A method for predicting highway pavement performance

Abstract

The present invention discloses a highway pavement performance prediction method, comprising the following steps: obtaining a target section data set, the target section data set comprising a pavement influencing factor data set and a pavement performance prediction index data set, and the time corresponding to each pavement influencing factor data in the pavement influencing factor data set and the time corresponding to each pavement performance prediction index data in the pavement performance prediction index data set correspond one to one; performing correlation analysis on the target section data set to obtain a pre-processed target section data set; performing influencing factor feature selection on the pre-processed target section data set using the Lasso method to obtain an influencing feature factor set; performing feature factor grey prediction on the influencing feature factor set using GM (1,1) to obtain a future feature factor set; and processing the future feature factor set using support vector regression SVR to obtain a future pavement prediction result. The present invention can improve the accuracy of highway pavement performance prediction.

Description

Technical Field

[0001] The invention belongs to the field of highway pavement performance prediction, and relates to a prediction method, and specifically to a highway pavement performance prediction method. Background Art

[0002] In the prediction of highway pavement performance at the project level, the traditional method is to use traditional mechanics or mechanical empirical models. It is necessary to analyze the changing laws based on historical data such as stress, strain, and displacement, and establish a relationship model between them and the decay of highway technical condition indicators. However, historical data such as stress, strain, and displacement are often missing. Although the empirical regression prediction model is simple to calculate and operate, the model construction requires a large amount of long-term measured data, and the accuracy is also affected by many factors. Its decay mechanism is difficult to accurately reflect the subjective and objective factors using a specific equation.

[0003] Therefore, based on the actual characteristics of project-level highways, a pavement performance decay model for project-level highways throughout their life cycle is developed and regularly optimized and updated, which is of great significance for the comparison of design schemes for project-level highway maintenance projects and the performance prediction of road network-level highways. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a highway pavement performance prediction method, which solves the problem of poor model prediction accuracy in the prior art.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A method for predicting highway pavement performance comprises the following steps:

[0007] S1. Obtain a target road section data set, where the target road section data set includes a road surface influencing factor data set and a road surface performance prediction index data set, and the time corresponding to each road surface influencing factor data set and the time corresponding to each road surface performance prediction index data set in the road surface performance prediction index data set correspond to each other;

[0008] S2, performing correlation analysis on the target road section data set to obtain a pre-processed target road section data set;

[0009] S3, using the Lasso method to select influencing factor features on the pre-processed target road section data set to obtain an influencing feature factor set;

[0010] S4, use GM (1,1) to perform grey prediction of the influencing characteristic factor set to obtain the future characteristic factor set;

[0011] S5. Use support vector regression SVR to process the future feature factor set to obtain the future road surface prediction result.

[0012] The beneficial effects of the above scheme are:

[0013] (1) The present invention adopts a combined model of feature factor search based on the four-stage adaptive Lasso method of increasing function f(j), grey prediction based on GM(1,1) feature factors, and support vector regression machine SVR based on slime mold algorithm (SMA) to predict project-level highway pavement performance. The model has the advantages of clear and complete theory, high feature selection accuracy, and easy SVR parameter optimization, thus achieving high-precision prediction of project-level highway pavement performance indicators.

[0014] (2) Compared with the traditional Lasso algorithm which has slow convergence and poor accuracy in feature factor selection, the four-stage adaptive Lasso method based on the increasing function f(j) proposed in the present invention adopts different penalty weights for different lag orders, has the advantages of fast parameter adjustment and convergence speed, and improves the accuracy and effectiveness of the model for parameter selection.

[0015] (3) The present invention adopts the slime mold algorithm (SMA) to optimize the support vector regression machine SVR penalty parameter C, the maximum error factor ε of the insensitive loss function and the kernel parameter γ, which has the advantages of complete and simple theory, strong optimization ability and easy implementation.

[0016] Further, step S2 specifically includes:

[0017] The Pearson correlation coefficient method is used to perform correlation analysis on the target road section data set to obtain the preprocessed target road section data set.

