Risk assessment method and system for bridge superstructure safety and storage medium

Abstract

The invention relates to a bridge superstructure safety risk assessment method and system and a storage medium. The method comprises the following steps: acquiring disease score deduction information of a technical condition of a bridge superstructure; obtaining bridge superstructure technical condition scoring data of the sample bridge, bridge superstructure score-deducted indexes and disease classification categories of the score-deducted indexes, and analyzing common indexes in the score-deducted indexes according to the score-deducted indexes; calculating a deduction value of the disease classification category of the common index at each scale based on an LSHADE algorithm, and calculating a safety risk score of the bridge superstructure of the sample bridge; dividing safety risk levels; constructing a bridge safety risk evaluation model by adopting a CatBoost ensemble learning method, and training the bridge safety risk evaluation model by utilizing a frequency reduction coefficient and the like of a sample bridge; and evaluating the safety risk level of the bridge superstructure by using the trained bridge safety risk evaluation model. The method is high in evaluation efficiency, low in cost and high in accuracy.

Description

Technical Field

[0001] This disclosure relates to the fields of machine learning and structural engineering technology, and in particular to a risk assessment method, system and storage medium for the safety of bridge superstructures. Background Technology

[0002] Bridges, as key nodes in transportation engineering, bear the heavy responsibility of maintaining smooth traffic flow and driving safety. With increasing service life, highway bridge structures inevitably suffer from the combined effects of environmental erosion, vehicle loads, and material aging, leading to a gradual degradation of their safety performance. Addressing this critical issue, how to achieve accurate and rapid assessment of bridge safety has become a current research hotspot and challenge.

[0003] Existing bridge safety evaluation methods mainly rely on regular manual inspections and limited monitoring means. Although these methods played an important role in the early operation and maintenance of bridges, with the increasing complexity of bridge structures, the continuous expansion of their scale, and the extension of their service life, it is difficult to achieve a comprehensive and accurate assessment of bridge safety by relying solely on manual inspections and basic monitoring data, thus limiting the scientific nature and efficiency of bridge safety management.

[0004] In terms of regular bridge inspections, my country mainly relies on the bridge technical condition assessment methods given in the standard "Standard for Technical Condition Assessment of Highway Bridges" (JTGTH21-2011) (hereinafter referred to as the Standard) to score and classify bridges. However, this method has obvious limitations: on the one hand, the assessment process is highly dependent on the apparent defects information obtained by manual visual inspection, which is highly subjective and makes it difficult to ensure the consistency of results; on the other hand, the assessment results only stay at the technical condition level and cannot effectively reflect the actual mechanical performance of the bridge, resulting in an inaccurate assessment of structural safety.

[0005] To address the problems existing in technical condition assessment methods, previous studies have explored aspects such as assessment index extraction and evaluation method improvement, aiming to enhance the objectivity of assessment results and their ability to reflect structural safety.

[0006] For example, the paper "On-site Inspection and Technical Condition Assessment of Highway Bridges" ([Shao Penglei. On-site Inspection and Technical Condition Assessment of Highway Bridges [D]. Zhengzhou University, 2016.]) organically combines the Analytic Hierarchy Process (AHP) and group decision-making method to establish a hierarchical model for evaluating the technical condition of beam bridges and proposes an improved method for assessing the technical condition of in-service bridges. However, although this method scientifically utilizes the AHP and group decision-making method to improve the relationship between technical condition assessment and safety, it is inevitably affected by the subjective judgment of experts, which may lead to certain differences in the bridge assessment results under different review groups or application scenarios.

[0007] For example, Chinese patent CN112668149A proposes a "parametric structural modeling and intelligent evaluation system for the technical condition of beam bridges". This system integrates on-site data collection and back-end evaluation, simplifies the operation process by using parametric modeling, and considers the evolution trend of defects in the scoring, taking into account both the evaluation standards and the actual development status, thereby improving the scientificity and operability of the evaluation.

[0008] For example, Chinese patent CN109102016A proposes "A Test Method for Bridge Technical Condition". By utilizing Long Short-Term Memory Recurrent Deep Network (LSTMRNN), a bridge technical condition prediction model that can automatically correct the weights of influencing factors is constructed, thus solving the problem that current bridge technical condition prediction methods are not accurate enough.

[0009] For example, Chinese patent CN118606847A proposes a "Method and System for Predicting the Technical Condition of Bridges Based on Machine Learning". This method involves preprocessing bridge inspection data, selecting features, and balancing the data. It uses random forest to evaluate the importance of features, combines the SMOTETomek algorithm to solve class imbalance, and finally uses the XGBoost model to predict the technical condition level of the bridge.

[0010] For example, Chinese patent CN107341282A proposes an "Improved Bridge Deterioration Assessment Method Based on the Technical Condition of the Previous Year". By coupling four key parameters, namely "technical condition score when the bridge was built", "time without deterioration", "statistical life of similar bridges", and "operation and use time", a model for the deterioration process of bridge technical condition that can describe the effects of environmental, load and material changes is established.

[0011] Although the above studies have introduced various improved methods in the assessment of bridge technical condition, enhancing the scientific rigor and convenience of the assessment, they still mainly rely on external defects or historical condition data to construct assessment models. This makes it difficult to directly reflect changes in structural mechanical performance, causing the assessment results to often focus on surface defect characteristics while neglecting internal stress mechanisms, making it difficult to achieve a comprehensive and accurate quantification of bridge safety.

[0012] In recent years, bridge monitoring technologies based on sensor networks and big data analysis have been gradually applied to bridge safety management. By collecting dynamic data such as bridge stress, vibration, and deformation, and combining them with long-term static data such as material aging and environmental impact, not only has comprehensive monitoring of the bridge's operational status been achieved, but also key data support has been provided for risk assessment.

[0013] For example, the paper "Data Storage and Early Warning Method for Bridge Health Monitoring Based on Big Data" published in the journal Science Technology and Engineering ([Ren Pu, Ding Youliang, Li Yadong, et al. Data Storage and Early Warning Method for Bridge Health Monitoring Based on Big Data[J]. Science Technology and Engineering, 2019, 19(12):5.DOI:CNKI:SUN:KXJS.0.2019-12-039.]) uses a multi-factor analysis method to mine the correlation between real-time sensor data, construct a bridge service performance evaluation model, and realize real-time early warning of safety status.

[0014] For example, the invention patent with publication number CN114282398A proposes "A Bridge Health Monitoring System and Method Based on Big Data". By integrating monitoring data, engineering data and social big data, and using tools such as WEKA for data mining, it can realize real-time damage prediction, model correction and multi-level early warning of bridges.

[0015] For example, the invention patent with publication number CN117648734A proposes "A Smart Monitoring Method and Evaluation System for Bridge Health". It combines theoretical simulation and monitoring point optimization to construct a monitoring and evaluation model. Based on strain deviation analysis, it realizes structural state identification and damage prediction, thereby improving the accuracy and intelligence level of bridge health assessment.

[0016] For example, the invention patent with publication number CN118549532A proposes "A method and device for monitoring the safety of highway bridges", which realizes real-time monitoring and safety assessment of bridge operation status by automatically collecting key structural response data and combining it with a time-varying bearing capacity calculation model.

[0017] However, existing methods for assessing the safety risks of bridge superstructures still suffer from low efficiency and require improvement in accuracy. Furthermore, for the large number of small- and medium-span bridges, current safety risk assessment methods are costly, hindering their large-scale application. Summary of the Invention

[0018] Therefore, it is necessary to provide a risk assessment method, system, and storage medium for the safety of bridge superstructures to address the problems of low assessment efficiency, low accuracy, and high cost.

