A method and system for predicting the technical condition of bridge infrastructure

Abstract

The application discloses a kind of prediction method and system for bridge infrastructure technical condition, belong to bridge structure health monitoring technical field, first data acquisition is carried out, historical maintenance behavior is quantified, and residual maintenance efficacy function is established, then machine learning regression model is used to predict the degradation amount under the given maintenance efficacy and environmental characteristics, and dynamic mapping mechanism is established, the degradation amount and maintenance efficacy are used to correct Markov intensity matrix in real time, the state transition probability calculation problem of non-equal interval detection data is solved by matrix exponential operation, finally, the model is closed-loop corrected using new measured data by introducing bayesian online updating framework.The application realizes the deep coupling of maintenance intervention and degradation mechanism, significantly improves the prediction accuracy of bridge under complex maintenance environment, and provides quantitative support for life cycle maintenance decision comparison and selection.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health monitoring technology, specifically to a method and system for predicting the technical condition of bridge infrastructure. Background Technology

[0002] As a core node of transportation infrastructure, the structural safety of bridges directly affects public safety and the stable operation of transportation networks. Their performance deteriorates over time due to environmental erosion, load-bearing effects, and material aging. Therefore, accurate prediction of the technical condition of bridge components is fundamental to ensuring structural safety and developing scientific maintenance plans.

[0003] However, existing methods for predicting the technical condition of bridge components generally suffer from the following problems in practical engineering applications: First, the problem of data sparsity and non-uniform intervals. Bridge periodic inspections are typically conducted every one to three years, and the inspection time span is highly variable due to factors such as funding and weather. Traditional time series models or discrete Markov models usually assume a fixed step size, making it difficult to effectively handle such non-uniformly spaced sparse data, thus limiting prediction accuracy. Second, insufficient quantification of the impact of maintenance interventions. Maintenance and repair activities have a significant "corrective" effect on bridge performance degradation, and the effects of maintenance measures have time-sensitive and spatial transmission characteristics. However, existing models often treat this as random disturbances or simple state resets, lacking explicit modeling of the decay law of maintenance effectiveness over time. This results in models being unable to accurately assess the marginal contribution of different maintenance strategies to the long-term service condition of bridges. Third, the lack of differentiation in component degradation mechanisms. Significant differences exist in material decay, load response, and environmental sensitivity among the bridge superstructure, substructure, and deck system. Existing forecasting methods often use a uniform model and fail to provide detailed models for the dominant influencing factors (such as traffic volume and corrosive environment) of different components.

[0004] Therefore, there is an urgent need for a bridge condition prediction scheme that can integrate quantitative maintenance interventions, adapt to non-equidistant data, and have dynamic adaptive capabilities, so as to achieve a leap from "passive management and maintenance" to "proactive and accurate prediction". Summary of the Invention

[0005] The technical problem to be solved by this invention is: how to solve the problems of heterogeneity of data sources, complexity of maintenance intervention, and correlation of system components in the existing process of predicting the technical condition of bridge components. It provides a prediction method for the technical condition of bridge infrastructure. By constructing a closed-loop architecture of "maintenance-condition" prediction and "dynamic-probability" transfer, the core logic is to use machine learning to capture complex degradation influencing factors and map them into the strength matrix of dynamic Markov chains, thereby realizing the state probability prediction under any time span, providing stable support for bridge structural safety monitoring and the formulation of scientific maintenance plans.

[0006] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0007] S1: Data collection and maintenance effectiveness quantification;

[0008] By acquiring historical bridge inspection data and maintenance event data through sensors or databases, a component-level maintenance effectiveness decay model is established to calculate the comprehensive residual maintenance effectiveness of each component at the predicted time.

