Two-stage project-level maintenance decision-making method based on evidence and demand

Abstract

The invention discloses a two-stage project-level maintenance decision-making method based on evidence and demands. The method comprises the following steps: S1, dividing standard road elements; s2, constructing a successful maintenance case database; s3, establishing an evidence K proximity model; s4, acquiring a set value decision result; s5, auditing, screening and making a decision of the first stage; s6, constructing an optimization model considering subjective maintenance requirements; and S7, carrying out second-stage decision making to determine a final maintenance scheme. According to the method, through two stages of set value decision making and optimization decision making, the review number of the road elements is effectively reduced, the labor efficiency of maintenance experts is improved, meanwhile, precious experience in a successful maintenance scheme library is scientifically utilized by a set value decision making result, 100% rejection decision making is achieved for the road elements with extremely abnormal detection information, and the decision making efficiency is improved. And the maintenance scheme of each element is reliable in evidence and optimal in maintenance demand, so that the decision-making efficiency is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of road maintenance plan decision-making, and particularly relates to a two-stage project-level maintenance decision-making method based on evidence and requirements. Background Art

[0002] Today, with the gradual improvement of the transportation system, road transportation plays an important role in all walks of life. With the action of loads and the environment, roads will be irreversibly damaged, resulting in a decline in transportation efficiency. How to make intelligent decisions on the maintenance plans of each section in the huge road system, improve decision-making benefits and traffic efficiency, and narrow the focus range of maintenance experts on difficult sections has positive significance for the road maintenance decision-making problem.

[0003] In recent years, with the gradual increase in the number of roads and the service time, the demand for road maintenance has become greater and greater, and intelligent maintenance decision-making methods based on computer assistance have gradually received attention. At present, the existing intelligent project-level maintenance decision-making technologies still have the problem that they cannot execute rejection decisions on road elements with anomaly detection information based on the evidence of successful maintenance cases, which leads to the need for experts to still conduct an overall review of the intelligent maintenance decision-making results, so that the benefits of intelligent decision-making technologies in reducing manual labor have not been significantly improved. In addition, the current decision-making technologies make decisions on normal detection road elements either only based on the evidence of successful maintenance cases or only based on the requirements of road maintenance management agencies, which may lead to the failure of the maintenance plans obtained by these technologies.

[0004] Therefore, an intelligent decision-making method that simultaneously considers the evidence of successful maintenance cases and maintenance requirements is crucial. Summary of the Invention

[0005] Based on the above background art and aiming at the deficiencies of the existing technology, the object of the present invention is to propose a two-stage project-level maintenance decision-making method based on evidence and requirements. By establishing an evidence K-nearest neighbor model that can output the mass function of each maintenance plan, "outlier plan", and plan set based on the evidence of successful maintenance cases to evaluate the evidence size of each maintenance plan adopted for all road elements on the road, using the Hods criterion to combine the mass function and the benefit matrix to determine the set-valued decision result of each road element, the road elements in the set P are executed rejection decisions and handed over to experts for review, the road elements not in the set P are used for optimizing model modeling, and using the integer programming algorithm to solve the model to obtain a maintenance plan that considers subjective maintenance requirements and is supported by the evidence of successful maintenance cases. Using the intelligent maintenance decision-making algorithm proposed by the present invention, it is possible to achieve a maintenance decision-making process with reduced expert review workload, reliable and optimal decision-making results.

[0006] Technical Solution: To achieve the above object of the invention, the technical solution adopted by the present invention is as follows:

[0007] A two-stage project-level maintenance decision-making method based on evidence and requirements, comprising the following steps:

[0008] S1. Divide standard road elements:

[0009] Set the standard length of the detection unit, and divide the management section into R road elements according to the standard length;

[0010] S2. Construct a successful maintenance case database:

[0011] Collect the successful maintenance plans of each road element in the management section over the years, and construct a successful maintenance case database composed of maintenance road elements. The maintenance road element samples record the maintenance plan and the detection information corresponding to the road element. Among them, the detection information contains F detection indicators, and the detection indicators of any maintenance road element sample in this database are represented by ;

[0012] S3. Establish an evidence K-nearest neighbor model:

[0013] Obtain the maintenance plans that appear in the successful maintenance case database, and assign a unique label to each plan. Let P q represent the corresponding plan, form a set of successful maintenance plans, add the "outlier plan" P0 to the set of successful maintenance plans to obtain a new set of plans P. Based on the detection information of the r-th road element and the successful maintenance case database, generate the mass function when the state of this road element is any maintenance plan or set of plans according to the following steps:

