A two-stage bridge disease analysis and risk assessment method and system

Abstract

The present invention relates to a two-stage analysis and risk assessment method and system for bridge defects, which belongs to the technical field of bridge defect analysis, and solves the problems of low bridge defect prediction accuracy and incomplete risk assessment in existing bridge defect analysis and risk assessment methods. The method includes the steps of: acquiring data of multiple bridges, including geometric characteristics, material characteristics, environmental factors and defect data; preprocessing the data, and performing data balancing on the preprocessed data; optimizing the characteristics of bridge defects using a two-stage analysis method, and constructing a multi-label classification model to predict the probability distribution of joint defects; based on the joint defect prediction probability, performing risk cluster analysis on bridges using a fuzzy C-means clustering method, dividing bridge risks into different categories and outputting risk clustering results. The present invention improves the accuracy of bridge defect prediction and the reliability of risk assessment, has stronger adaptability, and shows good application prospects and economic benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge disease analysis, and in particular to a two-stage bridge disease analysis and risk assessment method and system for bridge health monitoring, disease prediction and maintenance management. Background Art

[0002] As bridges age, their structural safety and durability are gradually affected by a variety of factors, including environmental conditions (such as climate change, temperature and humidity, salt damage, etc.), traffic loads (such as frequent heavy vehicle traffic), material aging, and geometric structural characteristics. These factors work together to make the types of bridge diseases complex and diverse, and different diseases are often interrelated and coupled. This complexity poses many challenges to the prediction and risk assessment of bridge diseases. Traditional manual detection methods are not only inefficient, but also difficult to accurately grasp the comprehensiveness and correlation of diseases. Traditional single-label prediction methods cannot effectively handle the multi-label characteristics of diseases, making it difficult to comprehensively assess the risk level of bridges.

[0003] Currently, there are some methods for bridge damage analysis and risk assessment, but they still have many limitations.

[0004] For example, the master's thesis "Design and Implementation of Bridge Health Monitoring and Defect Prediction System Based on BIM" proposed the use of time series methods LSTM and ARIMA to predict temperature, expansion joints, and settlement. However, this method only focuses on the time series prediction of some single features, and does not comprehensively consider the correlation and risk between multiple bridge diseases.

[0005] For example, the Chinese patent with publication number CN113569908A proposed a reverse calculation method for predicting bridge technical conditions and defects using the CPSO-BP neural network model, which accurately predicts the location, type and degree of bridge defects by optimizing the neural network structure and parameters. However, this method fails to provide the probability distribution of defect occurrence, makes it difficult to quantify the uncertainty of defect prediction, and ignores the potential correlation between defects.

[0006] For example, the master's thesis "Research on Bridge Group Risk Assessment Method Combining Fuzzy Clustering and Hierarchical Analysis" combines the fuzzy clustering algorithm with the hierarchical analysis method for optimization, which can not only achieve the classification of bridge operation risks, but also perform quantitative weight analysis. However, this method is mainly based on the evaluation system constructed by traditional expert experience, and fails to integrate the prediction information or probability of bridge diseases into the risk assessment model, and has limited quantitative ability.

[0007] For example, the paper "A New Risk Assessment Method for In-Service Bridges" published in the Journal of Highway and Transportation Research and Development uses the LKC method to conduct risk assessment on in-service bridges. However, this method is only applicable to macro-risk assessment and is not combined with the prediction of the probability of bridge disease occurrence, so it cannot provide refined assessment results.

[0008] The existing bridge disease analysis and risk assessment methods have the following main defects:

[0009] Insufficient damage prediction: Most existing bridge damage prediction methods focus on a few single damages (such as temperature deformation and settlement) or single variable prediction, and fail to fully cover multiple damages. At the same time, most methods are based on deterministic models or classification models, and fail to provide the probability distribution of damage prediction, making it difficult to evaluate the credibility of the prediction.