[0019] To solve the above problems, the present disclosure adopts the following technical solution: In a first aspect, this disclosure provides a risk assessment method for the safety of bridge superstructures, comprising the following steps: Obtain the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The system acquires the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deducted indicators for the sample bridge. Based on this, it analyzes the common indicators among the deducted indicators of the bridge superstructure. Using the LSHADE algorithm, it calculates the deduction values ​​of the defect classification categories of the common indicators at each scale. Based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale, it calculates the safety risk score of the bridge superstructure of the sample bridge. Based on the safety risk scores of the superstructure of the sample bridges, a clustering algorithm was used to classify the safety risk levels. A bridge safety risk assessment model is constructed using the CatBoost ensemble learning method. The model is trained by using the frequency reduction coefficient of the sample bridge, the structural parameters of the superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge. The output of the model is the safety risk level of the bridge superstructure. The safety risk level of the bridge superstructure is assessed using the trained bridge safety risk assessment model.

[0020] In a preferred embodiment, the step of calculating the deduction value of the common index's disease classification category on each scale based on the LSHADE algorithm includes: using the fitness function of maximizing the correlation between the frequency reduction coefficient of the sample bridge and the safety risk score of the bridge superstructure, using the LSHADE algorithm to calculate the deduction value correction coefficient of the common index's disease classification category, and calculating the deduction value of the common index's disease classification category on each scale based on the disease deduction information of the technical condition of the bridge superstructure and the deduction value correction coefficient.

[0021] In a preferred embodiment, the specific steps for training the bridge safety risk assessment model using the frequency reduction factor of the sample bridge, the structural parameters of the superstructure of the sample bridge, the defect classification categories of common indicators of the sample bridge, the deduction values ​​of the defect classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge include: Identify the main disease classification categories among the common disease classification categories of the sample bridges; Based on the main disease categories, the deduction values ​​of the main classification categories of common indicators of sample bridges, the frequency reduction coefficient of sample bridges corresponding to the main disease classification categories, the structural parameters of the superstructure of sample bridges, and the safety risk level of the superstructure of sample bridges, a bridge safety risk assessment model is trained.

[0022] In a preferred embodiment, the specific steps of training the bridge safety risk assessment model further include: evaluating and validating the ensemble learning model using model evaluation metrics, including accuracy, precision, recall, and F1 score.

[0023] In a preferred embodiment, the bridge dynamic characteristic parameters include the first-order natural frequency of the damaged bridge and the first-order natural frequency of the intact bridge; the bridge superstructure structural parameters include the cross-sectional shape of the bridge superstructure; and the frequency reduction factor of the sample bridge is the ratio of the first-order natural frequency of the damaged bridge to the first-order natural frequency of the intact bridge.

[0024] In a preferred embodiment, the clustering algorithm is a K-means clustering algorithm, and the number of categories in the K-means clustering algorithm is 4.

[0025] In a preferred embodiment, the analytic hierarchy process (AHP) includes: providing at least two judgment matrices for the original technical condition defect deduction information, wherein the elements of the judgment matrices represent the ratio of the impact of two technical condition assessment indicators on the bridge's technical condition; performing consistency checks on each judgment matrix; determining the weight vector of the judgment matrix for those that meet the consistency check; and determining the weight value of each technical condition assessment indicator based on the weight vector. The group decision method includes: calculating a reliability coefficient based on the weight vector obtained from the AHP; specifically, the numerical values ​​calculated by the AHP and group decision method are numerical values ​​calculated based on the weight values, reliability coefficients, and the original technical condition defect deduction information.

[0026] In a preferred embodiment, the loss function for training the bridge safety risk assessment model is:

[0027] in, Represents the total loss function; The total number of training samples; Indicates the Loss per training sample; Indicates the The input feature vector of each training sample; For the first The true value of each training sample; For the first The model's predicted values ​​for each training sample; Indicates the total number of decision trees; Indicates the sequence number of the decision tree; Indicates the A decision tree, For the first The regularization term of a decision tree.

[0028] Secondly, this disclosure provides a risk assessment system for the safety of bridge superstructures, including, The acquisition module is used to acquire the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The calculation module is used to acquire the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deduction indicators of the sample bridge, and analyze the common indicators among the bridge superstructure deduction indicators accordingly; it is used to calculate the deduction values ​​of the defect classification categories of the common indicators at each scale based on the LSHADE algorithm, and to calculate the safety risk score of the bridge superstructure of the sample bridge based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale. The risk level classification module is used to classify the safety risk level of the sample bridges based on the safety risk score of the bridge superstructure and using a clustering algorithm. The model building and training module is used to build a bridge safety risk assessment model using the CatBoost ensemble learning method. It is used to train the bridge safety risk assessment model using the frequency reduction coefficient of the sample bridge, the structural parameters of the superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge. The output of the model is the safety risk level of the bridge superstructure. The safety risk assessment module is used to assess the safety risk level of the bridge superstructure using a trained bridge safety risk assessment model.

[0029] Thirdly, this disclosure provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform a risk assessment method for the safety of a bridge superstructure as described in the first aspect.

[0030] This disclosure presents a method, system, and medium for risk assessment of bridge superstructure safety. Based on the technical condition deduction information obtained from the analytic hierarchy process (AHP) and group decision-making method, common indicators and deduction categories are obtained from sample bridges. The LSHADE algorithm is used to accurately calculate the deduction values ​​of the deduction categories of the common indicators at each scale, thus obtaining the safety risk scoring rule. Based on this rule, the safety risk score of the sample bridge is calculated. Safety risk levels are then classified according to the calculation. A frequency reduction coefficient is introduced to train a bridge safety risk assessment model constructed using the CatBoost ensemble learning method, resulting in a bridge safety risk assessment model capable of outputting safety risk levels. Risk assessment is then performed using this model. This design makes the assessment efficient and cost-effective. The entire process design improves the accuracy of safety risk assessment, making it fast and cost-effective for a large number of small and medium-span bridges. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating a method in one embodiment of the present disclosure; Figure 2 This is a statistical chart of common defects in the superstructure of a bridge, as shown in one embodiment of this disclosure. Figure 3 This is a graph showing the convergence curve of the optimal parameters in one embodiment of this disclosure; Figure 4 This is a distribution diagram of the optimal combination of correction coefficients for each deduction value in one embodiment of this disclosure; Figure 5 This is a classification boundary diagram of the safety risk level of the bridge superstructure in one embodiment of this disclosure; Figure 6 This is a schematic diagram of the confusion matrix for evaluating the effect in one embodiment of this disclosure; Figure 7 This is a schematic diagram of the ROC curve for evaluating the effect in one embodiment of this disclosure; Figure 8 This is a schematic diagram of the system structure in one embodiment of the present disclosure. Detailed Implementation

[0032] The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0033] Terminology Explanation: Technical condition assessment indicators are a series of metrics used to evaluate the performance of bridge structures, components, or the overall structure. They can be understood as damage categories considered in bridge safety risk assessments, reflecting the current condition of the bridge and potential future performance changes. For the bridge superstructure, which includes load-bearing components, general components, and supports, there are 11 technical condition assessment indicators for both load-bearing and general components: ① honeycombing / pitting, ② spalling / corner chipping, ③ voids / cavities, ④ concrete cover thickness, ⑤ steel reinforcement corrosion, ⑥ concrete carbonation, ⑦ concrete strength, ⑧ mid-span deflection, ⑨ structural displacement, and ⑩ damage to prestressed components. Cracks in the main beam. The technical condition assessment indicators for the aforementioned supports total three: ① aging and deterioration, cracking; ② defects, bulging; ③ positional movement, voids. It is understood that this document contains the aforementioned 25 technical condition assessment indicators, but the number is not limited to 25, nor is the content of the indicators limited to the examples below. These can be varied according to the technical condition assessment standards and / or other actual circumstances.