[0014] S3.1. Determine the value of the number K of the nearest samples of the target road element. Based on the Euclidean distance, obtain a set Φ r of K nearest maintenance road element samples whose detection indicators are similar to x in the successful maintenance history. If the label of the k-th nearest sample in Φ r is q, then the mass functions of this sample supporting the r-th road element on P q and P are respectively:

[0015]

[0016] In the formula, γ q =1 / d q d q is the average value of the Euclidean distances between the detection indicators of every two samples with the plan P q in the successful maintenance history, d r,k is the Euclidean distance between the detection indicators of the sample k and the road element r, e is the natural logarithm base, and α0, β, and are all hyperparameters;

[0017] S3.2. Let denote the set composed of samples with the same scenario label q in the nearest sample set Φ r ;

[0018] According to the DS fusion law, the fusion result of the r-th path element generated by different samples in q on the sets P

[0019]

[0020] In the formula, S k' denotes the k'-th sample with scenario label q in

[0021] S3.3. According to the DS fusion law, for each sample set r in Φ the fusion result of the r-th path element generated on the sets P q and P is as follows:

[0022]

[0023] where is the normalization factor, and the expression is as follows:

[0024]

[0025] In the formula, Q is the number of successful maintenance scenarios in the successful maintenance case database, and q' represents a label different from q.

[0026] S4. Obtain the set-valued decision result:

[0027] Construct the utility matrix Use the evidence K-nearest neighbor model to output the mass function of the r-th path element on each maintenance scenario, "outlier scenario", and scenario set on the road, and combine the mass function with the utility matrix using the Hods criterion to obtain the expected utility E r (A t ), and the set r (A t ) that maximizes the expected utility E is the final set-valued decision result of the r-th path element;

[0028] S5. Review, screen, and make the first-stage decision:

[0029] Analyze the set-valued decision results of all path elements in S4, and execute the rejection decision for the path elements in the set A t where A is a subset of P,t The decision results that are not the path elements of set P are retained until the decision-making in the first stage is completed;

[0030] S6. Construct an optimization model considering subjective maintenance requirements:

[0031] Based on the set-valued decision results in S5, construct an optimization model including an objective function, necessary constraint conditions, and maintenance requirement constraint conditions;

[0032] S7. Make a decision in the second stage to determine the final maintenance plan:

[0033] Use the integer programming algorithm to solve the optimization model. The final values of the decision variables of this model are the final maintenance plan, and the decision-making in the second stage is completed.

[0034] Preferably, in S3, the expression of set P is as follows:

[0035] P = {P0, P1, …, P q , …, P Q}

[0036] In the formula, P1 to P Q represent the successful maintenance plans in the successful maintenance case database, and P0 represents the "outlier plan".

[0037] Preferably, in S3, the mass function of the evidence K-nearest neighbor model on the "outlier plan" satisfies:

[0038] m(P0) ≡ 0

[0039] In the formula, m(P0) is the mass function of the "outlier plan";

[0040] If then let

[0041] Preferably, in S4, the utility matrix is a matrix composed of (2 Q+1 -1) × (Q + 1) real numbers, and it satisfies:

[0042]

[0043] The construction process of the utility matrix is as follows:

[0044] S4.1. Determine the imprecision tolerance γ, and the expression is as follows:

[0045]

[0046] In the formula, |A t|is a non-empty subset A t of the basis; is satisfied and is the weight vector;

[0047] S4.2. Maximize the information entropy to obtain the weight vector The expression is as follows:

[0048]

[0049] The above formula is constrained by

[0050] S4.3. Calculate the utility matrix The specific values of each element in are as follows:

[0051]

[0052] In the formula, represents the utility of the t-th set q when the state is the q-th category P ;

[0053] When the actual plan is the q-th maintenance plan, u (j)q is the j-th largest element in the set {u i,q |P i ∈A t}, u i,q is taken from the identity matrix

[0054] Preferably, in S4, the method for calculating the utility of the set containing the "outlier plan" is the same as that of other sets.

[0055] Preferably, in S4, when obtaining the expected utility E r (A t ), the expression of the Hoz rule combining the mass function and the utility matrix is as follows:

[0056]

[0057] In the formula, v is the parameter to be trained, and v ∈ [0, 1).