[0009] S2: Multi-source feature fusion and degradation prediction;

[0010] Construction including bridge components The current detection score, bridge structure type, bridge component type, environmental index, bridge age, load pressure, comprehensive residual maintenance effectiveness, and prediction time step. The multidimensional feature vectors, along with a pre-defined machine learning regression model, are used to predict the component's performance. Internal state degradation prediction value ;

[0011] S3: Construct the dynamic Markov transition matrix;

[0012] Based on component state degradation prediction values In conjunction with the comprehensive residual maintenance effectiveness, the basic state intensity matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. The corresponding prediction time step is calculated using the matrix exponential power series expansion method. Non-equal interval state transition probability matrix ;

[0013] S4: Probabilistic evolution prediction;

[0014] The current state probability vector of the bridge components is compared with the non-equal interval state transition probability matrix. By multiplying these values, we obtain the probability distribution and expected value of the component's state at each state level in the future, thus enabling the prediction of its technical condition.

[0015] Furthermore, in step S1, the calculation process for the overall residual maintenance effectiveness is as follows:

[0016] S11: For each maintenance measure The residual maintenance effectiveness function is defined as follows:

[0017] ;

[0018] in, This represents the initial effectiveness value of the maintenance measures. The attenuation coefficient is related to the environment. For the implementation time of maintenance measures, Indicates the predicted time;

[0019] S12: Considering the mutual influence between components, components Comprehensive residual maintenance efficacy The weighted sum of the effectiveness of all maintenance measures applied to this component:

[0020] ;

[0021] in, For maintenance measures Acting on components The effectiveness, For cross-component coordination coefficient, Indicates adjacent components For components Cross-component coordination coefficient, For maintenance measures Acting on adjacent components Its maintenance efficacy.

[0022] Furthermore, in step S3, the basic state strength matrix is... The nonlinear correction method is as follows:

[0023] Through mapping function Adjusting the basic state strength matrix off-diagonal elements The dynamic intensity matrix elements are obtained. :

[0024] ;

[0025] in, These are the row and column numbers of the matrix. For degradation shift sensitivity factor, Sensitivity factors for maintenance intervention;

[0026] For the basic state strength matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. :

[0027] ;

[0028] in, For the correction matrix, by Fill non-diagonal elements, diagonal elements are... Sure.

[0029] Furthermore, The following was obtained by maximum likelihood estimation using historical bridge inspection data:

[0030] ;

[0031] in, The total number of historical samples. and The first Second and third The results of the test For the first Predicted state degradation values ​​within each detection cycle. This refers to the comprehensive residual maintenance effectiveness during the corresponding time period.

[0032] Furthermore, in step S3, the corresponding prediction time step is calculated using matrix exponential series expansion. Non-equal interval state transition probability matrix as follows:

[0033] ;

[0034] in, It is an identity matrix.

[0035] Furthermore, in step S4, the specific processing procedure is as follows:

[0036] S41: First, determine the prediction start time. The state of the bridge components is used to construct the initial state probability vector. ;

[0037] S42: Next, the initial state probability vector and the non-equal interval state transition probability matrix Perform matrix multiplication to obtain future time steps. Probability distribution vector of a component at each state level Each element in the vector This indicates that the component is in After a certain time, the status level will be... The probability of , where, ;

[0038] S43: Finally, the probability distribution vector Standard score vector corresponding to each state level Multiply to calculate the expected score at the predicted time. That is, to obtain the predicted expected value.

[0039] Furthermore, the prediction method also includes the following steps:

[0040] S5: Bayesian online update and closed-loop correction;

[0041] New measured detection scores of bridge components are obtained, the posterior probability distribution of component status is updated through Bayesian deduction, and the parameters of the machine learning regression model are adjusted according to the residual between the measured value and the predicted expected value to achieve adaptive evolution of the model.

[0042] Furthermore, in step S5, the specific processing procedure is as follows:

[0043] S51: When obtaining new measured test scores for bridge components At that time, constructing with Likelihood function centered at the detection accuracy as the standard deviation:

[0044] ;

[0045] in, Indicates time The new actual test score, Indicates the condition level of bridge components. Indicates the state level The corresponding standard score or the center point of the score interval, The standard deviation of the detection bias of the detection method;

[0046] S52: Perform Bayesian derivation based on the predicted probability distribution to update the posterior probability distribution of the component state. :

[0047] ;

[0048] in, , Indicates the condition level of bridge components. This represents the total number of state levels. For posterior probability, The likelihood function value, This is the prior probability;

[0049] S53: Based on the residual between the new measured detection score and the predicted expected value, the prediction bias term of the machine learning regression model is adjusted in reverse to achieve adaptive evolution of the model.

