An asphalt pavement maintenance evaluation method based on a Bayesian sparse trust network

Abstract

The present application relates to the technical field of asphalt pavement maintenance evaluation, and particularly relates to an asphalt pavement maintenance evaluation method based on a Bayesian sparse trust network, comprising: taking a Page Rank dynamic trust relationship as a framework, constructing a trust network among indexes through initial trust, secondary trust and a decay factor, and realizing dynamic evolution of evaluation weights; introducing a Bayesian inference theory, probabilistically modeling the decay factor, secondary trust and noise variance, and using an MH algorithm for posterior sampling to quantize uncertainty of evaluation results; applying a regularization sparse constraint to suppress interference of weakly correlated indexes and focus on core indexes such as rut depth and crack rate. The three are organically coordinated to form an integrated evaluation framework of "dynamic weight evolution-uncertainty quantization-core index focusing", and finally, aiming at the engineering characteristics of multiple indexes, strong coupling and uncertain information of asphalt pavement preventive maintenance evaluation, the maintenance evaluation highly matched with the actual technical condition is realized.

Description

Technical Field

[0001] This invention relates to the field of asphalt pavement maintenance evaluation technology, and in particular to an asphalt pavement maintenance evaluation method based on Bayesian sparse trust networks. Background Technology

[0002] As of July 2025, my country's total highway mileage reached 5.4904 million kilometers, continuing to rank first in the world. High-grade highways primarily use asphalt pavement. The total length of highways under maintenance reached 5.2516 million kilometers, accounting for 99.4% of the total highway mileage. Highway maintenance has achieved full network coverage and routine management. Preventive maintenance of asphalt pavements, as a core technical means to delay pavement performance degradation, extend service life, and reduce life-cycle maintenance costs, has become a core component of high-grade highway maintenance management in my country. Industry engineering practice and academic research have both confirmed that scientific preventive maintenance can extend the service life of asphalt pavements by 3-5 years and reduce life-cycle maintenance costs by more than 30%. Accurate and reliable pavement preventive maintenance evaluation is the prerequisite and core basis for formulating maintenance strategies, optimizing resource allocation, and maximizing maintenance benefits.

[0003] During long-term service, the performance degradation of asphalt pavements is a complex nonlinear process involving the dynamic coupling of multiple factors such as pavement structural characteristics, traffic load, natural climate, and maintenance history. This determines that the preventive maintenance evaluation of asphalt pavements is essentially a multi-dimensional, strongly coupled, and highly uncertain multi-attribute decision-making problem. Around this core engineering need, scholars and engineers both domestically and internationally have conducted extensive research, and the development of related evaluation technologies has mainly gone through three stages:

[0004] The first stage is the single-index standard evaluation stage. This stage is based on the highway industry standards of various countries and mainly uses single pavement performance indicators such as pavement damage index, ride quality index, rut depth index, and skid resistance index as the core of evaluation, directly classifying maintenance levels according to the standard thresholds. This type of method is simple to operate and has strong engineering versatility, but it can only reflect the current apparent performance state of the pavement. It cannot comprehensively consider the impact of multiple dimensions such as structural characteristics, environmental loads, and maintenance economy on the long-term performance evolution of the pavement. The evaluation dimensions are fundamentally one-sided and it is difficult to support the medium- and long-term decision-making needs of preventive maintenance.

[0005] The second stage is the multi-attribute comprehensive evaluation stage. Addressing the limitations of single-index evaluation, this stage introduces multi-attribute decision-making methods such as the Analytic Hierarchy Process (AHP), Network Hierarchy Process (NHP), Entropy Weight Method, Matter-Element Extension Theory, Cloud Model, and Evidence Theory. This constructs a multi-dimensional evaluation system covering pavement performance, technical feasibility, and economic rationality, achieving a technological leap from "single-index judgment" to "multi-dimensional comprehensive evaluation." However, these methods still have significant engineering adaptability defects: AHP and NHP rely on expert experience to determine index weights, resulting in strong subjectivity and susceptibility to individual cognitive biases, leading to poor reproducibility of evaluation results; Entropy Weight Method can only reflect the dispersion of index data and cannot reflect the actual engineering importance of the indicators, easily leading to a disconnect between "data-driven" and "engineering experience"; While Extension Cloud Theory can simultaneously handle the fuzziness and randomness of evaluation information, it cannot quantify the uncertainty of evaluation results and cannot provide a risk dimension reference for maintenance decisions.

[0006] The third stage is the intelligent algorithm optimization and evaluation stage. With the development of artificial intelligence technology, machine learning algorithms such as neural networks, support vector machines, and random forests have been introduced into the field of asphalt pavement maintenance evaluation. Through data training, the evaluation model can be adaptively optimized, which reduces the subjectivity of weight determination to a certain extent. However, such methods have extremely high requirements for the quantity and quality of sample data. In highway engineering, complete test samples of the entire pavement life cycle are often difficult to obtain, and the model is prone to overfitting and insufficient generalization ability. At the same time, the "black box characteristic" of machine learning models leads to a lack of interpretability of evaluation results, making it impossible to clarify the contribution of each indicator to the evaluation results. This makes it difficult for front-line maintenance personnel to understand and apply the results, thus limiting the practicality of the project.

[0007] Despite the significant achievements in existing research, current evaluation methods for preventive maintenance of asphalt pavements still face numerous unresolved bottlenecks in practical engineering applications. These methods fail to adapt to the core scenarios encountered in engineering practice, such as dynamic interactions among multiple indicators, ambiguity and randomness in evaluation information, and significant differences in correlations between indicators. Specifically, these bottlenecks manifest in the following four aspects:

[0008] First, the evaluation system is one-sided, lacking key influencing factors. Existing methods mostly focus on pavement performance, technical and economic dimensions, failing to incorporate climate conditions—a core factor in the formation and performance degradation of asphalt pavement defects—into the evaluation system. my country has a vast territory, with significant differences in extreme temperatures, rainfall, and freeze-thaw cycles across different regions. The long-term effects of temperature and moisture are the core causes of rutting, cracking, potholes, and other defects in asphalt pavements. Ignoring the influence of climate conditions will directly lead to evaluation results that do not match the actual evolution of pavement performance, making it impossible to achieve "site-specific and precise" preventative maintenance.

[0009] Second, the fixed weighting model fails to reflect the dynamic interactions between indicators. Most existing methods employ a fixed weighting model, meaning that once the indicator weights are determined during the evaluation process, they remain unchanged. This fails to reflect the dynamic interactions and evolving importance of indicators across different road sections and service environments. Furthermore, the evaluation indicators for asphalt pavement are not independent but exhibit complex nonlinear coupling relationships. For example, rutting depth is directly related to traffic volume and high-temperature conditions, while crack rate is highly correlated with freeze-thaw cycles and rainfall. The fixed weighting model cannot adapt to these dynamic interactions between indicators, leading to a disconnect between weight allocation and actual engineering conditions, ultimately resulting in distorted evaluation results.

[0010] Third, there is insufficient capacity for processing uncertain information and a lack of quantitative characterization methods. The entire process of asphalt pavement maintenance evaluation involves multiple uncertainties, including random errors in test data, fuzziness in index quantification, subjective uncertainties in expert experience, and random fluctuations in pavement performance evolution. Existing methods either only handle a single type of uncertainty or completely ignore its impact, failing to systematically model the multiple uncertainties in the evaluation process, let alone quantify the degree of uncertainty's influence on the final evaluation results. This leads to unverifiable reliability of the evaluation results and even misjudgments where the evaluation level does not match the actual pavement condition, posing potential risks to maintenance decisions.

[0011] Fourth, the evaluation results lack risk warning capabilities and offer limited decision support. Existing methods only output a single evaluation level or comprehensive score, merely answering the question of "what is the current condition of the road surface?" They fail to address crucial decision-making questions such as "how reliable are the evaluation results?", "which road sections are at risk of inaccurate evaluations?", and "which indicators are the core control factors affecting road performance?" For high-risk road sections with high traffic volume and severe weather conditions, the evaluation results cannot identify potential risks or provide data support for subsequent supplementary inspections and key maintenance, making it difficult to meet the refined, intelligent, and risk-controllable management needs of modern highway maintenance.

[0012] To address the aforementioned issues, this application proposes a method for evaluating asphalt pavement maintenance based on Bayesian sparse trust networks. Summary of the Invention

[0013] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for evaluating asphalt pavement maintenance based on a Bayesian sparse trust network. Its core essence lies in "Page Rank dynamic trust network construction - Bayesian probability modeling and uncertainty quantification - The three-level linkage logic of "regularized sparse constraints" enables high-precision and high-stability evaluation of the preventive maintenance status of asphalt pavement.

[0014] To achieve the above objectives, the present invention adopts the following technical solution: an asphalt pavement maintenance evaluation method based on Bayesian sparse trust networks, comprising the following steps:

[0015] Step 1: Construction of Evaluation Index System: Following the principles of systematicness, scientificity, and practicality, construct an evaluation index system for preventive maintenance of asphalt pavement covering multiple dimensions, and clarify the definition, nature, and data source of each index;

[0016] Step 2, Data Preprocessing: Quantify and classify the indicator data, positiveize negative indicators, and normalize intervals to achieve dimensionless and homogeneous indicators, providing standardized data input for modeling;

[0017] Step 3: Dynamic Trust Network Construction: Quantify initial trust, secondary trust, and decay factor, construct a dynamic trust relationship network among indicators, and calculate the initial trust level and indicator weights.