[0058] Preferably, the training process of the parameter v is as follows:

[0059] S4.1. Let v = 0, the step size of v increase is 0.01, and let the maximum total utility on the successful maintenance case database be 0;

[0060] S4.2. Obtain the judgment value under the current value of v through the following expression, and enter S4.3;

[0061]

[0062] Wherein, N is the number of samples in the successful maintenance case database;

[0063] S4.3. Compare the obtained judgment value with the maximum total utility;

[0064] If the judgment value is greater than the maximum total utility, then set the maximum total benefit equal to the judgment value, update v, and enter S4.4;

[0065] S4.4. Let v = v + 0.01. If v < 1, then return to S4.2; otherwise, output the current v to complete the training of the parameter v.

[0066] Preferably, the objective function expression of the optimization model is as follows:

[0067]

[0068] Wherein, is the set-valued decision result of the r-th road element, and W is the set composed of weights, x r,q is the decision variable of the r-th road element corresponding to the q-th maintenance plan, w j is the j-th weight, O j is the j-th maintenance requirement;

[0069] When the r-th road element adopts the q-th plan, x r,q is equal to 1, otherwise it is equal to 0.

[0070] Preferably, the necessary constraint conditions of the objective function of the optimization model are:

[0071] x r,q ∈ {0, 1}

[0072]

[0073] Preferably, the maintenance requirement constraint conditions of the optimization model at least include budget constraint;

[0074] Among them, the constraint relationship between the budget constraint and the maintenance cost satisfies:

[0075]

[0076] Wherein, cost q is the cost of the q-th maintenance plan, and budget is the total budget of the maintenance cost.

[0077] Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0078] Through two stages of set-valued decision-making and optimization decision-making, in the set-valued decision-making stage, based on the evidence in the successful maintenance case database, the road elements with abnormal detection information are directly rejected, effectively reducing the number of reviews of road elements and improving the manual efficiency of maintenance experts. At the same time, the set-valued decision-making results also scientifically utilize the valuable experience in the successful maintenance plan library, and the decision-making result that one road element can obtain multiple feasible successful maintenance plans can be achieved; in the optimization decision-making stage, the set-valued decision-making results of the road elements not rejected in the decision-making are used as the value range of decision variables, so as to ensure that the maintenance plan obtained by the optimization decision-making not only meets the subjective needs of the road management agency, but also is supported by the evidence of successful maintenance cases. To sum up, the present invention can achieve a 100% rejection decision for road elements with extremely abnormal detection information, and make the maintenance plans of each road element not only reliable in evidence but also optimal in maintenance requirements, greatly improving the decision-making efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 is the multi-stage project-level maintenance decision-making flow chart;

[0080] Figure 2 is the structural diagram of the successful maintenance history plan library;

[0081] Figure 3 is the maintenance plan set diagram containing "outlier plans";

[0082] Figure 4 is the set-valued decision-making result diagram of road elements with abnormal detection information;

[0083] Figure 5 is the effect diagram of the optimization decision-making. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0084] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples of the Zhejiang Yongjin Expressway.

[0085] As Figures 1-5 shown, a two-stage project-level maintenance decision-making method based on evidence and requirements includes the following steps:

[0086] S1. Divide standard road elements: Set the standard length of the detection unit, and the detection unit is selected from the detection reports of the management section. In this embodiment, the standard length of the detection unit is set to 1 km, and the management section (i.e., the section within the management scope) is divided into R road elements according to the standard length. In this embodiment, R = 179.

[0087] S2. Construct a successful maintenance case database: Collect the successful maintenance plans of the managed sections over the years, and select the corresponding maintenance road element samples to form the Yongjin Expressway successful maintenance case database based on the principle that the specified pavement detection indicators in the "China Highway Technical Condition Assessment Standard" (JTG 5210-2018) have a positive improvement. It includes the maintenance road element samples of each road element. The maintenance road element samples record the maintenance plan and its corresponding detection information. The form of any maintenance road element sample in the successful maintenance history database is "detection information - maintenance plan". The detection information is various detection indicators of the maintenance road element in the maintenance year, and the maintenance plan is the maintenance plan actually adopted in the maintenance year; among them, the detection information contains F detection indicators, and the detection indicators of any maintenance road element sample in this database are used to represent.