[0050] Furthermore, in step S53, the specific processing procedure is as follows:

[0051] S531: Based on the residual between the newly measured detection score and the predicted expected value, adjust the machine learning regression model using the idea of ​​error backpropagation, and define the expected score bias. :

[0052] ;

[0053] S532: Will As a feedback signal, the state degradation prediction value output by the machine learning regression model is corrected. The correction formula is as follows:

[0054] ;

[0055] in, The corrected state degradation prediction value will be used in the next prediction cycle; This is the output function of the original machine learning regression model. The input is a multidimensional feature vector. The learning rate is used to control the contribution of the residuals to model correction.

[0056] The present invention also provides a prediction system for the technical condition of bridge infrastructure, for implementing the above-mentioned prediction method, comprising:

[0057] The data acquisition and quantification module is used to acquire historical bridge inspection data and maintenance event data through sensors or databases, establish a component-level maintenance effectiveness decay model, and calculate the comprehensive residual maintenance effectiveness of each component at the predicted time.

[0058] The fusion and prediction module is used to build a system containing bridge components. The current detection score, bridge structure type, bridge component type, environmental index, bridge age, load pressure, comprehensive residual maintenance effectiveness, and prediction time step. The multidimensional feature vectors, along with a pre-defined machine learning regression model, are used to predict the component's performance. Internal state degradation prediction value ;

[0059] Matrix construction module for predicting component-based state degradation values. In conjunction with the comprehensive residual maintenance effectiveness, the basic state intensity matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. The corresponding prediction time step is calculated using the matrix exponential power series expansion method. Non-equal interval state transition probability matrix ;

[0060] The evolution prediction module is used to combine the current state probability vector of bridge components with the non-equal interval state transition probability matrix. By multiplying these values, we obtain the probability distribution and expected value of the component's state at each state level in the future, thus enabling the prediction of its technical condition.

[0061] Compared with existing technologies, this invention has the following advantages: This method for predicting the technical condition of bridge infrastructure first collects data. By coupling maintenance behavior with degradation mechanisms, it quantifies historical maintenance behavior and establishes a residual maintenance effectiveness function, solving the problem that traditional models cannot quantify the time-dependent decay of maintenance measures and the synergistic gains between components. Subsequently, a machine learning regression model is used to predict the degradation amount under given maintenance effectiveness and environmental characteristics, and a dynamic mapping mechanism is established. The Markov intensity matrix is ​​corrected in real time using degradation amount and maintenance effectiveness. Matrix exponential operations are used to solve the problem of calculating the state transition probability of non-uniformly spaced detection data, adapting to non-uniformly spaced data in complex maintenance environments. Finally, a Bayesian online update framework is introduced to perform closed-loop correction of the model using new measured data, achieving adaptive model evolution. This invention achieves deep coupling between maintenance intervention and degradation mechanisms, significantly improving the prediction accuracy of bridges in complex maintenance environments and providing quantitative support for life-cycle maintenance decision-making. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the method for predicting the technical condition of bridge infrastructure in an embodiment of the present invention.

[0063] Figure 2 This is a graph showing the decay of the effectiveness of maintenance measures over time in the embodiments of the present invention;

[0064] Figure 3 This is a schematic diagram illustrating the calculation process of the state transition probability matrix in an embodiment of the present invention;

[0065] Figure 4 These are different prediction spans in the embodiments of the present invention. The following is a heatmap of the state transition of bridge components, in which (a) the predicted span =1 year, (b) predicted span =5 years;

[0066] Figure 5 This is a comparison chart showing the changes in bridge technical condition scores under different maintenance scenarios in this invention over future service time. Detailed Implementation

[0067] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0068] Example 1

[0069] like Figure 1 As shown, this embodiment provides a technical solution: a method for predicting the technical condition of bridge infrastructure, comprising the following steps:

[0070] Step S1: Data collection and quantification of maintenance effectiveness;

[0071] By acquiring historical bridge inspection data and maintenance event data through sensors or databases, a component-level maintenance effectiveness attenuation model is established to calculate the comprehensive residual maintenance effectiveness of each component at the predicted time. .