[0010] The correlation between defects is not considered: The coupling and mutual influence between defects are very important in actual bridge operation, but existing methods rarely capture these complex relationships through interaction features or joint modeling;

[0011] Limitations of risk assessment: Existing bridge risk assessment methods are usually based on qualitative expert experience or static disease information, lack a comprehensive analysis of the probability of disease occurrence, and fail to combine disease prediction results for dynamic quantitative risk assessment, resulting in insufficient prediction accuracy and applicability of the model. Summary of the invention

[0012] In view of the problems existing in the existing bridge disease analysis and risk assessment methods, such as insufficient prediction of bridge diseases, low accuracy of bridge disease prediction, failure to consider the potential correlation between diseases, and large limitations in risk assessment, the present invention proposes a two-stage bridge disease analysis and risk assessment method and system based on multi-label learning, which ultimately achieves accurate analysis and risk prediction of bridge diseases through data collection, preprocessing, feature optimization, multi-label classification, and risk assessment. The present invention improves the accuracy of bridge disease prediction and the reliability of risk assessment, has stronger adaptability, higher efficiency, and has good application prospects and economic benefits.

[0013] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0014] A two-stage bridge disease analysis and risk assessment method includes the following steps:

[0015] Step 1: Acquire data of multiple bridges, wherein the data includes geometric property data, material property data, environmental factor data, and disease data;

[0016] Step 2: Preprocess the acquired data, including disease type keyword extraction, missing value filling, standardization and outlier detection. Different processing is performed according to the type of data during preprocessing to ensure the integrity and consistency of the data and obtain the preprocessed data;

[0017] Step 3: Perform data balancing on the preprocessed data. During the data balancing, the multi-label problem is first converted into a multi-classification problem using the LabelPowerset method, and a corresponding balancing method is adopted for each type of sample according to the number of samples. Then, the balanced multi-classification labels are mapped back to the multi-label format.

[0018] Step 4: A two-stage analysis method is used to complete the feature optimization and multi-label classification prediction of bridge defects. The two-stage analysis method includes:

[0019] The first stage is single disease feature extraction and prediction: the feature importance evaluation method based on the random forest model is used to extract high-importance features related to single disease prediction, and then the multi-label classification algorithm CatBoost is used to separately model and predict each disease, and the prediction probability of the single disease is output;

[0020] The second stage is the disease interaction modeling and joint prediction: the single disease prediction probability generated in the first stage is used to construct the joint features between diseases, and the CatBoost model is used based on the joint features to model and predict the joint occurrence probability of bridge diseases, and the joint disease prediction probability is output;

[0021] Step 5: Based on the joint disease prediction probability, the fuzzy C-means clustering method is used to perform risk clustering analysis on the bridge, the bridge risks are divided into different categories, and finally the risk clustering results are output.

[0022] Accordingly, the present invention also proposes a two-stage bridge disease analysis and risk assessment system, comprising:

[0023] A data acquisition unit, used to acquire data of a plurality of bridges, wherein the data includes geometric characteristic data, material characteristic data, environmental factor data and disease data;

[0024] The data preprocessing unit is used to preprocess the acquired data, including disease type keyword extraction, missing value filling, standardization processing and outlier detection. During preprocessing, different processing is performed according to the type of data to ensure the integrity and consistency of the data and obtain the preprocessed data;

[0025] A data balancing unit is used to perform data balancing on the preprocessed data. During the data balancing, the multi-label problem is first converted into a multi-classification problem using the Label Powerset method, and a corresponding balancing method is adopted for each type of sample according to the number of samples, and then the balanced multi-classification labels are mapped back to the multi-label format;

[0026] The feature optimization and classification unit is used to complete the feature optimization and multi-label classification prediction of bridge defects using a two-stage analysis method, wherein the two-stage analysis method includes:

[0027] The first stage is single disease feature extraction and prediction: the feature importance evaluation method based on the random forest model is used to extract high-importance features related to single disease prediction, and then the multi-label classification algorithm CatBoost is used to separately model and predict each disease, and the prediction probability of the single disease is output;

[0028] The second stage is the disease interaction modeling and joint prediction: the single disease prediction probability generated in the first stage is used to construct the joint features between diseases, and the CatBoost model is used based on the joint features to model and predict the joint occurrence probability of bridge diseases, and the joint disease prediction probability is output;

[0029] The risk assessment unit is used to perform risk clustering analysis on the bridge based on the joint disease prediction probability by using a fuzzy C-means clustering method, classify the bridge risks into different categories, and finally output the risk clustering results.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) Comprehensive prediction of multiple types of defects and quantification of their occurrence probability: The present invention adopts a multi-label classification algorithm, which can simultaneously predict multiple types of defects (such as cracks, water erosion, spalling, etc.) and accurately calculate the probability of each defect. This method breaks through the limitation of insufficient prediction of defect types in the existing technology and provides more comprehensive results for bridge defect analysis. At the same time, by quantifying the probability of defects, it provides scientific data support for bridge risk assessment.