[0018] Further, step S3 specifically includes:

[0019] S31, establishing a preset regression model for the pavement influencing factor data and the pavement performance prediction index data in the preprocessing target road section data set;

[0020] S32, using the Lasso method to establish an adaptive Lasso estimation equation for fitting parameters in a preset regression model;

[0021] S33. Solve the adaptive Lasso estimation equation, determine the fitting parameters, and determine the set of influencing characteristic factors.

[0022] Furthermore, in step S31, the preset regression model is:

[0023]

[0024] Wherein, j represents the influencing factor, that is, the order of the preset regression model, j = 1, 2....., n; i represents the time length of the target road segment data set, i = 1, 2....., t, with t as the current time point, i represents the target road segment data set established from a previous time to the current time point t, β j represents the fitting parameter, ε i represents the error term;

[0025] The adaptive Lasso estimation equation is:

[0026]

[0027] Among them, β j represents the fitting parameters; represents the penalty function; t represents a non-negative adjustment parameter; w j represents adaptive weight; represents the fitting parameter β j An estimated value of

[0028] In the fitting parameter β j In the adaptive Lasso estimation equation, the adaptive weight w j The definition equation is:

[0029]

[0030] in, is the initial estimate, which can be obtained by ordinary least squares OLS regression; δ1 is the initial estimate , δ1>0; δ2 is the penalty parameter of order j, δ2>0; f(j) is an increasing function of order j.

[0031] In the adaptive weight w j In the definition equation of , the definition equation of the increasing function f(j) is:

[0032] f(j)=kj

[0033] Among them, k is a non-negative coefficient: Q1, Q2, and Q3 are the lower quartile, median, and upper quartile of sequence j, respectively.

[0034] The beneficial effect of the above further scheme is that different penalty weights are used for different lag orders, which has the advantages of fast parameter adjustment and convergence speed, and improves the accuracy and effectiveness of the model for parameter selection.

[0035] Further, step S4 specifically includes:

[0036] S41, constructing a 1-AGO sequence of each pavement influencing factor data in the influencing feature factor set;

[0037] S42, generating an adjacent mean sequence based on the 1-AGO sequence;

[0038] S43, constructing a grey differential equation GM(1,1) based on the adjacent mean sequence, and obtaining the least squares estimation parameter sequence of the grey differential equation;

[0039] S44, obtaining a 1-AGO estimation sequence based on the least squares estimation parameter sequence;

[0040] S45, perform cumulative reduction on the 1-AGO estimation sequence to obtain the future time series of each influencing characteristic factor,

[0041] S46. Combine the influencing characteristic factor set and the future time series to construct a future characteristic factor set.

[0042] Furthermore, after step S4, the method further includes:

[0043] Step S4 is tested based on the small residual probability P value test method and the variance ratio C value test method.

[0044] The beneficial effects of the above further scheme are: GM(1,1) has high computational efficiency, is suitable for the case of less data in the present invention, and has high prediction accuracy for first-order linear cumulative sequences.

[0045] Furthermore, step S5 specifically includes:

[0046] S51, constructing a training set based on the future feature factor set, wherein the training set includes an input feature vector and a prediction index;

[0047] S52, using a nonlinear transformation relationship to map the input feature vector to a high-dimensional linear space;

[0048] S53, constructing a function expression of support vector regression SVR according to the nonlinear transformation relationship, and solving the function expression;

[0049] S54. Determine the future road surface prediction result based on the function expression and the future characteristic factor set.

[0050] Further, step S53 specifically includes:

[0051] S531, converting the process of solving the function expression into the problem of finding the optimal penalty parameter, the optimal maximum error factor of the insensitive loss function and the optimal kernel parameter;

[0052] S532. Use the slime mold algorithm SMA to determine the optimal penalty parameter, the optimal insensitive loss function maximum error factor and the optimal kernel parameter to solve the function expression.

