Bridge safety assessment method based on multi-intelligent algorithm fusion

Abstract

The invention discloses a bridge safety assessment method based on multi-intelligent algorithm fusion, and belongs to the technical field of bridge engineering and intelligent algorithms. Aiming at the problems of insufficient multi-source data processing, lack of physical constraints and the like of a traditional evaluation technology, the method integrates multi-dimensional indexes through eight steps of multi-source data fusion preprocessing, physical information neural network construction, adaptive deep forest feature learning and the like, and generates a structural vulnerability index (SVI) to guide a maintenance decision. Innovation points comprise a five-algorithm deep coupling architecture, a physically embedded digital twinborn body, a dynamic weight proof theory and the like. The method has the advantages that high-precision recognition, interpretable prediction and visual decision making of bridge damage are achieved, evaluation precision and operation and maintenance efficiency are remarkably improved, and the method is suitable for full-life-cycle management of various bridges.

Description

Technical Field

[0001] The present invention relates to the technical fields of bridge engineering and intelligent algorithms, and particularly relates to a bridge safety assessment method based on the fusion of multiple intelligent algorithms. Background Art

[0002] At present, as a core component of transportation infrastructure, the safety assessment of bridges is crucial for public safety and economic operation. Traditional bridge safety assessment technologies mainly rely on single sensors (such as strain gauges, accelerometers) to monitor key parameters, and combine empirical formulas or finite element models for damage judgment, which have the following significant defects: Insufficient multi-source data fusion technology: Existing methods mostly independently process data such as strain, vibration, and vision, and do not effectively solve data heterogeneity and environmental noise interference (such as strain drift caused by temperature changes, vibration noise caused by traffic loads). For example, the traditional Hampel filter uses a fixed threshold and cannot suppress the influence of daily periodic temperature fluctuations on sensor data, resulting in an abnormal value misjudgment rate of more than 15%.

[0003] Modal decomposition technologies (such as traditional VMD) do not consider signal distribution matching and frequency interval constraints, are prone to modal aliasing, and it is difficult to accurately separate abnormal vibration modes caused by damage (such as the frequency overlap problem between the cable relaxation of a cable-stayed bridge and the vibration of the main girder).

[0004] Disconnection between physical constraints and data-driven models: Although pure data-driven models (such as LSTM, CNN) can fit historical data, they lack structural mechanics constraints, and the prediction results may violate physical laws (such as the nonlinear relationship between the mid-span displacement and load of a simply supported beam being ignored) under extreme working conditions (such as overloading, earthquake), resulting in insufficient assessment reliability.

[0005] Traditional finite element models rely on idealized material parameters and boundary conditions, have weak real-time update capabilities (the model update period is usually ≥ 24 hours), and cannot reflect material degradation (such as the decrease in elastic modulus caused by concrete carbonation) and geometric deformation (such as the reduction in cross-section stiffness caused by crack propagation) during the service process of bridges.

[0006] Single algorithms are difficult to cope with complex damage patterns: There are various types of bridge damage (such as steel structure fatigue cracks, concrete cracks, cable force loss). Traditional single classifiers (such as random forests) have limited learning capabilities for high-dimensional non-linear features (such as the kurtosis of vibration signals, the fractal dimension of cracks), and the damage identification accuracy is generally lower than 80%.

[0007] The parameter optimization process mostly adopts a single objective (such as only optimizing the recognition accuracy), ignoring the inference efficiency and model sensitivity, resulting in problems such as "high accuracy but long calculation time" or "high sensitivity leading to false alarms" in practical applications. Summary of the Invention

[0008] The object of the present invention is to provide a bridge safety assessment method based on the fusion of multiple intelligent algorithms, which solves the problems of multi-source data processing, physical constraint modeling, multi-algorithm collaboration and visual decision-making in traditional technologies.