[0072] In step S1, the residual maintenance effectiveness is calculated as follows: for each maintenance measure... The residual maintenance effectiveness function is defined as follows:

[0073] (1)

[0074] in, This represents the initial effectiveness value of the maintenance measures. The environmentally relevant attenuation coefficient was obtained through statistical analysis of bridge maintenance intervals under different environmental conditions. For the implementation time of maintenance measures, Indicates the predicted time. Considering the interactions between components, the components... Comprehensive residual maintenance efficacy The weighted sum of the effectiveness of all maintenance measures applied to this component:

[0075] (2)

[0076] in, For maintenance measures Acting on components The effectiveness, The cross-component coordination coefficient is influenced by the hierarchical protection relationship (such as the bridge deck system to the superstructure) and the mechanical transmission relationship (such as the bearings to the substructure) between bridge components. It is derived based on the spatial correlation between bridge components. Indicates adjacent components For components Cross-component coordination coefficient, For maintenance measures Acting on adjacent components Its maintenance efficacy.

[0077] Step S2: Multi-source feature fusion and degradation prediction;

[0078] Construction including bridge components Current test scores, environmental indices, load pressure, and overall residual maintenance effectiveness. and prediction time step The multidimensional feature vectors of the data are used to predict the component's performance using the multidimensional feature vectors and a pre-defined machine learning regression model. Internal state degradation prediction value .

[0079] In step S2, the system retrieves specific data for different bridge components i from the bridge maintenance system database. For qualitative indicators (such as bridge structure type, bridge component type, environmental index, etc.), they are transformed into continuous variables between [0,1] using One-Hot coding or a grading system. For quantitative physical indicators (such as current inspection score, bridge age, load pressure, comprehensive residual maintenance effectiveness, etc.), they are standardized using Min-Max scaling or Z-score to ensure that different data are on the same dimension. Finally, the parameters processed above are arranged in a predefined time order to form a multidimensional feature vector.

[0080] In step S2, the machine learning regression model employs ensemble learning algorithms such as XGBoost / LightGBM, leveraging their feature extraction capabilities for high-dimensional sparse data to predict the complex mapping relationship between environmental erosion, traffic load, and bridge technical condition degradation. The specific construction and training process of the machine learning regression model is as follows:

[0081] First, a multi-dimensional input feature vector is constructed based on existing data, including categorical features (such as bridge structure type, bridge component type, environmental index, etc.) and numerical features (such as current detection score, bridge age, load pressure, comprehensive residual maintenance effectiveness, time step, etc.). The time interval between two consecutive detections is used as a training sample, and the time interval between two historical detections is also included. As one of the input features, it ensures that the model can learn the degradation rate at different time steps. The label value corresponding to the model's input feature vector is the change in the bridge detection score within that time interval.

[0082] Next, a regression model is constructed and trained. The regression model employs an ensemble architecture based on decision trees (XGBoost / LightGBM), formed iteratively from multiple decision trees. The prediction residuals of the previous model serve as the fitting target for the next model, with the objective function being to minimize the mean squared error between the predicted value and the actual degradation. The residuals are iteratively fitted, and each decision tree forms a hierarchical structure through feature splitting nodes, with leaf nodes outputting the corresponding regression values. The main parameters for model training include: maximum tree depth, learning rate, number of weak learners, and minimum number of sample splits. These parameters are optimized using cross-validation.

[0083] Finally, after the model training is complete, the constructed real-time feature vectors are input into the trained model, which outputs the predicted state degradation values ​​within the corresponding prediction time step. The prediction result is then used as input for subsequent steps.

[0084] Step S3: Construct the dynamic Markov transition matrix;

[0085] Based on component state degradation prediction values With comprehensive residual maintenance efficacy For the basic state strength matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. The corresponding prediction time step is calculated using matrix exponential series expansion. Non-equal interval state transition probability matrix .

