Method for correcting inherent frequency predicted value of bridge under operation condition

Abstract

The invention discloses a method for correcting a predicted value of inherent frequency of a bridge under an operation condition, and belongs to the technical field of bridge monitoring, and the method comprises the following steps: S1, collecting bridge data; s2, data preprocessing; s3, environment and load interference decoupling is carried out; s4, constructing a model; s5, correcting the AI driving model; s6, verifying and iterating the mechanism; according to the method, breakthrough is made in three aspects of environmental interference stripping, parameter dynamic optimization and intelligent algorithm iteration, so that the prediction precision is improved, the engineering efficiency is optimized, the safety is ensured, the environmental interference stripping technology can effectively eliminate external environmental noise, and the accuracy of prediction data is ensured; a parameter dynamic optimization technology can dynamically adjust model parameters according to real-time data, so that a prediction result is closer to reality; the intelligent algorithm iteration continuously optimizes the model through machine learning and improves the intelligent level of prediction, and the method of updating posteriori distribution through Bayesian is adopted, so that the prediction model can be continuously corrected according to newly acquired data, and the prediction accuracy is ensured.

Description

Technical Field

[0001] This invention specifically relates to a method for correcting the predicted value of the natural frequency of a bridge under operational conditions, belonging to the field of bridge monitoring technology. Background Technology

[0002] In the field of bridge engineering, natural frequency is one of the key parameters for measuring the overall stability and dynamic characteristics of bridge structures. Currently, the prediction methods for the natural frequency of bridges mainly rely on mathematical models and physical experiments. However, due to many complex factors under actual operating conditions, such as changes in ambient temperature, humidity fluctuations, and dynamic loads brought by vehicle traffic, a single prediction method is often difficult to accurately reflect the true natural frequency of the bridge under operating conditions. Therefore, there is an urgent need for a method that can effectively correct the prediction results and improve the prediction accuracy in order to better understand and maintain the safety status of bridges. Existing prediction methods have revealed some problems in practical applications. First, the uncertainty of structural parameters is one of the main influencing factors. For example, parameters such as the hardness and elastic modulus of bridge concrete will fluctuate due to environmental changes, directly leading to an increase in the error of model prediction. Second, the actual operating environment of bridges differs greatly from theoretical assumptions. For example, changes in ambient temperature will affect the mechanical properties of materials, and mechanical models cannot accurately reproduce these changes. In addition, the influence of dynamic loads cannot be ignored. Different types of vehicles will generate different transient impacts when driving on bridges, which existing models cannot fully consider. To address the aforementioned technical issues, a method for correcting the predicted natural frequency of bridges under operational conditions is proposed. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for correcting the predicted values ​​of the natural frequency of bridges under operational conditions. This method achieves a triple breakthrough through environmental interference removal, dynamic parameter optimization, and intelligent algorithm iteration, combining the core advantages of prediction accuracy, engineering efficiency, and safety assurance. It employs Bayesian update of posterior distribution correction and sensitivity-oriented optimization to prioritize the correction of highly sensitive parameters, thereby reducing the error between the model response and the measured values. A method for correcting the predicted natural frequency of a bridge under operational conditions includes the following steps: S1. Bridge data acquisition: Deploy a multi-source sensor network for data acquisition; S2. Data preprocessing: perform data cleaning and anomaly handling, correct abnormal data and dynamically avoid interference. S3. Decouple environmental and load disturbances, construct a temperature-frequency transfer function, and deploy a distributed dynamic weighing system and machine vision fusion technology; S4. Build a model by modeling multimodal data and establishing a dynamic correction engine based on the core AI model; S5, AI-driven model correction, performs deep learning feature extraction, parameter uncertainty quantification, and response surface replacement model optimization; S6. Verification and iteration mechanism: perform cross-validation and establish a time-varying database of modal parameters.

[0004] In a further preferred embodiment, in step S1, an array of acceleration sensors is arranged at key points such as the mid-span and supports to collect vibration signals in the 0.5-20Hz main frequency band, identify modal frequencies and mode shape characteristics, and thus perform vibration monitoring; temperature and humidity sensors are integrated to record the thermal expansion and contraction effects of materials, thereby performing synchronous environmental monitoring; dynamic load capture is performed, and a distributed dynamic weighing system and machine vision are integrated to identify vehicle type, speed and axle load distribution in real time; and a drone and AI crack recognition system are used to achieve damage location accuracy of ±0.5m.