[0034] Scales are standards used to classify and assess the technical condition of bridges, and are also applicable to bridge safety risk assessment. In this article, scales refer to technical condition assessment indicators, common indicators, and defect classification categories. Commonly used scales range from 1 to 5, referred to as Category 1, Category 2, Category 3, Category 4, and Category 5, respectively. It is understood that the specific scale divisions may change due to standard updates. Each technical condition assessment indicator has its corresponding number of scales. Most technical condition assessment indicators have 5 scales, but some only have 4, and some have 3. For example, the indicator "honeycomb, pitted surface" has 3 scales, only having scales "1", "2", and "3", i.e., Category 1, Category 2, and Category 3. Similarly, the indicator "peeling, corner chipping" has 4 scales, only having scales "1", "2", "3", and "4", i.e., Category 1, Category 2, Category 3, and Category 4. Since the common indicators and disease classification categories in this article are based on technical condition assessment indicators, the scaling explanations and examples of technical condition assessment indicators in this paragraph also apply to common indicators and disease classification categories.

[0035] See Figure 1 This embodiment provides a risk assessment method for the safety of bridge superstructure, which includes the following steps: Obtain the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The system acquires the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deducted indicators for the sample bridge. Based on this, it analyzes the common indicators among the deducted indicators of the bridge superstructure. Using the LSHADE algorithm, it calculates the deduction values ​​of the defect classification categories of the common indicators at each scale. Based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale, it calculates the safety risk score of the bridge superstructure of the sample bridge. Based on the safety risk scores of the superstructure of the sample bridges, a clustering algorithm was used to classify the safety risk levels. A bridge safety risk assessment model is constructed using the CatBoost ensemble learning method. The model is trained by using the frequency reduction coefficient of the sample bridges, the structural parameters of the superstructure of the sample bridges, the disease classification categories of common indicators of the sample bridges, the deduction values ​​of the disease classification categories of common indicators of the sample bridges, and the safety risk level of the superstructure (safety risk score) of the sample bridges. The output of the model is the safety risk level of the bridge superstructure. The safety risk level of the bridge superstructure is assessed using the trained bridge safety risk assessment model.

[0036] The risk assessment method and its effectiveness for the safety of the bridge's superstructure are described in detail below. (See below) Figure 2 The method includes: Step 1: Obtain the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The acquisition in step one can be done directly or through calculation.

[0037] The analytic hierarchy process includes: providing at least two judgment matrices for the original technical condition deduction information, wherein the elements of the judgment matrices represent the ratio of the impact of two technical condition assessment indicators on the bridge's technical condition; performing consistency checks on each judgment matrix; determining the weight vector of the judgment matrix for judgment matrices that meet the consistency check; and determining the weight value of each technical condition assessment indicator based on the weight vector. The group decision-making method includes: calculating the credibility coefficient based on the weight vector obtained by the analytic hierarchy process; The numerical values ​​calculated using the Analytic Hierarchy Process (AHP) and group decision-making method are specifically calculated based on weight values, reliability coefficients, and the original technical condition defect deduction information. Specifically, the weight values ​​and reliability coefficients of the technical condition assessment index are multiplied to obtain the first product. All first products of the technical condition assessment index are summed to obtain the deduction coefficient for that index. This deduction coefficient is then multiplied by the corresponding deduction value in the original technical condition defect deduction information (referred to as the original deduction value) to obtain the deduction value in the defect deduction information for the obtained bridge superstructure technical condition. Typically, the defect deduction information for the bridge superstructure technical condition is presented in tabular form.

[0038] Specifically, the paper "On-site Inspection and Technical Condition Assessment of Highway Bridges" was used to obtain information on the defects and deductions of the technical condition of the bridge superstructure.

[0039] In step one, the results of the paper "On-site Inspection and Technical Condition Assessment of Highway Bridges" can be directly obtained as the deduction information for the technical condition of the bridge superstructure. Alternatively, the deduction information can be calculated using the analytic hierarchy process (AHP) and group decision method based on existing deduction values ​​(this method is described in the paper "On-site Inspection and Technical Condition Assessment of Highway Bridges"). Preferably, the existing deduction values ​​are those recorded in the standard "Standard for Technical Condition Assessment of Highway Bridges"; it is understood that the source of the existing deduction values ​​is not limited.

[0040] The analytic hierarchy process (AHP) mainly includes the following steps: First, for each technical condition assessment indicator in the standard "Standard for Technical Condition Assessment of Highway Bridges," at least two judgment matrices are provided. Specifically, each element of the judgment matrix represents the ratio of the importance of two technical condition assessment indicators to the bridge's technical condition assessment, i.e., the ratio of their impact on the bridge's technical condition. Except for the diagonal elements, all elements represent the ratio of the importance of two different technical condition assessment indicators to the bridge's technical condition assessment; that is, the diagonal equals 1, meaning the ratio of the importance of the same technical condition assessment indicator to the bridge's technical condition assessment is 1. Different judgment matrices are determined based on the opinions of different experts. Then, a consistency check is performed on each judgment matrix to determine if it conforms to consistency. If not, the corresponding judgment matrix is ​​redefined, or the judgment matrix is ​​discarded. Finally, for the judgment matrices that conform to the consistency check, a weight vector is determined. The weight vector is a column vector. Based on the weight vector, the weight value of each technical condition assessment indicator is determined. Specifically, all weight vectors are summarized, and the weight value of each technical condition assessment indicator is obtained accordingly. Here, the number of experts required to achieve consistency in the judgment matrices of the upper load-bearing components, the upper general components, and the supports is [number missing]. .

[0041] Example 1: An expert's judgment matrix for the upper load-bearing component, as shown in the following formula:

[0042] The first column, fourth row represents the ratio of the importance of the "concrete cover thickness" indicator to the bridge risk assessment to the importance of the "honeycomb and pitting" indicator to the bridge technical condition assessment.

[0043] The group decision-making method includes: obtaining the results for the upper load-bearing component / upper general component / support respectively based on the analytic hierarchy process. For simplicity, the weight vectors are not differentiated between the weight vectors of upper load-bearing components, upper general components, or supports. The weight vector is defined as follows: ,in, , This represents the total number of technical condition assessment indicators for the superstructure of bridges, including the upper load-bearing components, general components, and supports. Indicates the Experts The first of the experts (position) of the superstructure of the bridge, specifically the superstructure of the upper load-bearing members / general members / supports. The deduction value (judgment value) for each technical condition assessment indicator. Indicates the The sum of the standard deviations of the evaluation indicators for each technical condition of the superstructure load-bearing components, superstructure general components, and supports by the experts:

[0044] Will Normalization yields the coefficient of difference , This indicates the differences in the assessment of a certain technical condition indicator for the upper load-bearing component / upper general component / support, i.e., the... The difference coefficient of the weight vector of the expert, that is, the first expert There are three coefficients of difference among the experts, corresponding one-to-one with the upper load-bearing components, the upper general components, and the supports.

