Bridge condition prediction method based on fusion of category clustering and trend modeling

Abstract

The present application relates to a bridge technical condition prediction method based on category clustering and trend modeling fusion, belongs to the technical field of bridge state prediction and maintenance management, and solves the problems of limited generalization ability and low prediction accuracy of existing prediction models. The method first collects periodic detection data of multiple bridges and performs preprocessing; after feature encoding and dimension reduction, the bridges are grouped using the K-means clustering algorithm, and the Huber regression method is used to establish an intra-category degradation trend model; the XGBoost regression method is used to establish an intra-category technical condition prediction model; according to the intra-category technical condition prediction model and the intra-category degradation trend model, the target bridge's bridge technical condition prediction results for multiple future prediction years are generated through a dynamic weighted fusion formula; and finally, the model performance is evaluated. The present application can effectively improve the accuracy of bridge technical condition prediction, provide reliable long-term degradation trend prediction, and optimize bridge maintenance decisions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge state prediction and maintenance management, and particularly relates to a bridge technical condition prediction method based on category clustering and trend modeling fusion, and is particularly used for predicting the bridge technical condition grade and analyzing the degradation trend under the condition of limited inspection data, so as to provide intelligent technical support for bridge maintenance decision and maintenance optimization. BACKGROUND

[0002] With the continuous growth of road traffic volume and the aggravation of infrastructure aging, the safety and durability of bridge structures are facing great challenges. Bridge technical condition assessment and future degradation trend prediction are key links in bridge daily operation and maintenance and long-term health management, and are of great significance for ensuring road smoothness, prolonging bridge service life and reasonably formulating maintenance and repair strategies.

[0003] At present, the bridge management system at home and abroad mainly relies on periodic detection and manual inspection to obtain the technical condition grade of the bridge, and predicts the future technical condition of the bridge through expert experience method, regression model or single machine learning model.

[0004] For example, the paper "Bridge Technical Condition Prediction Based on Regression Analysis Method" published in the journal of "Beijing University of Civil Engineering and Architecture" proposes to use regression analysis method to fit the bridge technical condition degradation model, and then predict the development trend of the bridge technical condition.

[0005] For example, the paper "Research on Development Trend Prediction Method of In-service Bridge Technical Condition Grade" published in the journal of "Highway Traffic Technology (Application Technology Edition)" establishes the development trend model of various bridge technical condition grades with operation time by linear regression on the historical data of highway bridge technical condition grade evaluation.

[0006] For example, the paper "Research on Gray Markov Chain Model of Bridge Technical Condition Prediction" published in the journal of "Journal of Wuhan University of Technology (Transportation Science and Engineering Edition)" combines the gray model GM(1, 1) with Markov chain, and considers the overall change and local fluctuation of the bridge technical condition.

[0007] For example, the paper "Research on Degradation Law of Beam Bridge Based on Markov Process" published in the journal of "Journal of Jiamusi University (Natural Science Edition)" uses Markov theory and genetic algorithm of nonlinear programming to study the time-varying law of preventive maintenance technical condition of concrete highway beam bridge.

[0008] A bridge technology condition and disease prediction backstepping algorithm method based on CPSO-BP neural network model is proposed in Chinese patent CN113569908A "Bridge technology condition and disease prediction backstepping algorithm method based on deep learning", which accurately predicts the bridge disease location, type and degree by optimizing the neural network structure and parameters.

[0009] A method for predicting the service performance degradation of reinforced concrete bridges is proposed in Chinese patent CN111160528A "Method for predicting service performance degradation of reinforced concrete bridges", which accurately predicts the technical condition score of reinforced concrete bridges in the future several years by training and parameterizing multiple LSTM neural networks using annual detection report data.

[0010] A method for quickly identifying the bridge technology condition level based on natural language processing is provided in Chinese patent CN114036258A "Method for quickly identifying bridge technology condition level based on natural language processing", which includes organizing bridge health state information into text description, classifying technology condition level, and using machine learning algorithm to establish training model to realize quick identification.