[0005] In a further preferred embodiment, in step S2, high-frequency noise is separated by db4 wavelet packet decomposition while retaining structural characteristic frequencies, thereby completing noise filtering; outlier removal is performed by using a filter with a sliding window to remove outliers beyond three times the standard deviation, and data from periods without vehicles is automatically filtered based on load spectrum analysis to avoid transient impact effects.

[0006] More preferably, in step S2, for monitoring data exceeding the threshold, the arithmetic mean of the previous and next 5 data points is used to replace it, and monitoring data during periods without vehicle traffic is preferred to avoid frequency shift caused by the added mass of vehicles. Single-mode response is extracted by combining wavelet analysis to improve the availability of data in noisy environments.

[0007] In a further preferred embodiment, in step S3, a negative linear regression model of temperature and modal frequency is established using long-term monitoring data. Nonlinear principal component analysis is used to isolate the coupling effects of temperature and vehicle load. An artificial neural network is used to correct the mapping from environmental variables to frequencies. A distributed dynamic weighing system and machine vision fusion technology are deployed. A random traffic flow probability model is constructed to replace the traditional static load assumption. A deep convolutional neural network is applied to identify the characteristics of moving loads. The transient impact effect is reconstructed by combining the vehicle-bridge coupled vibration equation.

[0008] More preferably, in S4, the multimodal data modeling includes an environmental disturbance decoupling module, dynamic load feature extraction, and structural parameter uncertainty quantification; the dynamic correction engine adopts a hybrid deep learning architecture to optimize the surrogate model and is equipped with a correction stream transformer.

[0009] In a further preferred embodiment, in S5, a CNN-LSTM hybrid architecture is used for deep learning feature extraction to generate synthetic data-enhanced samples, thereby improving the robustness of the model under small sample conditions. The posterior distribution is updated by combining MCMC sampling with measured data, the parameter sensitivity is calculated, highly sensitive parameters are corrected first, a Kriging surrogate model is established, and the support vector machine kernel function is optimized using a genetic algorithm to achieve efficient parameter inversion.

[0010] In a further preferred embodiment, in step S5, the finite element model parameters are iteratively corrected in descending order of parameter sensitivity until the theoretical frequency and measured value converge, establishing a nonlinear mapping relationship between structural parameters and dynamic response. For example, a genetic algorithm is used to optimize the support vector machine model, significantly improving prediction efficiency and accuracy. The improved grey prediction model processes fluctuation data through data translation and geometric mean, reducing prediction errors. A bidirectional long short-term memory network combined with a local mean decomposition algorithm decomposes environmental and structural state data subsequences, enhancing the robustness of time series prediction.

[0011] In a further preferred embodiment, in step S6, a time-varying database of modal parameters is established, a blind source separation algorithm is used to track the impact of latent damage such as stiffness degradation on the frequency, and the surrogate model parameters are recalibrated quarterly to adapt to the material aging effect.

[0012] In a further preferred embodiment, in step S6, intelligent algorithms are used to fuse predictions, perform time-series decomposition, apply local mean decomposition to split the frequency sequence into environmental components and structural state components, perform bidirectional LSTM prediction, input environmental parameters and historical frequencies, output corrected frequency prediction values, perform static and dynamic test comparisons and modal confidence criteria, thereby performing multi-source cross-validation.