[0032] (2) Fusion of multidimensional data to improve model performance: By introducing multidimensional data such as traffic flow, climate conditions, material properties and service life, the present invention achieves a comprehensive analysis of bridge damage, overcoming the shortcomings of the traditional method of single data source; multidimensional data fusion fully considers the internal and external causes of bridge damage, improves the scientificity and accuracy of model prediction, and provides a more reliable basis for risk assessment and maintenance decision-making;

[0033] (3) Combining feature optimization with multi-label learning to improve prediction accuracy: The present invention uses an integrated learning algorithm to screen out key features related to defects, reduce data redundancy, and reduce model complexity. Combined with a multi-label learning algorithm, the present invention significantly improves the accuracy of the model while achieving simultaneous prediction of multiple types of defects, providing higher quality results for bridge defect prediction and risk assessment.

[0034] (4) Automated analysis process improves efficiency and practicality: Through modular design, the present invention realizes the automation of the entire process from data collection to risk assessment, significantly improving the analysis efficiency and avoiding the errors that may be caused by manual operation. The automated analysis process not only reduces costs, but also improves the reliability and consistency of the results, providing an efficient and convenient solution for bridge management and maintenance practices. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0036] Figure 1 This is a flow chart of a two-stage bridge disease analysis and risk assessment method according to one embodiment of the present invention;

[0037] Figure 2 This is a comparison chart of the importance of characteristics of each disease type;

[0038] Figure 3 This is a comparison chart of the independent occurrence probability distribution of each disease;

[0039] Figure 4 This is a comparison chart of the probability distribution of joint disease occurrence;

[0040] Figure 5 This is the clustering result diagram of the test data;

[0041] Figure 6 This is the distribution diagram of each risk bridge in the test data;

[0042] Figure 7 This is a schematic structural diagram of a two-stage bridge disease analysis and risk assessment system according to another embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should be understood that the specific embodiments described are only used to explain the present invention and are not used to limit the present invention.

[0044] The present invention proposes a bridge disease risk prediction method based on a two-stage analysis method. The first stage is the extraction and prediction of single disease features. The random forest model is used to screen the importance of key features related to bridge diseases, and then the CatBoost model is used to separately model and predict each disease to obtain the probability distribution of independent occurrence of each disease. In the second stage, the disease interaction features are constructed for the possible coupling relationship between diseases, and the integrated learning model is combined for joint prediction to output the probability distribution of the joint occurrence of each disease. On this basis, the fuzzy C-means clustering method is further used to perform risk clustering analysis on the bridge based on the results of the joint probability prediction, and the bridge risks are divided into different categories (such as low risk, medium risk and high risk). Finally, the risk clustering results are output, which provides a reliable basis for the scientific maintenance and precise management of bridges. Through this two-stage modeling method, the correlation between diseases is fully considered, the accuracy and scientificity of the prediction are improved, and a reliable basis is provided for the intelligent maintenance and management of bridges.

[0045] Specifically, Figure 1 As shown, this embodiment provides a two-stage bridge disease analysis and risk assessment method based on multi-label learning, which includes the following steps 1 to 5.

[0046] Step 1 (S100): Acquire data including geometric property data, material property data, environmental factor data and disease data of the bridge, including but not limited to structural dimensions, component types, material composition, traffic flow, climate conditions, service life and disease conditions of the bridge.

[0047] In this embodiment, the data of multiple bridges can be obtained from the bridge inspection database, which stores the inspection data of multiple bridges, such as the inspection data of 2,314 bridges in a province in southwest my country, mainly including three aspects: basic information of bridges, rating information of bridge technical conditions, and information of bridge diseases. For bridge disease analysis, multiple features including numerical data (such as traffic volume, rating, bridge length) and text data (such as bridge type, rating, structure type) need to be processed.

