Hollow slab girder bridge performance degradation prediction method and system based on road network-level bridge group long-period monitoring data

Abstract

The invention discloses a hollow slab girder bridge performance degradation prediction method and system based on road network level bridge group long-period monitoring data, and belongs to the technical field of bridge state monitoring, and the method comprises the following steps: constructing a hollow slab bridge group multi-source bridge state database based on bridge structure parameters, long-period traffic volume parameters and environmental load parameters; training to obtain a bridge state degradation prediction model based on a neural network by using the road network level region bridge long-period condition data; and utilizing the bridge state degradation prediction model to predict the technical condition of the bridge in the road network level area to obtain an upper structure score, a lower structure score, a bridge deck system score and a bridge technical condition score. According to the bridge state degradation prediction method based on multi-source data of the road network level hollow slab bridge group disclosed by the invention, prediction and evaluation of part-single bridge-road network service state degradation based on machine learning provide reference for medium and long-term maintenance planning of regional bridges.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bridge state monitoring, and particularly relates to a hollow slab bridge performance degradation prediction method based on long-period monitoring data of a bridge group at a road network level and a system thereof. BACKGROUND

[0002] Bridge technical condition scoring is an important basis for bridge service condition understanding by management and maintenance departments. Overly high scoring will cause a bridge to be evaluated as safe, resulting in missing the optimal maintenance opportunity and causing the bridge structural safety performance to be reduced. Overly low scoring will inevitably lead to excessive maintenance of the bridge, which greatly wastes financial and material resources. A large amount of bridge periodic detection data provides a data basis for regional bridge network level evaluation, but how to correctly depict the complex nonlinear logical relationship and internal connection between the data is still a problem to be solved. At the present stage, the traditional models describing the bridge technical condition degradation trend mainly include deterministic models and random models. Such models have many limitations in application. In recent years, with the promotion of intelligent transportation and the rise of artificial intelligence, machine learning is widely applied in the engineering field. The machine learning method has more advantages than the traditional evaluation method. Not only can the data deposited by a large number of detection reports be fully utilized, but also the fixed weight assignment based on the complex calculation formula for evaluating the bridge technical condition score can be avoided. The mapping relationship between the independent variables (evaluation parameters) and the dependent variables (bridge technical condition score) is described by using an implicit function, and the possible interaction between the variables is identified and extracted. The operability and realizability of the method are also more friendly to detection personnel and researchers. SUMMARY

[0003] The purpose of the present application is to overcome the deficiencies in the prior art and provide a hollow slab bridge performance degradation prediction method and system based on long-period monitoring data of a bridge group at a road network level.

[0004] To achieve the above-mentioned purpose, the present application is implemented by using the following technical solutions: On the one hand, the present application provides a hollow slab bridge performance degradation prediction method based on long-period monitoring data of a bridge group at a road network level, characterized in that it comprises the following steps: constructing a hollow slab bridge group multi-source bridge state database based on bridge structural parameters, long-period traffic volume parameters and environmental load parameters; training a bridge state degradation prediction model based on a neural network by using the multi-source bridge state database and long-period condition data of regional bridges at a road network level; predicting the technical condition of regional bridges at a road network level by using the bridge state degradation prediction model to obtain an upper structure score, a lower structure score, a bridge deck system score and a bridge technical condition score.

[0005] Further, the multi-source bridge state database comprises traffic volume, total length of bridge, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature.

[0006] Further, the correlation between the bridge performance degradation and the multi-source bridge state database is measured based on a Pearson coefficient, and a feature candidate set with high correlation to the bridge score is screened from each variable in the multi-source bridge state database, and the feature candidate set corresponding to the superstructure score is traffic volume, bridge length, bridge width, service life, and annual maximum temperature. The feature candidate set corresponding to the substructure score is bridge length, service life, annual maximum temperature, and annual minimum temperature. The feature candidate set corresponding to the bridge deck system score is traffic volume, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature. The feature candidate set corresponding to the bridge technical condition score is traffic volume, bridge length, number of lanes, service life, annual maximum temperature, and annual minimum temperature.