[0009] To achieve the above object, the present invention provides a bridge safety assessment method based on the fusion of multiple intelligent algorithms, including the following steps: S1. Clean the bridge sensor data, visual data and environmental data, and perform modal decomposition on the sensor data through improved variational mode decomposition to generate a high-dimensional feature vector containing vibration modal features, image spectrum features and environmental compensation strain; S2. Fusion the sensor time series features and finite element physical constraints through a dual-channel network, embed the beam bending equation as a physical loss term, and output the structure response prediction value; S3. Adopt a four-level cascaded forest structure, combine the dynamic adjustment of tree depth and the feature importance feedback mechanism to extract high-dimensional non-linear damage features; S4. Take the damage recognition accuracy, inference time and vulnerability index sensitivity as the objective functions, and generate a multi-objective optimal parameter combination through an improved firefly algorithm; S5. Establish a hierarchical Bayesian model, and calculate the confidence interval of the prediction result through Hamiltonian Monte Carlo sampling; S6. Dynamically adjust the multi-model weights according to the real-time prediction error, and prove the weight convergence through Lyapunov stability theory; S7. Real-time update the finite element model, combine CUDA parallel computing and phase field method simulation to realize the rendering of the structural stress nephogram and the prediction of crack evolution; S8. Integrate the structural strength, dynamic characteristics and fatigue life indexes, generate a comprehensive evaluation index through adaptive weights, and guide the hierarchical maintenance decision-making.

[0010] Preferably, in step S1, the outlier processing adopts an improved Hampel filter, and its threshold calculation formula is ; where is the standard deviation multiple, MAD is the median absolute deviation, t is the time, and the dynamic factor reflects the daily cycle environmental impact.

[0011] Preferably, in step S1, the mathematical optimization model of the improved variational mode decomposition is: ; Among them, is the set of each modal component obtained by decomposition, is the set of the central frequencies of each modal component; is the modal component, is the central frequency, KL The divergence constraint makes the simulated signal and the observed signal have the same distribution, and the frequency interval constraint avoids modal aliasing; is the total number of modal components, is the unit impulse function, is the partial differential operator, is the imaginary unit, is the complex exponential modulation factor, which is used to perform frequency modulation on the signal; is KL the weight coefficient of the divergence constraint term, is KL the divergence, which is used to measure the difference between two probability distributions and ; is the weight coefficient of the frequency interval constraint term; For the difference between two adjacent modal central frequencies and take the square of the norm.

[0012] Preferably, in step S2, the dual-channel network is the dual-channel architecture of the Physics-Informed Neural Network (PINN), which specifically includes: Sensor data channel: Extract the temporal features of strain and vibration signals through 1D-CNN; Physical constraint channel: Input finite element parameters and embed the beam bending equation as the physical loss term.

[0013] Preferably, in step S3, a four-level cascaded forest structure is adopted, combined with the tree depth dynamic adjustment and feature importance feedback mechanism, and the four-level cascaded forest structure is: The first layer: 10 random forests, with a feature random sampling rate of 0.8; The second layer: 8 completely random forests, with single-feature splitting; The third layer: 6 gradient boosting forests, with a learning rate of 0.05; The fourth layer: 4 XGBoost forests, with a maximum depth of 7.

[0014] Preferably, in step S4, taking the damage identification accuracy, inference time, and vulnerability index sensitivity as the objective function, generate a multi-objective optimal parameter combination through an improved firefly algorithm, which specifically includes: ; Among them, It is the first optimization objective function, used to measure the accuracy loss of the model; It is the second optimization objective function, used to measure the inference time of the model; is the third optimization objective function, which measures the sensitivity of the model to bridge damage; is the parameter to be optimized, For input The evaluation model, is the damage identification accuracy, is the inference time, and ▽SVI is the vulnerability index gradient.

[0015] Preferably, in step S5, a hierarchical Bayesian model is established, as shown below: Observation layer: ; in, is the observation data; are model parameters; is the observation noise variance; is a normal distribution; is the model prediction value; is the identity matrix; Parameter layer: ; in, is the parameter dimension; is the accuracy hyperparameter, controlling the model parameters The degree of concentration of the prior distribution, The precision corresponding to the i-th parameter; is the product distribution; Assume that the model parameters are normal distributions. The mean is 0 and the variance is Normal distribution; Hyperparameter layer: ; in, is the gamma distribution, and is the hyperparameter Provide a prior distribution and control its statistical properties; is the gamma distribution hyperparameter, Control the shape of the distribution, Control scale; is the gamma distribution hyperparameter, which controls the prior distribution of the observation noise accuracy; is the noise accuracy; Compute 95% confidence intervals for the posterior predictive distribution via Hamiltonian Monte Carlo sampling.