[0088] S3. Establish an evidence K-nearest neighbor model: As Figure 3 shown, add the "outlier plan" P0 to the set of maintenance plans obtained by sorting the Yongjin Expressway successful maintenance case database to form a new set of maintenance plans P = {P0, P1,..., P q ,..., P7}, where P q represents the q-th maintenance plan. Based on the detection information of the r-th road element and the successful maintenance case database, generate the mass function when the state of this road element is any maintenance plan or plan set according to the following steps;

[0089] S3.1. Determine the value of the number K of the nearest samples of the target road element. Based on the Euclidean distance, obtain the set Φ r of K nearest maintenance road element samples with detection indicators similar to x in the successful maintenance history. If the label of the k-th nearest sample in Φ r is q, then the mass functions of this sample supporting the r-th road element on P q and P are respectively:

[0090]

[0091] In the formula, γ q = 1 / d q , d q is the average value of the Euclidean distances between the detection indicators of every two samples with the plan of P q in the successful maintenance history, d r,k is the Euclidean distance between the detection indicators of sample k and road element r, e is the natural base, and α0, β, and γ q are all hyperparameters; in this embodiment, α0 and β are respectively set to 0.95 and 1;

[0092] S3.2. Let Denote the nearest neighbor sample set Φ r The set composed of samples with the same scenario label q in

[0093] According to the DS fusion law, The r-th path element generated by different samples in q And the fusion result of the mass function on P is as follows:

[0094]

[0095] In the formula, S k' Denotes The k'-th sample with scenario label q in

[0096] S3.3. According to the DS fusion law, for each sample set in Φ r The r-th path element generated In the set P q And the fusion result of the mass function on P is as follows:

[0097]

[0098] Among them, Is the normalization factor, and the expression is as follows:

[0099]

[0100] In the formula, Q is the number of successful maintenance scenarios in the successful maintenance case database, and q' represents a label different from q.

[0101] The mass function of the evidence K-nearest neighbor model on the "outlier scenario" satisfies:

[0102] m(P0) ≡ 0

[0103] In the formula, m(P0) is the mass function of the "outlier scenario";

[0104] If Then let Since no sample in the successful maintenance history belongs to P0, there is always

[0105] S4. Obtain the set-valued decision result: Construct the utility matrix Use the evidence K-nearest neighbor model to output the mass functions of all path elements on the road for each maintenance scenario, "outlier scenario", and scenario set, and use the Hods criterion to combine the mass function with the utility matrix To obtain the expected utility E r (A t) such that the expected utility E r (A t ) of the largest set is the final set-valued decision result; this result can be supported by the evidence of successful maintenance cases.

[0106] Utility matrix is a matrix composed of (2 Q+1 -1) × (Q + 1) real numbers, which satisfies:

[0107]

[0108] Utility matrix The construction process is as follows:

[0109] S4.1. Determine the imprecision tolerance γ, and the expression is as follows:

[0110]

[0111] In the formula, |A t | is the basis of the non-empty subset A t ; is the weight vector that satisfies and ;

[0112] S4.2. Maximize the information entropy to obtain the weight vector The expression is as follows:

[0113]

[0114] The above formula is subject to

[0115] S4.3. Calculate the specific values of the elements in the utility matrix , and the expression is as follows:

[0116]

[0117] In the formula, represents the utility of the t-th set q when the state is the q-th category P ;

[0118] When the actual solution is the q-th maintenance solution, u (j)q is the j-th largest element in the set {u i,q |P i ∈A t}, and u i,q is taken from the identity matrix

[0119] The utility calculation method for a set containing an "outlier scenario" is the same as that for other set utilities.

[0120] Obtain the expected utility E r (A t ) When, the expression of the Hodes criterion combining the mass function and the utility matrix is as follows:

[0121]

[0122] In the formula, v is the parameter to be trained, and v ∈ [0, 1).

[0123] The training process of the parameter v is as follows:

[0124] S4.1. Let v = 0, the step size of v increase is 0.01, and let the maximum total utility on the successful maintenance case database be 0;

[0125] S4.2. Obtain the judgment value under the current value of v through the following expression, and enter S4.3;

[0126]

[0127] In the formula, N is the number of samples in the successful maintenance case database;

[0128] S4.3. Compare the obtained judgment value with the maximum total utility;

[0129] If the judgment value is greater than the maximum total utility, let the maximum total benefit be equal to the judgment value, update v, and enter S4.4;

[0130] S4.4. Let v = v + 0.01. If v < 1, return to S4.2, otherwise output the current v, and complete the training of the parameter v.