[0007] Further, a bridge state degradation prediction model based on a neural network is trained, including the following steps: The typical degradation mode of the bridge structure is analyzed, the degradation feature variable is initially selected, the candidate feature set of the sensitive variable is obtained, and the complete training sample input is formed by encoding processing according to different data types; The topological structure of the neural network is determined, the data set samples are divided into two subsets, i.e., the training set samples and the verification set samples, according to the proportion of 70% and 30%, the network is trained, the accuracy of the verification set data after training is tested, and the bridge state degradation prediction model is output under the premise of meeting the accuracy condition; The bridge structure parameters (bridge length, bridge width, number of lanes, and service life) and the operation environment parameters (annual maximum temperature, annual minimum temperature, and traffic volume) that need to be evaluated and predicted are input into the bridge state degradation prediction model, so as to judge the future state of the bridge.

[0008] On the other hand, the present application provides a hollow slab bridge performance degradation prediction system based on long-period monitoring data of a bridge group at a road network level, comprising the following modules: A data module is used to construct a multi-source bridge state database of the hollow slab bridge group based on bridge structure parameters, long-period traffic volume parameters, and environmental load parameters; A model module is used to train a bridge state degradation prediction model based on a neural network by using the multi-source bridge state database and long-period condition data of regional bridges at a road network level; A prediction module is used to predict the technical condition of regional bridges at a road network level by using the bridge state degradation prediction model, and obtain the superstructure score, the substructure score, the bridge deck system score, and the bridge technical condition score.​

[0009] Furthermore, the multi-source bridge status database includes traffic volume, total bridge length, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature.

[0010] Furthermore, based on the Pearson coefficient to measure the correlation between bridge performance degradation and the multi-source bridge status database, the following feature candidate sets with high correlation to bridge scores are selected from each variable in the multi-source bridge status database: The candidate feature set corresponding to the superstructure score is: traffic volume, bridge length, bridge width, service life, and annual maximum temperature; The candidate feature set corresponding to the substructure score is: bridge length, service life, annual maximum temperature, and annual minimum temperature; The candidate feature set corresponding to the bridge deck system score is: traffic volume, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature; The candidate set of features corresponding to the bridge technical condition score is: traffic volume, bridge length, number of lanes, service life, annual maximum temperature, and annual minimum temperature.

[0011] Furthermore, a neural network-based bridge state degradation prediction model is trained, including the following steps: The typical degradation patterns of bridge structures are analyzed, and degradation characteristic variables are initially selected based on this. Candidate feature sets of sensitive variables are obtained, and encoding processing is performed for different data types to form complete training sample inputs. The topology of the neural network is determined, and the dataset samples are divided into two subsets according to the ratio of 70% and 30%, namely the training set samples and the validation set samples. The network is trained, and the accuracy of the validation set data after training is checked. Under the premise of meeting the accuracy conditions, the bridge state degradation prediction model is output. The bridge structural parameters (bridge length, bridge width, number of lanes, service life) and operating environment parameters (annual maximum temperature, annual minimum temperature, and traffic volume) that need to be evaluated and predicted are input into the bridge condition degradation prediction model to determine the future condition of the bridge.

[0012] Furthermore, the predicted condition of bridges can be classified according to the predicted score. According to the "Highway Bridge Technical Condition Assessment Standard JTG / T H21—2011", the technical condition level of bridges is divided into 5 categories: Category 1 [95,100], Category 2 [80,95), Category 3 [60,80), Category 4 [40,60), and Category 5 [0,40]. The predicted changes in the technical condition level of bridges at the road network level within 5 years correspond to the timing of component maintenance planning for specific bridges, thereby enabling the development of medium-term maintenance planning for component-bridge group maintenance objectives.

[0013] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The method and system for predicting bridge condition degradation based on multi-source data of road network-level hollow slab bridge groups provided by this invention utilize multi-source data such as bridge structural parameters, long-term traffic volume parameters, and environmental load parameters as model inputs. It analyzes the predictive effect of multi-dimensional features on the technical condition of a region over the years and verifies the results by combining historical bridge inspection data from different years. The results show that the predicted values ​​of the established multi-dimensional model are close to the measured values, meeting the accuracy requirements of practical engineering. This enables intelligent scientific decision-making for regional bridge operation and can provide a reference for research on evolution patterns and condition assessment based on historical inspection data. Attached Figure Description

[0014] Figure 1 A flowchart of the prediction method provided in an embodiment of the present invention.