[0016] Preferably, in step S7, the finite element model is updated in real time to a three-dimensional digital twin, and the construction steps include: Real-time stress contour rendering: Based on CUDA parallel computing, map strain values to RGB colors; Crack evolution prediction: Use the phase-field method to simulate crack propagation.

[0017] Preferably, in step S8, the formula for integrating the structural strength, dynamic characteristics, and fatigue life indicators is the structural vulnerability index SVI, and its expression is: ; where SSI is the structural strength index, DFI is the dynamic characteristics index, FLI is the fatigue life index, and the weights α, β, γ are adaptively adjusted through a linear model that includes the index change rate and environmental severity.

[0018] Preferably, integrate the strength, dynamics, and fatigue indicators, and the adaptive weights reflect the damage development speed and environmental impact to generate a quantitative evaluation result to guide maintenance decisions.

[0019] Therefore, the bridge safety assessment method based on the integration of multiple intelligent algorithms with the above structure has the following beneficial effects: (1) The present invention proposes a "five-algorithm deep coupling" framework (PINN, ADF, MOFA, Bayesian model, dynamic weight mechanism), which dynamically integrates the physics-informed neural network with the four-level cascade forest and multi-objective optimization algorithm for the first time, breaking through the performance bottleneck of traditional single algorithms; (2) Interpretability: PINN and the digital twin ensure that the prediction results conform to mechanical laws and support damage mechanism analysis; (3) Robustness: The adaptive mechanism and uncertainty quantification adapt to complex working conditions, and the false alarm rate is reduced to 2.1%; (4) Decision support: 3D visualization and the SVI index provide intuitive evaluation results, reducing the maintenance cost by 40% and increasing the early warning lead time by 320%.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0021] Figure 1 It is a structural schematic diagram of a bridge safety assessment method based on the integration of multiple intelligent algorithms of the present invention. Detailed Embodiments

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical terms or scientific terms used in this invention shall have the ordinary meanings understood by those with ordinary skills in the field to which this invention pertains. The "first", "second" and similar terms used in this invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0024] Embodiment

[0025] As Figure 1 shown, the present invention provides a bridge safety assessment method based on the fusion of multiple intelligent algorithms, including the following steps: S1. Clean the bridge sensor data, visual data and environmental data, and perform modal decomposition on the sensor data through improved variational mode decomposition to generate a high-dimensional feature vector containing vibration modal features, image spectrum features and environmental compensation strain; S2. Fusion the sensor time series features and finite element physical constraints through a dual-channel network, embed the beam bending equation as a physical loss term, and output the structural response prediction value; S3. Adopt a four-level cascaded forest structure, combine the dynamic adjustment of tree depth and the feature importance feedback mechanism to extract high-dimensional non-linear damage features; S4. Take the damage identification accuracy, inference time and vulnerability index sensitivity as the objective function, and generate a multi-objective optimal parameter combination through the improved firefly algorithm; S5. Establish a hierarchical Bayesian model, and calculate the confidence interval of the prediction result through Hamiltonian Monte Carlo sampling; S6. Dynamically adjust the multi-model weights according to the real-time prediction error, and prove the weight convergence through Lyapunov stability theory; S7. Update the finite element model in real time, combine CUDA parallel computing and phase field method simulation to realize the rendering of the structural stress nephogram and the prediction of crack evolution; S8. Integrate the structural strength, dynamic characteristics and fatigue life indexes, generate a comprehensive evaluation index through adaptive weights, and guide the hierarchical maintenance decision-making.