[0011] A method for intelligent evaluation of bridge technology condition based on machine learning is proposed in the paper "Method for intelligent evaluation of bridge technology condition based on machine learning" published in Chang'an University Journal (Natural Science Edition), which realizes accurate evaluation of bridge technology condition by constructing bridge state database and machine learning algorithm.

[0012] Most of these methods focus on modeling the historical data of single-span bridges, and cannot fully tap the commonality of structure and degradation behavior among similar bridges, resulting in limited model generalization ability, especially in terms of limited data volume and long-term trend prediction accuracy. On the other hand, existing technologies often ignore the differences between bridge categories and the degradation rules of different categories of bridges, and the modeling of degradation trends is mainly based on the average level of all bridges, which is difficult to reflect the real degradation process within the internal fine categories. In addition, traditional methods often only use a single short-term prediction model, and fail to realize the organic combination of short-term high-precision prediction and long-term trend extrapolation, resulting in actual prediction results that cannot effectively support fine bridge maintenance decisions.

[0013] Therefore, there is an urgent need for a new method for predicting the technology condition of bridges that can comprehensively utilize limited inspection data, combine bridge clustering and degradation trend modeling, and realize dynamic fusion of short-term prediction and long-term trend extrapolation, in order to improve the intelligent and scientific level of bridge health management and make up for the shortcomings of existing technologies. SUMMARY

[0014] In view of the low prediction accuracy of conventional data-driven models in the case of limited bridge inspection data or incomplete historical records, a bridge technical condition prediction method fusing structural mechanics mechanism is urgently needed. The present application proposes a bridge technical condition prediction method based on category clustering and trend modeling fusion. The method finally realizes the prediction of the technical condition of the bridge and the evaluation of the degradation trend under limited data through data acquisition and preprocessing, category feature clustering and trend analysis, intra-category prediction model establishment, joint prediction and dynamic weighted fusion, result evaluation and comparison verification. The method can effectively improve the accuracy of bridge technical condition prediction and provide reliable long-term degradation trend prediction, optimizing the maintenance decision of the bridge.

[0015] To solve the above technical problems, the present application adopts the following technical solution:

[0016] The bridge technical condition prediction method based on category clustering and trend modeling fusion comprises the following steps:

[0017] Step 1: Collect the periodic detection data of multiple bridges, including bridge code, construction year, detection year, evaluation grade, structural parameters and material properties, and preprocess the collected data, including missing value filling, outlier processing, feature calculation and standardization processing, to obtain structured pretreated data;

[0018] Step 2: Feature encoding and dimensionality reduction are performed on the structured pretreated data obtained in step 1, and then K-means clustering algorithm is used to group the bridges to obtain several bridge categories with similar structures and degradation behaviors, and then significant degradation bridge samples are selected in each bridge category, and Huber regression method is used to fit the degradation trend curve of bridge age-technical condition in each bridge category to establish the intra-category degradation trend model;

[0019] Step 3: In each bridge category, XGBoost regression method is used to establish the intra-category technical condition prediction model based on historical evaluation grades and static features;

[0020] Step 4: According to the intra-category technical condition prediction model and the intra-category degradation trend model corresponding to the bridge category where the target bridge is located, the bridge technical condition prediction results of the target bridge in multiple future prediction years are generated through a dynamic weighted fusion formula. In the dynamic weighted fusion formula, the weight of the intra-category technical condition prediction model is higher in short-term prediction, and the weight of the intra-category degradation trend model is higher in long-term prediction, and the sum of the weights of the intra-category technical condition prediction model and the intra-category degradation trend model is 1;

[0021] Step 5: The bridge technical condition prediction results obtained in step 4 are evaluated in terms of model performance using evaluation indexes, including accuracy, mean absolute error and root mean square error.