[0018] Step 4, Bayesian probabilistic modeling and Sparsity constraints: Prior distributions are defined for the core parameters of the model; an observation model and a joint posterior distribution are constructed; and sparsity constraints are introduced. Regularized sparse constraints suppress interference from weakly correlated indices;

[0019] Step 5, MCMC posterior inference: The Metropolis-Hastings (MH) algorithm is used to perform posterior parameter sampling, and the posterior estimation results of the index weights are calculated based on the effective sampled values;

[0020] Step 6, Multi-round interactive convergence mechanism: Through information updates and consensus level checks, the dynamic evolution and convergence of trust relationships are achieved, ensuring the stability and uniqueness of the evaluation results;

[0021] Step 7: Evaluation Result Calculation and Output: Calculate the comprehensive evaluation value, grade probability distribution and uncertainty probability of the road segment based on the optimal posterior weight, and output the final maintenance evaluation result.

[0022] Preferably, in step 1, the specific method is as follows:

[0023] A preventive maintenance evaluation index system for asphalt pavements was constructed, comprising 12 indicators across 5 core dimensions: pavement structure, pavement performance, pavement use, climate conditions, and maintenance benefits. The composition and significance of the indicators for each dimension are as follows:

[0024] Road surface structure dimension: including two indicators, road grade and construction difficulty, characterizes the inherent structural characteristics of asphalt pavement and determines the basic bearing capacity and maintenance difficulty of the pavement;

[0025] Road surface performance dimensions include four indicators: rut depth index, crack rate, pothole rate, and skid resistance index. These are the core dimensions for evaluating the performance status of asphalt pavement, directly reflecting the structural integrity and safety of the pavement, and serving as the core basis for preventive maintenance decisions.

[0026] Road surface usage dimension: including two indicators, road age and traffic volume, which characterize the usage intensity and service life of asphalt pavement and are important factors affecting the degradation of pavement performance;

[0027] Climate conditions dimension: including three indicators: maximum temperature, minimum temperature, and rainfall, which characterize the natural environment in which the road surface is located. The long-term effects of temperature and rainfall accelerate the degradation of road surface performance and are important environmental factors that need to be considered in preventive maintenance.

[0028] Maintenance benefit dimension: This is the cost-benefit coefficient, which characterizes the input-output ratio of preventive maintenance of asphalt pavement and is a core indicator for evaluating the rationality and economy of maintenance measures.

[0029] Preferably, in step 2, the specific method is as follows:

[0030] Because different indicators differ in their dimensions, orders of magnitude, and attributes (positive / negative), a three-step data preprocessing method is required to ensure that the indicators can be directly used in evaluation calculations:

[0031] Indicator Quantification and Grading: According to the "Technical Specification for Maintenance of Asphalt Pavement of Highway" (JTG 5142-2019), qualitative indicators (such as road grade and construction difficulty) are quantified and graded, and quantitative indicators (such as crack rate and traffic volume) are standardized and converted. All indicators are quantified into positive indicators (the larger the indicator value, the better the pavement condition or the higher the maintenance benefit).

[0032] Transforming Negative Indicators to Positive Values: For negative indicators such as crack rate, pothole rate, highest temperature, lowest temperature, and rainfall (the higher the value, the worse the road surface condition), an extreme value method is used for positive transformation. The formula is as follows:

[0033] In the formula, This represents the positive value of the indicator; The original value of the indicator; , These are the minimum and maximum values ​​of the original index, respectively. Note: The minimum temperature does not need to be processed by taking the absolute value. The lower (more negative) its original value, the greater the damage to the road surface caused by freeze-thaw cycles. It is a negative index and can be directly converted to a positive value using the formula above.

[0034] Normalization transformation: To adapt to the evaluation requirements of the Bayesian sparse dynamic trust network model, the quantified index values ​​are normalized to... The normalization formula is as follows: The evaluation values ​​of each indicator are within a reasonable range to ensure that the evaluation level of all road sections is consistent with the actual engineering results.

[0035] In the formula, For the first Road section, No. The normalized value of the indicator; For the first Road section, No. The quantified value of the indicator; , The first Minimum and maximum values ​​of the quantified indicator; This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0036] Preferably, in step 3, the specific method is as follows:

[0037] Using PageRank dynamic trust relationships as the core framework, this approach quantifies three core parameters and constructs a dynamic trust network among indicators, achieving a breakthrough from "fixed weights" to "dynamic weights."

[0038] Step 3.1: Definition and Calculation of Core Parameters

[0039] Initial trust is recorded as This characterizes the prior trust level of the evaluator towards each evaluation indicator in a non-interactive state, reflecting the inherent evaluation value of the indicator. Equal-weight initialization is used to ensure unbiasedness. The formula is:

[0040] In the formula, For the first The initial trust value for each indicator; The total number of evaluation indicators in this invention ; For indexing indicators;

[0041] The attenuation factor is denoted as Characteristic indicators To the indicator The attenuation level when trust is transferred, with a value range of [0,1], is calculated using the cosine similarity of the evaluation information vectors between indicators, and the formula is:

[0042] In the formula, As an indicator With indicators Trust decay factor; , Indicators , The evaluation information vector; Let cosine similarity be the similarity between the two vectors. This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0043] Secondary trust is recorded as Characteristic indicators With indicators The trust increment generated after information exchange reflects the dynamic interaction between indicators, and the formula is:

[0044] In the formula, As an indicator For indicators The secondary trust value is positive, indicating increased trust, and negative, indicating decreased trust. For interactive rounds, the initial round ; For road segment indexing; For the first In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; For: the In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; , For: respectively the first wheel, In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment;

[0045] Step 3.2: Building a Trust Relationship Network

[0046] The network nodes are composed of 12 evaluation indicators, and the edge weights between nodes are composed of "initial trust allocation + secondary trust increment".

[0047] Initial Trust Assignment In the absence of interaction, the metrics For indicators The initial trust value is calculated using the following formula:

[0048] In the formula, As an indicator For indicators The initial trust assignment value; As an indicator The initial trust value; The total number of evaluation indicators; This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0049] Overall trust level :index For indicators The overall trust strength, used as the edge weight of the trust network, is expressed by the following formula:

[0050] In the formula, As an indicator For indicators The initial trust assignment value; As an indicator For indicators Secondary trust value;

[0051] Step 3.3: Initial Trust Level and Weight Calculation

[0052] In the formula, For the first Trust levels for each indicator; For the first The decay factor of each indicator relative to the global trust mean. , For the first Global Trust Mean Vector of Each Indicator , The global trust mean of all indicators , The total trust matrix is ​​the first... List; As an indicator The initial trust value; As an indicator With indicators The attenuation factor; As an indicator For indicators Overall trust level; For indexing indicators;

[0053] Initial weights The normalized results of the trust levels of each indicator reflect the contribution of the indicator in the evaluation. The formula is:

[0054] In the formula, For the first The initial weights of each indicator, ; For the first Trust levels for each indicator; For the first Trust levels for each indicator; For indexing indicators.

[0055] Preferably, in step 4, the specific method is as follows:

[0056] Based on Bayesian inference theory, probabilistic modeling is performed on the core parameters of the model, and... Regularized sparse constraints are used to construct a joint posterior distribution, providing a mathematical foundation for subsequent parameter estimation.

[0057] Step 4.1: Setting the Prior Distribution of Core Parameters

[0058] Based on the physical characteristics of the parameters and engineering experience, a suitable probability distribution is selected, following the principles of "prioritizing engineering experience, using prior information without a priori assistance, and adapting to sparse constraints":

[0059] Attenuation factor The value range is [0,1]. Modeling is performed using a Beta distribution, and hyperparameters are set according to the correlation between indicators.

[0060] In the formula, It follows a Beta distribution. , Hyperparameters that determine the shape of the distribution; strong correlation: cosine similarity between indices. Beta distribution mean Correlation: Cosine similarity between indicators mean Weak correlation: cosine similarity between indicators mean ;

[0061] Secondary Trust To achieve Regularized sparsity constraints are modeled using a Laplace distribution, and the formula is as follows:

[0062] In the formula, It follows a Laplace distribution; This is the regularization strength parameter; As an indicator For indicators Secondary trust value;

[0063] Observation noise variance To characterize the degree of random fluctuation in the observed data, inverse Gamma weak information prior modeling is used.

[0064] In the formula, It follows an inverse Gamma distribution; , As hyperparameters, weak information priors are used to avoid dominating posterior inferences;

[0065] Regularization strength We use Gamma weak information prior modeling to support adaptive learning. The formula is:

[0066] In the formula, It follows a Gamma distribution; , For hyperparameters, weak information priors;

[0067] Step 4.2, Observation Model Construction

[0068] Construct a linear observation equation, connecting the model parameters with the observation data, using the following formula:

[0069]

[0070] In the formula, Observations for secondary trust; The true value of secondary trust; The observation noise is independent and identically distributed, following a mean of 0 and a variance of . The normal distribution, i.e. ,and ;

[0071] Step 4.3: Construction of Joint Posterior Distribution

[0072] According to Bayes' theorem, a joint posterior distribution is constructed by fusing the prior distribution of all parameters with the likelihood function of the observation model:

[0073] In the formula: Let be the prior distribution of the attenuation factor, and be the product of the prior values ​​of each index with respect to the attenuation factor Beta. Let be the prior distribution of secondary trust, and be the product of each index with respect to the Laplace prior of secondary trust. For the prior distribution of regularization intensity, we adopt Weak information prior; For the prior distribution of noise variance, we use Weak information prior; Let be the likelihood function of the observed data, and be the product of the secondary trust normal likelihoods of each index. For the observation data matrix; Proportional to, ignoring marginal likelihoods independent of parameters. ;

[0074] Step 4.4, Definition of Uncertainty Probability

[0075] The formula for quantifying the risk level of the evaluation results is:

[0076] In the formula, The probability of uncertainty in the evaluation results of a road segment is as close to 0 as the stability of the evaluation results. This represents the maximum posterior probability of the road segment's evaluation level.