[0026] Among them, data acquisition: 200 strain sensors (100 Hz) and 50 vibration accelerometers (500 Hz) collect data in real time, and the drone obtains crack images (0.1 mm accuracy); Preprocessing: Improve the Hampel filter to remove temperature drift, decompose the vibration signal by IVMD, and identify the abnormal resonance band of 3.2 - 4.0 Hz (pre - warning of cable force loss). The outlier processing uses an improved Hampel filter, and its threshold calculation formula is ; where, is the multiple of the standard deviation, MAD is the median absolute deviation, t is time, and the dynamic factor reflects the influence of the daily cycle environment.

[0027] PINN modeling: The dual - path network inputs sensor data and finite - element parameters, and inversely calculates the cable force loss rate (such as the 3rd cable loses 18.7%). The physical loss term ensures that the inversion result conforms to the beam theory.

[0028] The dual - path architecture of the physics - informed neural network (PINN) includes: Sensor data path: Extract the time - series features of strain and vibration signals through 1D - CNN; Physical constraint path: Input finite - element parameters and embed the beam bending equation as the physical loss term.

[0029] The mathematical optimization model of the improved variational mode decomposition is: ; where, is the set of each modal component obtained by decomposition, is the set of the central frequencies of each modal component; is the modal component, is the central frequency, KL The divergence constraint simulates that the signal and the observed signal have the same distribution, and the frequency interval constraint avoids modal aliasing; is the total number of modal components, is the unit impulse function, is the partial differential operator, is the imaginary unit, is the complex - exponential modulation factor used to frequency - modulate the signal; is KL the weight coefficient of the divergence constraint term, is KL the divergence, used to measure the difference between two probability distributions and ; is the weight coefficient of the frequency interval constraint term; For adjacent two modal central frequencies and Take the difference norm square.

[0030] ADF Feature Learning: The four - level forest extracts features such as crack width and main vibration frequency, and the gradient - boosted forest focuses on micro - crack identification. The Adaptive Deep Forest (ADF) adopts a four - level cascade structure: The first layer: 10 random forests, with a feature random sampling rate of 0.8; The second layer: 8 completely random forests, with single - feature splitting; The third layer: 6 gradient - boosted forests, with a learning rate of 0.05; The fourth layer: 4 XGBoost forests, with a maximum depth of 7.

[0031] MOFA Optimization: Balance the accuracy (98%) and inference time (180ms) to generate the optimal model parameter combination. The objective function of the multi - objective firefly optimization includes: ; where is the first optimization objective function, used to measure the accuracy loss of the model; is the second optimization objective function, used to measure the inference time of the model; is the third optimization objective function, measuring the sensitivity of the model to bridge damage; are the parameters to be optimized, is the input evaluation model, is the damage recognition accuracy rate, is the inference time, and ▽SVI is the vulnerability index gradient.

[0032] BUQ Analysis: Bayesian uncertainty quantification is achieved through a hierarchical Bayesian model; Observation layer: ; where is the observed data; are the model parameters; is the observation noise variance; is the normal distribution; is the model prediction value; is the identity matrix; Parameter layer: ; where is the parameter dimension; is the precision hyperparameter, controlling the concentration degree of the prior distribution of the model parameter , corresponding to the precision of the i - th parameter; is the product distribution; is the prior normal distribution, assuming the model parameter Follow a normal distribution with a mean of 0 and a variance of ; Hyperparameter layer: ; Among them, is a gamma distribution, and is a hyperparameter providing a prior distribution to control its statistical characteristics; is the hyperparameter of the gamma distribution, controlling the distribution shape, controlling the scale; is the hyperparameter of the gamma distribution, controlling the prior distribution of the observation noise precision; is the noise precision; Calculate the 95% confidence interval of the posterior predictive distribution through Hamiltonian Monte Carlo sampling.

[0033] The confidence interval of the remaining life prediction is [6.21, 8.15] years, providing probabilistic support for the maintenance plan.

[0034] Digital twin: The Unity3D engine renders the stress nephogram in real time, and the AR marks the position of the damaged cable to guide precise maintenance.

[0035] SVI calculation: The calculation formula of the structural vulnerability index SVI is: ; Among them, SSI is the structural strength index, DFI is the dynamic characteristic index, FLI is the fatigue life index, and the weights α, β, γ are adaptively adjusted through a linear model including the index change rate and environmental severity.