[0053] Further, step S532 specifically includes:

[0054] S5321, set initial parameters and initialize the population;

[0055] The initial parameters include population size, individual dimension, fitness function, maximum number of iterations, upper boundary of individual dimension and lower boundary of individual dimension;

[0056] S5322, calculating the initial fitness value;

[0057] S5323, calculating weight, first parameter and second parameter;

[0058] S5324, generating random numbers and pseudo-random numbers, determining the size of the random numbers and update parameters, and updating the individual position according to the determination result;

[0059] S5325, calculating the fitness value, and updating the optimization penalty parameter, the optimization insensitive loss function maximum error factor and the optimization kernel parameter;

[0060] S5326. Determine whether the termination conditions are met. If so, output the optimized penalty parameters, the optimized maximum error factor of the insensitive loss function, and the optimized kernel parameters. If not, repeat steps S5323-S5325.

[0061] Furthermore, in step S5321, the fitness function is:

[0062]

[0063] Among them, C, ε, and γ represent the optimal penalty parameter, the optimal maximum error factor of the insensitive loss function, and the optimal kernel parameter, respectively. represents the fitness value of the i-th individual, y g represents the actual value of the training set, It represents the predicted value of the SVR regression of the training set when taking the i-th individual, g represents the sample of the training set, and h represents the number of samples in the data set.

[0064] The beneficial effect of the above further scheme is: using the slime mold algorithm (SMA) to optimize the support vector regression machine SVR penalty parameter C, the insensitive loss function maximum error factor ε and the kernel parameter γ, it has the advantages of complete and simple theory, strong optimization ability and easy implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1A schematic diagram of a process for predicting highway pavement performance. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0067] like Figure 1 As shown, a highway pavement performance prediction method comprises the following steps:

[0068] S1. Obtain a target road section data set, where the target road section data set includes a road surface influencing factor data set and a road surface performance prediction index data set, and the time corresponding to each road surface influencing factor data set and the time corresponding to each road surface performance prediction index data set in the road surface performance prediction index data set correspond one to one.

[0069] In this embodiment, the acquired target road section data set may be a time series.

[0070] Optionally, the method for obtaining the pavement influencing factor data set is to analyze the basic information of the project-level target road section, maintenance history, climate environment, traffic flow, maintenance funds, etc. For example, annual rainfall, annual maximum temperature, annual minimum temperature, annual average temperature, permanent population, general public budget expenditure (transportation) and AADT can be used as pavement influencing factor data.

[0071] Optionally, the pavement performance prediction index dataset may include a pavement damage condition index PCI or a pavement ride quality index RQI.

[0072] S2. Perform correlation analysis on the target road section data set to obtain a preprocessed target road section data set.

[0073] In this embodiment, step S2 specifically includes:

[0074] The Pearson correlation coefficient method is used to perform correlation analysis on the target road section data set. The analysis content includes the correlation analysis between the pavement influencing factor data and the correlation analysis between the pavement influencing factor data and the pavement performance prediction index data, and the preprocessed target road section data set is obtained.

[0075] S3. Use the Lasso method to select influencing factor features on the preprocessed target road section data set to obtain an influencing feature factor set.

[0076] In this embodiment, step S3 specifically includes:

[0077] S31, establishing a preset regression model for the pavement influencing factor data and the pavement performance prediction index data in the preprocessing target road section data set;

[0078] S32, using the Lasso method to establish an adaptive Lasso estimation equation for fitting parameters in a preset regression model;

[0079] S33. Solve the adaptive Lasso estimation equation, determine the fitting parameters, and determine the set of influencing characteristic factors.

[0080] In this embodiment, in step S31, the preset regression model is:

[0081]

[0082] Where i represents the influencing factor, i.e., the order of the preset regression model, j = 1, 2....., n; i represents the time length of the target road segment data set, i = 1, 2....., t, with t as the current time point, i represents the target road segment data set established from a previous time to the current time point t, β j represents the fitting parameter, ε i represents the error term,

[0083] The adaptive Lasso estimation equation is:

[0084]

[0085] Among them, β j represents the fitting parameters; represents the penalty function; t represents a non-negative adjustment parameter; w j represents adaptive weight; represents the fitting parameter β j The estimated value of .