[0013] Beneficial effects: This invention achieves breakthroughs in three aspects: environmental interference removal, dynamic parameter optimization, and intelligent algorithm iteration. These breakthroughs aim to improve prediction accuracy, optimize engineering efficiency, and ensure safety. Environmental interference removal technology effectively eliminates external environmental noise, guaranteeing the accuracy of prediction data. Dynamic parameter optimization technology dynamically adjusts model parameters based on real-time data, making prediction results closer to reality. Intelligent algorithm iteration continuously optimizes the model through machine learning, enhancing the intelligence level of prediction. The use of Bayesian posterior distribution updates continuously corrects the prediction model based on newly acquired data, ensuring prediction accuracy. Sensitivity-oriented optimization technology prioritizes correcting parameters that have a significant impact on prediction results, further improving the model's prediction accuracy. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method for correcting the predicted value of the natural frequency of a bridge under operational conditions in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1 This embodiment uses a newly built high-speed railway bridge spanning a river as the monitoring object. The main bridge adopts a steel-concrete hybrid continuous beam structure, with the main span and side spans being steel box girders and the remaining segments being prestressed concrete box girders. A steel-concrete joint structure is set between the steel beams and the concrete beams. The bridge site is located in a typical alluvial plain and floodplain area, with a generally flat and open terrain, with slight local undulations. The river is approximately 2.3 km wide, with the main channel approximately 600 m wide, and the water flow is gentle, meeting the conditions for a Class IV navigation channel. The main beam arrangement is (102+4×178+102) m, with the beam height varying with the span. The beam height at mid-span is approximately 8.5 m, and the beam height at the mid-support is approximately 13 m. Equal-height sections are set to ensure structural stress and overall stability.

[0017] Please see Figure 1 As shown, in this engineering context, this embodiment of the invention provides a method for correcting the predicted value of the natural frequency of a bridge under operational conditions, comprising the following steps: S1. Bridge data acquisition: Deploy a multi-source sensor network for data acquisition, and arrange accelerometer arrays at key points such as mid-span and supports to collect vibration signals in the 0.5-20Hz main frequency band, identify modal frequencies and vibration mode characteristics, and thus perform vibration monitoring; integrate temperature and humidity sensors to record the thermal expansion and contraction effects of materials, and thus perform synchronous environmental monitoring; perform dynamic load capture, integrate a distributed dynamic weighing system and machine vision to identify vehicle type, speed and axle load distribution in real time, and use drones and AI crack recognition systems to achieve damage location accuracy of ±0.5m; S2. Data preprocessing: Data cleaning and anomaly handling are performed to correct abnormal data and dynamically avoid interference. High-frequency noise is separated by db4 wavelet packet decomposition while retaining structural characteristic frequencies, thereby completing noise filtering. Outlier removal is performed by using a filter with a sliding window to remove outliers beyond three standard deviations. Data during periods without vehicles is automatically filtered based on load spectrum analysis to avoid transient impact effects. S3. Decoupling of environmental and load disturbances, constructing a temperature-frequency transfer function, and deploying a distributed dynamic weighing system and machine vision fusion technology. A negative linear regression model of temperature and modal frequency is established through long-term monitoring data. Nonlinear principal component analysis is used to remove the coupling effect of temperature and vehicle load. The mapping correction from environmental variables to frequency is achieved through artificial neural networks. A random traffic flow probability model is constructed to replace the traditional static load assumption. A deep convolutional neural network is applied to identify the characteristics of moving loads. The transient impact effect is reconstructed by combining the vehicle-bridge coupled vibration equation. S4. Model building: Modeling is performed using multimodal data, and a dynamic correction engine is built based on the core AI model. Multimodal data modeling includes environmental interference decoupling module, dynamic load feature extraction, and structural parameter uncertainty quantification. The dynamic correction engine adopts a hybrid deep learning architecture to optimize the proxy model and is equipped with a correction stream transformer. S5, AI-driven model correction, performs deep learning feature extraction, parameter uncertainty quantification, and response surface replacement model optimization. It adopts a CNN-LSTM hybrid architecture for deep learning feature extraction, generates synthetic data augmentation samples, improves model robustness under small sample conditions, updates the posterior distribution by combining MCMC sampling with measured data, calculates parameter sensitivity, prioritizes the correction of highly sensitive parameters, establishes a Kriging surrogate model, and uses genetic algorithms to optimize the support vector machine kernel function to achieve efficient parameter inversion; S6. Verification and iteration mechanism: cross-validation is performed, and a time-varying database of modal parameters is established. The blind source separation algorithm is used to track the impact of latent damage such as stiffness degradation on frequency. The surrogate model parameters are recalibrated quarterly to adapt to material aging effects.