[0048] Step 2 (S200): preprocess the acquired data, including disease type keyword extraction, missing value filling, standardization and outlier detection, and perform different processing according to the type of data (numerical or categorical) during preprocessing to ensure the integrity and consistency of the data, and obtain preprocessed data after preprocessing.

[0049] Disease type keyword extraction: Use keyword matching technology to automatically extract disease types from the disease description in the data. By building a disease type keyword dictionary, regular expressions and keyword matching methods are used to extract. Specifically, let the description text be , the keyword set is , the disease type set is , then the disease extraction function is:

[0050] (1)

[0051] By segmenting the text, all words are extracted. The words in the disease keyword dictionary are further matched to associate the disease with the predefined type. In this embodiment, for example, the input text: "severe cracks, peeling" matches the keywords: "cracks", "peeling", and the output disease type is: "cracks, peeling". There are ten types of diseases that are finally output: water erosion, cracks, peeling, corrosion, deformation, exposed tendons, honeycomb, hollowing, whitening, and web cracking.

[0052] Missing value filling: fill in the mean of all numerical data to complete the missing data.

[0053] Standardization: Use label encoding to convert categorical data, and represent each disease as 0 and 1 for easy model processing. Specifically, all disease types are converted, and different diseases are converted into multi-label format. The variables generated by this encoding only contain 0 and 1, and different types of data are stored in the feature space. Among them, the standardization calculation formula is:

[0054] (2)

[0055] in: Represents the original data, represents the mean, represents the standard deviation, Represents the data after standardization. This method subtracts the feature value from its mean and then divides it by its standard deviation to ensure that the mean of all features is 0 and the standard deviation is 1.

[0056] Outlier detection: Missing value detection is performed on the standardized data, and interpolation methods are used to handle abnormal data to improve the stability of the model.

[0057] Step 3 (S300): The data preprocessed in step 2 is further subjected to data balancing. During the data balancing process, the Label Powerset method is first used to convert the multi-label problem into a multi-classification problem, and corresponding balancing methods are adopted for each type of sample according to the number of samples. The balanced multi-classification labels are then mapped back to the multi-label format to retain the characteristics of the multi-label problem.

[0058] First, the preprocessed data of multi-label (disease) is encoded into a binary matrix format. The multi-label data is converted into a single multi-classification label by the Label Powerset method, and each label combination (i.e., Powerset) is mapped to an independent multi-classification label. The sample distribution of each multi-classification label is counted to provide a basis for subsequent balancing processing.

[0059] According to the number of samples, the multi-classification labels are divided into a minority sample category (sample number < 6) and a normal sample category (sample number >= 6). In this embodiment, there are 72 categories, of which 54 sample categories are minority samples and 18 sample categories are normal samples. A corresponding balancing method is adopted for each type of sample.

[0060] For the minority sample categories with a small number of samples, the random oversampling technique is used to randomly oversample the minority sample categories to expand the samples, increase the number of samples of these categories, and ensure that they are not ignored in the data set.

[0061] For the normal sample category with a normal number of samples, the synthetic minority over-sampling technique (SMOTE) is applied to generate synthetic samples, and the balancing process is performed by synthesizing new samples. SMOTE generates new adjacent samples based on existing samples, which not only increases the number of minority class samples, but also enhances the diversity of data. Specifically, the formula for SMOTE to generate new samples based on minority class samples is:

[0062] (3)

[0063] in, Represents two different minority class sample vectors; represents the random coefficient; Indicates that the generated new sample is the synthetic sample.

[0064] The processed few-sample category data and normal sample category data are merged into a complete dataset.

[0065] According to the multi-classification labels after oversampling, re-map back to the multi-label format to ensure that the balanced data meets the requirements of the multi-label problem. The balanced results of the diseases in this embodiment are shown in Table 1.

