Bridge state evaluation and prediction method based on multi-source information

Abstract

The invention discloses a bridge state evaluation and prediction method based on multi-source information. The method comprises the steps of establishment of a bridge state database, bridge degradation modeling based on machine learning and bridge management and maintenance strategy optimization based on a genetic algorithm. The bridge state database is based on a detection report and monitoring information of a bridge, the direct interference effect of vehicle loads on bridge degradation is considered, and data fusion and data regularization are carried out; carrying out machine learning model training and verification on the extracted key parameters and the data set to generate a series of degradation models of bridge members and structural layers, and carrying out interpretability analysis on the models based on an SHAP algorithm to obtain a feature importance sequence of the parameters; and performing state evaluation and degradation trend prediction on the bridge based on the degradation model, and realizing multi-objective optimization of a bridge management and maintenance strategy by adopting a genetic algorithm. Compared with the prior art, the intelligent degree and the result accuracy of bridge state evaluation are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of bridge status assessment and prediction, and specifically is a bridge status assessment and prediction method based on multi-source information, which realizes intelligent assessment and intelligent prediction of the status of in-service bridges. Background Art

[0002] As the hub of transportation engineering, bridges are inevitably subject to structural degradation due to the influence of the structure itself and the external environment, which can lead to component failure, reduced bridge bearing capacity, and even the collapse of the entire bridge. Therefore, it is necessary to inspect and evaluate the structures of in-service bridges, which is of great significance to ensure the health of bridges.

[0003] The current widely used condition assessment method in the engineering community is based on structural health monitoring (SHM). This health monitoring method integrates knowledge from multiple fields, improving the reliability of predictive assessments. However, the relevant theoretical and technical details are still immature. These issues include: 1) lack of unified monitoring standards and significant variations in scale; 2) monitoring content covers a wide range of technical indicators but lacks specificity; and 3) the condition assessment system based on monitoring is imperfect. These issues result in a lack of representativeness, specificity, and accuracy in bridge condition assessments, making them unable to meet the needs of modern bridge condition assessment and prediction.

[0004] In line with the trend of interdisciplinary and technological convergence, the integration of machine learning, deep learning, big data, and artificial intelligence has enabled the transformation and upgrading of bridge condition assessment and prediction methods towards a more systematic, scientific, and intelligent approach. These technologies effectively address the shortcomings of SHM monitoring methods, preventing subjective factors from influencing bridge assessments. They also enable big data analysis and processing of massive, multi-source, targeted data, improving the accuracy and relevance of bridge condition assessments and the applicability of prediction methods. This invention builds on the advantages of interdisciplinary technologies. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a bridge status assessment and prediction method based on multi-source information, which is used to solve the problems in the prior art of bridge detection and assessment, such as the heavy workload, difficult data processing, complex assessment methods, and the significant influence of subjective factors during the assessment. The method aims to achieve efficient, reliable, and intelligent assessment and analysis of operating bridges in order to make accurate maintenance decisions.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A bridge condition assessment and prediction method based on multi-source information mainly includes the following steps: (1) Bridge status database integrating multi-source data Establish a database based on regular bridge inspection reports and monitoring information, consider the direct interference effect of vehicle loads on bridge degradation, determine the characteristic parameters that affect bridge degradation, and provide a detailed and reliable data foundation for bridge degradation modeling through data preprocessing and regularization; (2) Establishment of bridge degradation model Establish a bridge degradation model based on multi-source information; explore the intrinsic relationships between parameters in massive data; conduct interpretable analysis on the model to predict the degradation trend of the bridge and provide a decision-making basis for subsequent bridge management and maintenance; (3) Optimization of management and maintenance strategies In order to comprehensively consider the safety and cost indicators of bridges, a non-dominated sorting genetic algorithm with an elite strategy is used to analyze and refine the bridge evaluation results, generate a set of alternative maintenance plans that meet the specified objectives, and provide a reference for the optimization of maintenance decisions.

[0007] Furthermore, the specific steps of step (1) are: (1.1) Collect a large number of bridge inspection reports and supplement them with other data; (1.2) Obtain key monitoring information of the bridge based on the bridge dynamic weighing system and vehicle load identification system; (1.3) Perform data regularization on the acquired detection and monitoring information, converting the original multi-source data from unstructured and semi-structured to structured data for easy use in data modeling and pattern mining; (1.4) Perform database feature subset selection to define the parameter set most relevant to the overall technical condition of the bridge; (1.5) Normalize the data of the selected feature subset and normalize all attribute values ​​to the interval [0,1]; (1.6) Finally, a bridge status database is formed by integrating multi-source data.

[0008] Furthermore, other data in step (1.1) include bridge construction drawings, technical condition assessment tables, and traffic flow statistics tables.