[0086] In this embodiment, the fitting parameter β j In the adaptive Lasso estimation equation, the adaptive weight w j The definition equation is:

[0087]

[0088] in: is the initial estimate, which can be obtained by ordinary least squares OLS regression; δ1 is the initial estimate , δ1>0; δ2 is the penalty parameter of order j, δ2>0; f(j) is an increasing function of order j.

[0089] In the adaptive weight w j In the definition equation of , the definition equation of the increasing function f(j) is:

[0090] f(j)=kj

[0091] Among them, k is a non-negative coefficient: Q1, Q2, and Q3 are the lower quartile, median, and upper quartile of sequence j, respectively.

[0092] Then the fitting parameter β j The non-negative adjustment parameter λ in the adaptive Lasso estimation equation t , adaptive weight w j The initial estimate in the definition equation The penalty parameter δ1 of the order j and the penalty parameter δ2 of the order j are grid searched to obtain the fitting parameters based on the adaptive Lasso estimation.

[0093] Finally, after the above feature selection, the set of influencing feature factors can be determined

[0094] S4. Use GM (1,1) to perform grey prediction of the influencing characteristic factor set to obtain the future characteristic factor set.

[0095] In this embodiment, step S4 specifically includes:

[0096] S41, constructing a 1-AGO sequence of each pavement influencing factor data in the influencing feature factor set;

[0097] S42, generating an adjacent mean sequence based on the 1-AGO sequence;

[0098] S43, constructing a grey differential equation GM(1,1) based on the adjacent mean sequence, and obtaining the least squares estimated parameter sequence of the grey differential equation;

[0099] S44, obtaining a 1-AGO estimation sequence based on the least squares estimation parameter sequence;

[0100] S45. Perform cumulative reduction on the 1-AGO estimation sequence to obtain the future time series of each influencing characteristic factor.

[0101] S46. Combine the influencing characteristic factor set and the future time series to construct a future characteristic factor set.

[0102] For example, the time series of each road surface influencing factor data in the influencing feature factor set is constructed as

[0103] For example, the time series is generated The 1-AGO sequence can be in:

[0104] For example, the adjacent mean sequence generated based on the 1-AGO sequence can be in:

[0105] For example, the grey differential equation constructed based on the adjacent mean sequence can be: And it can make The least squares estimated parameter sequence of the grey differential equation can be obtained as

[0106] Exemplarily, the obtained 1-AGO estimation sequence may be:

[0107] For example, the 1-AGO estimation sequence is subjected to cumulative reduction to obtain a time estimation sequence affecting the characteristic factor v, which can be:

[0108] Optionally, when k ≥ t, the prediction of each influencing characteristic factor v at a certain time point in the future can be achieved, that is, the future time series can be obtained.

[0109] Optionally, after step S4, the method further comprises:

[0110] Step S4 is tested based on the small residual probability P value test method and the variance ratio C value test method.

[0111] Exemplarily, testing step S4 based on a small residual probability P value test method may specifically include:

[0112] First, construct the residual sequence Then calculate the average relative error Then the small residual probability P value is calculated: p=(1-MRE(k))*100%.

[0113] Exemplarily, the variance ratio C value test method for testing step S4 may specifically include:

[0114] First, calculate the time series The mean Secondly, calculate the time series Variance Next, calculate the variance of the residual sequence ε(k) Finally, calculate the variance ratio C value:

[0115] Optionally, the GM(1,1) model evaluation is performed using a combination of a small residual probability P value and a variance ratio C value, and the evaluation results can be graded into: good, qualified, barely qualified, and unqualified. As shown in Table 1, Table 1 is a grading standard for the evaluation results of the GM(1,1) model evaluation using a combination of a small residual probability P value and a variance ratio C value.

[0116] Table 1 Classification criteria for P value and C value evaluation results

[0117] P-value C Value Model Evaluation ＞0.95 ＜0.35 good ＞0.8 ＜0.5 qualified ＞0.7 ＜0.65 Barely qualified ≤0.7 ≥0.65 Failure

[0118] S5. Use support vector regression SVR to process the future feature factor set to obtain the future road surface prediction result.