[0022] Compared with the prior art, the present application has the following beneficial effects:

[0023] (1) Improve prediction accuracy and applicability: the present application overcomes the shortcomings of low prediction accuracy and poor generalization ability of traditional single model prediction by combining category clustering and trend modeling, significantly improving the accuracy and robustness of bridge technical condition prediction, especially suitable for actual engineering scenarios with limited inspection data;

[0024] (2) Realize short-term and long-term dynamic fusion prediction: the dynamic weighted fusion of intra-class technical condition prediction model and intra-class degradation trend model is adopted, which takes into account the high accuracy of short-term prediction and the rationality of long-term trend extrapolation, effectively making up for the shortcomings of existing methods in long-term degradation prediction;

[0025] (3) Reflect category differences and reveal degradation rules: by clustering the structure, material and other characteristics of the bridge, and fitting the degradation trend within the category, the degradation differences and evolution rules of different categories of bridges can be fully revealed, improving the engineering reference value of the prediction results;

[0026] (4) The method is scientific and reliable, and has strong engineering practicability: the present application combines various cross-validation and evaluation indexes to fully verify the scientificity and superiority of the method, which is easy to integrate into a bridge health management system, improving the intelligent level and management efficiency of bridge maintenance decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the example embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0028] Figure 1 The flow chart of the bridge technical condition prediction method based on category clustering and trend modeling fusion according to one of the embodiments of the present application;

[0029] Figure 2 The clustering result graph of the bridge category;

[0030] Figure 3 The technical condition grade prediction curve graph of a certain bridge;

[0031] Figure 4 The future average degradation trend prediction graph of each category of bridges. DETAILED DESCRIPTION

[0032] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application. It should be understood that the described specific embodiments are only used to explain the present application, and are not used to limit the present application.

[0033] The present application proposes a bridge class clustering and trend modeling combined prediction method. The method mainly groups the bridges by using the K-means clustering algorithm on the preprocessed bridge data, and fits the degradation trend curve of the bridge age-technical condition within the category by using the Huber regression method. Then, in each bridge category, a machine learning regression method is used to establish an intra-class technical condition prediction model to realize the short-term inspection data-driven local prediction capability. Finally, joint prediction and dynamic weighted fusion are performed. For the target bridge, the intra-class technical condition prediction model and the intra-class degradation trend model are combined, and a dynamic weighted fusion scheme is used to generate the bridge technical condition prediction results for future years. Through this method of clustering first and then combined prediction, the characteristics of different categories of bridges are 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.

[0034] Specifically, as shown in Figure 1 The present application provides a bridge technical condition prediction method based on category clustering and trend modeling fusion. The method mainly includes the following steps 1 to step 5.

[0035] Step 1 (S100), data acquisition and preprocessing: collecting periodic detection data of multiple bridges, the periodic detection data including bridge code, construction year, detection year, evaluation grade, structure parameter and material characteristics, and preprocessing the collected periodic detection data, including missing value filling, outlier processing, feature calculation and standardization processing, to obtain structured preprocessed data after preprocessing.

[0036] In this embodiment, the periodic inspection data, i.e., bridge data, can be obtained from the bridge periodic inspection database. This database stores inspection data for multiple bridges. Taking five consecutive years (e.g., 2020-2024) of periodic inspection data as an example, the collected bridge data file includes the following main fields: bridge code, construction year, inspection year, rating level, total span length, maximum single-span span, bridge deck clear width, superstructure material name, and design load level. The rating level is mapped from text to numerical values ​​(e.g., "Class I" corresponds to 1, Class II to 2, and so on). For the obtained periodic inspection data, for fields such as construction year that are missing, the median of similar bridges is used to fill in the missing values. Then, based on the original data fields, the span ratio (total span length / maximum single-span span), width-to-span ratio (bridge deck clear width / maximum single-span span), material degradation coefficient (assigned from a table based on structure type), and load factor (numerically converted according to design load level) are automatically calculated. For all numerical features, standardization is performed using the mean and standard deviation of the same feature across all bridges to ensure consistent feature distribution across different years and bridges. For categorical features, one-hot encoding is used to generate 0-1 variables to adapt to model training. Outliers are addressed... The correction is performed using the box plot method shown in the following formula:

[0037] (1);

[0038] in, For the first The values ​​of each sample after outlier correction; and They are the 1st and 3rd quartiles, respectively. .