[0015] Figure 2 This is a density curve of regional technical condition scoring provided for an embodiment of the present invention.

[0016] Figure 3 The annual temperature variation diagrams for various regions are provided for embodiments of the present invention.

[0017] Figure 4 Traffic volume composition diagram provided for embodiments of the present invention.

[0018] Figures 5-8 The diagram shows the relationship between the scores of various structures and related quantities provided in the embodiments of the present invention.

[0019] Figure 9 This is a schematic diagram of a neural network architecture provided in an embodiment of the present invention.

[0020] Figure 10 The diagram shows the MP neuron model provided in the embodiments of the present invention.

[0021] Figure 11 A flowchart of BP neural network operation provided for an embodiment of the present invention.

[0022] Figure 12 The topology diagram of the BP neural network model provided in the embodiment of the present invention is shown.

[0023] Figures 13-16 The image shows the model training results provided in the embodiments of the present invention.

[0024] Figures 17-20 The image shows the road network prediction results provided in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0026] like Figure 1 As shown, this embodiment of the invention provides a method for predicting the performance degradation of hollow slab beam bridges based on long-term monitoring data of road network-level bridge groups, characterized by including the following steps: Construct a multi-source bridge state database for hollow slab bridge groups based on bridge structural parameters, long-cycle traffic volume parameters, and environmental load parameters; Using the multi-source bridge status database and long-term status data of regional bridges at the road network level, a bridge status degradation prediction model based on a neural network is trained and obtained. The bridge condition degradation prediction model is used to predict the technical condition of bridges in a road network-level area, and the scores for the superstructure, substructure, bridge deck system, and bridge technical condition are obtained.

[0027] In this embodiment, the multi-source bridge status database includes traffic volume, total bridge length, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature.

[0028] Based on bridge inspection reports from a prefecture-level city between 2011 and 2020, a comprehensive maintenance evaluation model was established. The data is primarily concentrated in three regions: Region 1, Region 2, and Region 3. Therefore, bridge data was divided into groups according to these regions. Statistics were compiled on the regions and cumulative number of bridges inspected, yielding an overview of the periodic inspection samples for each region. A total of 572 bridges were inspected across the three regions: Region 1 had 262 bridges (45.8%), Region 2 had 100 bridges (17.5%), and Region 3 had 210 bridges (36.7%).

[0029] The historical temperature and climate parameters of the three regions up to 2023 were analyzed. Since the regions are close to each other and the average temperature is not much different, the highest and lowest temperatures of each region were considered as climate data parameters.

[0030] Among ordinary national and provincial highway bridges, as of 2020, there was one bridge in Region 2 with a service life of over 30 years; in Region 3, there were two bridges with a service life of 20-30 years, three bridges with a service life of 30-40 years, and one bridge with a service life of over 50 years. The remaining 565 bridges, accounting for 98.7%, were built after 2000. Considering the lack of data for 2021-2024, these 565 bridges were categorized into five groups based on their service life: 1-4 years, 5-8 years, 9-12 years, 13-16 years, and 17-20 years.

[0031] Traffic volume is an external determinant of factors affecting road surface performance changes. The planning, scientific selection and evaluation of highway construction projects, as well as highway maintenance and management, all require traffic volume as the most basic data support. Therefore, traffic volume analysis is necessary. According to the traffic volume and composition statistics of national and provincial trunk highways in a certain city in 2024, the average annual mixed traffic volume of ordinary national and provincial highways was 12,855 vehicles / day. In terms of the composition of motor vehicle traffic, passenger cars accounted for 77%, and freight cars accounted for 23%.