[0036] Combining SSI (strength degradation), DFI (frequency shift), and FLI (fatigue consumption), an adaptive weight generates SVI = 0.62, triggering a decision to limit load and pass.

[0037] Finally, integrate the strength, dynamic, and fatigue indicators. The adaptive weights reflect the damage development speed and environmental impact, generating a quantitative evaluation result to guide the maintenance decision.

[0038] Taking a 300m-span cable-stayed bridge as an example: Data acquisition: 200 strain sensors (100Hz) and 50 accelerometers (500Hz) sample in real time, and the drone obtains crack images with a precision of 0.1mm.

[0039] Preprocessing verification: Improve the Hampel filter to eliminate the strain drift caused by the daily temperature difference (20°C) (the error is reduced from 18% to 3.5%), and the IVMD separates the characteristic frequency of 3.6Hz of the cable relaxation.

[0040] PINN Inversion: The dual-channel network inputs sensor data and finite element parameters to invert the cable force loss of Cable No. 3 by 18.7%. The physical loss term ensures that the error between the inversion result and the measured cable force is <1%.

[0041] ADF Feature Extraction: The four-level forest extracts features such as the fractal dimension of cracks (1.72) and the main vibration frequency shift (from 4.2 Hz to 3.8 Hz). The damage identification accuracy is 98.6%.

[0042] SVI Decision-making: Integrate SSI (strength degradation of 15%), DFI (frequency shift of 9.3%), and FLI (fatigue consumption of 28%) to generate SVI = 0.62, triggering the maintenance decision of restricting the load of the main span by 50%.

[0043] Therefore, through the deep coupling of technologies such as data preprocessing, physics-informed neural network, deep forest, multi-objective optimization, uncertainty quantification, and digital twin, the present invention realizes high-precision identification, interpretable prediction, and visual decision-making of bridge damage, significantly improving the evaluation accuracy and operation and maintenance efficiency, and is applicable to the whole life cycle management of various bridges.

[0044] And the present invention forms a complete technical chain from multi-source data preprocessing (S1) to digital twin decision-making (S8): Data layer: Improve the Hampel filter (dynamic threshold formula) and IVMD (KL divergence constraint) to suppress environmental noise; Model layer: The ADF four-level forest (random forest → XGBoost) extracts 23-dimensional non-linear features, and MOFA optimizes the objective function group (accuracy / inference time / SVI gradient); Application layer: The digital twin renders the stress nephogram through CUDA, and the SVI index (α, β, γ adaptive weights) guides hierarchical maintenance. The integrity of the technical link is better than the traditional solution.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A bridge safety assessment method based on the fusion of multiple intelligent algorithms, characterized in that, It includes the following steps: S1. Clean the bridge sensor data, visual data and environmental data, and perform modal decomposition on the sensor data through improved variational mode decomposition to generate a high-dimensional feature vector containing vibration modal features, image spectrum features and environmentally compensated strain; S2. Fusion the sensor time series features and finite element physical constraints through a dual-channel network, embed the beam bending equation as a physical loss term, and output the structural response prediction value; S3. Adopt a four-level cascaded forest structure, combine the dynamic adjustment of tree depth and the feature importance feedback mechanism to extract high-dimensional non-linear damage features; S4. Take the damage identification accuracy, inference time and vulnerability index sensitivity as the objective function, and generate a multi-objective optimal parameter combination through an improved firefly algorithm; S5. Establish a hierarchical Bayesian model, and calculate the confidence interval of the prediction result through Hamiltonian Monte Carlo sampling; S6. Dynamically adjust the multi-model weights according to the real-time prediction error, and prove the weight convergence through Lyapunov stability theory; S7. Update the finite element model in real time, combine CUDA parallel computing and phase field method simulation to realize the rendering of the structural stress nephogram and the prediction of crack evolution; S8. Integrate the structural strength, dynamic characteristics and fatigue life indicators, generate a comprehensive evaluation index through adaptive weights, and guide the hierarchical maintenance decision-making.