[0039] Step 2 (S200): Clustering and Trend Analysis of Category Features: Feature encoding and dimensionality reduction are performed on the preprocessed bridge data, i.e., the structured preprocessed data obtained in Step 1. Then, the K-means clustering algorithm is used to group the bridges to obtain several bridge categories with similar structures and degradation behaviors. Within each bridge category, bridge samples with significant degradation are selected. Then, the Huber regression method is used to fit the degradation trend curve of bridge age-technical condition within each bridge category to establish a degradation trend model within the category.

[0040] Step 2.1: Construct a bridge feature matrix based on the preprocessed bridge data It includes all bridge features after standardized coding, encompassing structural, material, operational, and static characteristics (such as bridge age, span ratio, material degradation coefficient, load factor, and recent rating). Principal component analysis (PCA) is used to analyze the bridge feature matrix. Dimensionality reduction, reducing the dimensionality According to the cumulative variance contribution rate, the target is to capture the main difference information, and the PCA dimension reduction process is expressed by the formula:

[0041] (2) ;

[0042] wherein, is the projection matrix composed of the first principal components; is the feature matrix after PCA dimension reduction, representing the projection characteristics of the bridge in the principal component space.

[0043] Step 2.2: Grouping of bridges is achieved by K-means clustering of PCA results. The number of clustering categories can be adaptively selected according to the actual number of bridges and the characteristic difference, preferably 3-6 categories, and the number of categories is set to in this embodiment. K-means clustering algorithm is used to group the bridges, and four bridge categories with similar structure and degradation behavior are obtained, namely cluster 0, cluster 1, cluster 2 and cluster 3. The objective function of the K-means clustering algorithm is:

[0044] (3) ;

[0045] wherein, is the PCA dimension reduction feature of the bridge , , is the th cluster center, is the number of categories, is the number of bridges participating in clustering, represents the square of the Euclidean distance between and the cluster center .

[0046] The clustering results can be visualized by the scatter plot of the first two dimensions of the principal components, as shown in Figure 2 .

[0047] Step 2.3: Next, the intra-class degradation trend is modeled, and the intra-class degradation trend model is established. First, the bridge samples with significant degradation are selected within each bridge class, and the selection criteria are that the maximum value of the bridge evaluation grade in the bridge class is greater than or equal to 4 in the past five years, and the score fluctuation range exceeds 1 point; then, for the selected bridge samples, regression samples are constructed, and the Huber regression method is used to fit the degradation trend curve of the bridge age-technical condition within each bridge class, to obtain the intra-class degradation trend model, and the calculation formula is:

[0048] (4) ;

[0049] wherein, For the bridge In the year The bridge technology condition assessment grade; For the bridge In the year The bridge age; For the regression coefficient; For the residual.

[0050] Step 3 (S300), establish the technology condition prediction model within the category: within each bridge category, based on the historical assessment grades and static characteristics of multiple years, use XGBoost regression method to establish the technology condition prediction model within the category, and realize the local prediction ability driven by short-term inspection data.