[0066] Table 1 Data of bridge defects before and after balancing treatment

[0067] Diseases Data before balancing After balancing the data Flaking 155 11308 Deformation 23 1649 Water erosion 159 9699 Whitening 13 1609 Web crack 16 1629 empty 41 1689 corrosion 52 3268 honeycomb 23 1659 crack 562 19348 exposed tendons 105 9699

[0068] Step 4 (S400): A two-stage analysis method is used to complete feature optimization and multi-label classification prediction of bridge defects, wherein the two-stage analysis method specifically includes: the first stage is single defect feature extraction and prediction: a feature importance evaluation method based on a random forest model is used to extract high-importance features related to single defect prediction, and then each defect is separately modeled and predicted using a multi-label classification algorithm CatBoost, and a single defect prediction probability is output; the second stage is defect interaction modeling and joint prediction: the single defect prediction probability generated in the first stage is used to construct joint features between defects, and based on the joint features, a CatBoost model is used to model and predict the joint occurrence probability of bridge defects, and a joint defect prediction probability is output.

[0069] Phase 1: Single disease feature extraction and prediction.

[0070] In this stage, the feature importance evaluation function of the random forest model is used to quantify the contribution of each feature to disease prediction and select high-importance features according to the set threshold. The random forest model calculates the importance of features based on information gain or mean square error reduction by evaluating the importance scores of features in all trees. Specifically, the feature importance calculation formula is as follows:

[0071] (4)

[0072] in, Representation characteristics Importance score of represents the total number of trees in the model; Indicates The tree has features The set of nodes to be split; Representation Node Split gain (such as information gain, mean square error reduction, etc.); Represents the index of the decision tree.

[0073] To filter features, we use the importance score of the features Setting the Threshold , filter to meet The high contribution features of are taken as the extracted high importance features. If no feature meets the screening conditions , then all features are retained to ensure model stability. The value range of , the specific value can be determined by cross-validation tuning, for example . For each disease, important features are selected and used in subsequent disease prediction modeling to improve the interpretability and performance of the model. The comparison results of the importance of features of each disease type are as follows Figure 2 shown.

[0074] Then, the multi-label classification algorithm CatBoost model was used to model each disease separately and predict the independent occurrence probability of each disease. The formula for the probability of each disease occurring independently, that is, the prediction probability of a single disease, is as follows:

[0075] (5)

[0076] in, Represents the input features, that is, the data in the dataset except the label; Indicates The weight vector of the established disease model; Indicates bias of the disease model; Represents the Sigmoid function, which is defined as:

[0077] (6)

[0078] The prediction result is , that is, the input feature Lower Disease The probability of independent occurrence.

[0079] According to formula (5), multi-label classification modeling is performed on the extracted high-importance features, and the probability of disease occurrence is predicted one by one. The accuracy of the probability prediction value of each disease is as high as more than 99%. The independent occurrence probability distribution of each disease is compared. Figure 3 As shown, Figure 3 The probability distribution of each disease type in the data is roughly expressed as two peak areas: the peak close to 0 indicates that the model predicts that the disease will not occur in most samples (or the probability is extremely low); the peak close to 1 indicates that the model predicts that the disease has a high probability of occurrence in some samples. Figure 3 It can be seen that the model has a higher confidence level in areas close to 0 and 1, indicating that the model's predictions for most diseases in the samples are clear. In the intermediate probability area of ​​0.4 to 0.6, the distribution density is lower, indicating that the model predicts fewer samples as "uncertain".

[0080] Phase II: disease interaction modeling and joint prediction.

[0081] In this stage, the interaction features between diseases are introduced to construct joint features, and the probability distribution of the joint occurrence of multiple diseases is comprehensively calculated. The original features are combined with the interaction features to form joint features. It can be expressed as:

[0082] (7)

[0083] in, Represents interactive features, which are generated by interactively combining the original features. The interactive combination may include nonlinear transformations, such as wait; Indicates the number of disease types.

[0084] Next, we use the joint features of diseases to build a joint prediction model, quantify the correlation between diseases, and improve the accuracy of prediction. Based on the joint features, we build a joint prediction model and calculate the joint disease prediction probability:

[0085] (8)

[0086] in, Represents the probability of all diseases occurring together.

[0087] The joint prediction is trained using CatBoost multi-output classifiers, and the risk of bridge damage is further analyzed by combining the probability distribution of the damage. Figure 4 As shown in the figure, overall, the model can accurately identify the probability of joint disease occurrence for most samples and performs well in the high confidence interval.