[0119] In this embodiment, step S5 specifically includes:

[0120] S51, constructing a training set based on the future feature factor set, wherein the training set includes an input feature vector and a prediction index;

[0121] S52, using a nonlinear transformation relationship to map the input feature vector to a high-dimensional linear space;

[0122] S53, constructing a function expression of support vector regression SVR according to the nonlinear transformation relationship, and solving the function expression;

[0123] S54. Determine the future road surface prediction result based on the function expression and the future characteristic factor set.

[0124] In this embodiment, step S53 specifically includes:

[0125] S531, converting the process of solving the function expression into the problem of finding the optimal penalty parameter, the optimal maximum error factor of the insensitive loss function and the optimal kernel parameter;

[0126] S532. Use the slime mold algorithm SMA to determine the optimal penalty parameter, the optimal insensitive loss function maximum error factor and the optimal kernel parameter to solve the function expression.

[0127] In this embodiment, step S532 specifically includes:

[0128] S5321, set initial parameters and initialize the population;

[0129] The initial parameters include population size, individual dimension, fitness function, maximum number of iterations, upper boundary of individual dimension and lower boundary of individual dimension;

[0130] S5322, calculating the initial fitness value;

[0131] S5323, calculating weight, first parameter and second parameter;

[0132] S5324, generating random numbers and pseudo-random numbers, determining the size of the random numbers and update parameters, and updating the individual position according to the determination result;

[0133] S5325, calculating the fitness value, and updating the optimization penalty parameter, the optimization insensitive loss function maximum error factor and the optimization kernel parameter;

[0134] S5326. Determine whether the termination conditions are met. If so, output the optimized penalty parameters, the optimized maximum error factor of the insensitive loss function, and the optimized kernel parameters. If not, repeat steps S5323-S5325.

[0135] In this embodiment, in step S5321, the fitness function is:

[0136]

[0137] Among them, C, ε, and γ represent the optimal penalty parameter, the optimal maximum error factor of the insensitive loss function, and the optimal kernel parameter, respectively. represents the fitness value of the i-th individual, y g represents the actual value of the training set, It represents the predicted value of the SVR regression of the training set when taking the i-th individual, g represents the sample of the training set, and h represents the number of samples in the data set.

[0138] Exemplarily, the training set constructed based on the future feature factor set may be T={(x1, y1), (x2, y2), …, (x h ,y h )}, where h is the number of samples in the training set, h <t,x i (x i ∈R q , i=1,2,…,h) is the input feature vector of the i-th sample, x i =[x i1 ,x i2 …x iq ] T , output value y i is the prediction index, y i ∈R.

[0139] For example, a nonlinear transformation is used Input the training set into the feature vector x i ∈R q Mapped to a high-dimensional linear space, the constructor expression is: Among them, ω is the weight and b is the bias term.

[0140] Adopt ε-insensitive loss function and consider slack variables ξ, ξ * , transform the process of solving f(x) into solving the following optimization problem:

[0141]

[0142]

[0143]

[0144] ξ,ξ * ≥0

[0145] Among them, C is the penalty parameter and ε is the maximum error factor of the insensitive loss function.

[0146] For example, the Lagrange multiplier method is adopted, and the Lagrange multiplier α is introduced i , Through the saddle point condition we get:

[0147]

[0148]

[0149]

[0150] in: is the defined kernel function.

[0151] Thus, we can get

[0152] For example, a radial basis function (RBF) is used as the kernel function k(x i ,x), its expression is: k RBF (x i ,x)=exp(-γ||x i -x|| 2 ), where: kernel parameters

[0153] Exemplarily, using the slime mold algorithm SMA to determine the optimal penalty parameter, the optimal insensitive loss function maximum error factor, and the optimal kernel parameter may specifically include:

[0154] Set parameters (population size pop, individual dimension dim, fitness function fitn, maximum number of iterations M, upper boundary of individual dimension maxx j , the lower boundary of the individual dimension minx j ), initialize the population X pop×dim .

[0155] Calculate the initial fitness value fitn i , where i is the individual, i≤pop, and j is the individual dimension, j≤dim.