[0077] Preferably, in step 5, the specific method is as follows:

[0078] The Metropolis-Hastings (MH) algorithm is used for posterior sampling of MCMC to obtain effective sampled values ​​of parameters, calculate the posterior estimation results of index weights, and set the combustion period. Valid sample count :

[0079] Step 5.1, Parameter Initialization

[0080] Set initial values ​​for each parameter to provide a starting point for sampling:

[0081] Attenuation factor matrix The secondary trust matrix is ​​calculated from the initial cosine similarity between indicators. Initialize to a 0 matrix, regularization strength noise variance ;

[0082] Step 5.2, MH sampling by parameter

[0083] For each parameter, candidate values ​​are generated sequentially, the acceptance probability is calculated, and the parameter is updated according to random rules:

[0084] Attenuation factor sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability:

[0085]

[0086] In the formula, represents the acceptance probability of candidate values ​​for the attenuation factor, with a value range of [0,1]. It is a minimum value function; For the first Round candidate decay factor; For the first Current decay factor; Let be the likelihood function of the candidate decay factor; Let be the likelihood function of the current decay factor; the likelihood function follows a normal distribution, as shown in the formula: ; The prior probability of the candidate attenuation factor, i.e., the candidate attenuation factor. The probability density under the Beta prior distribution; The prior probability of the current decay factor, i.e., the current decay factor. The probability density under the corresponding Beta prior distribution; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0087] Secondary trust sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability:

[0088] In the formula, The probability of accepting the secondary trust candidate value; It is a minimum value function; No. Secondary trust value generated during round iteration; For the first Secondary trust value generated during round iteration; Let be the likelihood function of the observed values ​​under the candidate secondary trust; Let the likelihood function be the observed value under the current secondary trust. The prior probability of candidate secondary trust; Let be the prior probability of the current secondary trust. For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0089] Regularization intensity sampling: from the proposed distribution Generate candidate values ,set up Calculate the acceptance probability:

[0090] In the formula, The acceptance probability of a candidate value for regularization strength; It is a minimum value function; For the first Candidate values ​​for regularization strength in round iterations; For the first The current value of the regularization strength in each iteration; Let be the joint prior probability of all secondary trusts under the candidate regularization strength; Let be the joint prior probability of all secondary trusts under the current regularization strength; The secondary trust matrix for all indicator pairs; The prior probability of the candidate regularization strength; The prior probability of the current regularization strength; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0091] Noise variance sampling: from the proposed distribution Generate candidate values ,set up Calculate the acceptance probability:

[0092] In the formula, The acceptance probability of a candidate value for the noise variance; It is a minimum value function; For the first Candidate values ​​for noise variance in each iteration; For the first The current value of the noise variance in the current round of iteration; Let be the joint likelihood function of all observations under the candidate noise variance; This is the joint likelihood function of all observations under the current noise variance; The matrix of secondary trust observations for all indicator pairs; The secondary trust matrix for all indicator pairs; is the prior probability of the candidate noise variance; This represents the prior probability of the current noise variance; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0093] Step 5.3: Valid Sampling Preservation and Posterior Calculation

[0094] Discarding during the burning period: When the number of iterations is ≤3000, the sampled values ​​are unstable and are discarded directly;

[0095] Valid sample saving: When the number of iterations > 3000, the subsequent 7000 sample values ​​are saved as valid samples;

[0096] Posterior estimation: Calculate the posterior mean, standard deviation, and 95% confidence interval of the parameters based on the effective sample;

[0097] Posterior mean of trust level:

[0098] In the formula, For the first The posterior mean of the trust level of each indicator; The number of valid samples; For the first The first sampling Indicator trust level;

[0099] Posterior mean of indicator weights:

[0100] In the formula, For the first The posterior mean of the weights of each indicator; For the observation data matrix; The total number of evaluation indicators; For summation operators; For the first The posterior mean of the trust level of each indicator; Index for target evaluation indicators; For the first Trust level of each indicator; For the first The posterior mean of the trust level of each indicator; For traversing the evaluation index; For the first Trust level of each indicator.

[0101] Preferably, in step 6, the specific method is as follows:

[0102] A multi-round interaction mechanism is designed to achieve dynamic evolution of trust relationships and consensus convergence, ensuring stable and unique evaluation results.

[0103] Step 6.1: Setting the number of interaction rounds

[0104] Maximum number of interaction rounds If the convergence condition is met in advance, the process terminates to balance convergence and computational efficiency.

[0105] Step 6.2, Information Update Logic

[0106] Each round of interaction updates the evaluation information matrix based on the weights and evaluation information from the previous round, using the following formula:

[0107] In the formula, For the first The evaluation information matrix of the wheel; For the first The evaluation information matrix of the wheel; For the first The average global evaluation information of the round is derived from the first round. The weights of the indicators in each round are calculated from the evaluation information matrix to achieve the guiding role of global information; 0.05 is the global information adjustment coefficient, which controls the degree of influence of global information on the updating of evaluation information; The term represents a small random disturbance, simulating the random fluctuations in evaluation information during actual engineering projects. It follows a standard normal distribution;

[0108] Step 6.3, Convergence Termination Conditions

[0109] The convergence status is tested using consensus bias and the global consensus index:

[0110] Consensus Bias Calculation: The first Consensus bias of individual indicators The formula representing the degree of deviation between the evaluation information of this indicator and the overall evaluation information is:

[0111] In the formula, For the first Indicators, No. Evaluation value of the road segment; The mean of the global evaluation information, the first Evaluation value of the road segment; This represents the total number of road segments; The total number of indicators;

[0112] Global Consensus Index Calculation: Global Consensus Index The weighted average of the consensus deviations for each indicator is obtained by weighting the indicators by their respective weights. It represents the overall consensus level of the entire evaluation system, and the formula is as follows:

[0113] In the formula, For the first The weight of each indicator;

[0114] Termination condition: Set consensus identification coefficient If the consensus deviation of all indicators meets the following conditions: If the interaction ends prematurely, it will terminate early; otherwise, iterate until... .

[0115] Preferably, in step 7, the specific method is as follows:

[0116] Based on the converged optimal posterior weights, the comprehensive evaluation results of the road segment are calculated, and a quantitative evaluation conclusion is output:

[0117] Step 7.1: Calculation of Comprehensive Evaluation Value of Road Section

[0118] The weighted summation method is used to calculate the comprehensive evaluation value of each road segment. The formula is as follows:

[0119] In the formula, For the first The comprehensive evaluation value of each road segment; For the first The optimal posterior weights of each indicator; For the first The section of road, the first Normalized values ​​of each indicator; For road segment indexing; For indexing indicators;

[0120] Step 7.2, Calculation of the probability distribution of the grade

[0121] Based on the comprehensive evaluation value and the evaluation grading criteria (Excellent, Good, Average, Below Average, Poor), the posterior probability of each grade is calculated using the following formula:

[0122] In the formula, For the first The road section belongs to the grade The posterior probability; The evaluation grades are 1 = Excellent, 2 = Good, 3 = Average, 4 = Poor, and 5 = Very Poor. For level The median values ​​are: Excellent = 95, Good = 85, Average = 75, Difficult = 65, Poor = 50. For level Standard deviation, empirical value ; For all rating levels, To sum over 5 levels;

[0123] Step 7.3, Output Results

[0124] Basic outputs: comprehensive evaluation value, probability distribution of grade, and final evaluation grade for each road segment. The highest level;

[0125] Risk output: Uncertain probability of each road segment ;

[0126] Auxiliary outputs: posterior mean, standard deviation and 95% confidence interval of indicator weights, core indicator identification results, and indicators with weights ≥ 0.1.

[0127] By adopting the above technical solution: using PageRank dynamic trust relationships as a framework, a trust network is constructed among indicators through initial trust, secondary trust, and decay factor to achieve dynamic evolution of evaluation weights; Bayesian inference theory is introduced to probabilistically model the decay factor, secondary trust, and noise variance, and the Metropolis-Hastings (MH) algorithm is used for posterior sampling to quantify the uncertainty of the evaluation results; and applying... Regularized sparsity constraints suppress interference from weakly correlated indicators, focusing on core indicators such as rutting depth and crack rate. These three elements work together to form an integrated evaluation framework of "dynamic weight evolution, uncertainty quantification, and core indicator focus." Ultimately, this framework addresses the engineering characteristics of asphalt pavement preventive maintenance evaluation—characterized by multiple indicators, strong coupling, and uncertain information—achieving a maintenance evaluation highly aligned with actual on-site technical conditions. This provides a scientific and comprehensive quantitative basis for maintenance planning and resource allocation.

[0128] Compared with the prior art, the present invention has the following beneficial effects:

[0129] 1. The Bayesian Sparse Dynamic Trust Network (BSDTN) model constructed in this invention, through "dynamic trust network construction - Bayesian probability inference - The "regularized sparse constraint" three-level linkage architecture breaks through the technical limitations of fixed weights in traditional evaluation methods. Relying on the PageRank dynamic trust network, it realizes the dynamic evolution of evaluation weights according to the interaction relationships between indicators, completely eliminating subjective human interference in the weight determination process; through... Regularized sparsity constraints effectively suppress interference from weakly correlated indicators, keeping their weights below 0.05 and highlighting the dominant role of core indicators such as pavement performance. The total weight of pavement performance indicators reaches 0.591, ensuring the scientific rationality of the evaluation system. Engineering examples have verified that the method of this invention provides evaluation results for six test road sections with different technical levels and service environments that are completely consistent with actual road conditions. The evaluation level identification accuracy reaches 100%, with no misjudgments. In contrast, the traditional cloud model improvement method misjudged two road sections under the same test conditions, fully demonstrating that the evaluation accuracy and anti-interference ability of this method in complex engineering scenarios are significantly superior to existing technologies.