[0088] In order to verify the effectiveness of the proposed model, the CatBoost model was used to model the bridge data and predict the probability of all diseases. In this embodiment, the most satisfactory results were obtained by comparing this method with other classical machine learning methods. The calculation results of each model on the two evaluation indicators of average accuracy and Hamming loss are shown in Table 2.

[0089] Table 2 Comparison results of various models

[0090] Model Average accuracy Hamming loss Random Forest 0.979 0.003 Decision Tree 0.909 0.011 SVM 0.872 0.018 KNN 0.960 0.007 AdaBoost 0.456 0.082 XGBoost 0.977 0.003 CatBoost 0.982 0.002

[0091] Step 5 (S500): Based on the joint disease prediction probability calculated in step 4, a risk clustering analysis is performed on the bridge through cluster analysis (fuzzy C-means clustering method), and the bridge risks are divided into different categories. Finally, the risk clustering results are output, and high-risk areas are automatically identified, thereby providing a dynamic quantitative risk assessment basis for the bridge operation period and providing a decision-making basis for subsequent bridge maintenance and repair.

[0092] Specifically, when using the fuzzy C-means clustering method for risk clustering analysis, the risk data of the bridge is converted into a multi-dimensional risk assessment model, and the fuzzy membership function is used to analyze the distribution of bridges between different risk categories. The formula for fuzzy clustering is:

[0093] (9)

[0094] in, Indicates the number of samples; represents the number of clusters; is the membership degree, indicating that the sample Belongs to cluster possibility; represents the cluster center; Indicates individual data; represents the membership matrix; represents the cluster center matrix; is the fuzzy coefficient, usually taken Using the joint disease prediction probability and the fuzzy clustering method, the bridge risks are clustered and the risk categories of the bridges are calculated. The final risk category of the bridge is The output formula is as follows:

[0095] (10)

[0096] in, represents the joint disease prediction probability With cluster center The clustering results of the test data are as follows Figure 5 As shown in Figure 2, the distribution of risk bridges in the test data is as follows: Figure 6 As shown in the figure, according to statistics, the proportion of bridges with three or more diseases at the same time in the test data is about 11%, while the proportion of bridges with less than three diseases or no diseases is about 89%, indicating that the bridge risk distribution effect is obvious.

[0097] The advantages of fuzzy C-means clustering in risk prediction are its flexible classification, ability to handle data uncertainty and comprehensive multi-dimensional features, which provides scientificity and reliability for bridge risk assessment. Based on the results of fuzzy C-means clustering, the accuracy of risk assessment is optimized and more accurate decision support is provided for bridge maintenance.

[0098] Multi-label learning technology can more comprehensively capture the complexity and correlation of bridge diseases by simultaneously predicting the probability of occurrence of multiple bridge diseases, thereby improving the coverage and accuracy of disease analysis. Compared with traditional single-label prediction methods, multi-label learning can effectively explore the potential relationship between diseases and provide a more comprehensive assessment of bridge health status. In addition, fuzzy clustering technology can finely classify bridges in risk assessment, and analyze the distribution of bridge risks between different categories based on fuzzy membership functions, which solves the limitations of rigid classification in bridge risk assessment. The combination of the two can achieve a deep integration of bridge disease analysis and risk assessment, providing strong support for precise maintenance and scientific decision-making of bridges.

[0099] like Figure 7 As shown, another embodiment of the present invention provides a two-stage bridge disease analysis and risk assessment system based on multi-label learning, the system comprising:

[0100] A data acquisition unit 100 is used to acquire data of a plurality of bridges, the data including geometric characteristic data, material characteristic data, environmental factor data and disease data;

[0101] The data preprocessing unit 200 is used to preprocess the data acquired by the data acquisition unit, including disease type keyword extraction, missing value filling, standardization processing and abnormal value detection. During preprocessing, different processing is performed according to the type of data to ensure the integrity and consistency of the data and obtain preprocessed data;

[0102] The data balancing unit 300 is used to perform data balancing on the preprocessed data output by the data preprocessing unit. During the data balancing, the multi-label problem is first converted into a multi-classification problem using the Label Powerset method, and a corresponding balancing method is adopted for each type of sample according to the number of samples, and then the balanced multi-classification labels are mapped back to the multi-label format;