[0018] As a technical optimization scheme of the present invention, in S2, for monitoring data exceeding the threshold, the arithmetic mean of the previous and next 5 data points is used to replace it, and monitoring data during periods without vehicle traffic is preferred to avoid frequency shift caused by the added mass of vehicles. Combined with wavelet analysis to extract single-mode response, the availability of data in noisy environments is improved.

[0019] As a technical optimization scheme of the present invention, in S5, the parameters of the finite element model are iteratively corrected in order of high to low parameter sensitivity until the theoretical frequency and the measured value converge, and a nonlinear mapping relationship between structural parameters and dynamic response is established. For example, a genetic algorithm is used to optimize the support vector machine model, which significantly improves the prediction efficiency and accuracy. The improved grey prediction model processes fluctuation data through data translation transformation and geometric mean to reduce prediction error. A bidirectional long short-term memory network combined with a local mean decomposition algorithm decomposes the environmental and structural state data subsequences to enhance the robustness of time series prediction.

[0020] As a technical optimization scheme of the present invention, in S6, intelligent algorithm fusion prediction is adopted to perform time series decomposition. Local mean decomposition is applied to split the frequency sequence into environmental components and structural state components. Bidirectional LSTM prediction is performed. Environmental parameters and historical frequencies are input, and the corrected frequency prediction value is output. Static and dynamic test comparison and modal confidence criteria are performed to perform multi-source cross-validation.

[0021] Example 2 This invention also provides a method for correcting the predicted value of the natural frequency of a bridge under operational conditions, comprising the following steps: S1. Multimodal data acquisition and preprocessing; S2. Intelligent decoupling of environmental and load disturbances; S3, AI-driven model correction; S4. Dynamic verification and closed-loop optimization.

[0022] As a technical optimization scheme of the present invention, in S1, acceleration sensors are arranged at key locations such as the mid-span and supports of the bridge to form a spatial measuring point network, which collects vibration signals in the 0.5-20Hz main frequency band, integrates temperature and humidity sensors, records environmental time history data, eliminates the thermal expansion and contraction effect, and captures vehicle type, speed and axle load distribution in real time through the fusion of distributed dynamic weighing system and machine vision; Hampel filter is used to remove outliers beyond 3 times the standard deviation with a sliding window (width ≥ 10 sampling points), wavelet packet decomposition (db4 wavelet basis) is applied to separate high-frequency noise, retain structural characteristic frequencies, filter data during periods without vehicles, or remove transient impact effects based on load spectrum analysis.

[0023] As a technical optimization of the present invention, in S2, a temperature compensation model is established, and a frequency-temperature transfer function is constructed: ; Where k is the fitting sensitivity coefficient (typical value -0.03% / ℃), which is determined by linear regression of long-term monitoring data; Correction for measured frequency: ; Vehicle load dynamic modeling is performed, and the vehicle-bridge coupled dynamic equations are as follows: ; Input the random traffic load time history P(t) identified by WIM; Frequency variation coefficient control: ; Data resampling is triggered when β > 2% to ensure that load disturbances are controllable.

[0024] As a technical optimization scheme of the present invention, in S3, deep learning feature extraction is performed, and a CNN-LSTM hybrid architecture is adopted. The CNN-LSTM hybrid architecture includes a CNN layer and an LSTM layer. The CNN layer extracts the spatial features of vibration signals (such as spectral peaks and energy distribution); the LSTM layer learns the frequency temporal evolution law and captures the material aging trend; and synthetic data is generated to enhance the sample and improve the robustness of the model under small sample conditions. Quantify parameter uncertainty and define elastic modulus. The prior distribution with equal parameters is used; the posterior distribution is updated by combining MCMC sampling with measured data. Perform sensitivity-guided correction and calculate parameter sensitivity. Prioritize correcting highly sensitive parameters; Establishing a parameter-frequency mapping using a Kriging proxy model: ; Genetic algorithms are used to optimize the kernel function of support vector machines, achieving efficient parameter inversion.