[0032] Throughout a bridge's lifespan, structural performance inevitably deteriorates due to internal material degradation and the combined effects of external loads and the environment. It is generally believed that there is a correlation between the structural condition of a bridge and a series of independent variables. This model describing the degree of structural degradation and its future trend is called a bridge degradation prediction model. It is used to predict the condition of bridge components or the entire bridge at a specific future time. Typically, a technical condition score is used to measure the structural condition Y, with sensitive parameters affecting the bridge's degradation trend as independent variables X, such as bridge span, bridge type, and bridge service life. This mathematical model can be expressed as: ; Based on the integrated multi-source data mentioned above, each individual bridge is represented and stored using a key-value pair method consisting of feature attributes and attribute values. The former specifies the specific features, and the latter assigns corresponding data to those features. Bridge individual attributes include information from three aspects: road network level (traffic volume, etc.), bridge level (service life, bridge length, etc.), and component level (component technical condition). Different attributes have different data formats, such as textual and numerical types. Textual data refers to the names or labels of the research objects, while numerical data indicates orders of magnitude relationships between data points.

[0033] Based on the analysis of bridge degradation patterns and existing research, a bridge condition database reflecting the degradation patterns of bridge technical conditions is established. The database inputs 11 characteristic indicators: traffic volume, total bridge length, bridge width, number of lanes, service life, annual maximum temperature, annual minimum temperature, superstructure score, substructure score, bridge deck system score, and bridge technical condition score. Traffic volume, total bridge length, bridge width, number of lanes, service life, annual maximum temperature, and annual minimum temperature are input variables, while superstructure score, substructure score, bridge deck system score, and bridge technical condition score are output variables. The training set for the database requires comparison of the accuracy between the input and output labels.

[0034] Bridge group data is divided into regions 1, 2, and 3; traffic volume is the annual average daily traffic volume of the corresponding national and provincial highways, with missing years estimated using a fitting formula; total bridge length and width indicate the corresponding bridge attributes; the number of lanes includes two-lane, three-lane, and four-lane, with 2 representing two lanes, 3 representing three lanes, and 4 representing four lanes; service life is the year of opening to traffic corresponding to the scheduled maintenance date; the annual highest and lowest temperatures represent regional operating environment data; the superstructure score, substructure score, bridge deck system score, and bridge technical condition score are all based on a percentage system. A portion of the data is shown in Table 1 below.

[0035] Table 1 Long-term multi-source monitoring database Traffic volume Total length of bridge Width of bridge Number of lanes Service life Annual maximum temperature Annual minimum temperature Superstructure rating Substructure rating Deck system rating Bridge technical condition rating 13941 34.3 12.75 2 9 36.8 -6 96.69 91.27 90.69 93.32 13941 34.3 12.75 2 9 36.8 -6 98.69 92.05 90.69 94.44 13941 34.3 12.85 2 10 36.8 -6 92.72 95.6 86.53 92.63 13941 34.3 12.85 2 10 36.8 -6 94.92 95.27 94.17 94.91 13941 28.3 12.85 2 9 36.8 -6 91.51 87.95 97.22 87.22 13941 28.3 12.85 2 9 36.8 -6 91.68 87.67 100 83.05 13941 25.5 12.85 2 9 36.8 -6 90.94 95.96 98.33 94.43 13941 25.5 12.85 2 9 36.8 -6 92.72 93.12 98.33 94 11158 28.2 17.2 4 15 36.8 -6 86.82 93.5 78.38 87.81 11158 27.7 17.2 4 15 36.8 -6 78.33 93.46 82.06 85.13 11158 27.6 17.2 4 15 36.8 -6 74.61 95.48 76.91 83.42 11158 27 17.2 4 15 36.8 -6 90.4 98.25 76.91 90.85 11158 34.2 17 4 15 36.8 -6 93.52 98.29 88.68 94.46 11158 27.7 17 4 15 36.8 -6 85.64 93.24 80.15 87.58 11158 27.6 17 4 15 36.8 -6 78.5 93.37 76.91 84.13 11158 34.2 17 4 15 36.8 -6 92.08 93.53 82.06 90.65 11158 31 23 4 15 36.8 -6 87.96 89.78 80.59 87.22 11158 35.3 17 4 15 36.8 -6 89.53 95.12 76.91 89.24 11158 55.6 17 4 15 36.8 -6 89.49 88.15 76.91 86.44 11158 34 17 4 15 36.8 -6 87.24 93.53 76.91 87.69 11158 27.6 17 4 15 36.8 -6 96.62 93.74 83.82 92.91 13941 53 25 2 9 36 -7 73.68 100 92.3 87.93 13941 53 25 2 9 36 -7 80.67 100 88 89.87 13941 44.582 26 2 9 36 -7 77.57 99.85 92.3 89.43 13941 44.582 26 2 9 36 -7 79.34 100 88.21 89.38 13941 33 25 2 9 36 -7 88.86 100 92.3 94 13941 33 25 2 9 36 -7 87.52 97.45 93.57 92.7 13941 95 25.4 2 9 36 -7 94.75 98.47 100 97.29 13941 95 25.4 2 9 36 -7 94.17 96.35 92.3 94.67 The Pearson coefficient is the most commonly used correlation coefficient to measure the correlation between two interval variables. This coefficient has a direction: if one variable increases as the other increases, it is called a positive correlation, indicated by a positive sign; if one variable increases and the other decreases, it is called a negative correlation, indicated by a negative sign. The formula for calculating the Pearson coefficient is: ; In the formula: x i , y i Variables X , Y The i-th sample value; x , y These are the sample means of the two variables, respectively; n The sample size is given. The Pearson coefficient is applicable when the two variables have a linear relationship and are both continuous variables obtained from measurements; the populations of both variables follow a normal distribution or a unimodal symmetrical distribution that is close to normal; the observations of the two variables are paired, and each pair of observations must be independent of each other.