[0045] Then, similarity calculations are performed, assuming any two weight vectors among the upper load-bearing component / upper general component / support are... and , find the first Weight vectors and the Weight vectors The angle between the two experts (the first and second experts) is expressed by the cosine of the angle. The position and the first The similarity judged by (a number of experts), using Indicates the Similarity among experts, similarity Indicates the The sum of the cosine values ​​of the weight vectors of the individual expert and all other experts, minus 1. Normalization is performed. As a similarity coefficient, used Indicates the The similarity coefficient of the weight vectors of the superstructure load-bearing components / superstructure general components / supports by the experts, that is, the first... There are three similarity coefficients among the experts, corresponding one-to-one with the upper load-bearing components, upper general components, and supports.

[0046] Let the first The experts' reliability coefficient for the upper load-bearing components / upper general components / supports is: ,but for:

[0047] Obviously, the first Credibility coefficient of the experts There are three in total, corresponding one-to-one with the upper load-bearing components, the upper general components, and the supports.

[0048] The weight values ​​of the technical condition assessment indicators obtained by the analytic hierarchy process are multiplied by all the confidence coefficients obtained by the group decision method, and the products are added together to obtain the deduction coefficient of the technical condition assessment indicator. Multiplying the deduction coefficient by the corresponding original deduction value yields the new deduction value. Finally, the deduction information of all technical conditions is obtained. As an example, the deduction information of the bridge superstructure obtained in this step is shown in Table 1.

[0049] Table 1

[0050] Continued from Table 1

[0051] Step 2: Obtain bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and defect classification categories of the deducted indicators from multiple sample bridges. Analyze the common indicators among the deducted indicators of the bridge superstructure. Based on the defect deduction information of the bridge superstructure's technical condition, use the frequency reduction coefficient of the sample bridge and the maximum correlation between the frequency reduction coefficient and the bridge superstructure safety risk score as the fitness function. Utilize the LSHADE algorithm to calculate the deduction value correction coefficient for the defect classification categories of common indicators. Based on the defect deduction information and the deduction value correction coefficient, calculate the deduction value of the defect classification category of the common indicators at each scale. Establish safety risk defect deduction information based on the defect classification categories of the common indicators and their deduction values ​​at each scale. Calculate the safety risk score of the bridge superstructure of the sample bridges based on the safety risk defect deduction information. The disease deduction information for safety risks includes the disease classification categories of common indicators, the deduction values ​​of the disease classification categories of common indicators on each scale, and in some embodiments, common indicators may also be included.

[0052] It is understood that the deducted indicators are technical condition indicators subject to deduction; common indicators are those frequently deducted from the deducted indicators, also known as common deducted indicators. Common indicators refer to technical condition assessment indicators whose probability of occurrence is greater than a certain preset probability threshold in the technical condition scores of the superstructure of all sample bridges. It is understood that the defect classification categories of common indicators serve as indicators in the safety risk scoring rules.

[0053] Here, it is possible that one or more common indicators among the deducted indicators do not have at least two specific disease classification categories, and the disease classification category of the common indicator is itself.

[0054] In this embodiment, step two involves adjusting the deduction values ​​for the disease classification categories of all common indicators. In some embodiments, the deduction values ​​are only adjusted for the disease classification categories of common indicators that have at least two specific disease classifications (i.e., the disease classification is not itself).

[0055] In this embodiment, the bridge dynamic characteristic parameters include the theoretical natural frequencies and the measured natural frequencies of the bridge (i.e., the first-order natural frequencies of an intact bridge and the first-order natural frequencies of a damaged bridge), and the bridge superstructure structural parameters include the cross-sectional shape of the bridge superstructure. It is understood that different defect classifications have different impacts on structural safety. The defect classification category of the deducted indicator is the classification of the deducted indicator, including causal classification, orientation classification, and / or morphological classification. For example, the classification of cracks in the main beam includes transverse cracks, longitudinal cracks, diagonal cracks, and network cracks. The sample bridge is a bridge with technical condition scoring data for its superstructure, used as a sample for calculating the deduction correction coefficient using the LSHADE algorithm.

[0056] Using the maximization of the correlation between the frequency reduction coefficient of the sample bridges and the safety risk score data of the bridge superstructure as the fitness function, the optimal combination of deduction value correction coefficients is iteratively solved using the LSHADE algorithm (Linear Population Size Reduction Adaptive Differential Evolution, an improved differential evolution (DE) algorithm). Based on the deduction information of the bridge superstructure defects and the optimal deduction value correction coefficients, the corrected deduction values ​​of the defect categories of common indicators at each scale are obtained. In other words, based on the relevant information of the technical condition score, the deduction values ​​of the defect categories of common indicators are corrected to obtain the deduction values ​​used for safety risk assessment, thus obtaining the safety risk scoring system.

[0057] The following is a specific example of step two. As an embodiment, the sample bridge comes from the bridge digital management platform developed by Guangxi Jiaoke Group Co., Ltd. At a certain time period, the platform contains periodic inspection data for 3202 bridges, mainly including bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition scoring data, deducted indicators for the bridge superstructure, and the defect classification categories of the deducted indicators. Statistical results of some common deducted indicators for the superstructure of large, medium, and small bridges are shown below. Figure 2 As shown, Figure 2The first column corresponds to the main beam crack, and the second column's "surface defects" correspond to "stripping, corner chipping," "voids, holes," "honeycomb, pitting," and "support defects" correspond to the support's "defects." According to the statistics of the bridge digital management platform, for common reinforced concrete beam bridges: for the superstructure load-bearing components, D11 "main beam cracks," D2 "stripping, corner chipping," D3 "voids, holes," and D5 "reinforcement corrosion" are the most common deduction indicators, which are common defects; the superstructure general components mainly include diaphragms, wet joints, and hinge joints, with D12 "honeycomb, pitting," D13 "stripping, corner chipping," D14 "voids, holes," D16 "reinforcement corrosion," and D22 "main beam cracks" being the most common deduction indicators; for plate support defects, D23, D24, and D25 are all common deduction indicators.

[0058] Currently, the standards specify the same deduction values ​​for different types of cracks. However, cracks can be divided into structural cracks and non-structural cracks. Further, the classification categories mainly include transverse cracks, longitudinal cracks, diagonal cracks, and network cracks. In reality, different types of cracks do not have entirely the same impact on structural safety. Therefore, it is necessary to redefine the deduction value correction coefficient and the deduction value for safety risk scoring for different types of cracks. To more objectively correct the deduction values ​​for various defect classification categories, a frequency reduction coefficient is introduced (…). This parameter reflects the safety of the bridge, and the frequency reduction factor is defined as follows:

[0059] in, , where is the first-order natural frequency of the bridge after damage, and is the frequency obtained from the actual bridge dynamic characteristic test, i.e., the actual natural frequency. is the first-order natural frequency of the intact bridge structure, and is the theoretical natural frequency obtained from numerical simulation.

[0060] To reduce computational load, when calculating the safety risk score of the superstructure of the sample bridges, a subset of sample bridges can be selected, with the requirement that the selected subset of bridges should cover as many technical condition levels as possible.

[0061] Furthermore, the number of sample bridges and the sample bridges used to determine common indicators are not required to be the same as the number of sample bridges used to calculate the deduction values ​​of the common indicators' disease classification categories on each scale. Usually, the sample bridges used to calculate the deduction values ​​are a subset of the sample bridges used to determine common indicators. When calculating the safety risk score of the superstructure of the sample bridges based on the deduction values, the number of sample bridges is not limited, nor is it limited to whether they are from the sample bridges used to determine common indicators and calculate the deduction values ​​of the common indicators' disease classification categories on each scale. Preferably, the technical condition score data of the superstructure of all sample bridges used in this step (calculating the safety risk score of the superstructure of the sample bridges) can cover all technical condition levels or cover the first four technical condition levels. The existing technical condition levels are divided into 5 levels (bridge technical condition assessment levels are divided into five categories: Category I, Category II, Category III, Category IV, and Category V), which is the "Technical Condition Level Classification" in the standard "Highway Bridge Technical Condition Assessment Standard" (JTGT H21-2011).