[0009] Furthermore, the vehicle load identification system in step (1.2) identifies the vehicle load based on computer vision technology and target tracking algorithms. The bridge weigh-in-motion system is a technology used to accurately measure vehicle weight in real time during normal driving. It utilizes modern sensors, data acquisition, signal processing, and data analysis technologies to monitor and record key data such as vehicle weight, number of axles, and wheelbase in real time.

[0010] Furthermore, the monitoring information in step (1.2) is the traffic volume and vehicle load of the bridge, and the direct interference effect of the vehicle load on the degradation of the bridge state is mainly considered.

[0011] Furthermore, the database feature subset selected in step (1.4) is a set of parameters most relevant to the overall technical condition of the bridge, such as bridge age, bridge type, bridge length, span, design load, traffic volume, vehicle load, maintenance record, ambient temperature and humidity, and air pollutant concentration.

[0012] Furthermore, the specific steps of step (2) are: (2.1) Extract bridge degradation characteristics and determine the input and output parameters of the machine learning model; (2.2) Use expert evaluation method / analytic hierarchy process to determine the weight coefficient of each indicator; (2.3) Build a machine learning model; (2.4) Model training and optimization for bridge dynamic database; (2.5) The bridge dynamic assessment index is multiplied by the weight coefficient to obtain the bridge dynamic technical condition score; (2.6) Classify bridge conditions according to the bridge dynamic technical status score; (2.7) Explain the machine learning model and rank the importance of each input and output parameter feature; (2.8) Predict long-term bridge performance; (2.9) Finally, a bridge degradation model based on multi-source data information is formed.

[0013] Furthermore, in step (2.1), multiple input and output parameters are established. The inputs are the parameter set most relevant to the overall technical condition of the bridge, such as bridge age, bridge type, bridge length, span, design load, traffic volume, vehicle load, maintenance record, ambient temperature and humidity, and air pollutant concentration. The outputs are the superstructure score, substructure score, and bridge deck score. The overall technical condition index of the bridge is calculated based on the indicator weights from the three output results, and the technical condition rating of the bridge is finally obtained.

[0014] Furthermore, the machine learning model in step (2.3) is constructed by selecting an XGBoost ensemble learning model, which can achieve faster and more accurate prediction and evaluation compared to traditional neural networks; the XGBoost model establishes a statistically based bridge degradation model from the component level and the structural level.

[0015] Furthermore, in step 2.2, a theoretical method combining triangular fuzzy numbers and the analytic hierarchy process is used to calculate the weights of the components of the bridge technical condition, which solves the uncertainty of the traditional scoring method and makes the constructed results more reasonable.

[0016] Furthermore, step (2.7) uses the SHAP algorithm to perform interpretable analysis on the model and derive the importance ranking of each characteristic parameter, thereby determining the factors that have the most adverse impact on bridge degradation and providing a more scientific and reliable decision-making basis for bridge maintenance.

[0017] Furthermore, the specific steps of step (3) are: (3.1) Perform algorithm optimization; (3.2) Determine bridge safety and cost indicators; (3.3) Establish a set of alternative maintenance solutions that meet the specified objectives; (3.4) Ultimately, optimize the management and maintenance strategy for bridges.

[0018] Furthermore, the step (3.1) uses the NSGA-II algorithm to find the optimal solution.

[0019] The beneficial effects of the present invention are as follows: The present invention establishes a bridge status database through data fusion and data regularization, trains and verifies the extracted key parameters and data sets into a machine learning model, generates a series of degradation models for bridge components and structural layers, performs interpretable analysis on the machine learning model, obtains the feature importance ranking of key parameters, realizes bridge status assessment and degradation trend prediction, and implements multi-objective optimization of bridge management and maintenance strategies based on genetic algorithms. Compared with the existing technology, the present invention realizes dynamic evaluation based on bridge inspection reports and bridge monitoring information, improves the ability to predict bridge diseases and take reasonable management and maintenance decisions, significantly improves the intelligence level of evaluation and the reliability of results, and provides important support and decision-making basis for the long-term maintenance and management of bridges. Ensuring the safe service and operation of bridge projects has important practical significance for bridge disease treatment and maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a general flow chart of a bridge condition assessment and prediction method based on multi-source information; Figure 2 Flowchart for multi-source data preprocessing; Figure 3 Design and modeling flowchart for the XGBoost model; Figure 4 A flowchart for interpretable machine learning; Figure 5 Model diagram for bridge technical condition assessment; Figure 6 Graph of the multi-objective optimization process based on genetic algorithm. DETAILED DESCRIPTION

[0021] The present invention will be described in detail below with reference to the accompanying drawings and specific implementations.