[0036] The correlation between bridge ratings and various variables was analyzed based on statistical correlation coefficients. It can be seen that the bridge superstructure is significantly affected by traffic volume, but least affected by annual minimum temperature and number of lanes; the correlation coefficients for the remaining variables are all above 0.15. Temperature changes significantly affect the substructure, and service life is a crucial factor influencing substructure performance degradation. The primary influencing factor for the bridge deck system is the number of lanes, with a correlation coefficient exceeding -0.6, and it is also affected by temperature. Notably, the overall bridge rating decreases with increasing traffic volume and number of lanes, indirectly reflecting the unique contribution of traffic volume to bridge deterioration. Correlation analysis reveals the dependencies between bridge age, length, width, number of lanes, traffic volume, service life, and environmental parameters and the bridge deck system rating, superstructure rating, substructure rating, and overall rating. It also reveals the unique contribution of each variable to bridge performance degradation, laying the foundation for the deduction and prediction of long-term bridge structural performance. However, some correlation coefficients are at low levels, i.e., low or very low correlation. This is partly because the differences in the description habits and cognitive levels of different inspectors in bridge defects lead to inaccuracy and instability of the scoring data, excessive data variability, and the correlation expression is significantly affected by noise. Another reason is that the Pearson correlation coefficient is only sensitive to linear data and is difficult to quantify and identify the nonlinear relationship that may exist between variables, thus some correlation coefficients are low.

[0037] By comprehensively considering the different correlation patterns among the data, we deeply explored the inherent characteristics and correlations among bridge data. It was found that traffic volume, service life, and temperature parameters all played important roles. Using 0.15 as the critical value for correlation coefficient, insensitive variables with correlation coefficients below 0.15 were eliminated in subsequent bridge condition degradation modeling. This allowed us to select a candidate set of features highly correlated with bridge scores, reducing model dimensionality and computational complexity, and minimizing unnecessary calculations in subsequent model training and prediction processes.

[0038] The final candidate sets of variables, arranged according to their correlation coefficients, are as follows: [Superstructure: traffic volume, bridge length, bridge width, service life, maximum temperature], [Substructure: bridge length, service life, maximum temperature, minimum temperature], [Bridge deck system: traffic volume, bridge width, number of lanes, service life, maximum temperature, minimum temperature], [Overall: traffic volume, bridge length, number of lanes, service life, maximum temperature, minimum temperature].