[0086] In step S3, the basic state strength matrix The correction logic is as follows: through the mapping function Adjusting the basic state strength matrix off-diagonal elements The elements of the dynamic intensity matrix are obtained as follows:

[0087] (3)

[0088] in, These are the row and column numbers of the matrix. For degradation shift sensitivity factor, As a sensitivity factor for maintenance intervention, and . The following can be obtained by maximum likelihood estimation using historical detection data:

[0089] (4)

[0090] in, The total number of historical samples. and The first Second and third The results of the test For the first Predicted state degradation values ​​within each detection cycle. This refers to the comprehensive residual maintenance effectiveness during the corresponding time period.

[0091] For the basic state strength matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. :

[0092] (5)

[0093] in, For the correction matrix, by Fill non-diagonal elements, diagonal elements are... Sure.

[0094] Finally, the corresponding prediction time step is calculated using matrix exponential series expansion. Non-equal interval state transition probability matrix :

[0095] (6)

[0096] in, It is the identity matrix. For any non-equal interval time step, the non-equal interval transition probability matrix is... Numerical matrix exponents are used in actual calculations.

[0097] Step S4: Probabilistic evolution prediction;

[0098] Combine the current state probability vector with the non-equal interval transition probability matrix Multiplying these yields the probability distribution and expected value of the component's state at each future time step. .

[0099] In step S4, firstly, the prediction start time is determined. The state of the component is used to construct the initial state probability vector. For example, the start time The component's actual test score was 88 points, classifying it as Category II. Secondly, the initial state probability vector and the non-equal interval state transition probability matrix Perform matrix multiplication to obtain future time steps. Probability distribution vector of a component at each state level Each element in the vector ( ) represents that the component is in After a certain time, the status level will be... The probability. Finally, the probability distribution vector. Standard score vector corresponding to each state level Multiply the values ​​(e.g., [95,85,75,65,50]) to calculate the expected score at the predicted time. .

[0100] Following step S4, Bayesian online updates and closed-loop corrections are also included, specifically comprising the following steps:

[0101] Step S51: When obtaining the detection score of a new bridge component At that time, constructing with Likelihood function centered at the detection accuracy as the standard deviation:

[0102] (7)

[0103] in, Indicates time The actual test score, Indicates the condition level of bridge components, for example: (corresponding to categories one through five respectively) Indicates the state level The corresponding standard score or the center point of the score range, This represents the standard deviation of the detection bias of the detection method.

[0104] Step S52: Perform Bayesian derivation based on the predicted probability distribution to update the posterior probability distribution of the component state. :

[0105] (8)

[0106] in, Indicates time The actual detection score, , Indicates the condition level of bridge components. This represents the total number of state levels. The posterior probability refers to the probability obtained after obtaining the detection score. Under these conditions, the component is in state level The corrected probability. The likelihood function value, derived from formula (7), refers to the value assumed when the component is in state level. At that time, the detection score was observed. The probability density. This is the prior probability, i.e., the probability obtained in step S4 through the Markov transition matrix (non-equal interval transition probability matrix). The probability distribution of each state level is predicted without correction based on measured values. The posterior probability distribution obtained by this formula serves as the optimal estimate of the bridge's state at the current moment, and is used for risk assessment and decision support.

[0107] Step S53: Based on the residual between the measured value and the predicted expected value, adjust the prediction bias term of the machine learning regression model in reverse to achieve adaptive evolution of the model.

[0108] Specifically: Based on the residuals between the measured values ​​and the predicted expected values, the machine learning regression model is adjusted using the concept of "error backpropagation," and the expected score bias is defined. :

[0109] (9)

[0110] in, The posterior probability is derived from formula (8). As a feedback signal, the output of the machine learning model is used in subsequent predictions. A correction factor will be added. This enables the model to evolve adaptively.