[0022] like Figure 1 As shown, the present invention provides a multi-source information-based bridge condition assessment and prediction method, including the establishment of a bridge condition database, machine learning-based bridge degradation modeling, and genetic algorithm-based bridge maintenance decision-making. The bridge condition database is established through data fusion and regularization based on bridge inspection reports and monitoring information. The extracted key parameters and data sets are then used to train and validate a machine learning model to generate a series of degradation models for bridge components and structural layers. Based on the degradation models, bridge condition assessment and degradation trend prediction are performed, and a genetic algorithm is used to achieve multi-objective optimization of bridge maintenance strategies.

[0023] The establishment of the bridge status database in the described method includes collecting data sources (inspection reports and monitoring information), data preprocessing and regularization, determining feature subsets (bridge age, bridge type, bridge length, span, design load, traffic volume, vehicle load, maintenance records, ambient temperature and humidity, air pollutant concentration, etc.), and data normalization.

[0024] The establishment of a bridge degradation model includes determining characteristic indicators, determining weight coefficients for each indicator, selecting a machine learning model, conducting model training, testing and tuning, model interpretability analysis, and obtaining the final bridge degradation prediction results.

[0025] The bridge maintenance decision optimization part is mainly based on genetic algorithms, which analyzes and refines the bridge evaluation results, generates a set of alternative maintenance plans that meet the specified goals, and provides a reference for the optimization of maintenance decisions.

[0026] like Figure 2 As shown in the figure, multi-source data preprocessing provides data support for training the bridge degradation model and constructs an objective and complete bridge condition database. The process begins with the acquisition of data sources, including bridge inspection reports, technical condition assessment forms, bridge monitoring data, traffic flow statistics, vehicle load statistics, and basic bridge structural information. Bridge construction drawings and other data sources are also used as supplementary data sources. Text recognition and extraction are then performed on the acquired semi- and unstructured data. The required feature attributes and their values ​​are automatically retrieved from each spreadsheet, forming a set of key-value pairs and encapsulating them into data objects corresponding to each individual bridge. Attribute subsets are then selected from the database to define the most relevant parameter sets for bridge degradation, such as bridge age, bridge type, bridge length, span, design load, traffic volume, vehicle load, maintenance records, ambient temperature and humidity, and air pollutant concentrations, to determine the model input parameters. Finally, the different data types in the attribute subset are encoded into a format that can be understood by computers. The data types mainly include numerical, nominal, and Boolean types. Through normalization methods, all attribute values ​​are dimensioned to the interval [0,1], and finally a bridge status database based on multi-source information is obtained.

[0027] like Figure 3 As shown, the XGBoost-based ensemble learning model features feature splitting calculation, node splitting sorting, and missing value processing, improving training speed and prediction accuracy. Leveraging the advantages of the XGBoost model, this paper establishes a statistically based bridge degradation model at both the component and structural levels. By extracting bridge degradation features and determining the model's input and output parameters, a multi-input and multi-output approach is employed. Inputs include bridge age, length, bridge type, traffic volume, vehicle load, and maintenance records, while outputs include the bridge's superstructure score, substructure score, and deck system score. The three outputs are calculated based on indicator weights to produce an overall bridge technical condition score, ultimately yielding a bridge technical condition grade. The XGBoost model then performs predictions on new input data by taking a weighted average of the prediction results from multiple decision trees. Within each decision tree, XGBoost automatically calculates feature importance and selects the optimal features and nodes for splitting, ultimately generating a network structure for the bridge degradation model. Finally, the XGBoost model is trained and tested, with the bridge status database divided into training, validation, and test sets in a 70%, 15%, and 15% ratio. Through the above series of processes, the degradation trend of bridges can be evaluated and predicted efficiently and accurately, providing a scientific decision-making basis for bridge management and maintenance.

[0028] like Figure 4 As shown, interpretable machine learning, based on the SHAP algorithm and the established XGBoost model, is used to determine the contribution of each input feature to the model output, specifically the impact of factors such as bridge age, length, span, design load, traffic volume, vehicle load, and maintenance records on the technical condition of the bridge, thereby deriving a feature importance ranking for each input parameter. The SHAP algorithm assigns a Shapley value to each feature by combining feature subsets and calculating the average marginal contribution of different feature subsets to the prediction results. This value represents the contribution of the feature factor. By performing interpretable analysis on the XGBoost model, it is possible to identify the factors that most adversely affect bridge degradation, thereby providing an effective and accurate basis for bridge maintenance decisions.

[0029] like Figure 5 As shown, the bridge technical condition assessment consists of a score for each bridge component and the weight of each component relative to the bridge's technical condition. This paper uses a theoretical approach combining triangular fuzzy numbers and the analytic hierarchy process to calculate the weights of the components involved in the bridge's technical condition, resolving the uncertainty inherent in traditional scoring methods and resulting in more reasonable results.