[0131] S5. Review and screen and make the first-stage decision: Analyze the set-valued decision result in S4, and make a rejection decision for the path element of set A t For the path element of set P, remind the decision maker to conduct manual review and determination during the rejection decision. For the set A t The path elements that are not the path elements of set P are retained. After judging all the path elements, complete the first-stage decision; Next, use Figure 4 For example, as Figure 4 Ⅰ shows, randomly select 6 path elements. As Figure 4 Shown by the box annotation in Ⅱ, maliciously modify the detection information of two of the path elements, so that the path element becomes abnormal in detection information compared with the path elements in the successful maintenance case database. As Figure 4As shown in FIGS. III and IV, the set-valued decision result of the method proposed by the present invention for detecting information-abnormal road elements is the entire set P, which means that the present invention can effectively reject decisions for such abnormal road elements.

[0132] S6. Construct an optimization model considering subjective maintenance requirements: Based on the set-valued decision result of S5, specifically the set-valued decision result corresponding to the un-rejected road elements, construct an optimization model including an objective function, necessary constraint conditions, and maintenance requirement constraint conditions.

[0133] The expression of the objective function of the optimization model is as follows:

[0134]

[0135] In the formula, is the set-valued decision result of the r-th road element, and W is the set composed of weights, x r,q is the decision variable of the r-th road element corresponding to the q-th maintenance plan, w j is the j-th weight, and O j is the j-th maintenance requirement;

[0136] When the r-th road element adopts the q-th plan, x r,q is equal to 1, otherwise it is equal to 0.

[0137] The necessary constraint conditions of the objective function of the optimization model are:

[0138] x r,q ∈{0,1}

[0139]

[0140] The maintenance requirement constraint conditions refer to a series of maintenance requirements that the road management department may subjectively need. The maintenance requirement constraint conditions of the optimization model at least include budget constraints;

[0141] Among them, the relationship between the budget constraint and the maintenance cost constraint is satisfied as:

[0142]

[0143] In the formula, cost q is the cost of the q-th maintenance plan, and budget is the total budget of the maintenance cost.

[0144] S7. Make decisions in the second stage to determine the final maintenance plan: Use the integer programming algorithm to solve the optimization model. The final values of the decision variables are the final maintenance plans, completing the decisions in the second stage. The final maintenance plans have two important properties: evidence support from successful maintenance cases and optimized maintenance requirements. The improvement effects of these plans on the pavement condition index (PCI) under the same budget compared with the plans obtained from the model that only uses evidence from successful maintenance cases are as Figure 5 shown. The results indicate that, on the premise of ensuring evidence support, the improvement of PCI by the plan proposed in the present invention can be increased by 13%.

Claims

1. A two-stage project-level maintenance decision-making method based on evidence and requirements, characterized in that It includes the following steps: S1. Divide the standard road elements: Set the standard length of the detection unit, and divide the management section into R road elements according to the standard length; S2. Construct a successful maintenance case database: Collect the successful maintenance plans of the management section over the years, and use the successful maintenance plans of multiple road elements to build a successful maintenance case database, which includes the maintenance road element samples of each road element. The maintenance road element samples record the maintenance plans and their corresponding detection information, where the detection information includes F detection indicators. The detection indicators of any maintenance road element sample in this database are represented by ; S3. Establish an evidence K-nearest neighbor model: Obtain the maintenance plans that appear in the successful maintenance case database, and let Assign a unique label P to each plan q represent the corresponding plan, form a set of successful maintenance plans, add the "outlier plan" P0 to the set of successful maintenance plans to obtain a new plan set P, and based on the detection information of the r-th road element and the successful maintenance case database, generate the mass function when the road element state is any one maintenance plan or plan set according to the following steps; S3.

1. Determine the value of the number K of the nearest samples of the target road element, and obtain the set Φ of K nearest maintenance road element samples with detection indicators similar to x in the successful maintenance history based on the Euclidean distance r , if Φ r the label of the k-th nearest sample in is q, then the corresponding sample supports the r-th road element at P q The mass functions on P and P are respectively:[[]]END]] where γ q = 1 / d q , d q is the average of the Euclidean distances between every two sample detection indices in the successful maintenance history with the plan being P q , d r,k is the Euclidean distance between the detection indices of sample k and road element r, e is the natural base, and a0, β, and are all hyperparameters; S3.

2. Let represent the set composed of samples with the same scenario label q in the nearest sample set Φ r ; According to the DS fusion law, the fusion result of the r-th path element generated by different samples in the set P q and the mass function on P is as follows: where S k' represents the sample with the k'-th scheme label q in S3.