[0111] (10)

[0112] in, This is the corrected state degradation prediction value, used in the next prediction cycle. This is the output function of the original machine learning regression model. The input is a multidimensional feature vector (including current score, maintenance effectiveness, environmental index, etc.). The learning rate is used to control the contribution of the residuals to model correction. It is usually set to 0.1 to 0.3 to prevent drastic fluctuations in the model caused by a single detection error. The expected score deviation is derived from formula (9).

[0113] This embodiment also provides a prediction system for the technical condition of bridge infrastructure, used to implement the above-mentioned prediction method, including:

[0114] The data acquisition and quantification module is used to acquire historical bridge inspection data and maintenance event data through sensors or databases, establish a component-level maintenance effectiveness decay model, and calculate the comprehensive residual maintenance effectiveness of each component at the predicted time.

[0115] The fusion and prediction module is used to build a system containing bridge components. The current detection score, bridge structure type, bridge component type, environmental index, bridge age, load pressure, comprehensive residual maintenance effectiveness, and prediction time step. The multidimensional feature vectors, along with a pre-defined machine learning regression model, are used to predict the component's performance. Internal state degradation prediction value ;

[0116] Matrix construction module for predicting component-based state degradation values. In conjunction with the comprehensive residual maintenance effectiveness, the basic state intensity matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. The corresponding prediction time step is calculated using the matrix exponential power series expansion method. Non-equal interval state transition probability matrix ;

[0117] The evolution prediction module is used to combine the current state probability vector of bridge components with the non-equal interval state transition probability matrix. By multiplying these values, we obtain the probability distribution and expected value of the component's state at each state level in the future, thus enabling the prediction of its technical condition.

[0118] Example 2

[0119] This embodiment provides a method for predicting the technical condition of bridge infrastructure. The method is described in conjunction with the prediction system in Embodiment 1. The core implementation process of the prediction method is as follows:

[0120] 1. Data Preprocessing and Maintenance Effectiveness Quantification Stage. First, the system extracts historical inspection sequences and maintenance records of the target bridge from the bridge management database. For each maintenance event, the system quantifies it using a preset attenuation function. For example... Figure 2 As shown, the system generates a logical curve of the decay of the effectiveness of maintenance measures over time according to formula (1). The initial effectiveness reaches its peak after maintenance is completed, and then decays over time, which intuitively reflects the timeliness of maintenance behavior.

[0121] Implementation Case: Taking a reinforced concrete bridge as an example, this bridge is 15 years old. The current inspection score obtained by the highway management office during regular inspections is 85 points (the overall bridge score is a weighted average of the scores of the superstructure, substructure, and bridge deck system; to simplify the process, this embodiment directly calculates the score based on the entire bridge). The main beam underwent maintenance two years ago. The system allocates its initial effectiveness. Environmental attenuation coefficient According to the formula Calculate the current comprehensive residual maintenance effectiveness. This value, along with the current score (85 points) and other parameters, forms the feature vector input to the prediction engine.

[0122] 2. Calculation stage of conditional degradation rate and dynamic intensity matrix. For example... Figure 3 As shown, firstly, the machine learning module predicts the future based on feature vectors. Predicted state degradation values ​​within 3 years (e.g.) This means that under the "current maintenance conditions," the estimated score after 3 years is [score]. point.

[0123] Secondly, the dynamic intensity matrix correction unit utilizes and For the basic state strength matrix Make corrections. Set the sensitivity factor. and Calculate the corrected instantaneous transfer rate In this embodiment, the calculated corrected instantaneous transfer rate increased from 0.15 to 0.22, reflecting the increase in degradation rate under the current environment.

[0124] 3. Non-equidistant transition probability generation stage. Utilizing the matrix exponent formula. The state transition probability matrix after 3 years is calculated. This step does not rely on a fixed step size, perfectly adapting to the randomness of bridge detection. The calculated state probability distribution of the bridge after 3 years is as follows: remain a Class II bridge: 45%; degenerate to a Class III bridge: 52% (high probability interval); degenerate to a Class IV bridge: 3%; predicted expected score: 79.2 points.