[0130] 2. This invention utilizes a Bayesian inference framework and the Metropolis-Hastings (MH) algorithm's MCMC posterior sampling to fully solve for the posterior probability distribution of model parameters and evaluation results. It not only outputs a deterministic pavement evaluation level but also accurately quantifies the risk level of the evaluation results through a defined uncertainty probability index. In engineering examples, this method can accurately identify high-risk road sections with large fluctuations in traffic volume and temperature, such as road sections 1 and 4, with an uncertainty probability of 0.010. This allows maintenance management units to be alerted in advance to conduct supplementary inspections and focused monitoring of such road sections, providing a quantitative reference for risk dimensions completely lacking in existing methods for preventative maintenance decisions, significantly improving the scientific rigor and foresight of maintenance decisions. Furthermore, the multi-round interactive consensus convergence mechanism designed in this invention enables rapid and stable convergence of evaluation results, requiring only two rounds of interaction to meet the convergence condition. This ensures the stability and uniqueness of the evaluation results while also considering computational efficiency in engineering applications.

[0131] 3. This invention constructs a comprehensive evaluation system encompassing five core dimensions and twelve sub-indicators: pavement structure, pavement performance, pavement use, climate conditions, and maintenance benefits. It comprehensively covers all factors influencing the preventive maintenance effectiveness of asphalt pavements, including inherent structural attributes, service performance status, service intensity, natural environmental erosion, and the economic rationality of maintenance. In particular, it incorporates climate indicators such as maximum and minimum temperatures and rainfall, which play a crucial role in the development of asphalt pavement defects. This solves the problem of existing technical evaluation systems being disconnected from the evolution mechanism of pavement defects, and can comprehensively and objectively reflect the true service status and preventive maintenance needs of asphalt pavements.

[0132] 4. The 12 indicators used in the evaluation system of this invention can all be obtained through conventional channels such as routine highway maintenance inspections, industry standard data, and engineering settlement statistics. No additional complex specialized testing equipment is required, significantly reducing the hardware threshold and data acquisition costs for technology application. The prior distribution settings of the model parameters fully integrate current highway industry standards and engineering practice experience, possessing clear physical meaning and engineering basis, avoiding the problem of poor engineering adaptability of purely mathematical models. The complete model calculation process can be implemented through programming with conventional engineering calculation software such as MATLAB, with standardized operation procedures that are easy for engineering technicians to master and apply. Verification through examples of highways of different technical grades, different climate zones, and road sections with different service years has shown that this method exhibits excellent adaptability and stability. It can provide accurate and reliable quantitative basis for the formulation of preventive maintenance plans for asphalt pavements and the optimal allocation of maintenance resources in different scenarios, possessing extremely high engineering promotion and application value. Attached Figure Description

[0133] Figure 1 This is a schematic diagram illustrating the construction and execution of the Bayesian sparse dynamic trust network model in this invention;

[0134] Figure 2 This is a schematic diagram of the hierarchical structure of the evaluation index system in an embodiment of the present invention;

[0135] Figure 3 This is a schematic diagram comparing the expected scores of the BSDTN model for six road segments in an embodiment of the present invention;

[0136] Figure 4 This is a schematic diagram of the probability distribution of different road segments of the BSDTN in an embodiment of the present invention;

[0137] Figure 5 This is a schematic diagram comparing the probability levels of different road sections in an embodiment of the present invention;

[0138] Figure 6 This is a schematic diagram comparing the probabilities of five methods for road segment 1 in an embodiment of the present invention;

[0139] Figure 7 This is a schematic diagram comparing the probabilities of five methods for road segment 4 in an embodiment of the present invention. Detailed Implementation

[0140] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the scope of protection of the present invention. The embodiments described in this invention are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0141] Example 1

[0142] This study evaluated six asphalt pavement sections of different grades within Gansu Province, covering four technical levels: expressways, Class I highways, Class II highways, and Class III highways. The sections were distributed across areas with varying climatic conditions (significant differences in temperature and rainfall), with road ages ranging from 2 to 6 years and traffic volumes ranging from 9.2 × 10³ to 32.6 × 10³, demonstrating good sample representativeness. The original data required for the evaluation came from the "Gansu Province Highway Pavement Performance Test Report (2023)" and the supporting test data of the "Technical Specification for Preventive Maintenance of Highway Asphalt Pavement" (JTG / T 5142-2019), including pavement structure parameters, pavement performance indicators, traffic load data, climate observation records, and maintenance cost statistics. All data were obtained through on-site testing, laboratory experiments, and final settlement statistics, ensuring their authenticity and reliability.

[0143] An evaluation method for asphalt pavement maintenance based on Bayesian sparse trust networks includes the following steps:

[0144] Step 1: Construction of Evaluation Index System: Following the principles of systematicness, scientificity, and practicality, construct an evaluation index system for preventive maintenance of asphalt pavement covering multiple dimensions, and clarify the definition, nature, and data source of each index;

[0145] Step 2, Data Preprocessing: Quantify and classify the indicator data, positiveize negative indicators, and normalize intervals to achieve dimensionless and homogeneous indicators, providing standardized data input for modeling;

[0146] Step 3: Dynamic Trust Network Construction: Quantify initial trust, secondary trust, and decay factor, construct a dynamic trust relationship network among indicators, and calculate the initial trust level and indicator weights.

[0147] Step 4, Bayesian probabilistic modeling and Sparsity constraints: Prior distributions are defined for the core parameters of the model; an observation model and a joint posterior distribution are constructed; and sparsity constraints are introduced. Regularized sparse constraints suppress interference from weakly correlated indices;

[0148] Step 5, MCMC posterior inference: The Metropolis-Hastings (MH) algorithm is used to perform posterior parameter sampling, and the posterior estimation results of the index weights are calculated based on the effective sampled values;

[0149] Step 6, Multi-round interactive convergence mechanism: Through information updates and consensus level checks, the dynamic evolution and convergence of trust relationships are achieved, ensuring the stability and uniqueness of the evaluation results;

[0150] Step 7: Evaluation Result Calculation and Output: Calculate the comprehensive evaluation value, grade probability distribution and uncertainty probability of the road segment based on the optimal posterior weight, and output the final maintenance evaluation result.

[0151] The overall construction and execution process of the Bayesian sparse dynamic trust network model used in this embodiment is as follows: Figure 1 As shown, this fully presents the closed-loop execution logic of the entire process of this invention: "standardized data input - dynamic trust relationship evolution - Bayesian joint posterior distribution construction - MCMC posterior sampling - multi-round consensus convergence verification - multi-dimensional result output". It clearly demonstrates the linkage relationship of the three core innovative modules: "dynamic weight evolution - uncertainty quantification - core indicator focus". It solves the problems of fragmented process and disconnected core modules in existing evaluation methods, and provides clear and intuitive process guidance for engineering and technical personnel to understand and apply this method.

[0152] Specifically, in step 1, the specific method is as follows:

[0153] A preventive maintenance evaluation index system for asphalt pavements was constructed, comprising 12 indicators across 5 core dimensions: pavement structure, pavement performance, pavement use, climate conditions, and maintenance benefits. The composition and significance of the indicators for each dimension are as follows:

[0154] Road surface structure dimension: including two indicators, road grade and construction difficulty, characterizes the inherent structural characteristics of asphalt pavement and determines the basic bearing capacity and maintenance difficulty of the pavement;

[0155] Road surface performance dimensions include four indicators: rutting depth index, crack rate, pothole rate, and skid resistance index. These are the core dimensions for evaluating the performance status of asphalt pavement, directly reflecting the structural integrity and safety of the pavement, and serving as the core basis for preventive maintenance decisions.

[0156] Road surface usage dimension: including two indicators, road age and traffic volume, which characterize the usage intensity and service life of asphalt pavement and are important factors affecting the degradation of pavement performance;

[0157] Climate conditions dimension: including three indicators: maximum temperature, minimum temperature, and rainfall, which characterize the natural environment in which the road surface is located. The long-term effects of temperature and rainfall accelerate the degradation of road surface performance and are important environmental factors that need to be considered in preventive maintenance.

[0158] Maintenance benefit dimension: This is the cost-benefit coefficient, which characterizes the input-output ratio of preventive maintenance of asphalt pavement and is a core indicator for evaluating the rationality and economy of maintenance measures.

[0159] The evaluation index system for preventive maintenance of asphalt pavement constructed in this embodiment clarifies the codes, names, properties, and data sources of the 5 major dimensions and 12 sub-indicators of the evaluation system, and constructs a systematic three-level evaluation architecture of "target layer - dimension layer - indicator layer" as follows: Figure 2 This addresses the industry pain points of existing technologies, such as the one-sided evaluation dimensions and the lack of core influencing factors of pavement distress, such as climate conditions. It achieves full coverage of all factors in the pavement chain, including inherent properties, service status, environmental erosion, and maintenance economy, providing a scientific and comprehensive indicator basis for subsequent model calculations. This ensures that the evaluation system is highly compatible with the pavement performance evolution mechanism and the actual decision-making needs of engineering, as shown in Table 1 below.

[0160] Table 1 Evaluation Index System for Preventive Maintenance of Asphalt Pavements

[0161] Specifically, in step 2, the specific method is as follows:

[0162] Because different indicators differ in their dimensions, orders of magnitude, and attributes (positive / negative), a three-step data preprocessing method is required to ensure that the indicators can be directly used in evaluation calculations:

[0163] Indicator Quantification and Grading: According to the "Technical Specification for Maintenance of Asphalt Pavement of Highway" (JTG 5142-2019), qualitative indicators (such as road grade and construction difficulty) are quantified and graded, and quantitative indicators (such as crack rate and traffic volume) are standardized and converted. All indicators are quantified into positive indicators (the larger the indicator value, the better the pavement condition or the higher the maintenance benefit).

[0164] Transforming Negative Indicators to Positive Values: For negative indicators such as crack rate, pothole rate, highest temperature, lowest temperature, and rainfall (the higher the value, the worse the road surface condition), an extreme value method is used for positive transformation. The formula is as follows:

[0165] In the formula, This represents the positive value of the indicator; The original value of the indicator; , These are the minimum and maximum values ​​of the original index, respectively. Note: The minimum temperature does not need to be processed by taking the absolute value. The lower (more negative) its original value, the greater the damage to the road surface caused by freeze-thaw cycles. It is a negative index and can be directly converted to a positive value using the formula above.