[0156] In this embodiment, minx C =1,maxx C =200, minx ε =0.1,maxx ε =2, minx γ= 0.01, maxx γ = 20, M = 200. For the penalty parameter C, the maximum error factor ε of the insensitive loss function, and the kernel parameter γ in the optimization support vector regression machine SVR, dim = 3.

[0157] Exemplarily, the fitness function can be:

[0158]

[0159] Wherein, the meaning on the right side of the formula is to take the mean square error of the actual value and the predicted value y of the dataset SVR regression when taking the i-th individual (c, ε, and γ), g represents the samples in the training set, and h represents the number of samples in the training set. i 、 The mean square error of, g represents the samples in the training set, and h represents the number of samples in the training set.

[0160] Exemplarily, the first parameter can be a, the second parameter can be b, the weight can be W, and

[0161]

[0162] a = arctanh(1 - (m / M))

[0163] b = 1 - m / M

[0164] Where m is the number of iterations, m ≤ M, BF(m) is the optimal fitness of the population at the m-th iteration, WF(m) is the worst fitness of the population at the m-th iteration, F i (m) is the fitness value of individual i at the m-th iteration; rand([1, dim]) is a 1-row and dim-column random number vector.

[0165] Exemplarily, the random number can be r, the pseudo-random number can be rand, and the update parameter can be z.

[0166] In this embodiment, z = 0.03, generate the random numbers r and rand, and the judgment of the magnitude relationship between r and the position update parameter z can be:

[0167] If r < z, then according to X ij (m + 1) = rand · (maxx j - minx j ) + minx j Update the individual position; otherwise, update the parameters p, va, vb:

[0168] p = tanh(|F m (i)| - BF m )

[0169] va = 2a · rand([1, dim]) - a

[0170] vb = 2b·rand([1, dim]) - b

[0171] Optionally, if r < p, then according to X ij (m + 1) = X BFj (m) + va·(W ij (m)·X Aj (m) - X Bj (m)) to update the individual position, where A and B are two randomly selected individuals, and X BFj (m) is the position of the optimal fitness individual at the m-th iteration.

[0172] Optionally, if r ≥ p, then according to X ij (m + 1) = vb·X ij (m) to update the individual position.

[0173] And, the fitness value can be calculated, the global optimal solution can be updated, and it can be determined whether the end condition is satisfied.

[0174] If it is satisfied, then the optimized penalty parameter, the maximum error factor of the optimized insensitive loss function, and the optimized kernel parameter are output. If it is not satisfied, then steps S5323 - S5325 are repeated.

[0175] Optionally, the end condition can be reaching the maximum number of iterations M.

[0176] In this embodiment, the slime mold algorithm (SMA) is used to optimize the support vector regression machine SVR penalty parameter C, the maximum error factor ε of the insensitive loss function, and the kernel parameter γ, which has the advantages of complete and simple theory, strong optimization ability, and easy implementation.

[0177] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations without departing from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the invention.

Claims

1. A method for predicting highway pavement performance, characterized in that: The method comprises: S1. Acquire a target road section data set, wherein the target road section data set includes a road surface influencing factor data set and a road surface performance prediction index data set, and the time corresponding to each road surface influencing factor data set and the time corresponding to each road surface performance prediction index data set in the road surface performance prediction index data set correspond to each other; S2, performing correlation analysis on the target road section data set to obtain a pre-processed target road section data set; S3, using the Lasso method to perform influencing factor feature selection on the pre-processed target road section data set to obtain an influencing feature factor set; The step S3 specifically includes: S31, establishing a preset regression model for the pavement influencing factor data and the pavement performance prediction index data in the pre-processed target road section data set; S32, using the Lasso method to establish an adaptive Lasso estimation equation for the fitting parameters in the preset regression model; S33, solving the adaptive Lasso estimation equation, determining the fitting parameters, and determining the set of influencing characteristic factors; In step S31, the preset regression model is: in, Represents the influencing factors, which is the order of the preset regression model. ; Indicates the time length of the target road segment dataset. ,by For the current time point, Represents the target road segment dataset established from a previous time to the current time point t, represents the fitting parameters, represents the error term; The adaptive Lasso estimation equation is: in, represents the fitting parameters; represents the penalty function; represents a non-negative adjustment parameter; represents adaptive weight; Represents the fitting parameters An estimated value of In the adaptive Lasso estimation equation, the adaptive weight The definition equation is: in, is the initial estimate, which can be obtained by ordinary least squares OLS regression; is the initial estimate The penalty parameter, ; is the penalty parameter of order j, ; is an increasing function of order j; In the adaptive weight In the definition equation, the increasing function The definition equation is: Among them, k is a non-negative coefficient: , Q1, Q2, and Q3 are the lower quartile, median, and upper quartile of sequence j, respectively; S4, using GM (1,1) to perform characteristic factor grey prediction on the influencing characteristic factor set to obtain a future characteristic factor set; S5. Use support vector regression SVR to process the future feature factor set to obtain a future road surface prediction result.