[0051] For each bridge category, within each category, the historical continuous four-year assessment grades And static characteristics Of the bridge Are input, and the next year's assessment grade Is output, wherein the static characteristics refer to the bridge body properties that do not change with the detection year, including but not limited to bridge age, span ratio, material degradation coefficient, load factor, structure material type, etc. These characteristics are usually used as model input together with the dynamic scores of previous years (such as technology condition grade) when modeling. XGBoost regression model is used to establish the technology condition prediction model within the category, and the prediction model formula is as follows:

[0052] (5) ;

[0053] Wherein, Represents the predicted value of the technology condition grade of the bridge In Year; Is the number of weak learners; Is the Th regression tree; Is the input feature vector, which is composed of the assessment grades Of the bridge In the four years before the prediction year and the static characteristics Of the bridge, which are spliced to form the complete input of the technology condition prediction model within the category.

[0054] Step 4 (S400), joint prediction and dynamic weighted fusion: for the target bridge, combining the intra-class technical condition prediction model and the intra-class degradation trend model corresponding to the bridge class to which the target bridge belongs, a bridge technical condition prediction result of the target bridge in multiple future prediction years is generated through a dynamic weighted fusion formula, and in the dynamic weighted fusion formula, the weight of the intra-class technical condition prediction model is higher in short-term prediction, that is, the intra-class technical condition prediction model is mainly used in short-term prediction, and the weight of the intra-class degradation trend model is higher in long-term prediction, that is, the weight of the intra-class degradation trend model is gradually increased in long-term prediction, and the intra-class degradation trend model is mainly used, and the sum of the weight of the intra-class technical condition prediction model and the weight of the intra-class degradation trend model is always 1.

[0055] Suppose the prediction year is the future year , the output of the intra-class technical condition prediction model is , and the output of the intra-class degradation trend model is , where is the bridge age of the bridge in the future year , then the dynamic weighted fusion formula can be expressed as:

[0056] (6).

[0057] Wherein, is the predicted value of the bridge technical condition level of the target bridge in the future year ; and is a dynamic weight, which is determined adaptively according to the prediction year and the fitting result of historical data, and is used to increase the weight of the output of the intra-class technical condition prediction model in short-term prediction and increase the weight of the output of the intra-class degradation trend model in long-term prediction. The setting of the dynamic weight makes the fusion model be able to flexibly adjust the contribution of each component according to the prediction year, so that the short-term prediction relies on the data-driven model (XGBoost) to obtain high precision, and the long-term prediction focuses on the robustness of the degradation trend model to prevent extrapolation distortion. The short-term prediction in the embodiment refers to the prediction of the bridge technical condition level of the target bridge when the prediction year is less than or equal to the year threshold (for example, 3), and the long-term prediction refers to the prediction of the bridge technical condition level of the target bridge when the prediction year is greater than the year threshold.

[0058] ​Optionally, in the dynamic weighted fusion formula, the weight of the intra-category technology condition prediction model monotonically decreases with the increase of the prediction year, while the weight of the intra-category degradation trend model monotonically increases with the increase of the prediction year. For example, in the first three years of the prediction years, the intra-category technology condition prediction model is dominant, so the weights of the intra-category technology condition prediction model are set to 0.9, 0.7, and 0.5 respectively. After three years, the weight of the intra-category degradation trend model gradually increases, and the dynamic weights... The formula is expressed as follows:

[0059] (7);

[0060] In this embodiment, the weights of the technical condition prediction model within the category are set to 0.9, 0.7, and 0.5 for the first three years, reflecting a high degree of trust in short-term historical data. From the fourth year onwards, the weights are reduced by 0.1 each year until they reach 0, ensuring that the prediction results have a reasonable physical evolution trend in the long term. This weighting scheme can not only ensure the applicability of the model in different prediction years, but also be flexibly adjusted according to actual needs.

[0061] After each prediction, the bridge technical condition prediction results are added to the historical data, and the predictions for subsequent years are recursively made. Figure 3 The figure shows a prediction curve of the technical condition level of a bridge over the next 10 years. The "fusion prediction" shown by the red curve in the figure is the prediction result of the bridge's technical condition obtained by using the method of this invention.

