change prediction model is trained.

8. A prediction method for the influence of maintenance operations on the long-term performance of the road surface according to claim 1, characterized in that, The specific content of step S9 is: Input the point to be predicted j The optimized selection feature value is input into the pavement performance change prediction model, and the output time stage t The corresponding pavement performance change △ y p j,t ; According to the point location j The pavement performance before maintenance operations y j,0 , the pavement performance at the predicted point location after the maintenance operation cycle t is obtained : 。

Images