[0025] As a technical optimization scheme of the present invention, in S4, multi-source cross-validation is performed, static and dynamic test comparison is carried out, the deviation between GPS measured deflection and model prediction needs to be [6][10<1%], modal confidence criteria are performed, MAC value is calculated, the mode shape matching degree is required to be >0.9, a self-iterative mechanism is adopted, the proxy model parameters are updated every quarter, and the material performance degradation is adaptive.

[0026] Working principle: Accelerometer arrays are deployed at key points such as mid-span and supports to collect vibration signals in the 0.5-20Hz main frequency band, identify modal frequencies and mode shape characteristics, and thus perform vibration monitoring; temperature and humidity sensors are integrated to record the thermal expansion and contraction effects of materials, thereby performing synchronous environmental monitoring; dynamic load capture is performed, and a distributed dynamic weighing system and machine vision are integrated to identify vehicle type, speed and axle load distribution in real time; drones and AI crack recognition systems are used to achieve damage location accuracy of ±0.For 5m, high-frequency noise was separated using db4 wavelet packet decomposition, preserving structural characteristic frequencies to achieve noise filtering. Outlier removal was performed by using a filter with a sliding window to remove outliers exceeding three standard deviations. Data from periods without vehicle traffic was automatically selected based on load spectrum analysis to avoid transient impact effects. For monitoring data exceeding the threshold, the arithmetic mean of the previous and next five data points was used for replacement. Monitoring data from periods without vehicle traffic was prioritized to avoid frequency shifts caused by vehicle-added mass. Single-mode responses were extracted using wavelet analysis to improve data usability in noisy environments. A negative linear regression model of temperature and modal frequency was established using long-term monitoring data, employing nonlinear principal component analysis. This study analyzes and isolates the coupled effects of temperature and vehicle load, using artificial neural networks to correct the mapping from environmental variables to frequencies. It deploys a distributed dynamic weighing system and machine vision fusion technology, constructing a stochastic traffic flow probability model to replace traditional static load assumptions. Deep convolutional neural networks are applied to identify moving load characteristics, and transient impact effects are reconstructed using the vehicle-bridge coupled vibration equation. Multimodal data modeling includes an environmental disturbance decoupling module, dynamic load feature extraction, and structural parameter uncertainty quantification. The dynamic correction engine employs a hybrid deep learning architecture for surrogate model optimization and incorporates a correction flow transformer. A CNN-LSTM hybrid architecture is used for deep learning feature extraction. Synthetic data is generated to enhance sample quality and improve model robustness under small sample conditions. The posterior distribution is updated using MCMC sampling combined with measured data. Parameter sensitivity is calculated, and highly sensitive parameters are prioritized for correction. A Kriging surrogate model is established, and a genetic algorithm is used to optimize the support vector machine kernel function for efficient parameter inversion. Finite element model parameters are iteratively corrected in descending order of parameter sensitivity until the theoretical frequency and measured values ​​converge. A nonlinear mapping relationship between structural parameters and dynamic response is established. For example, using a genetic algorithm to optimize the support vector machine model significantly improves prediction efficiency and accuracy. The grey prediction model is improved by processing fluctuating data through data translation and geometric averaging to reduce prediction errors. This study utilizes a bidirectional long short-term memory network (LSTM) combined with a local mean decomposition (LMD) algorithm to decompose environmental and structural state data subsequences, enhancing the robustness of time-series predictions. A time-varying modal parameter database is established, and a blind source separation algorithm is employed to track the impact of latent damage such as stiffness degradation on frequencies. The surrogate model parameters are recalibrated quarterly to adapt to material aging effects. Intelligent algorithms are used for fusion prediction, and time-series decomposition is performed. LMD is applied to split the frequency sequence into environmental and structural state components, and bidirectional LSTM prediction is conducted. Environmental parameters and historical frequencies are input, and the corrected frequency prediction values ​​are output. Static and dynamic tests are performed for comparison, and modal confidence criteria are applied for multi-source cross-validation.