[0062] Step 5 (S500): Result evaluation and comparative verification: The model performance of the bridge technical condition prediction results obtained in Step 4 is evaluated using evaluation indicators, including accuracy, mean absolute error and root mean square error.

[0063] Accuracy ( The formula for calculating ) is:

[0064] (8);

[0065] in, For indicator functions, For the first The predicted value of the technical condition level of each bridge. For the first The actual value of the technical condition level of each bridge;

[0066] Mean absolute error ( The formula for calculating ) is:

[0067] (9);

[0068] Root mean square error (RMSE) The formula for calculating ) is:

[0069] (10);

[0070] wherein, is the total number of prediction samples.

[0071] Further, taking the periodic detection data of 2711 bridges as original data, cross validation is carried out on the 2711 bridges by using leave-one-out method, and the effect is compared with that of the traditional mean method and single regression method, so as to verify the effectiveness and superiority of the joint prediction method. The experimental results are shown in Table 1.

[0072] Table 1 Experimental results

[0073]

[0074] Figure 4 The future average degradation trend prediction graph of each type of bridge is shown. The average degradation trend refers to the joint prediction result (such as dynamic weighted fusion prediction) of the future years of the bridge technical condition grade of all bridges in the clustering category, and the trend curve obtained by taking the average value for each year. The "fusion mean" in the figure is the annual mean value of the fusion bridge technical condition grade prediction value of all bridges in the category, which is used to reflect the overall degradation evolution trend of the bridge.

[0075] In summary, the present application provides a bridge technical condition prediction method based on category clustering and trend modeling fusion. The method finally realizes the prediction of the technical condition of the bridge and the evaluation of the degradation trend under limited data through data acquisition and preprocessing, category feature clustering and trend analysis, category prediction model establishment, joint prediction and dynamic weighted fusion, result evaluation and comparison verification. By modeling and clustering different categories of bridges, the degradation trend of bridges in different categories can be fully mined. Finally, the joint prediction method is used to fuse the technical condition prediction model and the degradation trend model in the category, which can effectively improve the accuracy of the bridge technical condition prediction and provide reliable long-term degradation trend prediction, and optimize the maintenance decision of the bridge. Compared with the traditional method, the present application has strong practicability, high efficiency and broad application prospect. The present application has the advantages of fully mining the structure information of limited detection data, fusing the degradation evolution law of the category, considering both short-term and long-term prediction, and strong universality, and provides a scientific basis for bridge maintenance decision and risk assessment.

[0076] The above is only a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can make equivalent replacement or change according to the technical scheme and inventive concept of the present application within the scope disclosed by the present application, which belongs to the protection scope of the present application.