[0039] For the above candidate set of variables, a database of all data is first established. Then, some data are not related to all parameters of specific components, such as the superstructure. This can be understood as insensitive parameters being detrimental to the prediction results. Only the most sensitive parameters need to be selected to make the prediction results most accurate. Therefore, a candidate set of features is established. It can be seen that four models are actually established: superstructure, lower structure, bridge deck, and overall structure. Both components and the overall structure are predicted to facilitate the analysis of internal relationships and verification.

[0040] Numerous neurons (MP neurons) are the basic structural components of a neural network model and the most basic processing unit. They can generally be divided into three categories: input neurons, output neurons, and hidden neurons. Input neurons are responsible for receiving external information and data, output neurons output the results of internal system processing, and hidden neurons are located between input and output neurons, constructing the mapping relationship between training sample input data and output data. The complex logical operations and relationship recognition of the entire neural network are realized in the hidden layer.

[0041] Neurons can both receive input signals from other neurons and transmit output signals to other neurons. The way signals are transmitted is called a connection, and different connections are defined by different weights. When a neuron receives the total input value, it calculates the corresponding output parameters through an activation function and compares them with a threshold. The learning process of a neural network is equivalent to continuously adjusting the connection weights and thresholds.

[0042] Neural networks utilize activation functions to modify the outputs of neurons during information transmission, thereby altering the overall network output. The presence of activation functions endows neural networks with powerful learning and nonlinear mapping capabilities, making them a crucial component of neural networks. Without activation functions, regardless of the number of layers, a neural network is merely a cascaded perceptron, with the output being a linear combination of the inputs. Therefore, the choice of activation function is paramount. Commonly used activation functions include Sigmoid, Tanh, and ReLU. These three activation functions have different characteristic forms and produce different network training and prediction results for different data types.

[0043] The backpropagation algorithm is used to calculate the parameters of each neuron in the neural network. This algorithm can propagate forward layer by layer according to the error feedback. Each neuron adjusts the connection weights along the negative gradient direction and iterates multiple times until the error condition index is met. At this time, the network can better map the functional relationship between data.

[0044] The basic process of building a bridge condition degradation prediction model can be divided into three stages: training sample preparation stage, network model construction stage, and condition degradation prediction stage.

[0045] In the training sample preparation stage, the typical degradation patterns of bridge structures are first analyzed, and degradation feature variables are initially selected accordingly to obtain the key sensitive variable feature set. Then, encoding processing is performed for different data types to form complete training sample input.

[0046] In the neural network model construction stage, the topology of the neural network is first determined, then the network is trained using the training sample set, and the accuracy of the validation set data after training is checked. The model is output under the premise of meeting the accuracy conditions.

[0047] In the state degradation prediction stage, the bridge structural parameters and operating environment parameters that need to be evaluated and predicted can be input into different neural network models to determine the future state of the bridge.

[0048] First, the grid parameters are initialized, using the variable feature set data from the bridge condition database as input and the bridge technical condition score as output, where i (=1,2,...,n) is the sample number. A single-hidden-layer BP neural network is constructed as shown in the figure below, with 9 neurons in the input layer, 1 neuron in the output layer, and 10 neurons in the hidden layer (default setting). Then, the constructed network is used to calculate the output value and compare it with the theoretical value. By continuously adjusting the connection weights and thresholds between each layer, the network output is made closer to the theoretical output. The dataset samples are divided into two subsets at a ratio of 70% and 30%, namely the training set and the validation set. The training set minimizes the error between the predicted and expected values ​​by continuously adjusting the weights between neurons in the neural network, while the validation set is used to determine the optimal weights, mainly to verify the reliability of the model.

[0049] To eliminate the influence of different dimensions and units among variables, the original data was normalized to the [0,1] interval. This not only avoids the influence of outlier samples but also accelerates the learning rate of the neural network and improves training accuracy. After iteration, the distribution of data points and the goodness of fit were compared. The established neural network model showed high generalization and prediction capabilities on subsets, with correlation coefficients R reaching 0.86435, 0.90711, 0.88759, and 0.90408, respectively. These results indicate that the neural network model has a good fitting effect on randomly combined datasets and can better reflect the true relationship between bridge structural state and variables. It has great potential and research value in the application field of evaluating and predicting the future state of bridges.

