Cable-stayed bridge structure displacement effect analysis method and system

Abstract

The invention discloses a cable-stayed bridge structure displacement effect analysis method and system, and the method comprises the following steps: S1, data collection and processing: collecting monitoring data related to the displacement response of a cable-stayed bridge, and carrying out the preprocessing of the data, and obtaining a data set; s2, data set expansion: performing spectral analysis on time series data of the cable-stayed bridge by taking a timestamp as an index, performing feature engineering and time-frequency feature extraction to realize expansion of an existing feature number, expanding the expansion data into a data set, and randomly dividing the data set into a training set and a test set; s3, constructing a model to obtain a structure displacement prediction model; and S4, model testing. According to the method, the learnable normalization and anti-normalization technologies are combined, dynamic circulation closing of processes such as data, models, model explanation and explanatory feedback is achieved, the generalization performance of the prediction model is improved, and practical application of a cable-stayed bridge structure displacement effect analysis algorithm model is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction, and more specifically, to a method for analyzing displacement effects of a cable-stayed bridge structure. Background Art

[0002] Real-time monitoring of cable-stayed bridge structural performance is essential to ensure structural safety and functionality. Traditional monitoring methods rely on structural analysis and physical experimental methods, which have limitations in dealing with complex loading conditions and require a complete understanding of structural conditions, loading forms and working conditions, and rely heavily on subjective experience. In recent years, data-driven multivariate time series models have emerged, such as the use of moving average models (ARIMA) to monitor abnormal events or structural changes in post-tensioned prestressed concrete box girder bridges, which to some extent overcome the reliance on subjective experience.

[0003] However, for structural monitoring data indexed by timestamps, it is inevitable that under complex working conditions, data will jump frequently, and statistical characteristics such as mean and variance will change dynamically over time, making it difficult to remove noise. Such data changes in data distribution pose great challenges to the construction of structural displacement prediction models, and traditional time series modeling algorithms cannot be fully applied.

[0004] On the other hand, traditional modeling and prediction methods only predict the displacement of the target structure by fitting historical monitoring data, and realize end-to-end input and output of monitoring indicators to structural displacement. The internal mechanism of the model is unclear, which is not conducive to the effective diagnosis of the rationality of the model prediction; at the same time, there is a lack of quantitative or qualitative analysis between different factors (such as temperature, wind, load, stress, etc.) and the displacement of the target structure. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for analyzing the displacement effect of a cable-stayed bridge structure. First, data collection is performed, and a cable-stayed bridge structure displacement effect prediction data set is constructed through data denoising, normalization and anti-normalization operations; feature engineering and spectrum analysis are performed on the constructed structural displacement prediction data, and the data set is expanded through feature construction and time-frequency feature extraction; a real-time prediction model for upstream and downstream and multi-directional structural displacements of a cable-stayed bridge is constructed using the feature-engineered data set, and the model decision layer is fused; then test verification and hyperparameter optimization are performed, and performance evaluation is performed; an improved SHAP method is used to perform an interpretative analysis on the structural displacement prediction model, and according to the analysis results, the influence weights of the input features are dynamically adjusted, the prediction model is dynamically updated, and a trade-off is made between efficiency and accuracy to obtain the optimal model; finally, engineering measurement and application are performed.

[0006] The technical solution adopted by the present invention to solve this technical problem is: a method for analyzing displacement effect of a cable-stayed bridge structure, comprising the following steps:

[0007] S1. Data collection and processing: Collect monitoring data related to the displacement response of the cable-stayed bridge and pre-process the data to obtain a data set;

[0008] S2, feature engineering: perform spectrum analysis on the time series data of the cable-stayed bridge with timestamp as index, perform feature engineering and time-frequency feature extraction, expand the existing feature number, expand the expanded data into the data set, and randomly divide the data set into training set and test set;

[0009] S3, model construction: Establish a multi-objective output prediction algorithm wrapper suitable for real-time prediction of upstream, downstream, X, Y, and Z directions of cable-stayed bridge structures. Use the training set and basic algorithms such as CatBoost, RandomForest, and decision trees as meta-learning algorithms to build structural displacement prediction algorithm chains respectively, and perform decision-layer fusion on the constructed integrated learning model to obtain a structural displacement prediction model.