[0062] Here, 239 bridges were selected from 3202 bridges in the bridge digital management platform for dynamic characteristic testing. These included 92 Class I bridges, 76 Class II bridges, 44 Class III bridges, and 27 Class IV bridges. Based on Table 1 from Step 1, the common deduction indicators' defect categories and corresponding initial deduction values ​​were obtained, as shown in Table 2. For the superstructure, general superstructure, and supports, there are a total of 19 common deduction indicators' defect categories. Table 2 shows these 19 defect categories, their codes, and the deduction values ​​before LSHADE algorithm correction.

[0063] Table 2

[0064] An optimization model for the deduction value correction coefficient is constructed using the LSHADE differential evolution algorithm. By setting a fitness function, the deduction values ​​of the common indicators' disease classification categories on each scale are compared with the... The optimization objective is to maximize the correlation coefficient (Pearson correlation coefficient, i.e., Spearman correlation coefficient) between the two factors. The optimal combination of deduction correction coefficients is obtained through multiple runs. The specific steps are as follows: (1) For each component of the bridge superstructure, a defect classification deduction value matrix, also known as the bridge defect quantitative scoring matrix, is constructed. The row vectors represent the defect classification categories, and the column vectors correspond to the deduction values ​​at different scales. According to Table 2, the bridge defect quantitative scoring matrix is ​​constructed using a matrix representation method, with each column vector representing a different defect category. , , A matrix of deduction values ​​for the disease classification categories of the upper load-bearing components, upper general components, and supports is shown below, representing a one-to-one correspondence.

[0065]

[0066]

[0067]

[0068] (2) For the initial disease classification deduction value matrix, a deduction value correction coefficient is assigned to each disease classification category, which is called the initial disease classification deduction value matrix. The deduction value correction coefficients for the disease classification categories of the upper load-bearing components are as follows: , , , , , 、 , The correction coefficients for the deduction values ​​of the general component defects classification are as follows: , , , , , 、 , The correction coefficients for the deduction values ​​of the bearing defects classification categories are as follows: , , Since the focus of this study is to adjust the deduction values ​​of different types of cracks, referring to the deduction value setting for non-structural cracks in the "Technical Condition Assessment Standard for Highway Bridges (Draft for Comments)" (JTG H21-2017), the range of the deduction value correction coefficient for crack disease classification is set to [0.5, 1.5]. At the same time, in order to make appropriate corrections to the deduction value correction coefficients for other disease classifications, the range of the deduction value correction coefficient is limited to [0.8, 1.2]. Understandably, the following (3) also follows this setting range.

[0069] (3) For the bridge superstructure, according to the calculation process of bridge technical condition scoring in the specification, the safety risk score of the sample bridge is quickly calculated using Matlab software. That is, the score of the safety risk score of the bridge superstructure is calculated based on the deduction value of the disease classification category of the common indicators at each scale, and the safety risk score is obtained. The initial safety risk score is obtained based on the initial deduction value matrix of the disease classification category.

[0070] The calculation process is as follows: First, sort the deduction values ​​of the disease classification categories according to the size of the deduction values, and then calculate the deduction value and score of each component of the sample bridge. Then, the score for each component is calculated based on the number of components. Finally, the score of the entire superstructure is determined based on the weight of the components.

[0071] The main calculation formulas are as follows:

[0072]

[0073]

[0074]

[0075] in, This indicates the component number of the bridge superstructure. The values ​​are 1, 2, and 3. The value 1 corresponds to the upper load-bearing component. The value 2 corresponds to the upper general component. The value 3 corresponds to a support; express The component number on a class of components indicates the first... One component; and All indicate the first Class Components The classification of common disease categories on each component is understandable. It is 1 to Intermediate variables; For the first Class Components The first component Deduction values ​​for each disease category; Indicates the Class Components On the first component The deduction value for each disease category. Indicates the Class Components On the first component The conversion deduction value for each disease classification category; Indicates the Class Components The total number of disease classification categories for common indicators on each component; For the bridge superstructure Class Components Safety risk score for each component; For the bridge superstructure Safety risk score for this type of component; For the first The average safety risk score of each component in the class of components; For the first The minimum safety risk score of each component in a class of parts; For the first A coefficient that determines the number of components in a given type of component.

[0076] (4) Based on the initial safety risk score of each sample bridge calculated in (3), the control parameters (scaling factor) are dynamically adjusted using the LSHADE algorithm through a historical memory mechanism. F and crossover probability CR The adaptive population reduction strategy effectively balances global exploration and local exploitation capabilities, significantly improving the convergence speed and solution accuracy of complex optimization problems. First, the deduction value correction coefficient vector is set as follows:

[0077] Initially, the deduction value correction coefficient vector This is the vector of parameters to be optimized.

[0078] Given the first Features of individual sample bridges The output score of the model equipped with the LSHADE algorithm is :

[0079] in, Indicates the Safety risk score of the superstructure of the sample bridge. The function representing the calculation of the safety risk score of the bridge superstructure. This represents the total number of bridges in the sample. Indicates the sample bridge number, i.e., the first bridge. One sample bridge, features This includes information on the classification of defects and the number of components of the sample bridges. In this embodiment, 239 bridges are used as sample bridges.

[0080] Record of actual measurements The sequence is , Indicates the A sample bridge After removing invalid bridge samples, calculate the Spearman correlation coefficient:

[0081] in, This represents the calculated Spearman correlation coefficient. The function representing the calculation of the Spearman correlation coefficient.

[0082] The mean of the deduction value correction coefficient vector and standard deviation Regularization is defined as:

[0083]

[0084] in, This represents the total number of categories of defects in the bridge superstructure; in this embodiment, it is set to 19. The serial number indicates the category of defects in the bridge superstructure, and the number of defects in the bridge superstructure is indicated by the serial number. Disease classification categories Indicates the Correction coefficient for the deduction value of each disease classification category.

[0085] Then the above average value and standard deviation The fitness function is synthesized according to weights and is to be minimized. The calculation formula is as follows:

[0086] in, Punishment related to rank As a regularization penalty, a coefficient of 0.05 is used to balance the regularization intensity and prevent excessive oscillation of the parameter vector during the optimization process.

[0087] The fitness function is primarily optimized by minimizing the fitness function and using a global optimization method to fine-tune the bridge scoring model parameters, achieving a balance between prediction accuracy and parameter stability. The convergence condition mainly targets the optimization stability of the 19 deduction value correction coefficients, controlled by two core parameters: one is the function tolerance (Function Tolerance = 1 / e^(- ... -4 The algorithm is considered convergent when the average change of the fitness function over 50 consecutive generations is less than this threshold; secondly, it has a maximum number of iterations (MaxGenerations=10000), which forcibly limits the upper limit of the algorithm's execution to prevent infinite loops. To improve the accuracy of parameter optimization, the model was run 30 times, and the results are as follows. Figure 3 As shown, the optimal deduction correction coefficients for each disease classification type are as follows: Figure 4 As shown, Figure 4 In the chart, the blue bars represent the mean of the correction factors for each deduction value, and the black vertical lines represent the standard deviation. Because the model was run 30 times, the black vertical line represents the fluctuation range of the correction coefficient for the corresponding disease.