[0030] like Figure 6As shown, the multi-objective optimization process based on genetic algorithm can effectively screen out the optimal maintenance plan. In the present invention, chromosomes are the available maintenance plans for bridges, that is, the bridge objects and time nodes of the maintenance effect. The maintenance plan needs to be weighed in multiple dimensions, so two objective functions are set. Function 1 is a safety index that characterizes the overall state of the bridge, and function 2 is a cost index that characterizes the investment in bridge maintenance. The ranking of each alternative plan under the two objective functions is calculated separately. Considering the concepts of dominance and non-domination, the plans are continuously eliminated during the screening process until the termination conditions are met, thereby obtaining the optimal maintenance plan with multi-objective balance. This process can systematically, comprehensively and efficiently screen out the optimal bridge maintenance plan and achieve a balance between safety goals and economic goals.

Claims

1. A bridge condition assessment and prediction method based on multi-source information, Features: The main steps include: (1) Bridge status database integrating multi-source data Establish a database based on regular bridge inspection reports and monitoring information, consider the direct interference of vehicle loads on bridge degradation, determine the characteristic parameters that affect bridge degradation, and provide a detailed and reliable data basis for bridge degradation modeling after data preprocessing and regularization; (2) Establishment of bridge degradation model Establish a bridge degradation model based on multi-source information; mine the intrinsic relationship of parameters in massive data; Conduct interpretability analysis on the model to predict the degradation trend of the bridge status and provide a decision-making basis for the subsequent bridge management and maintenance; (3) Optimization of management and maintenance strategies In order to comprehensively consider the safety and cost indicators of bridges, a non-dominated sorting genetic algorithm with an elite strategy is used to analyze and refine the bridge evaluation results, generate a set of alternative maintenance plans that meet the specified objectives, and provide a reference for the optimization of maintenance decisions.

2. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 1, Features: The specific steps of step (1) are: (1.1) Collect a large number of bridge inspection reports and supplement them with other data; (1.2) Obtain key monitoring information of the bridge based on the bridge dynamic weighing system and vehicle load identification system; (1.3) Regularize the acquired detection and monitoring information, and convert the original multi-source data from unstructured and semi-structured to structured data for easy use in data modeling and pattern mining; (1.4) Select a subset of database features to define the set of parameters most relevant to the overall technical condition of the bridge; (1.5) Normalize the data of the selected feature subset and normalize all attribute values ​​to the interval [0,1]; (1.6) Finally, a bridge status database is formed by integrating multi-source data.

3. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 2, Features: Other data in step (1.1) include bridge construction drawings, technical condition assessment tables, and traffic flow statistics tables.

4. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 2, Features: The monitoring information in step (1.2) is the traffic volume and vehicle load of the bridge, and the direct interference effect of the traffic volume and vehicle load on the bridge degradation is mainly considered.

5. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 1, Features: The specific steps of step (2) are: (2.1) Extract bridge degradation characteristics and determine the input and output parameters of the machine learning model; (2.2) Use expert evaluation method / hierarchy analysis method to determine the weight coefficient of each indicator; (2.3) Build a machine learning model; (2.4) Model training and optimization for bridge dynamic database; (2.5) The bridge dynamic assessment index is multiplied by the weight coefficient to obtain the bridge dynamic technical condition score; (2.6) Classify the bridge condition according to the bridge dynamic technical condition score; (2.7) Explain the machine learning model and rank the importance of each input parameter feature; (2.8) Predict long-term bridge performance; (2.9) Finally, a bridge degradation model based on multi-source data information is formed.

6. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 5, Features: In the step (2.1), multiple inputs and multiple outputs of parameters are established, with the parameter set most relevant to the overall technical condition of the bridge, such as bridge age, bridge type, bridge length, span, design load, traffic volume, vehicle load, maintenance record, ambient temperature and humidity, and air pollutant concentration, as input, and the upper structure score, lower structure score, and bridge deck score as output. The overall technical condition index of the bridge is calculated based on the indicator weights from the three output results, and finally the technical condition rating of the bridge is obtained.

7. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 6, Features: In the step (2.3), the machine learning model is constructed by selecting the XGBoost integrated learning model.

8. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 7, Features: In the step 2.2, a theoretical method combining triangular fuzzy numbers and analytic hierarchy process is used to calculate the weights of the components of the bridge technical status.

9. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 8, Features: The step (2.7) uses the SHAP algorithm to perform interpretability analysis on the black box model, and obtains the importance ranking of each characteristic parameter, thereby determining the factors that have the most adverse impact on bridge degradation.

10. A bridge condition assessment and prediction method based on multi-source information as claimed in claim 1, Features: The specific steps of step (3) are: (3.1) Use NSGA-Ⅱ algorithm to find the optimal solution; (3.2) Determine bridge safety and cost indicators; (3.3) Establish a set of alternative maintenance solutions that meet the specified objectives; (3.4) Ultimately, the bridge maintenance strategy is optimized.

Images