[0125] In addition, such as Figure 4 As shown in (a) and (b), by changing The system can generate state transition heatmaps for different prediction spans based on the value of . As can be seen from the graph, with... Increase the non-equal interval state transition probability matrix The value on the main diagonal (indicating that the bridge condition remains unchanged) decreases, while the value in the area to the right of the main diagonal (indicating the probability that the bridge condition will deteriorate) increases accordingly.

[0126] 4. Bayesian Online Update and Closed-Loop Optimization Stage. If a new on-site measured score (e.g., 82 points) is obtained during the prediction period, the Bayesian update module uses this as observational evidence to calculate the posterior probability distribution to correct the current prediction state. Simultaneously, the system calculates the residual between the measured value and the expected predicted value, adjusting the bias term of the machine learning model in reverse to enable continuous learning of the specific bridge degradation characteristics.

[0127] 5. Scenario Simulation and Decision Support Stage

[0128] Users input different future maintenance plans (such as "Plan A: Implement specific maintenance measures" and "Plan B: Natural degradation") through a visual interface. The system calculates the residual effectiveness evolution curves for each plan and outputs the results as follows: Figure 5 The diagram shows the risk envelope for future years. Based on this, the system can recommend corresponding maintenance plans and output a quantitative comparison report.

[0129] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the technical condition of bridge infrastructure, characterized in that, Includes the following steps: S1: Data collection and maintenance effectiveness quantification; By acquiring historical bridge inspection data and maintenance event data through sensors or databases, a component-level maintenance effectiveness decay model is established to calculate the comprehensive residual maintenance effectiveness of each component at the predicted time. S2: Multi-source feature fusion and degradation prediction; Construction including bridge components The current detection score, bridge structure type, bridge component type, environmental index, bridge age, load pressure, comprehensive residual maintenance effectiveness, and prediction time step. The multidimensional feature vectors, along with a pre-defined machine learning regression model, are used to predict the component's performance. Internal state degradation prediction value ; S3: Construct the dynamic Markov transition matrix; Based on component state degradation prediction values In conjunction with the comprehensive residual maintenance effectiveness, the basic state intensity matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. ; The corresponding prediction time step is calculated using the matrix exponential power series expansion method. Non-equal interval state transition probability matrix ; S4: Probabilistic evolution prediction; The current state probability vector of the bridge components is compared with the non-equal interval state transition probability matrix. By multiplying these values, we obtain the probability distribution and expected value of the component's state at each state level in the future, thus enabling the prediction of its technical condition.

2. The method for predicting the technical condition of bridge infrastructure according to claim 1, characterized in that, In step S1, the calculation process for the overall residual maintenance effectiveness is as follows: S11: For each maintenance measure The residual maintenance effectiveness function is defined as follows: ； in, This represents the initial effectiveness value of the maintenance measures. The attenuation coefficient is related to the environment. For the implementation time of maintenance measures, Indicates the predicted time; S12: Considering the mutual influence between components, components Comprehensive residual maintenance efficacy The weighted sum of the effectiveness of all maintenance measures applied to this component: ； in, For maintenance measures Acting on components The effectiveness, For cross-component coordination coefficient, Indicates adjacent components For components Cross-component coordination coefficient, For maintenance measures Acting on adjacent components Its maintenance efficacy.

3. The method for predicting the technical condition of bridge infrastructure according to claim 2, characterized in that, In step S3, the basic state strength matrix is... The nonlinear correction method is as follows: Through mapping function Adjusting the basic state strength matrix off-diagonal elements The dynamic intensity matrix elements are obtained. : ； in, These are the row and column numbers of the matrix. For degradation shift sensitivity factor, Sensitivity factors for maintenance intervention; For the basic state strength matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. : ； in, For the correction matrix, by Fill non-diagonal elements, diagonal elements are... Sure.

4. The method for predicting the technical condition of bridge infrastructure according to claim 3, characterized in that, The following was obtained by maximum likelihood estimation using historical bridge inspection data: ； in, The total number of historical samples. and The first Second and third The results of the test For the first Predicted state degradation values ​​within each detection cycle. This refers to the comprehensive residual maintenance effectiveness during the corresponding time period.