[0088] (5) Calculate the deduction value of the common indicators’ disease classification category at each scale based on the deduction value correction coefficient finally determined by the LSHADE algorithm. Based on the common indicators’ disease classification category and the deduction value of the common indicators’ disease classification category at each scale, recalculate the bridge superstructure safety risk score of the sample bridge.

[0089] Step 3: Determine the number of bridge safety risk levels based on relevant data and set them as the cluster number for the K-means algorithm. Perform cluster analysis on the recalculated safety risk scores of the bridge superstructure. Divide each safety risk level according to the clustering results, thus obtaining the scoring interval corresponding to each safety risk level. Based on the interval of the safety risk score of each sample bridge, complete the risk level assessment of the sample bridge, thereby constructing a safety risk level dataset for the bridge superstructure. Each safety risk level is a dataset. Each dataset includes the frequency reduction coefficient of the sample bridge, the structural parameters of the bridge superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the corrected bridge superstructure score of the sample bridge. Obviously, it also includes information on the safety risk level to which the corrected bridge superstructure (safety risk score) of the sample bridge belongs.

[0090] The safety risk level range of bridges was determined. After obtaining the revised safety risk score, the bridge safety risk assessment level and corresponding characteristics were determined by referring to the relationship between bridge technical condition assessment classification and load-bearing capacity degradation in the "Highway Bridge and Culvert Maintenance Specification" (JTG 5120-2021), and the bridge safety risk assessment level was divided into five levels. The K-means clustering algorithm was used to set the number of clusters according to the classification level to obtain the score range under different safety risk levels. Based on the revised safety risk score, the sample bridges were classified into safety levels, and a risk level dataset of the bridge superstructure was constructed. The formula for K-means clustering is:

[0091] in, This represents the total squared error. The number of clusters; For the first A cluster is a set of sample points. For the first The center of the cluster (the mean of all samples within the cluster); These are sample data points.

[0092] In this embodiment, the sample bridges mainly focus on bridges with technical condition levels one to four. Therefore, as an embodiment and not a limitation, the maximum bridge safety risk assessment level in this sample interval is set to four. Thus, the number of clusters is set to "4" during sample clustering, and K-means clustering is used for cluster analysis to obtain the scoring interval thresholds for different safety risk levels. The safety risk level classification effect is as follows: Figure 5 As shown, for the range below the minimum score of the sample, it is directly classified as Level 5 safety risk, meaning there are a total of five safety risk levels. It is worth noting that when new sample bridges with a technical condition level of five are added, the entire risk assessment range will be dynamically adjusted. The bridge safety risk assessment criteria are shown in Table 3.

[0093] Table 3

[0094] Step 4: A bridge safety risk assessment model is constructed using the CatBoost (Categorical Boosting) ensemble learning method. The model is trained using the frequency reduction coefficient of the sample bridges, the structural parameters of the superstructure of the sample bridges, the disease classification categories of common indicators of the sample bridges, the deduction values ​​of the disease classification categories of common indicators of the sample bridges, and the safety risk level to which the safety risk score of the superstructure of the sample bridges belongs. The output of the model is the safety risk level of the bridge superstructure. This step uses the CatBoost ensemble learning method to build an ensemble learning model, which is then used to evaluate the safety risk level of the bridge.

[0095] First, the Optuna optimization method (Optuna, a hyperparameter optimization framework) is used to tune the hyperparameters of the CatBoost model, including tree depth, learning rate, and regularization coefficient. The frequency reduction coefficient of the sample bridges is then used. This document describes a bridge safety risk assessment model that trains upon obtaining the following information: the structural parameters of the superstructure of the sample bridge, the classification of common defects in the sample bridge, the deduction values ​​for each defect classification, and the safety risk level to which the superstructure safety risk score of the sample bridge belongs. The bridge cross-sections in the sample bridges in this embodiment mainly include solid slab beams, hollow slab beams, box beams, and T-beams. The common defect classifications selected are transverse cracks in the superstructure load-bearing components, transverse cracks in general superstructure components, and excessive voiding of supports.

[0096] Preferably, since the number of components in different bridges is not exactly the same, in order to improve the applicability of the input parameters, the concept of defect scale ratio is proposed, and the calculation formula is as follows:

[0097]

[0098]

[0099] As can be understood, each type of component has several parts. This indicates that it has the first position among the upper load-bearing components. The number of components with similar defects. This refers to the total number of components in the upper load-bearing structure. This indicates that the upper general component has the first The number of components with similar defects. This refers to the total number of components in the upper general components. Indicates that it has the first position in the support. The number of components with similar defects. This represents the total number of components in the support.

[0100] A percentage threshold is set, and the disease classification categories with a high percentage (above the percentage threshold) are selected as the main disease classification categories. That is, among the disease classification categories of common indicators, the disease classification categories that meet the percentage threshold are selected as the main disease classification categories. The main disease classification categories replace the disease classification categories of common indicators as the input of the ensemble learning model. Dynamic characteristics are used as the main indicator and apparent diseases are used as the auxiliary indicator to determine the safety of bridges. However, there are many diseases in actual bridges, so it is necessary to screen out the diseases that have a greater impact on bridge safety to improve the simplicity of bridge assessment.

[0101] Based on the bridge superstructure risk level dataset obtained in step three, the decision tree framework is improved through iterative gradient training, and the objective function is optimized to reduce model residuals. The objective function for training the bridge safety risk assessment model is a multiclass log loss function, and the loss function for training the bridge safety risk assessment model is defined as follows:

[0102] in, Represents the total loss function; The total number of training samples; Indicates the Loss per training sample; Indicates the The input feature vector (independent variable) of each training sample; For the first The true value of each training sample; For the first The model's predicted values ​​for each training sample; Indicates the total number of decision trees; Indicates the sequence number of the decision tree; Indicates the Tree base learner, also known as the first A decision tree; For the first The regularization term of a decision tree.

[0103] This step includes training error measurement, which measures the difference between the model's predicted values ​​and the true values. For multi-class classification problems, let the common error be... If there are multiple safety risk level classifications, then the multi-class log loss function can be expressed as:

[0104] in, It is an indicator function if and only if the first... The true label of each sample The value is 1 if the condition is met, and 0 otherwise. Indicates the category index number, from 0 to ; The model represents the first The output of the nth sample The raw category score refers to the direct output of the error measurement model for each safety risk level classification category; It also represents the category index number, from 0 to... ; Model for the first The output of the nth sample Original score for each category.

[0105] The loss function can also be expressed as:

[0106] in, The model represents the first The output of the nth sample Category raw score, Indicates the Each sample is predicted as a category. The probability of.

[0107] The regularization term is used to control model complexity to avoid overfitting, and its calculation formula is as follows:

[0108] in, The penalty coefficient representing the number of leaf nodes in the decision tree; Indicates the The total number of leaf nodes in the decision tree; express L 2. Regularization weights; expressL 1. Regularization weights; Indicates the The first decision tree The output weights of each leaf node.

[0109] Then, the optimized decision tree model is integrated into the CatBoost model. During the integration process, the prediction results of multiple decision tree models are integrated through gradient boosting to obtain the final ensemble learning model. The final ensemble model can accurately evaluate the safety risk level of bridges, making the evaluation results more objective and reducing the potential risks of bridges.

[0110] Furthermore, following step four, the following steps are also included: Step 5: Evaluate and validate the ensemble learning model using model evaluation metrics, including but not limited to accuracy, precision, recall, and F1 score. The model that meets the requirements of the corresponding evaluation metrics will be used as the trained bridge safety risk assessment model.