[0166] Normalization transformation: To adapt to the evaluation requirements of the Bayesian sparse dynamic trust network model, the quantified index values ​​are normalized to... The normalization formula is as follows: The evaluation values ​​of each indicator are within a reasonable range to ensure that the evaluation level of all road sections is consistent with the actual engineering results.

[0167] In the formula, For the first Road section, No. The normalized value of the indicator; For the first Road section, No. The quantified value of the indicator; , The first Minimum and maximum values ​​of the quantified indicator; This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0168] The original data of the evaluation indicators for the six test road sections in this embodiment cover all technical levels from expressways to Class III highways. They include typical engineering scenarios with different extreme temperatures, rainfall, traffic volume, and road age, and have strong scenario representativeness. This provides real and compliant input data for data preprocessing and is also used to verify the universality and engineering adaptability of the method of this invention for asphalt pavements of different regions, different levels, and different service conditions, as shown in Table 2.

[0169] Table 2. Original data of evaluation indicators for asphalt pavement of 6 highways

[0170] The evaluation standards for quantitative and qualitative indicators in this embodiment strictly conform to the requirements of the current highway maintenance industry standards, providing a unified and standardized execution standard for indicator quantification, grading, and standardized conversion, and providing standardized basic data for model input as shown in Tables 3 and 4 below.

[0171] Table 3 Evaluation Standards for Quantitative Indicators

[0172] Table 4 Qualitative Indicator Evaluation Level Standards

[0173] Specifically, in step 3, the specific method is as follows:

[0174] Using PageRank dynamic trust relationships as the core framework, this approach quantifies three core parameters and constructs a dynamic trust network among indicators, achieving a breakthrough from "fixed weights" to "dynamic weights."

[0175] Step 3.1: Definition and Calculation of Core Parameters

[0176] Initial trust is recorded as This characterizes the prior trust level of the evaluator towards each evaluation indicator in a non-interactive state, reflecting the inherent evaluation value of the indicator. Equal-weight initialization is used to ensure unbiasedness. The formula is:

[0177] In the formula, For the first The initial trust value for each indicator; The total number of evaluation indicators in this invention ; For indexing indicators;

[0178] The attenuation factor is denoted as Characteristic indicators To the indicator The attenuation level when trust is transferred, with a value range of [0,1], is calculated using the cosine similarity of the evaluation information vectors between indicators, and the formula is:

[0179] In the formula, As an indicator With indicators Trust decay factor; , Indicators , The evaluation information vector; Let cosine similarity be the similarity between the two vectors. This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0180] Secondary trust is recorded as Characteristic indicators With indicators The trust increment generated after information exchange reflects the dynamic interaction between indicators, and the formula is:

[0181] In the formula, As an indicator For indicators The secondary trust value is positive, indicating increased trust, and negative, indicating decreased trust. For interactive rounds, the initial round ; For road segment indexing; For the first In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; For: the In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; , For: respectively the first wheel, In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment;

[0182] Step 3.2: Building a Trust Relationship Network

[0183] The network nodes are composed of 12 evaluation indicators, and the edge weights between nodes are composed of "initial trust allocation + secondary trust increment".

[0184] Initial Trust Assignment In the absence of interaction, the metrics For indicators The initial trust value is calculated using the following formula:

[0185] In the formula, As an indicator For indicators The initial trust assignment value; As an indicator The initial trust value; The total number of evaluation indicators; This is a road segment index, with a value range of 1 to M. In this invention, M=6, meaning 6 road segments. The index is a value range of 1 to N. In this invention, N=12, that is, 12 indicators.

[0186] Overall trust level :index For indicators The overall trust strength, used as the edge weight of the trust network, is expressed by the following formula:

[0187] In the formula, As an indicator For indicators The initial trust assignment value; As an indicator For indicators Secondary trust value;

[0188] Step 3.3: Initial Trust Level and Weight Calculation

[0189] In the formula, For the first Trust levels for each indicator; For the first The decay factor of each indicator relative to the global trust mean. , For the first Global Trust Mean Vector of Each Indicator , The global trust mean of all indicators , The total trust matrix is ​​the first... List; As an indicator The initial trust value; As an indicator With indicators The attenuation factor; As an indicator For indicators Overall trust level; For indexing indicators;

[0190] Initial weights The normalized results of the trust levels of each indicator reflect the contribution of the indicator in the evaluation. The formula is:

[0191] In the formula, For the first The initial weights of each indicator, ; For the first Trust levels for each indicator; For the first Trust levels for each indicator; For indexing indicators.

[0192] Specifically, in step 4, the specific method is as follows:

[0193] Based on Bayesian inference theory, probabilistic modeling is performed on the core parameters of the model, and... Regularized sparse constraints are used to construct a joint posterior distribution, providing a mathematical foundation for subsequent parameter estimation.

[0194] Step 4.1: Setting the Prior Distribution of Core Parameters

[0195] Based on the physical characteristics of the parameters and engineering experience, a suitable probability distribution is selected, following the principles of "prioritizing engineering experience, using prior information without a priori assistance, and adapting to sparse constraints":

[0196] Attenuation factor The value range is [0,1]. Modeling is performed using a Beta distribution, and hyperparameters are set according to the correlation between indicators.

[0197] In the formula, It follows a Beta distribution. , Hyperparameters that determine the shape of the distribution; strong correlation: cosine similarity between indices. Beta distribution mean Correlation: Cosine similarity between indicators mean Weak correlation: cosine similarity between indicators mean ;

[0198] Secondary Trust To achieve Regularized sparsity constraints are modeled using a Laplace distribution, and the formula is as follows:

[0199] In the formula, It follows a Laplace distribution; This is the regularization strength parameter; As an indicator For indicators Secondary trust value;

[0200] Observation noise variance To characterize the degree of random fluctuation in the observed data, inverse Gamma weak information prior modeling is used.

[0201] In the formula, It follows an inverse Gamma distribution; , As hyperparameters, weak information priors are used to avoid dominating posterior inferences;

[0202] Regularization strength We use Gamma weak information prior modeling to support adaptive learning. The formula is:

[0203] In the formula, It follows a Gamma distribution; , For hyperparameters, weak information priors;

[0204] Step 4.2, Observation Model Construction

[0205] Construct a linear observation equation, connecting the model parameters with the observation data, using the following formula:

[0206]

[0207] In the formula, Observations for secondary trust; The true value of secondary trust; The observation noise is independent and identically distributed, following a mean of 0 and a variance of . The normal distribution, i.e. ,and ;

[0208] Step 4.3: Construction of Joint Posterior Distribution

[0209] According to Bayes' theorem, a joint posterior distribution is constructed by fusing the prior distribution of all parameters with the likelihood function of the observation model:

[0210] In the formula: Let be the prior distribution of the attenuation factor, and be the product of the prior values ​​of each index with respect to the attenuation factor Beta. Let be the prior distribution of secondary trust, and be the product of each index with respect to the Laplace prior of secondary trust. For the prior distribution of regularization intensity, we adopt Weak information prior; For the prior distribution of noise variance, we use Weak information prior; Let be the likelihood function of the observed data, and be the product of the secondary trust normal likelihoods of each index. For the observation data matrix; Proportional to, ignoring marginal likelihoods independent of parameters. ;

[0211] Step 4.4, Definition of Uncertainty Probability

[0212] The formula for quantifying the risk level of the evaluation results is:

[0213] In the formula, The probability of uncertainty in the evaluation results of a road segment is as close to 0 as the stability of the evaluation results. This represents the maximum posterior probability of the road segment's evaluation level.

[0214] Specifically, in step 5, the specific method is as follows:

[0215] The Metropolis-Hastings (MH) algorithm is used for posterior sampling of MCMC to obtain effective sampled values ​​of parameters, calculate the posterior estimation results of index weights, and set the combustion period. Valid sample count :

[0216] Step 5.1, Parameter Initialization

[0217] Set initial values ​​for each parameter to provide a starting point for sampling:

[0218] Attenuation factor matrix The secondary trust matrix is ​​calculated from the initial cosine similarity between indicators. Initialize to a 0 matrix, regularization strength noise variance ;

[0219] Step 5.2, MH sampling by parameter

[0220] For each parameter, candidate values ​​are generated sequentially, the acceptance probability is calculated, and the parameter is updated according to random rules:

[0221] Attenuation factor sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability:

[0222]

[0223] In the formula, represents the acceptance probability of candidate values ​​for the attenuation factor, with a value range of [0,1]. It is a minimum value function; For the first Round candidate decay factor; For the first Current decay factor; Let be the likelihood function of the candidate decay factor; Let be the likelihood function of the current decay factor; the likelihood function follows a normal distribution, as shown in the formula: ; The prior probability of the candidate attenuation factor, i.e., the candidate attenuation factor. The probability density under the Beta prior distribution; The prior probability of the current decay factor, i.e., the current decay factor. The probability density under the corresponding Beta prior distribution; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0224] Secondary trust sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability:

[0225] In the formula, The probability of accepting the secondary trust candidate value; It is a minimum value function; No. Secondary trust value generated during round iteration; For the first Secondary trust value generated during round iteration; Let be the likelihood function of the observed values ​​under the candidate secondary trust; Let the likelihood function be the observed value under the current secondary trust. The prior probability of candidate secondary trust; Let be the prior probability of the current secondary trust. For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0226] Regularization intensity sampling: from the proposed distribution Generate candidate values ,set up Calculate the acceptance probability:

[0227] In the formula, The acceptance probability of a candidate value for regularization strength; It is a minimum value function; For the first Candidate values ​​for regularization strength in round iterations; For the first The current value of the regularization strength in each iteration; Let be the joint prior probability of all secondary trusts under the candidate regularization strength; Let be the joint prior probability of all secondary trusts under the current regularization strength; The secondary trust matrix for all indicator pairs; The prior probability of the candidate regularization strength; The prior probability of the current regularization strength; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0228] Noise variance sampling: from the proposed distribution Generate candidate values ,set up Calculate the acceptance probability:

[0229] In the formula, The acceptance probability of a candidate value for the noise variance; It is a minimum value function; For the first Candidate values ​​for noise variance in each iteration; For the first The current value of the noise variance in the current round of iteration; Let be the joint likelihood function of all observations under the candidate noise variance; This is the joint likelihood function of all observations under the current noise variance; The matrix of secondary trust observations for all indicator pairs; The secondary trust matrix for all indicator pairs; is the prior probability of the candidate noise variance; This represents the prior probability of the current noise variance; For the MH sampling iteration round index; Mark candidate values ​​with superscripts;

[0230] Step 5.3: Valid Sampling Preservation and Posterior Calculation

[0231] Discarding during the burning period: When the number of iterations is ≤3000, the sampled values ​​are unstable and are discarded directly;

[0232] Valid sample saving: When the number of iterations > 3000, the subsequent 7000 sample values ​​are saved as valid samples;

[0233] Posterior estimation: Calculate the posterior mean, standard deviation, and 95% confidence interval of the parameters based on the effective sample;

[0234] Posterior mean of trust level:

[0235] In the formula, For the first The posterior mean of the trust level of each indicator; The number of valid samples; For the first The first sampling Indicator trust level;

[0236] Posterior mean of indicator weights:

[0237] In the formula, For the first The posterior mean of the weights of each indicator; For the observation data matrix; The total number of evaluation indicators; For summation operators; For the first The posterior mean of the trust level of each indicator; Index for target evaluation indicators; For the first Trust level of each indicator; For the first The posterior mean of the trust level of each indicator; For traversing the evaluation index; For the first Trust level of each indicator.

[0238] The posterior calculation results of the evaluation index weights obtained after steps 3 to 5 in this embodiment are shown in Table 5, which intuitively verifies the three core technical effects of the present invention: First, by using L1 regularization sparsity constraints, the weights of weakly correlated indicators such as pavement structure and climate conditions are stably controlled within 0.05, effectively suppressing the interference of weakly correlated indicators on the evaluation results; Second, it highlights the core leading role of the pavement performance dimension, with the total weight of the four pavement performance indicators reaching 0.591, which is completely consistent with the industry consensus that pavement performance is the core basis for preventive maintenance decisions in engineering practice; Third, through Bayesian posterior inference, it realizes the quantitative characterization of the uncertainty of weight estimation, and gives the fluctuation range and confidence interval of the weights, breaking through the technical bottleneck that the traditional fixed weight method cannot quantify the reliability of weights.

[0239] Table 5. Posterior calculation results of evaluation index weights

[0240] Specifically, in step 6, the specific method is as follows:

[0241] A multi-round interaction mechanism is designed to achieve dynamic evolution of trust relationships and consensus convergence, ensuring stable and unique evaluation results.

[0242] Step 6.1: Setting the number of interaction rounds

[0243] Maximum number of interaction rounds If the convergence condition is met in advance, the process terminates to balance convergence and computational efficiency.

[0244] Step 6.2, Information Update Logic

[0245] Each round of interaction updates the evaluation information matrix based on the weights and evaluation information from the previous round, using the following formula:

[0246] In the formula, For the first The evaluation information matrix of the wheel; For the first The evaluation information matrix of the wheel; For the first The average global evaluation information of the round is derived from the first round. The weights of the indicators in each round are calculated from the evaluation information matrix to achieve the guiding role of global information; 0.05 is the global information adjustment coefficient, which controls the degree of influence of global information on the updating of evaluation information; The term represents a small random disturbance, simulating the random fluctuations in evaluation information during actual engineering projects. It follows a standard normal distribution;

[0247] Step 6.3, Convergence Termination Conditions

[0248] The convergence status is tested using consensus bias and the global consensus exponent.

[0249] Consensus Bias Calculation: The first Consensus bias of individual indicators The formula representing the degree of deviation between the evaluation information of this indicator and the overall evaluation information is:

[0250] In the formula, For the first Indicators, the first Evaluation value of the road segment; The mean of the global evaluation information, the first Evaluation value of the road segment; This represents the total number of road segments; The total number of indicators;

[0251] Global Consensus Index Calculation: Global Consensus Index The weighted average of the consensus deviations for each indicator is obtained by weighting the indicators by their respective weights. It represents the overall consensus level of the entire evaluation system, and the formula is as follows:

[0252] In the formula, For the first The weight of each indicator;

[0253] Termination condition: Set consensus identification coefficient If the consensus deviation of all indicators meets the following conditions: If the interaction ends prematurely, it will terminate early; otherwise, iterate until... .

[0254] Specifically, in step 7, the specific method is as follows:

[0255] Based on the converged optimal posterior weights, the comprehensive evaluation results of the road segment are calculated, and a quantitative evaluation conclusion is output:

[0256] Step 7.1: Calculation of Comprehensive Evaluation Value of Road Section

[0257] The weighted summation method is used to calculate the comprehensive evaluation value of each road segment. The formula is as follows:

[0258] In the formula, For the first The comprehensive evaluation value of each road segment; For the first The optimal posterior weights of each indicator; For the first The section of road, the first Normalized values ​​of each indicator; For road segment indexing; For indexing indicators;

[0259] Step 7.2, Calculation of the probability distribution of the grade

[0260] Based on the comprehensive evaluation value and the evaluation grading criteria (Excellent, Good, Average, Below Average, Poor), the posterior probability of each grade is calculated using the following formula:

[0261] In the formula, For the first The road section belongs to the grade The posterior probability; The evaluation grades are 1 = Excellent, 2 = Good, 3 = Average, 4 = Poor, and 5 = Very Poor. For level The median values ​​are: Excellent = 95, Good = 85, Average = 75, Difficult = 65, Poor = 50. For level Standard deviation, empirical value ; For all rating levels, To sum over 5 levels;

[0262] Step 7.3, Output Results

[0263] Basic outputs: comprehensive evaluation value, probability distribution of grade, and final evaluation grade for each road segment. The highest level;

[0264] Risk output: Uncertain probability of each road segment ;

[0265] Auxiliary outputs: posterior mean, standard deviation and 95% confidence interval of indicator weights, core indicator identification results, and indicators with weights ≥ 0.1.

[0266] After calculations in steps 6 and 7 of this embodiment, the evaluation results of the six road segments obtained based on the BSDTN model are shown in Table 6. This fully verifies the engineering application value of the method of this invention: First, the final evaluation level of all road segments is completely consistent with the actual road conditions on site, achieving a 100% accuracy rate in evaluation level identification, proving the accuracy of the method; Second, through a custom uncertain probability index, road segments 1 and 4, which have high traffic volume and significant temperature fluctuations, are accurately identified as high-risk road segments, quantifying the risk level of the evaluation results and filling the industry gap in the lack of risk dimension decision-making reference in existing evaluation methods; Third, the posterior probability distribution of the full evaluation level of each road segment is output, fully presenting the probabilistic characteristics of the evaluation results, providing a more comprehensive quantitative basis for maintenance resource allocation and graded maintenance decisions.

[0267] Table 6 Evaluation results of 6 road segments based on the BSDTN model

[0268] The expected scores for all road segments are stably distributed as follows: The range closely matches the threshold range of Level III (Medium) evaluation grade, such as... Figure 3 This complements and verifies the quantitative calculation results in Table 6, intuitively demonstrating the stability of the evaluation results of the method of this invention; the posterior probability distribution of the five levels (excellent, good, average, subpar, and poor) for the six road segments is presented in a bar chart, with the probability of level III (average) being absolutely dominant for all road segments, which is completely consistent with the final evaluation results. Figure 4 At the same time, it clearly shows the differences in probability distribution among different road sections. For example, the probability of excellent and good grades in road section 4 is significantly higher than that in other road sections. This is highly consistent with the original data characteristics of the road section having a shorter road age and better pavement performance indicators, thus verifying the accurate identification capability of the method of this invention for subtle differences in pavement performance.

[0269] To further verify the technical superiority of the method of the present invention, four existing mainstream evaluation methods were selected for horizontal comparison with the method of the present invention. The evaluation results of the five methods are shown in Table 7.

[0270] Table 7 Comparison of Evaluation Results of Five Methods

[0271] Through horizontal comparison, the technical superiority of the method of this invention is intuitively demonstrated: First, traditional methods have resulted in varying degrees of misjudgment of evaluation levels, while the evaluation results of the method of this invention are consistent with the actual situation on all road sections. Its anti-interference ability and evaluation stability in complex engineering scenarios are significantly better than existing technologies. Second, the method of this invention can simultaneously output the uncertainty probability of the evaluation results, realizing the quantification of evaluation reliability, while all traditional methods cannot achieve a systematic characterization of uncertainty. Third, the level probability distribution of the method of this invention is more in line with the actual performance degradation law of the road surface, avoiding the problems of extreme probability concentration and multi-level probability dispersion distortion that occur in traditional methods. The evaluation results have stronger engineering reference value and robustness.

[0272] The probability level comparison of each road segment in this embodiment is as follows: Figure 5 The paper presents a direct comparison of the posterior probability distribution characteristics of five methods on six road segments at level III (intermediate). It clearly demonstrates that the intermediate probability distribution of the method of this invention is stable and closely matches the actual engineering situation, while the intermediate probability of traditional method 1 is significantly low and the probability of method 2 is excessively concentrated at 1.0, both of which have problems with probability distribution distortion. The visualization proves that the rationality and robustness of the evaluation results of the method of this invention are significantly better than the existing mainstream evaluation techniques.