[0050] Based on the hollow slab beam maintenance prediction model developed above, a road network-level verification was carried out using bridge inspection reports from a certain region. The model predicted the changes in the technical condition of 20 bridges across 3 lines over 5 years. However, the model only considers the natural degradation of bridge performance and does not consider the maintenance status during the maintenance planning period. This model can be matched with the timing of component maintenance planning for specific bridges, thereby enabling mid-term maintenance planning for component-bridge group maintenance objectives.

[0051] According to the "Standard for Technical Condition Assessment of Highway Bridges JTG / T H21—2011", the technical condition of bridges is divided into five categories: Category 1 [95,100], Category 2 [80,95), Category 3 [60,80), Category 4 [40,60), and Category 5 [0,40]. The condition of a bridge component is the weighted sum of its individual parts.

[0052] from Figures 17-20 The data shows that the overall technical condition of the 20 bridges on the route has gradually deteriorated over the past five years. A comparison with the latest bridge inspection technical condition in 2024 reveals that nine hollow slab bridges have undergone repairs, such as the Yinhe Bridge, which was repaired on November 1, 2020, maintaining its technical condition at Class 2. The technical condition assessment of 18 of the 20 hollow slab bridges was accurate, achieving a 90% accuracy rate. The performance of the superstructure and substructure has deteriorated rapidly. The substructure was predicted to have Class 4 components in 2024, requiring component maintenance. In particular, the changes in the superstructure are consistent with the development of hollow slab beam defects. The technical condition of the bridge deck system has gradually deteriorated, but overall, it remains at Class 3.

[0053] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A hollow slab bridge performance degradation prediction method based on long-period monitoring data of a bridge group at a road network level, characterized in that, The method comprises the following steps: constructing a hollow slab bridge group multi-source bridge state database based on bridge structure parameters, long-period traffic volume parameters and environmental load parameters; training a bridge state degradation prediction model based on a neural network using the multi-source bridge state database and long-period regional bridge condition data at a road network level to obtain the bridge state degradation prediction model; predicting the technical condition of a bridge at a road network level using the bridge state degradation prediction model to obtain an upper structure score, a lower structure score, a bridge deck system score and a bridge technical condition score.

2. The hollow slab bridge performance degradation prediction method based on long-period monitoring data of a bridge group at a road network level according to claim 1, characterized in that, The multi-source bridge state database includes traffic volume, total bridge length, bridge width, number of lanes, service life, annual maximum temperature and annual minimum temperature.

3. The hollow slab bridge performance degradation prediction method based on long-period monitoring data of bridge groups at the road network level according to claim 2, characterized in that, Pearson coefficients are used to measure the correlation between bridge performance degradation and the multi-source bridge state database, and the feature candidate set with a higher correlation to the bridge score is selected from each variable in the multi-source bridge state database as follows: The feature candidate set corresponding to the upper structure score is traffic volume, bridge length, bridge width, service life and annual maximum temperature; The feature candidate set corresponding to the lower structure score is bridge length, service life, annual maximum temperature and annual minimum temperature; The feature candidate set corresponding to the bridge deck system score is traffic volume, bridge width, number of lanes, service life, annual maximum temperature and annual minimum temperature; The feature candidate set corresponding to the bridge technical condition score is traffic volume, bridge length, number of lanes, service life, annual maximum temperature and annual minimum temperature.

4. The hollow slab bridge performance degradation prediction method based on long-period monitoring data of bridge groups at the road network level according to claim 1, characterized in that, The bridge state degradation prediction model based on the neural network is trained, including the following steps: analyze the typical degradation mode of the bridge structure, select the degradation feature variables accordingly, obtain the candidate feature set of sensitive variables, and perform encoding processing on different data types to form complete training sample inputs; determine the topology structure of the neural network, divide the data set samples into two subsets, i.e., training set samples and validation set samples, in a 70:30 ratio, perform network training, and perform precision testing on the validation set data after training, and output the bridge state degradation prediction model under the condition of meeting the precision; input the bridge structure parameters (bridge length, bridge width, number of lanes and service life), operating environment parameters (annual maximum temperature, annual minimum temperature and traffic volume) to be evaluated and predicted into the bridge state degradation prediction model to determine the future state of the bridge.