[0111] In this embodiment, the prediction results for the model evaluation metrics are: accuracy 0.946, precision 0.945, recall 0.929, and F1 score 0.937. Additionally, the confusion matrix and ROC curve (Receiver Operating Characteristic curve) of the prediction results are as follows: Figure 6 and Figure 7 As shown, Figure 6 The gradient colors represent different accuracy rates, with darker colors indicating higher accuracy. The percentage represents the proportion of correctly classified samples out of the total number of samples. Figure 7 The graph represents the relationship between the true positive rate and the false positive rate for safety risk levels. "Average" indicates the average value of the curves for safety risk levels one, two, three, and four in the graph. It represents the overall average performance obtained after uniformly calculating the sample bridges across all safety risk levels, reflecting the model's ability to distinguish between different sample bridges. The red line segment for safety risk level three is covered by the line segment for safety risk level four. AUC (Area Under the Curve) represents the area under the ROC curve, ranging from 0 to 1. It measures the classification model's ability to distinguish between samples; a higher value indicates a better classification ability.

[0112] Step Six: Assess the safety risk level of the bridge superstructure using the trained bridge safety risk assessment model. Inputs include the frequency reduction factor of the bridge superstructure to be assessed, the structural parameters of the bridge superstructure, the main defect classification categories / common indicator defect classification categories, and the deduction values ​​for the main defect classification categories / common indicator defect classification categories. Output is the safety risk level of the bridge superstructure.

[0113] See Figure 8 A risk assessment system for the safety of bridge superstructure is provided, comprising: The acquisition module is used to acquire the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The calculation module is used to acquire the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deduction indicators of the sample bridge, and analyze the common indicators among the bridge superstructure deduction indicators accordingly; it is used to calculate the deduction values ​​of the defect classification categories of the common indicators at each scale based on the LSHADE algorithm, and to calculate the safety risk score of the bridge superstructure of the sample bridge based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale. The risk level classification module is used to classify the safety risk level of the sample bridges based on the safety risk score of the bridge superstructure and using a clustering algorithm. The model building and training module is used to build a bridge safety risk assessment model using the CatBoost ensemble learning method. It is used to train the bridge safety risk assessment model using the frequency reduction coefficient of the sample bridge, the structural parameters of the superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge. The output of the model is the safety risk level of the bridge superstructure. The safety risk assessment module is used to assess the safety risk level of the bridge superstructure using a trained bridge safety risk assessment model.

[0114] In this embodiment, the step of calculating the deduction values ​​of the disease classification categories of the common indicators on each scale based on the LSHADE algorithm includes: using the frequency reduction coefficient of the sample bridge and the maximum correlation between the bridge superstructure safety risk score as the fitness function, using the LSHADE algorithm to obtain the deduction value correction coefficient of the disease classification categories of the common indicators, and calculating the deduction value of the disease classification categories of the common indicators on each scale based on the disease deduction information of the technical condition of the bridge superstructure and the deduction value correction coefficient.

[0115] In this embodiment, the model building and training module includes: The first determining unit is used to determine the main disease classification category in the disease classification category of common indicators of the sample bridge; The training unit is used to train the bridge safety risk assessment model based on the main disease categories, the deduction values ​​of the main classification categories of common indicators of sample bridges, the frequency reduction coefficient of sample bridges corresponding to the main disease classification categories, the structural parameters of the superstructure of sample bridges, and the safety risk level of the superstructure of sample bridges.

[0116] In this embodiment, the model building and training module is also used to evaluate and validate the ensemble learning model using model evaluation metrics, which include accuracy, precision, recall and F1 score.

[0117] In this embodiment, the bridge dynamic characteristic parameters include the first-order natural frequency of the damaged bridge and the first-order natural frequency of the intact bridge; the bridge superstructure structural parameters include the cross-sectional shape of the bridge superstructure; and the frequency reduction factor of the sample bridge is the ratio of the first-order natural frequency of the damaged bridge to the first-order natural frequency of the intact bridge.

[0118] In this embodiment, the clustering algorithm is the K-means clustering algorithm, and the number of categories in the K-means clustering algorithm is 4.

[0119] In this embodiment, the analytic hierarchy process (AHP) includes: providing at least two judgment matrices for the original technical condition defect deduction information, where each element of the judgment matrix represents the ratio of the impact of two technical condition assessment indicators on the bridge's technical condition; performing consistency checks on each judgment matrix; determining the weight vector of the judgment matrix for those that meet the consistency check; and determining the weight value of each technical condition assessment indicator based on the weight vector. The group decision method includes: calculating a reliability coefficient based on the weight vector obtained from the AHP; specifically, the numerical values ​​calculated by the AHP and group decision method are numerical values ​​calculated based on the weight values, reliability coefficients, and the original technical condition defect deduction information.

[0120] In this embodiment, the loss function for training the bridge safety risk assessment model is:

[0121] in, Represents the total loss function; The total number of training samples; Indicates the Loss per training sample; Indicates the The input feature vector of each training sample; For the first The true value of each training sample; For the first The model's predicted values ​​for each training sample; Indicates the total number of decision trees; Indicates the sequence number of the decision tree; Indicates the A decision tree, For the first The regularization term of a decision tree.

[0122] In specific implementation, the risk assessment system for the safety of bridge superstructure can be implemented by referring to one of the risk assessment methods for the safety of bridge superstructure in any of the above embodiments. The specific implementation steps will not be repeated here.

[0123] An electronic device can be implemented according to the method of this disclosure, the electronic device comprising: a memory; one or more processors; one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing a risk assessment method for the safety of a bridge superstructure according to any of the above embodiments.

[0124] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0125] Memory can include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Furthermore, memory can include any combination of computer-readable storage media; memory can be a semiconductor memory chip, a magnetic disk, or an optical disk.

[0126] The memory stores executable code, which, when processed by the processor, can cause the processor to execute some or all of the methods described above.

[0127] This disclosure also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the steps of the risk assessment method for the safety of a bridge superstructure as described in any of the above embodiments.

[0128] The effects of the risk assessment method, system, and medium for the safety of bridge superstructure disclosed herein are as follows: Based on the technical condition defect deduction information obtained from the analytic hierarchy process (AHP) and group decision-making method, common indicators and defect classifications are obtained from sample bridges. The LSHADE algorithm is used to accurately calculate the deduction values ​​for each defect classification of the common indicators on each scale, thus obtaining the safety risk scoring rule. Based on this rule, the safety risk score of the sample bridge is calculated. The calculated safety risk level is then classified, and a bridge safety risk assessment model constructed using the CatBoost ensemble learning method is trained to obtain a bridge safety risk assessment model capable of outputting safety risk levels. This disclosure forms a method for assessing the safety risk of bridge superstructures. Based on accurate technical condition defect deduction information, corresponding deduction values ​​for safety risks are set, and the defect classifications of common indicators are determined. This approach balances computational efficiency and the accuracy of deduction values. A safety risk assessment model is trained, and risk assessment is performed using this model. By introducing the frequency reduction coefficient as a mechanical performance indicator, the evaluation results of the bridge are made more objective and accurate. This disclosure has high assessment efficiency and low cost, and the design of the entire process improves the accuracy of safety risk assessment. For a large number of small and medium-span bridges, safety risk assessment is quick and cost-effective, making it easy to promote and apply on a large scale.

[0129] In this disclosure, a scoring rule for safety risk analysis is obtained based on a scoring rule for the technical condition of the bridge. Specifically, based on the scoring rule for the technical condition of the bridge superstructure, an accurate scoring rule suitable for safety risk analysis is obtained by acquiring relevant information of the sample bridge and using the LSHADE algorithm, filling a gap in the prior art. This disclosure also uses the designed safety risk scoring rule to calculate the safety risk score of the sample bridge, and then classifies the safety risk level based on the calculated safety risk score; a bridge safety risk assessment model is constructed and trained, based on which the safety risk level can be quickly and accurately assessed.