[0273] In this embodiment, the probability comparison of the five methods for road segment 1 and road segment 4 is as follows: Figure 6 ,like Figure 7 The results clearly demonstrate that road segment 1, under traditional method 3, exhibited a misjudgment where the probability of a "good" grade was dominant; method 1 showed problems with dispersed probabilities across multiple grades and the absence of a clearly dominant grade; road segment 4, under traditional method 2, also exhibited a misjudgment where the probability of a "good" grade was dominant; and the remaining traditional methods all showed varying degrees of probability distribution anomalies. In contrast, the probability distribution of the method presented in this invention closely matches the actual road conditions, verifying that the method of this invention maintains extremely high evaluation accuracy and strong anti-interference capabilities even on complex, high-risk road segments, with the medium grade being absolutely dominant, while simultaneously preserving the probability information and uncertainties of other grades.

[0274] This invention uses PageRank dynamic trust relationships as a framework, constructing a trust network among indicators through initial trust, secondary trust, and decay factors to achieve dynamic evolution of evaluation weights. It introduces Bayesian inference theory to probabilistically model the decay factor, secondary trust, and noise variance, and employs the Metropolis-Hastings (MH) algorithm for posterior sampling to quantify the uncertainty of evaluation results. Regularized sparsity constraints suppress interference from weakly correlated indicators, focusing on core indicators such as rutting depth and crack rate. These three elements work together to form an integrated evaluation framework of "dynamic weight evolution, uncertainty quantification, and core indicator focus." Ultimately, this framework addresses the engineering characteristics of asphalt pavement preventive maintenance evaluation—characterized by multiple indicators, strong coupling, and uncertain information—achieving a maintenance evaluation highly aligned with actual on-site technical conditions. This provides a scientific and comprehensive quantitative basis for maintenance planning and resource allocation.

[0275] In summary, this invention utilizes "PageRank dynamic trust network construction - Bayesian probability modeling and uncertainty quantification" to achieve this. The three-level linkage logic of "regularized sparse constraints" enables high-precision and high-stability evaluation of the preventive maintenance status of asphalt pavement.

[0276] The descriptions and practices disclosed in this invention are readily apparent and understandable to those skilled in the art, and various modifications and refinements can be made without departing from the principles of this invention. Therefore, any modifications or improvements made without departing from the spirit of this invention should also be considered within the scope of protection of this invention.

Claims

1. A method for evaluating asphalt pavement maintenance based on Bayesian sparse trust networks, characterized in that, Includes the following steps: Step 1: Construction of Evaluation Index System: Following the principles of systematicness, scientificity, and practicality, construct an evaluation index system for preventive maintenance of asphalt pavement covering multiple dimensions, and clarify the definition, nature, and data source of each index; Step 2, Data Preprocessing: Quantify and classify the indicator data, positiveize negative indicators, and normalize intervals to achieve dimensionless and homogeneous indicators, providing standardized data input for modeling; Step 3: Dynamic Trust Network Construction: Quantify initial trust, secondary trust, and decay factor, construct a dynamic trust relationship network among indicators, and calculate the initial trust level and indicator weights; Step 4, Bayesian probabilistic modeling and Sparsity constraints: Prior distributions are defined for the core parameters of the model; an observation model and a joint posterior distribution are constructed; and sparsity constraints are introduced. Regularized sparse constraints suppress weak index interference; Step 5, MCMC posterior inference: The MH algorithm is used to perform posterior sampling of parameters, and the posterior estimation results of index weights are calculated based on the effective sampled values; Step 6, Multi-round interactive convergence mechanism: Through information updates and consensus level checks, the dynamic evolution and convergence of trust relationships are achieved, ensuring the stability and uniqueness of the evaluation results; Step 7: Evaluation Result Calculation and Output: Calculate the comprehensive evaluation value, grade probability distribution and uncertainty probability of the road segment based on the optimal posterior weight, and output the final maintenance evaluation result.

2. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 1, the specific method is as follows: A preventive maintenance evaluation index system for asphalt pavements was constructed, comprising 12 indicators across 5 core dimensions: pavement structure, pavement performance, pavement use, climate conditions, and maintenance benefits. The composition and significance of the indicators for each dimension are as follows: Road surface structure dimension: including two indicators, road grade and construction difficulty, characterizes the inherent structural characteristics of asphalt pavement and determines the basic bearing capacity and maintenance difficulty of the pavement; Road surface performance dimensions include four indicators: rutting depth index, crack rate, pothole rate, and skid resistance index. These are the core dimensions for evaluating the performance status of asphalt pavement, directly reflecting the structural integrity and safety of the pavement, and serving as the core basis for preventive maintenance decisions. Road surface usage dimension: including two indicators, road age and traffic volume, which characterize the usage intensity and service life of asphalt pavement and are important factors affecting the degradation of pavement performance; Climate conditions dimension: including three indicators: maximum temperature, minimum temperature, and rainfall, which characterize the natural environment in which the road surface is located. The long-term effects of temperature and rainfall accelerate the degradation of road surface performance and are important environmental factors that need to be considered in preventive maintenance. Maintenance benefit dimension: This is the cost-benefit coefficient, which characterizes the input-output ratio of preventive maintenance of asphalt pavement and is a core indicator for evaluating the rationality and economy of maintenance measures.

3. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 2, the specific method is as follows: Data preprocessing is completed in three steps to ensure that the indicators can be directly used in evaluation calculations: Indicator Quantification and Grading: Qualitative indicators are quantified and graded, and quantitative indicators are standardized and transformed. All indicators are quantified into positive indicators. Positive Conversion of Negative Indicators: For negative indicators such as crack rate, pothole rate, maximum temperature, minimum temperature, and rainfall, the extreme value method is used for positive conversion. The formula is as follows: ； In the formula, This represents the positive value of the indicator; The original value of the indicator; , These are the minimum and maximum values ​​of the original index, respectively. Normalization transformation: To adapt to the evaluation requirements of the Bayesian sparse dynamic trust network model, the quantified index values ​​are normalized to... The normalization formula is as follows: The evaluation values ​​of each indicator are within a specified range to ensure that the evaluation level of all road sections is consistent with the actual engineering results. ； In the formula, For the first Road section, No. The normalized value of the indicator; For the first Road section, No. The quantified value of the indicator; The first Minimum and maximum values ​​of the quantified indicator; This is the road segment index, with a value range of 1 to M. Here, M=6, meaning there are 6 road segments. This is an index of indicators, with values ​​ranging from 1 to N. Here, N=12, meaning 12 indicators.

4. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, Step 3, the specific method is as follows: Using PageRank dynamic trust relationships as the core framework, this paper quantifies three core parameters and constructs a dynamic trust network among indicators, achieving a breakthrough from fixed weights to dynamic weights. Step 3.1: Definition and Calculation of Core Parameters Initial trust is recorded as This characterizes the prior trust level of the evaluator towards each evaluation indicator in a non-interactive state, reflecting the inherent evaluation value of the indicator. Equal-weight initialization is used to ensure unbiasedness. The formula is: ； In the formula, For the first The initial trust value for each indicator; The total number of evaluation indicators, For indexing indicators; The attenuation factor is denoted as Characteristic indicators To the indicator The attenuation level when trust is transferred, with a value range of [0,1], is calculated using the cosine similarity of the evaluation information vectors between indicators, and the formula is: ； In the formula, As an indicator With indicators Trust decay factor; , Indicators , The evaluation information vector; Let cosine similarity be the similarity between the two vectors. This is the road segment index, with a value range of 1 to M. Here, M=6, meaning there are 6 road segments. This is an index of indicators, with values ​​ranging from 1 to N. Here, N=12, meaning 12 indicators. Secondary trust is recorded as Characteristic indicators With indicators The trust increment generated after information exchange reflects the dynamic interaction between indicators, and the formula is: ； In the formula, As an indicator For indicators The secondary trust value is positive, indicating increased trust, and negative, indicating decreased trust. For interactive rounds, the initial round ; For road segment indexing; For the first In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; For: the In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; , For: respectively the first wheel, In the first round of interaction, The first indicator, the first Evaluation information values ​​for each road segment; Step 3.2: Building a Trust Relationship Network The network nodes are composed of 12 evaluation indicators, and the edge weights between nodes are composed of "initial trust allocation + secondary trust increment". Initial Trust Assignment In the absence of interaction, the metrics For indicators The initial trust value is calculated using the following formula: ； In the formula, As an indicator For indicators The initial trust assignment value; As an indicator The initial trust value; The total number of evaluation indicators; This is the road segment index, with a value range of 1 to M. Here, M=6, meaning there are 6 road segments. This is an index of indicators, with values ​​ranging from 1 to N. Here, N=12, meaning 12 indicators. Overall trust level :index For indicators The overall trust strength, used as the edge weight of the trust network, is expressed by the following formula: ； In the formula, As an indicator For indicators The initial trust assignment value; As an indicator For indicators Secondary trust value; Step 3.3: Initial Trust Level and Weight Calculation ； In the formula, For the first Trust levels for each indicator; For the first The decay factor of each indicator relative to the global trust mean. , For the first Global Trust Mean Vector of Each Indicator , The global trust mean of all indicators , The total trust matrix is ​​the first... List; As an indicator The initial trust value; As an indicator With indicators The attenuation factor; As an indicator For indicators Overall trust level; For indexing indicators; Initial weights The normalized results of the trust levels of each indicator reflect the contribution of the indicator in the evaluation. The formula is: ； In the formula, For the first The initial weights of each indicator, ; For the first Trust levels for each indicator; For the first Trust levels for each indicator; For indexing indicators.

5. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 4, the specific method is as follows: Based on Bayesian inference theory, probabilistic modeling is performed on the core parameters of the model, and... Regularized sparse constraints are used to construct a joint posterior distribution, providing a mathematical foundation for subsequent parameter estimation. Step 4.1: Setting the Prior Distribution of Core Parameters Based on the physical characteristics of the parameters and engineering experience, a suitable probability distribution is selected, following the principles of "prioritizing engineering experience, using prior information without a priori assistance, and adapting to sparse constraints": Attenuation factor The value range is [0,1]. Modeling is performed using a Beta distribution, and hyperparameters are set according to the correlation between indicators. ； In the formula, It follows a Beta distribution. , Hyperparameters that determine the shape of the distribution; Strong correlation: cosine similarity between indicators Beta distribution mean Correlation: Cosine similarity between indicators mean Weak correlation: cosine similarity between indicators mean ; Secondary Trust To achieve Regularized sparsity constraints are modeled using a Laplace distribution, and the formula is as follows: ； In the formula, It follows a Laplace distribution; This is the regularization strength parameter; As an indicator For indicators Secondary trust value; Observation noise variance To characterize the degree of random fluctuation in the observed data, inverse Gamma weak information prior modeling is used. ； In the formula, It follows an inverse Gamma distribution; , As hyperparameters, weak information priors are used to avoid dominating posterior inferences; Regularization strength We use Gamma weak information prior modeling to support adaptive learning. The formula is: ； In the formula, It follows a Gamma distribution; , For hyperparameters, weak information priors; Step 4.2, Observation Model Construction Construct a linear observation equation, connecting the model parameters with the observation data, using the following formula: ； In the formula, Observations for secondary trust; The true value of secondary trust; The observation noise is independent and identically distributed, following a mean of 0 and a variance of . The normal distribution, i.e. ,and ; Step 4.3: Construction of Joint Posterior Distribution According to Bayes' theorem, a joint posterior distribution is constructed by fusing the prior distribution of all parameters with the likelihood function of the observation model: ； In the formula: Let be the prior distribution of the attenuation factor, and be the product of the prior values ​​of each index with respect to the attenuation factor Beta. Let be the prior distribution of secondary trust, and be the product of each index with respect to the Laplace prior of secondary trust. For the prior distribution of regularization intensity, we adopt Weak information prior; For the prior distribution of noise variance, we use Weak information prior; Let be the likelihood function of the observed data, and be the product of the secondary trust normal likelihoods of each index. For the observation data matrix; Proportional to, ignoring marginal likelihoods independent of parameters. ; Step 4.4, Definition of Uncertainty Probability The formula for quantifying the risk level of the evaluation results is: ； In the formula, The uncertainty probability of the road segment evaluation results; This represents the maximum posterior probability of the road segment's evaluation level.

6. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 5, the specific method is as follows: The MH algorithm is used for MCMC posterior sampling to obtain effective sampled values ​​of parameters, calculate the posterior estimation results of index weights, and set the combustion period. Valid sample count : Step 5.1, Parameter Initialization Set initial values ​​for each parameter to provide a starting point for sampling: Attenuation factor matrix The secondary trust matrix is ​​calculated from the initial cosine similarity between indicators. Initialize to a 0 matrix, regularization strength noise variance ; Step 5.2, MH sampling by parameter For each parameter, candidate values ​​are generated sequentially, the acceptance probability is calculated, and the parameter is updated according to random rules: Attenuation factor sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability: ； In the formula, represents the acceptance probability of candidate values ​​for the attenuation factor, with a value range of [0,1]. It is a minimum value function; For the first Round candidate decay factor; For the first Current decay factor; Let be the likelihood function of the candidate decay factor; Let be the likelihood function of the current decay factor; the likelihood function follows a normal distribution, as shown in the formula: ; The prior probability of the candidate attenuation factor, i.e., the candidate attenuation factor. The probability density under the Beta prior distribution; The prior probability of the current decay factor, i.e., the current decay factor. The probability density under the corresponding Beta prior distribution; For the MH sampling iteration round index; Mark candidate values ​​with superscripts; Secondary trust sampling: for each From the proposed distribution Generate candidate values Calculate the acceptance probability: ； In the formula, The probability of accepting the secondary trust candidate value; It is a minimum value function; No. Secondary trust value generated during round iteration; For the first Secondary trust value generated during round iteration; Let be the likelihood function of the observed values ​​under the candidate secondary trust; Let the likelihood function be the observed value under the current secondary trust. The prior probability of candidate secondary trust; The prior probability of the current secondary trust; For the MH sampling iteration round index; Mark candidate values ​​with superscripts; Regularization intensity sampling: from the proposed distribution Generate candidate values ,set up Calculate the acceptance probability: ； In the formula, The acceptance probability of a candidate value for regularization strength; It is a minimum value function; For the first Candidate values ​​for regularization strength in round iterations; For the first The current value of the regularization strength in each iteration; Let be the joint prior probability of all secondary trusts under the candidate regularization strength; Let be the joint prior probability of all secondary trusts under the current regularization strength; The secondary trust matrix for all indicator pairs; The prior probability of the candidate regularization strength; The prior probability of the current regularization strength; For the MH sampling iteration round index; Mark candidate values ​​with superscripts; Noise variance sampling: from the proposal Generate candidate values ,set up Calculate the acceptance probability: ； In the formula, The acceptance probability of a candidate value for the noise variance; It is a minimum value function; For the first Candidate values ​​for noise variance in each iteration; For the first The current value of the noise variance in the current round of iteration; Let be the joint likelihood function of all observations under the candidate noise variance; This is the joint likelihood function of all observations under the current noise variance; The matrix of secondary trust observations for all indicator pairs; The secondary trust matrix for all indicator pairs; is the prior probability of the candidate noise variance; This represents the prior probability of the current noise variance; For the MH sampling iteration round index; Mark candidate values ​​with superscripts; Step 5.3: Valid Sampling Preservation and Posterior Calculation Discarding during the burning period: When the number of iterations is ≤3000, the sampled values ​​are unstable and are discarded directly; Valid sample saving: When the number of iterations > 3000, the subsequent 7000 sample values ​​are saved as valid samples; Posterior estimation: Calculate the posterior mean, standard deviation, and 95% confidence interval of the parameters based on the effective sample; Posterior mean of trust level: ； In the formula, For the first The posterior mean of the trust level of each indicator; The number of valid samples; For the first The first sampling Indicator trust level; Posterior mean of indicator weights: ； In the formula, For the first The posterior mean of the weights of each indicator; For the observation data matrix; The total number of evaluation indicators; For summation operators; For the first The posterior mean of the trust level of each indicator; Index for target evaluation indicators; For the first Trust level of each indicator; For the first The posterior mean of the trust level of each indicator; For traversing the evaluation index; For the first Trust level of each indicator.

7. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 6, the specific method is as follows: A multi-round interaction mechanism is designed to achieve dynamic evolution of trust relationships and consensus convergence, ensuring stable and unique evaluation results. Step 6.1: Setting the number of interaction rounds Maximum number of interaction rounds If the convergence condition is met in advance, the process terminates to balance convergence and computational efficiency. Step 6.2, Information Update Logic Each round of interaction updates the evaluation information matrix based on the weights and evaluation information from the previous round, using the following formula: ； In the formula, For the first The evaluation information matrix of the wheel; For the first The evaluation information matrix of the wheel; For the first The average global evaluation information of the round is derived from the first round. The weights of the indicators and the evaluation information matrix of the round are calculated to achieve the guiding role of global information; 0.05 is the global information adjustment coefficient, which controls the degree of influence of global information on the updating of evaluation information; The term represents a small random disturbance, simulating the random fluctuations in evaluation information during actual engineering projects. It follows a standard normal distribution; Step 6.3, Convergence Termination Conditions The convergence status is tested using consensus bias and the global consensus index: Consensus Bias Calculation: The first Consensus bias of individual indicators The formula representing the degree of deviation between the evaluation information of this indicator and the overall evaluation information is: ； In the formula, For the first Indicators, the first Evaluation value of the road segment; The mean of the global evaluation information, the first Evaluation value of the road segment; This represents the total number of road segments; The total number of indicators; Global Consensus Index Calculation: Global Consensus Index The weighted average of the consensus deviations for each indicator is obtained by weighting the indicators by their respective weights. It represents the overall consensus level of the entire evaluation system, and the formula is as follows: ； In the formula, For the first The weight of each indicator; Termination condition: Set consensus identification coefficient If the consensus deviation of all indicators meets the following conditions: If the interaction ends prematurely, it will terminate early; otherwise, iterate until... .

8. The asphalt pavement maintenance evaluation method based on Bayesian sparse trust network according to claim 1, characterized in that, In step 7, the specific method is as follows: Based on the converged optimal posterior weights, the comprehensive evaluation results of the road segment are calculated, and a quantitative evaluation conclusion is output: Step 7.1: Calculation of Comprehensive Evaluation Value of Road Section The weighted summation method is used to calculate the comprehensive evaluation value of each road segment. The formula is as follows: ； In the formula, For the first The comprehensive evaluation value of each road segment; For the first The optimal posterior weights of each indicator; For the first The section of road, the first Normalized values ​​of each indicator; For road segment indexing; For indexing indicators; Step 7.2, Calculation of the probability distribution of the grade Based on the comprehensive evaluation value and the evaluation grading criteria (Excellent, Good, Average, Below Average, Poor), the posterior probability of each grade is calculated using the following formula: ； In the formula, For the first The road section belongs to the grade The posterior probability; The evaluation grades are 1 = Excellent, 2 = Good, 3 = Average, 4 = Poor, and 5 = Very Poor. For level The median values ​​are: Excellent = 95, Good = 85, Average = 75, Difficult = 65, Poor = 50. For level Standard deviation, empirical value ; For all rating levels, To sum over 5 levels; Step 7.3, Output Results Basic outputs: comprehensive evaluation value, probability distribution of grade, and final evaluation grade for each road segment. The highest level; Risk output: Uncertain probability of each road segment ; Auxiliary outputs: posterior mean, standard deviation and 95% confidence interval of indicator weights, core indicator identification results, and indicators with weights ≥ 0.1.

Images