5. The method for predicting the technical condition of bridge infrastructure according to claim 4, characterized in that, In step S3, the corresponding prediction time step is calculated using matrix exponential series expansion. Non-equal interval state transition probability matrix as follows: ； in, It is an identity matrix.

6. The method for predicting the technical condition of bridge infrastructure according to claim 1, characterized in that, In step S4, the specific processing procedure is as follows: S41: First, determine the prediction start time. The state of the bridge components is used to construct the initial state probability vector. ; S42: Next, the initial state probability vector and the non-equal interval state transition probability matrix Perform matrix multiplication to obtain future time steps. Probability distribution vector of a component at each state level Each element in the vector This indicates that the component is in After a certain time, the status level will be... The probability of , where, ; S43: Finally, the probability distribution vector Standard score vector corresponding to each state level Multiply to calculate the expected score at the predicted time. That is, to obtain the predicted expected value.

7. The method for predicting the technical condition of bridge infrastructure according to claim 5, characterized in that, The prediction method further includes the following steps: S5: Bayesian online update and closed-loop correction; New measured detection scores of bridge components are obtained, the posterior probability distribution of component status is updated through Bayesian deduction, and the parameters of the machine learning regression model are adjusted according to the residual between the measured value and the predicted expected value to achieve adaptive evolution of the model.

8. The method for predicting the technical condition of bridge infrastructure according to claim 7, characterized in that, In step S5, the specific processing procedure is as follows: S51: When obtaining new measured test scores for bridge components At that time, constructing with Likelihood function centered at the detection accuracy as the standard deviation: ； in, Indicates time The new actual test score, Indicates the condition level of bridge components. Indicates the state level The corresponding standard score or the center point of the score interval, The standard deviation of the detection bias of the detection method; S52: Perform Bayesian derivation based on the predicted probability distribution to update the posterior probability distribution of the component state. : ； in, , Indicates the condition level of bridge components. This represents the total number of state levels. For posterior probability, The likelihood function value, This is the prior probability; S53: Based on the residual between the new measured detection score and the predicted expected value, the prediction bias term of the machine learning regression model is adjusted in reverse to achieve adaptive evolution of the model.

9. The method for predicting the technical condition of bridge infrastructure according to claim 8, characterized in that, In step S53, the specific processing procedure is as follows: S531: Based on the residual between the newly measured detection score and the predicted expected value, adjust the machine learning regression model using the idea of ​​error backpropagation, and define the expected score bias. : ； S532: Will As a feedback signal, the state degradation prediction value output by the machine learning regression model is corrected. The correction formula is as follows: ； in, The corrected state degradation prediction value will be used in the next prediction cycle; This is the output function of the original machine learning regression model. The input is a multidimensional feature vector. The learning rate is used to control the contribution of the residuals to model correction.

10. A predictive system for the technical condition of bridge infrastructure, characterized in that, The prediction method as described in any one of claims 1 to 9 includes: The data acquisition and quantification module is used to acquire historical bridge inspection data and maintenance event data through sensors or databases, establish a component-level maintenance effectiveness decay model, and calculate the comprehensive residual maintenance effectiveness of each component at the predicted time. The fusion and prediction module is used to build a system containing bridge components. The current detection score, bridge structure type, bridge component type, environmental index, bridge age, load pressure, comprehensive residual maintenance effectiveness, and prediction time step. The multidimensional feature vectors, along with a pre-defined machine learning regression model, are used to predict the component's performance. Internal state degradation prediction value ; Matrix construction module for predicting component-based state degradation values. In conjunction with the comprehensive residual maintenance effectiveness, the basic state intensity matrix Perform nonlinear corrections to construct the instantaneous dynamic intensity matrix. The corresponding prediction time step is calculated using the matrix exponential power series expansion method. Non-equal interval state transition probability matrix ; The evolution prediction module is used to combine the current state probability vector of bridge components with the non-equal interval state transition probability matrix. By multiplying these values, we obtain the probability distribution and expected value of the component's state at each state level in the future, thus enabling the prediction of its technical condition.

Images