3. According to the DS fusion law, Φ r in each sample set The r-th path element generated in the set P q and the fusion result of the mass function on P is as follows: Among them, is a normalization factor, and the expression is as follows: In the formula, Q is the number of successful maintenance plans in the successful maintenance case database, and q' represents a label different from q; S4. Obtain the set-valued decision result: Construct a utility matrix Use the evidence K-nearest neighbor model to output the mass function of the r-th road element on the road for each maintenance plan, "outlier plan", and the set of plans, and combine the mass function with the utility matrix using the Hods criterion Obtain the expected utility E r (A t ) such that the expected utility E r (A t ) is the largest set is the final set-valued decision result for the r-th road element; S5. Review, screen and make a decision in the first stage: Analyze the set-valued decision results of all path elements in S4 for set A t Execute a rejection decision for the path elements that are in set P for set A t Retain the decision results of the path elements that are not in set P until the decision-making in the first stage is completed; S6. Construct an optimization model considering subjective maintenance requirements: Based on the set-valued decision result of S5, construct an optimization model including an objective function, necessary constraint conditions and maintenance requirement constraint conditions; S7. Make a decision in the second stage to determine the final maintenance plan: Use the integer programming algorithm to solve the optimization model. The final value of the input model decision variable is the final maintenance plan, and the decision in the second stage is completed.

2. The two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, wherein In S3, the expression of the set P is as follows: P = {P0, P1, …, P q , …, P Q} In the formula, P1 to P Q represent the successful maintenance plans in the successful maintenance case database, and P0 represents the "outlier plan".

3. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, characterized in that In S3, the mass function of the evidence K-nearest neighbor model on the "outlier plan" satisfies: m(P0)≡0 In the formula, m(P0) is the mass function of the "outlier plan"; If then let 4. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, characterized in that In S4, the utility matrix is a matrix composed of (2 Q+1 - 1) × (Q + 1) real numbers, which satisfies: Utility matrix The construction process is as follows: S4.

1. Determine the imprecision tolerance γ, and the expression is as follows: where, |A t | is the basis of the non-empty subset A t ; is the weight vector that satisfies and ; S4.

2. Maximize the information entropy to obtain the weight vector The expression is as follows: The above formula is constrained by S4.

3. Calculate the utility matrix The specific values of each element in are as follows: The expression is as follows: In the formula, represents the utility of the t-th set q when the state is the q-th category P ; When the actual solution is the q-th curing solution, u (j)q is the j-th largest element in the set {u i,q | P i ∈ A t}, and u i,q is taken from the identity matrix 5. The two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, wherein In S4, the utility calculation method of the set containing the "outlier plan" is the same as that of other sets.

6. The two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, wherein, In S4, the expected utility E is obtained r (A t ) When, the expression of the Hoz rule combining the mass function and the utility matrix is as follows: In the formula, v is a parameter to be trained, and v∈[0,1).

7. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 6, characterized in that The training process of the parameter v is as follows: S4.

1. Let v = 0, the step size of v increasing is 0.01, and let the maximum total utility on the successful maintenance case database be 0; S4.

2. Obtain the judgment value under the current value of v through the following expression, and enter S4.3; In the formula, N is the number of samples in the successful maintenance case database; S4.

3. Compare the obtained judgment value with the maximum total utility; If the judgment value is greater than the maximum total utility, then let the maximum total benefit be equal to the judgment value, update v, and enter S4.4; S4.

4. Let v = v + 0.

01. If v < 1, then return to S4.2, otherwise output the current v, and complete the training of the parameter v.

8. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 1, characterized in that In S6, the expression of the objective function of the optimization model is as follows: In the formula, is the set-valued decision result of the r-th road element, and W is the set composed of weights, x r,q is the decision variable of the r-th road element corresponding to the q-th maintenance plan, w j is the j-th weight, O j is the j-th maintenance requirement; When the r-th path element adopts the q-th scheme, x r,q is equal to 1, otherwise it is equal to 0.

9. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 8, characterized in that The necessary constraint conditions of the objective function of the optimization model are: x r,q ∈{0,1 10. A two-stage project-level maintenance decision-making method based on evidence and requirements according to claim 8, characterized in that The maintenance requirement constraint conditions of the optimization model include at least the budget constraint; Among them, the constraint relationship between the budget constraint and the maintenance cost satisfies: where cost q is the cost of the q-th maintenance plan, and budget is the total budget for maintenance costs.

Images