[0103] The feature optimization and classification unit 400 is used to complete the feature optimization and multi-label classification prediction of bridge defects using a two-stage analysis method, wherein the two-stage analysis method includes:

[0104] The first stage is single disease feature extraction and prediction: the feature importance evaluation method based on the random forest model is used to extract high-importance features related to single disease prediction, and then the multi-label classification algorithm CatBoost is used to separately model and predict each disease, and the prediction probability of the single disease is output;

[0105] The second stage is the disease interaction modeling and joint prediction: the single disease prediction probability generated in the first stage is used to construct the joint features between diseases. Based on the joint features, the CatBoost model is used to model and predict the joint occurrence probability of bridge diseases, and the joint disease prediction probability is output;

[0106] The risk assessment unit 500 is used to perform risk clustering analysis on bridges based on the joint disease prediction probability output by the feature optimization and classification units, using the fuzzy C-means clustering method, classify bridge risks into different categories, and finally output risk clustering results.

[0107] The specific implementation of each unit in this embodiment can refer to the above-mentioned bridge disease two-stage analysis and risk assessment method embodiment, which will not be described one by one here. It should be noted that the system provided in this embodiment is only illustrated by the division of the above-mentioned functional units. In actual applications, the above-mentioned functions can be assigned to different functional units or modules as needed, that is, the internal structure can be divided into different functional units or modules to complete all or part of the functions described above.

[0108] In summary, the present invention provides a two-stage bridge disease analysis and risk assessment method and system based on multi-label learning. Through the steps of data acquisition, data preprocessing, data balancing, feature optimization and classification, and risk assessment, a reliable bridge disease analysis model, namely a multi-label learning model, is constructed, which provides important decision support for bridge management and maintenance. The present invention introduces multi-label learning technology and adopts a multi-step processing framework, including data acquisition, preprocessing, data balancing, feature optimization and classification, and risk assessment, which comprehensively improves the accuracy of bridge disease prediction and risk assessment. The contribution analysis of key features is optimized by an integrated learning algorithm (such as CatBoost), and the risk category division is realized by combining fuzzy C-means clustering technology, providing accurate risk assessment and scientific decision support for bridge maintenance. Compared with traditional methods, the present invention has strong practicality, high efficiency and broad application prospects.

[0109] The above is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and inventive concept of the present invention within the scope disclosed by the present invention, which shall fall within the protection scope of the present invention.

Claims

1. A two-stage bridge disease analysis and risk assessment method, characterized in that: The following steps are involved: Step 1: Acquire data of multiple bridges, wherein the data includes geometric property data, material property data, environmental factor data, and disease data; Step 2: Preprocess the acquired data, including disease type keyword extraction, missing value filling, standardization and outlier detection. Different processing is performed according to the type of data during preprocessing to ensure the integrity and consistency of the data and obtain the preprocessed data; Step 3: Perform data balancing on the preprocessed data. During the data balancing, the multi-label problem is first converted into a multi-classification problem using the LabelPowerset method, and a corresponding balancing method is adopted for each type of sample according to the number of samples. Then, the balanced multi-classification labels are mapped back to the multi-label format. Step 4: A two-stage analysis method is used to complete the feature optimization and multi-label classification prediction of bridge defects. The two-stage analysis method includes: The first stage is single disease feature extraction and prediction: the feature importance evaluation method based on the random forest model is used to extract high-importance features related to single disease prediction, and then the multi-label classification algorithm CatBoost is used to separately model and predict each disease, and the prediction probability of the single disease is output; The second stage is the disease interaction modeling and joint prediction: the single disease prediction probability generated in the first stage is used to construct the joint features between diseases, and the CatBoost model is used based on the joint features to model and predict the joint occurrence probability of bridge diseases, and the joint disease prediction probability is output; Step 5: Based on the joint disease prediction probability, the fuzzy C-means clustering method is used to perform risk clustering analysis on the bridge, the bridge risks are divided into different categories, and finally the risk clustering results are output.

2. A two-stage bridge disease analysis and risk assessment method according to claim 1, characterized in that: When preprocessing the acquired data, keyword matching technology is used on the disease description in the data to automatically extract the disease type; Perform mean filling on all numerical data to fill in missing data; Use label encoding method to convert categorical data; The standardized data were checked for missing values ​​and the abnormal data were processed using interpolation methods.