[0130] Specifically, a scoring system reflecting bridge safety was constructed. Drawing on the current standards' bridge technical condition defect deduction rules, the defect deduction table was reconstructed using the analytic hierarchy process (AHP) and group decision-making method, quantifying expert subjective judgments with scientific weights. Based on this, the LSHADE differential evolution algorithm was used to further refine the deduction values ​​for common defect categories, introducing a data-driven objective optimization mechanism to form a more reasonable and accurate safety risk scoring standard. This combined subjective and objective approach comprehensively reflects the impact of defects on structural safety, provides more scientific and reasonable deduction suggestions, and offers reliable data support for bridge safety risk assessment.

[0131] Specifically, the current methods for assessing the technical condition of bridges in the existing standards are somewhat inaccurate in reflecting the actual safety risks of bridges. To address this deficiency, this disclosure introduces the concept of bridge safety risk levels, classifies safety risk levels into five levels based on relevant standards, and uses the K-means clustering algorithm to perform cluster analysis on the corrected scoring data, dividing the scoring intervals corresponding to each risk level.

[0132] Specifically, to achieve accurate and rapid evaluation of the safety of bridge superstructures, the bridge superstructure safety risk assessment model constructed in this disclosure realizes accurate assessment of structural safety based primarily on dynamic characteristics and secondarily on apparent defects. On the one hand, it effectively reduces the interference of subjective factors in traditional methods based on manual judgment; on the other hand, by introducing a frequency reduction factor as a quantitative indicator of structural performance, it can more comprehensively reflect the degree of damage and degradation of the bridge superstructure, thereby improving the accuracy and reliability of bridge superstructure safety risk assessment.

[0133] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] The embodiments described above are merely illustrative of several implementations of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this disclosure, and these all fall within the protection scope of this disclosure. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A risk assessment method for the safety of bridge superstructure, characterized in that, Includes the following steps: Obtain the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The system acquires the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deducted indicators for the sample bridge. Based on this, it analyzes the common indicators among the deducted indicators of the bridge superstructure. Using the LSHADE algorithm, it calculates the deduction values ​​of the defect classification categories of the common indicators at each scale. Based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale, it calculates the safety risk score of the bridge superstructure of the sample bridge. Based on the safety risk scores of the superstructure of the sample bridges, a clustering algorithm was used to classify the safety risk levels. A bridge safety risk assessment model is constructed using the CatBoost ensemble learning method. The model is trained by using the frequency reduction coefficient of the sample bridge, the structural parameters of the superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge. The output of the model is the safety risk level of the bridge superstructure. The safety risk level of the bridge superstructure is assessed using the trained bridge safety risk assessment model.

2. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The calculation of the deduction values ​​of the common indicators' disease classification categories on each scale based on the LSHADE algorithm includes: using the maximum correlation between the frequency reduction coefficient of the sample bridge and the safety risk score of the bridge superstructure as the fitness function, using the LSHADE algorithm to obtain the deduction value correction coefficient of the common indicators' disease classification categories, and based on the disease deduction information of the technical condition of the bridge superstructure and the deduction value correction coefficient, obtaining the deduction values ​​of the common indicators' disease classification categories on each scale.

3. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The specific steps for training the bridge safety risk assessment model using the frequency reduction factor of the sample bridge, the structural parameters of the superstructure of the sample bridge, the defect classification categories of common indicators of the sample bridge, the deduction values ​​of the defect classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge include: Identify the main disease classification categories among the common disease classification categories of the sample bridges; Based on the main disease categories, the deduction values ​​of the main classification categories of common indicators of sample bridges, the frequency reduction coefficient of sample bridges corresponding to the main disease classification categories, the structural parameters of the superstructure of sample bridges, and the safety risk level of the superstructure of sample bridges, a bridge safety risk assessment model is trained.

4. The risk assessment method for the safety of bridge superstructure according to claim 3, characterized in that, The specific steps for training the bridge safety risk assessment model also include: evaluating and validating the ensemble learning model using model evaluation metrics, which include accuracy, precision, recall, and F1 score.

5. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The bridge dynamic characteristic parameters include the first-order natural frequency of the damaged bridge and the first-order natural frequency of the intact bridge; the bridge superstructure structural parameters include the cross-sectional shape of the bridge superstructure; the frequency reduction factor of the sample bridge is the ratio of the first-order natural frequency of the damaged bridge to the first-order natural frequency of the intact bridge.

6. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The clustering algorithm is the K-means clustering algorithm, and the number of clusters in the K-means clustering algorithm is 4.

7. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The analytic hierarchy process (AHP) includes: providing at least two judgment matrices for the original technical condition deduction information, where each element of the judgment matrix represents the ratio of the impact of two technical condition assessment indicators on the bridge's technical condition; performing consistency checks on each judgment matrix; determining the weight vector of the judgment matrix for those that meet the consistency check; and determining the weight value of each technical condition assessment indicator based on the weight vector. The group decision method includes: calculating a reliability coefficient based on the weight vector obtained from the AHP; specifically, the numerical values ​​calculated using the AHP and group decision methods are calculated based on the weight values, reliability coefficients, and the original technical condition deduction information.

8. The risk assessment method for the safety of bridge superstructure according to claim 1, characterized in that, The loss function for training the bridge safety risk assessment model is: ； in, Represents the total loss function; The total number of training samples; Indicates the first Loss per training sample; Indicates the first The input feature vector of each training sample; For the first The true value of each training sample; For the first The model's predicted values ​​for each training sample; Indicates the total number of decision trees; Indicates the sequence number of the decision tree; Indicates the first A decision tree, For the first The regularization term of a decision tree.

9. A risk assessment system for the safety of bridge superstructure, characterized in that, include, The acquisition module is used to acquire the technical condition deduction information of the bridge superstructure. The technical condition deduction information includes technical condition assessment indicators and the deduction value of each technical condition assessment indicator at each scale. The deduction value of each technical condition assessment indicator at each scale is a value calculated by the analytic hierarchy process and the group decision method. The calculation module is used to acquire the bridge dynamic characteristic parameters, bridge superstructure structural parameters, bridge superstructure technical condition score data, bridge superstructure deduction indicators, and the defect classification categories of the deduction indicators of the sample bridge, and analyze the common indicators among the bridge superstructure deduction indicators accordingly; it is used to calculate the deduction values ​​of the defect classification categories of the common indicators at each scale based on the LSHADE algorithm, and to calculate the safety risk score of the bridge superstructure of the sample bridge based on the defect classification categories of the common indicators and the deduction values ​​of the defect classification categories of the common indicators at each scale. The risk level classification module is used to classify the safety risk level of the sample bridges based on the safety risk score of the bridge superstructure and using a clustering algorithm. The model building and training module is used to build a bridge safety risk assessment model using the CatBoost ensemble learning method. It is used to train the bridge safety risk assessment model using the frequency reduction coefficient of the sample bridge, the structural parameters of the superstructure of the sample bridge, the disease classification categories of common indicators of the sample bridge, the deduction values ​​of the disease classification categories of common indicators of the sample bridge, and the safety risk level of the superstructure of the sample bridge. The output of the model is the safety risk level of the bridge superstructure. The safety risk assessment module is used to assess the safety risk level of the bridge superstructure using a trained bridge safety risk assessment model.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to perform a risk assessment method for the safety of a bridge superstructure as described in any one of claims 1 to 8.

Images