2. The bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S1, the outlier processing uses an improved Hampel filter, and its threshold calculation formula is ; Among them, is the multiple of the standard deviation, MAD is the median absolute deviation, t is the time, and the dynamic factor reflects the impact of the daily cycle environment.

3. The bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, wherein: In step S1, the mathematical optimization model of the improved variational mode decomposition is: ； Among them, is the set of each decomposed modal component, is the set of the center frequencies of each modal component; is the modal component, is the center frequency, KL The divergence constraint makes the simulated signal and the observed signal have the same distribution, and the frequency interval constraint avoids mode aliasing; is the total number of modal components, is the unit impulse function, is the partial differential operator, is the imaginary unit, is the complex exponential modulation factor, which is used to frequency-modulate the signal; is KL the weight coefficient of the divergence constraint term, is KL the divergence, which is used to measure the difference between two probability distributions and ; is the weight coefficient of the frequency interval constraint term; takes the and difference between two adjacent modal center frequencies the square of the norm.

4. The bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S2, the dual-channel network is the dual-channel architecture of the physics-informed neural network PINN, which specifically includes: Sensor data channel: Extract the time series features of strain and vibration signals through 1D-CNN; Physical constraint channel: Input the finite element parameters, and embed the beam bending equation as a physical loss term.

5. A bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S3, a four-level cascaded forest structure is adopted, combined with the dynamic adjustment of tree depth and the feature importance feedback mechanism, and the four-level cascaded forest structure is: The first layer: 10 random forests, with a feature random sampling rate of 0.8; The second layer: 8 completely random forests, with single-feature segmentation; The third layer: 6 gradient boosting forests, with a learning rate of 0.05; The fourth layer: 4 XGBoost forests, with a maximum depth of 7.

6. The bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S4, taking the damage identification accuracy, inference time and vulnerability index sensitivity as the objective function, and generating a multi-objective optimal parameter combination through an improved firefly algorithm, specifically including: ； Among them, is the first optimization objective function, which is used to measure the precision loss of the model; is the second optimization objective function, which is used to measure the inference time consumption of the model; is the third optimization objective function, which measures the sensitivity of the model to bridge damage; is the parameter to be optimized, is the input of the evaluation model, is the damage recognition accuracy rate, is the inference time, and ▽SVI is the vulnerability index gradient.

7. A bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S5, establish a hierarchical Bayesian model, specifically as follows: Observation layer: ; Among them, is the observed data; is the model parameter; is the variance of the observation noise; is the normal distribution; is the model predicted value; is the identity matrix; Parameter layer: ; Among them, is the parameter dimension; is the precision hyperparameter that controls the concentration degree of the prior distribution of the model parameter , and corresponds to the precision of the i-th parameter; is the product distribution; is the prior normal distribution, assuming that the model parameter follows a normal distribution with a mean of 0 and a variance of . Hyperparameter layer: ; Among them, is a gamma distribution, and is a hyperparameter provides a prior distribution to control its statistical characteristics; is the hyperparameter of the gamma distribution, controls the shape of the distribution, controls the scale; is the hyperparameter of the gamma distribution, which controls the prior distribution of the observation noise precision; is the noise precision; Calculate the 95% confidence interval of the posterior predictive distribution through Hamiltonian Monte Carlo sampling.

8. A bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S7, update the finite element model in real time to a three-dimensional digital twin, and its construction steps include: Real-time stress nephogram rendering: Based on CUDA parallel computing, map the strain value to RGB color; Crack evolution prediction: Use the phase field method to simulate crack propagation.

9. A bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: In step S8, the formula for integrating the structural strength, dynamic characteristics and fatigue life indicators is the structural vulnerability index SVI, and its expression is: ； Among them, SSI is the structural strength index, DFI is the dynamic characteristic index, FLI is the fatigue life index, and the weights α, β, and γ are adaptively adjusted through a linear model that includes the index change rate and environmental severity.

10. The bridge safety assessment method based on the fusion of multiple intelligent algorithms according to claim 1, characterized in that: Integrating strength, dynamic, and fatigue indicators, the adaptive weights reflect the damage development rate and environmental impact, generating a quantitative evaluation result to guide maintenance decisions.

Images