Claims

1. A bridge technology condition prediction method based on the fusion of category clustering and trend modeling, characterized in that, The method comprises the following steps: Step 1: collecting periodic detection data of a plurality of bridges, including bridge code, construction year, detection year, evaluation grade, structural parameter and material characteristic, and preprocessing the collected data, including missing value filling, abnormal value processing, feature calculation and standardization processing, to obtain structured pretreated data; Step 2: performing feature coding and dimension reduction on the structured pretreated data obtained in step 1, and then grouping the bridges by using a K-means clustering algorithm to obtain a plurality of bridge categories with similar structures and degradation behaviors, and screening out bridge samples with significant degradation in each bridge category, and then fitting a degradation trend curve of bridge age-technical condition in each bridge category by using a Huber regression method to establish an intra-category degradation trend model; Step 3: in each bridge category, an intra-category technical condition prediction model is established by using an XGBoost regression method based on historical evaluation grades and static features of a plurality of years; Step 4: according to the intra-category technical condition prediction model and the intra-category degradation trend model corresponding to the bridge category to which the target bridge belongs, a bridge technical condition prediction result of the target bridge in a plurality of prediction years in the future is generated by using a dynamic weighted fusion formula, in which the weight of the intra-category technical condition prediction model is higher in short-term prediction, the weight of the intra-category degradation trend model is higher in long-term prediction, and the sum of the weights of the intra-category technical condition prediction model and the intra-category degradation trend model is 1; Step 5: the bridge technical condition prediction result obtained in step 4 is evaluated in terms of model performance by using evaluation indexes, wherein the evaluation indexes include accuracy, mean absolute error and root mean square error. 2.The bridge technology condition prediction method based on the fusion of category clustering and trend modeling according to claim 1, wherein, The preprocessing in step 1 comprises: filling the numerical features with mean values; encoding the category features into one-hot variables for model training; Outliers The correction is made using the box plot method shown in the following equation: （1）； wherein, is the value of the i-th sample after outlier correction; is the value of the i-th sample after outlier correction; and are the 1st and 3rd quartiles, respectively, . 3.The bridge technology condition prediction method based on fusion of category clustering and trend modeling according to claim 1 or 2, characterized in that, Step 2 comprises: Step 2.1: Constructing the bridge feature matrix based on the structured pre-processed data obtained from Step 1 , and performing dimensionality reduction on the bridge feature matrix using the principal component analysis method . The PCA dimensionality reduction process is as follows: （2）； wherein, is a projection matrix composed of the first principal components; is a feature matrix after dimensionality reduction by the principal component analysis method; Step 2.2: grouping the bridges by using a K-means clustering algorithm, and the objective function of the K-means clustering algorithm is: （3）； wherein, is the PCA dimensionality reduction feature of the bridge, is the i-th cluster center, is the number of classes, is the number of bridges participating in the clustering, denotes the square of the Euclidean distance between the bridge and the cluster center .​​ Step 2.3: in each bridge category, screening out bridge samples with significant degradation, fitting a degradation trend curve of bridge age-technical condition in the bridge category by using a Huber regression method for the bridge samples with significant degradation, and obtaining an intra-category degradation trend model.

4. The bridge technical condition prediction method based on the fusion of category clustering and trend modeling according to claim 3, characterized in that, The intra-category degradation trend model is: （4）； wherein, is the bridge in years of the bridge technology condition assessment level; is the bridge in years of the bridge age; is the regression coefficient; is the residual.

5. The bridge condition prediction method based on fusion of category clustering and trend modeling according to claim 3, characterized in that, The screening standard for screening out bridge samples with significant degradation in the bridge category is that the maximum value of the evaluation grade of the bridge in the bridge category in the past five years is greater than or equal to 4, and the fluctuation range of the evaluation grade is more than 1.

6. The bridge technical condition prediction method based on fusion of category clustering and trend modeling according to claim 1 or 2, characterized in that, The intra-category technical condition prediction model is: （5）； wherein, represents a bridge In the predicted value of the technical status level in 2015; is the number of weak learners; is the regression tree; is the input feature vector.

7. The bridge technical condition prediction method based on fusion of category clustering and trend modeling according to claim 1 or 2, characterized in that, The dynamic weighted fusion formula is: （6）； wherein, is a predicted value of the future bridge technical condition rating of the target bridge in year ; is a dynamic weight; is an output of the within-class technical condition prediction model; is an output of the within-class deterioration trend model.

8. The bridge technical condition prediction method based on fusion of category clustering and trend modeling according to claim 7, characterized in that, Dynamic weights The formula is as follows: （7）； wherein is the forecast year. 9.The bridge technology condition prediction method based on fusion of category clustering and trend modeling according to claim 1 or 2, characterized in that, The calculation formula of the accuracy is: （8）； wherein, is an indicator function, is a predicted value of the technical condition class of the bridge, is an actual value of the technical condition class of the bridge; The calculation formula of the mean absolute error is: （9）； The calculation formula of the root mean square error is: （10）； wherein is the total number of prediction samples.

10. The bridge technical condition prediction method based on fusion of category clustering and trend modeling according to claim 1 or 2, characterized in that, The periodic detection data are collected from a bridge detection database.

Images