3. A two-stage bridge disease analysis and risk assessment method according to claim 1 or 2, characterized in that: In step 3, when taking corresponding balancing processing methods for each type of samples, for the minority sample category with a small number of samples, the random oversampling technology is used to expand the samples; for the normal sample category with a normal number of samples, the synthetic minority class oversampling technology is used to generate synthetic samples; the processed minority sample category data and normal sample category data are merged into a complete data set.

4. A two-stage bridge disease analysis and risk assessment method according to claim 3, characterized in that: The formula for generating the synthetic sample using the synthetic minority class oversampling technique is: x new =x i +λ·(x j -x i ),λ∈[0,1] (3) Among them, x i , x j represents two different minority class sample vectors; λ represents the random coefficient; x new represents the generated synthetic samples.

5. A two-stage bridge disease analysis and risk assessment method according to claim 1 or 2, characterized in that: In step 4, when extracting high-importance features, the random forest model calculates the importance of the feature by evaluating the importance score of the feature in all trees based on information gain or mean square error reduction. The feature importance calculation formula is as follows: Among them, I j represents the importance score of feature j; T represents the total number of trees in the model; S t (j) represents the set of nodes split by feature j in the tth tree; ΔG(s, t) represents the split gain of node s; t represents the index of the decision tree.

6. A two-stage bridge disease analysis and risk assessment method according to claim 5, characterized in that: According to the importance score of the feature I j Set the threshold τ to filter the content of I j The high contribution features with a value greater than τ are taken as the extracted high-importance features, where the value range of the threshold τ is [0.01, 0.1].

7. A two-stage bridge disease analysis and risk assessment method according to claim 1 or 2, characterized in that: The formula for the single disease prediction probability is: Among them, x represents a high-importance feature; w i represents the weight vector of the i-th disease model established; b i represents the bias of the i-th disease model; σ(z) represents the Sigmoid function.

8. A two-stage bridge disease analysis and risk assessment method according to claim 7, characterized in that: The joint features x between the diseases constructed combined It is expressed as: x combined =[P(y1|x), P(y2|x),..., P(y k |x), F int ] (7) Among them, F int represents interactive features, which are generated by interactively combining the original features; k represents the number of disease types.

9. A two-stage bridge disease analysis and risk assessment method according to claim 8, characterized in that: The formula for the prediction probability of the combined disease is: Among them, P(y|x) represents the probability of all diseases occurring together.

10. A two-stage bridge disease analysis and risk assessment system, characterized in that: include: A data acquisition unit, used to acquire data of a plurality of bridges, wherein the data includes geometric characteristic data, material characteristic data, environmental factor data and disease data; The data preprocessing unit is used to preprocess the acquired data, including disease type keyword extraction, missing value filling, standardization processing and outlier detection. During preprocessing, different processing is performed according to the type of data to ensure the integrity and consistency of the data and obtain the preprocessed data; A data balancing unit is used to perform data balancing on the preprocessed data. During the data balancing, the multi-label problem is first converted into a multi-classification problem using the LabelPowerset method, and a corresponding balancing method is adopted for each type of sample according to the number of samples, and then the balanced multi-classification labels are mapped back to the multi-label format; The feature optimization and classification unit is used to complete the feature optimization and multi-label classification prediction of bridge defects using a two-stage analysis method, wherein the two-stage analysis method includes: The first stage is single disease feature extraction and prediction: the feature importance evaluation method based on the random forest model is used to extract high-importance features related to single disease prediction, and then the multi-label classification algorithm CatBoost is used to separately model and predict each disease, and the prediction probability of the single disease is output; The second stage is the disease interaction modeling and joint prediction: the single disease prediction probability generated in the first stage is used to construct the joint features between diseases, and the CatBoost model is used based on the joint features to model and predict the joint occurrence probability of bridge diseases, and the joint disease prediction probability is output; The risk assessment unit is used to perform risk clustering analysis on the bridge based on the joint disease prediction probability by using a fuzzy C-means clustering method, classify the bridge risks into different categories, and finally output the risk clustering results.

Images