2. The method according to claim 1, characterized in that The step S2 specifically includes: The Pearson correlation coefficient method is used to perform correlation analysis on the target road section data set to obtain the pre-processed target road section data set.

3. The method according to claim 1, characterized in that The step S4 specifically includes: S41, constructing a 1-AGO sequence of each road surface influencing factor data in the influencing feature factor set; S42, generating an adjacent mean sequence based on the 1-AGO sequence; S43, constructing a grey differential equation GM(1,1) based on the adjacent mean sequence, and obtaining a least squares estimation parameter sequence of the grey differential equation; S44, obtaining a 1-AGO estimation sequence based on the least squares estimation parameter sequence; S45, performing cumulative reduction reduction on the 1-AGO estimation sequence to obtain the future time series of each of the influencing characteristic factors, S46: Combining the influencing characteristic factor set and the future time series to construct a future characteristic factor set.

4. The method according to claim 1, characterized in that After step S4, the method further includes: The step S4 is tested based on the small residual probability P value test method and the variance ratio C value test method.

5. The method according to claim 1, characterized in that The step S5 specifically includes: S51, constructing a training set based on the future feature factor set, wherein the training set includes an input feature vector and a prediction index; S52, mapping the input feature vector to a high-dimensional linear space using a nonlinear transformation relationship; S53, constructing a function expression of support vector regression SVR according to the nonlinear transformation relationship, and solving the function expression; S54: Determine the future road surface prediction result according to the function expression and the future characteristic factor set.

6. The method according to claim 5, characterized in that The step S53 specifically includes: S531, converting the process of solving the function expression into a problem of finding an optimal penalty parameter, an optimal maximum error factor of an insensitive loss function, and an optimal kernel parameter; S532: Use a slime mold algorithm (SMA) to determine the optimization penalty parameter, the maximum error factor of the optimization insensitive loss function, and the optimization kernel parameter to solve the function expression.

7. The method according to claim 6, characterized in that The step S532 specifically includes: S5321, set initial parameters and initialize the population; The initial parameters include population size, individual dimension, fitness function, maximum number of iterations, upper boundary of individual dimension and lower boundary of individual dimension; S5322, calculating the initial fitness value; S5323, calculating weight, first parameter and second parameter; S5324, generating random numbers and pseudo-random numbers, determining the size of the random numbers and update parameters, and updating the individual position according to the determination result; S5325, calculating the fitness value, and updating the optimization penalty parameter, the maximum error factor of the optimization insensitive loss function and the optimization kernel parameter; S5326. Determine whether the end condition is met. If so, output the optimization penalty parameter, the optimization insensitive loss function maximum error factor and the optimization kernel parameter. If not, repeat steps S5323-S5325.

8. The method according to claim 7, characterized in that In step S5321, the fitness function is: in, They represent the optimal penalty parameter, the optimal maximum error factor of the insensitive loss function and the optimal kernel parameter respectively. Indicates The fitness value of each individual, represents the actual value of the training set, Indicates the first The predicted value of SVR regression of the training set when there are individuals, represents the samples of the training set, and h represents the number of samples in the data set.

Images