[0027] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for correcting the predicted natural frequency of a bridge under operational conditions, characterized in that, Includes the following steps: S1. Bridge data acquisition: Deploy a multi-source sensor network for data acquisition; S2. Data preprocessing: perform data cleaning and anomaly handling, correct abnormal data and dynamically avoid interference. S3. Decouple environmental and load disturbances, construct a temperature-frequency transfer function, and deploy a distributed dynamic weighing system and machine vision fusion technology; S4. Build a model by modeling multimodal data and establishing a dynamic correction engine based on the core AI model; S5, AI-driven model correction, performs deep learning feature extraction, parameter uncertainty quantification, and response surface replacement model optimization; S6. Verification and iteration mechanism: perform cross-validation and establish a time-varying database of modal parameters.

2. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S1, an array of acceleration sensors is arranged at key points such as the mid-span and supports to collect vibration signals in the 0.5-20Hz main frequency band, identify modal frequencies and mode shape characteristics, and thus perform vibration monitoring; a temperature and humidity sensor is integrated to record the thermal expansion and contraction effect of materials, thereby performing synchronous environmental monitoring. Dynamic load capture is performed, integrating a distributed dynamic weighing system with machine vision to identify vehicle type, speed and axle load distribution in real time. A drone and AI crack recognition system are used to achieve damage location accuracy of ±0.5m.

3. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S2, high-frequency noise is separated by db4 wavelet packet decomposition while retaining structural characteristic frequencies, thereby completing noise filtering; outlier removal is performed by using a filter with a sliding window to remove outliers beyond three standard deviations, and data during periods without vehicles is automatically screened based on load spectrum analysis to avoid transient impact effects.

4. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S2, monitoring data exceeding the threshold is replaced by the arithmetic mean of the previous and next 5 data points. Monitoring data from periods without vehicle traffic are preferred to avoid frequency shifts caused by the added mass of vehicles. Single-mode response is extracted by combining wavelet analysis to improve the availability of data in noisy environments.

5. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S3, a negative linear regression model of temperature and modal frequency is established through long-term monitoring data. Nonlinear principal component analysis is used to remove the coupling effect of temperature and vehicle load. An artificial neural network is used to realize the mapping correction from environmental variables to frequency. A distributed dynamic weighing system and machine vision fusion technology are deployed. A random traffic flow probability model is constructed to replace the traditional static load assumption. A deep convolutional neural network is applied to identify the characteristics of moving loads. The transient impact effect is reconstructed by combining the vehicle-bridge coupled vibration equation.

6. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S4, multimodal data modeling includes an environmental disturbance decoupling module, dynamic load feature extraction, and structural parameter uncertainty quantification; the dynamic correction engine adopts a hybrid deep learning architecture to optimize the surrogate model and is equipped with a correction stream transformer.

7. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S5, a CNN-LSTM hybrid architecture is used for deep learning feature extraction to generate synthetic data-enhanced samples, thereby improving the robustness of the model under small sample conditions. The posterior distribution is updated by combining MCMC sampling with measured data, the parameter sensitivity is calculated, highly sensitive parameters are corrected first, a Kriging surrogate model is established, and the support vector machine kernel function is optimized using a genetic algorithm to achieve efficient parameter inversion.

8. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S5, the finite element model parameters are iteratively corrected in descending order of parameter sensitivity until the theoretical frequency and measured value converge, establishing a nonlinear mapping relationship between structural parameters and dynamic response. For example, a genetic algorithm is used to optimize the support vector machine model, significantly improving prediction efficiency and accuracy. The improved grey prediction model processes fluctuation data through data translation and geometric mean to reduce prediction errors. A bidirectional long short-term memory network combined with a local mean decomposition algorithm decomposes environmental and structural state data subsequences, enhancing the robustness of time series prediction.

9. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S6, a time-varying database of modal parameters is established, and a blind source separation algorithm is used to track the impact of latent damage such as stiffness degradation on the frequency. The surrogate model parameters are recalibrated quarterly to adapt to the material aging effect.

10. The method for correcting the predicted natural frequency of a bridge under operational conditions as described in claim 1, characterized in that: In S6, intelligent algorithms are used to fuse predictions and perform time-series decomposition. Local mean decomposition is applied to split the frequency sequence into environmental components and structural state components. Bidirectional LSTM prediction is performed, with environmental parameters and historical frequencies as inputs, and the corrected frequency prediction values ​​as outputs. Static and dynamic tests are performed for comparison and modal confidence criteria, thereby performing multi-source cross-validation.

Images