5. The hollow slab bridge performance degradation prediction method based on long-period monitoring data of bridge groups at the road network level according to claim 1, characterized in that, The predicted condition of the bridge can be classified according to the prediction score, and according to the “Highway Bridge Technical Condition Evaluation Standard JTG / T H21-2011”, the bridge technical condition level is divided into five categories, i.e., category 1 [95, 100], category 2 [80, 95), category 3 [60, 80), category 4 [40, 60) and category 5 [0, 40); the change in the technical condition level of the predicted road network level bridge within five years is corresponding to the specific bridge component maintenance planning opportunity, so that the medium-term maintenance planning is carried out for the component-bridge group maintenance target.

6. A hollow slab bridge performance degradation prediction system based on long-period monitoring data of a bridge group at a road network level, characterized by, The method comprises the following modules: a data module for constructing a hollow slab bridge group multi-source bridge state database based on bridge structure parameters, long-period traffic volume parameters and environmental load parameters; A model module is configured to train a bridge state degradation prediction model based on a neural network by using the multi-source bridge state database and the bridge long-period condition data of the road network level region. A prediction module is configured to predict the technical condition of the bridge of the road network level region by using the bridge state degradation prediction model to obtain the superstructure score, the substructure score, the bridge deck system score and the bridge technical condition score.

7. The hollow slab bridge performance degradation prediction system based on long-period monitoring data of bridge groups at road network level according to claim 6, characterized in that, The multi-source bridge state database includes traffic volume, bridge total length, bridge width, lane number, service life, annual maximum temperature and annual minimum temperature. 8.The system for hollow slab bridge performance degradation prediction based on long-period monitoring data of bridge groups at road network level according to claim 6, characterized in that, Pearson coefficient is used to measure the correlation between bridge performance degradation and the multi-source bridge state database, and the feature candidate set with high correlation to the bridge score is selected from each variable in the multi-source bridge state database. The feature candidate set corresponding to the superstructure score includes traffic volume, bridge length, bridge width, service life and annual maximum temperature. The feature candidate set corresponding to the substructure score includes bridge length, service life, annual maximum temperature and annual minimum temperature. The feature candidate set corresponding to the bridge deck system score includes traffic volume, bridge width, lane number, service life, annual maximum temperature and annual minimum temperature. The feature candidate set corresponding to the bridge technical condition score includes traffic volume, bridge length, lane number, service life, annual maximum temperature and annual minimum temperature. 9.The hollow slab bridge performance degradation prediction system based on long-period monitoring data of bridge groups at road network level according to claim 6, characterized in that, The bridge state degradation prediction model based on the neural network is trained, including the following steps: Analyze the typical degradation mode of the bridge structure, select the degradation feature variable, obtain the candidate feature set of the sensitive variable, and encode the different data types to form a complete training sample input; Determine the topology structure of the neural network, divide the data set sample into two subsets according to the proportion of 70% and 30%, that is, the training set sample and the verification set sample, train the network, and test the accuracy of the verification set data after training, and output the bridge state degradation prediction model under the condition of meeting the accuracy; Bridge structure parameters (bridge length, bridge width, lane number and service life), operating environment parameters (annual maximum temperature, annual minimum temperature and traffic volume) are input into the bridge state degradation prediction model to judge the future state of the bridge.

10. The hollow slab bridge performance degradation prediction system based on long-period monitoring data of bridge groups at road network level according to claim 6, characterized in that, The bridge prediction condition can be classified according to the prediction score, and according to the "Highway Bridge Technical Condition Evaluation Standard JTG / T H21-2011", the bridge technical condition level is divided into five categories, category 1 [95, 100], category 2 [80, 95), category 3 [60, 80), category 4 [40, 60) and category 5 [0, 40); the technical condition level change of the predicted road network level bridge within 5 years is corresponding to the specific bridge component maintenance planning opportunity, so as to carry out the medium-term maintenance planning for the component-bridge group maintenance target.

Images