[0010] S4. Model testing: The structural displacement prediction model is tested using the test set, and the improved cross-validation is used to evaluate the model performance to obtain comprehensive and stable prediction results;

[0011] S5. Hyperparameter tuning: The hyperparameters of the structural displacement prediction model are tuned by the grid search method to determine the optimal model parameter configuration;

[0012] S6. Performance evaluation: Use statistical performance indicators (such as R 2 , MAPE, AIC, BIC) to evaluate the prediction accuracy and generalization ability of the model;

[0013] S7. Model interpretation and update: Apply Weighted SHAP to interpret model predictions and identify the contribution of each feature to the prediction results; adjust the contribution weight of the input features according to the relative size of the contribution of each feature, and dynamically update the prediction algorithm; that is, use Weighted SHAP analysis to provide individual and global explanations of the model output for each feature, while enhancing the transparency of the model's prediction results, and further utilize the explanation results to achieve dynamic updates of the prediction model, realizing a dynamic cycle of data, modeling prediction, explanation analysis to model update.

[0014] S8. Result verification and application: Apply the trained model to the test set that has not participated in the training to verify the model's predictive performance; that is, verify the practicality and effectiveness of the model through actual bridge monitoring data to provide decision support for bridge maintenance and management.

[0015] As a further solution of the present invention, the monitoring data includes influencing factors such as temperature, wind speed, load, stress, etc.;

[0016] The data preprocessing specifically includes noise reduction, normalization and denormalization operations to improve the quality and efficiency of model training.

[0017] As a further solution of the present invention, the present invention aims to solve the problems of frequent dynamic changes in cable-stayed bridge monitoring data, large data jumps, and difficulty in removing noise, and establishes a cable-stayed bridge structure displacement monitoring data denoising algorithm that combines variational mode decomposition and bidirectional upper short-term memory network to denoise the monitoring data.

[0018] As a further solution of the present invention, hyperparameters are tuned including learning rate, number of trees, maximum depth of trees, minimum child weight, sample sampling rate and feature sampling rate, etc.; meanwhile, for decision layer fusion, the objects of hyperparameter tuning also include basic model weights.

[0019] The present invention also provides a system for analyzing the displacement effect of a cable-stayed bridge structure, comprising:

[0020] The data acquisition module is used to collect the environmental and operating condition data of the cable-stayed bridge in real time, including temperature, wind speed, humidity, load, etc. Specifically, the data acquisition module is integrated with the sensor data interface deployed on the bridge structure to dynamically receive input signals from various sensors in real time and obtain dynamic monitoring data indexed by timestamps, providing data support for the construction of the cable-stayed bridge structure displacement prediction model;

[0021] The data preprocessing module is used to preprocess the collected raw data, including time stamp conversion, timestamp denoising, normalization and denormalization of time series monitoring data, etc., to improve data quality;

[0022] The feature extraction and selection module is used to automatically identify and extract features related to the displacement response of the cable-stayed bridge; including: performing spectrum analysis on the input time series monitoring data, extracting the frequency domain features of the monitoring data, and constructing features on the input time series monitoring data, such as calculating the temperature difference and dimensional gradient of different main beam segments, to achieve feature construction and expansion, and expand sample diversity;

[0023] The model training and management module is used for the training configuration management of the cable-stayed bridge structure displacement prediction model, including the initialization and adjustment of hyperparameters, real-time feedback of model training loss and prediction accuracy, and the saving and serialization export of the optimal model checkpoint at the stage;

[0024] The explanatory analysis module combines the Weighted SHAP method to analyze the qualitative and quantitative relationship between the model prediction results and each input feature time, ensuring that users can intuitively understand the impact trend of each monitoring feature on the structural displacement, as well as the specific impact degree;

[0025] As a further solution of the present invention, the explanatory analysis module is combined with the model training and management module, and according to the evaluation results provided by the explanatory analysis, the influence weights of the input features are dynamically adjusted, and the cable-stayed bridge structure displacement prediction model is updated, thereby further improving the model prediction accuracy.

[0026] As a further solution of the present invention, it also includes:

[0027] A user interface module (UI) provides a user interaction interface, allowing the user to input parameters, start analysis, view results and interpretative analysis reports; the user interface module includes dynamic charts and dashboards for displaying real-time data and forecast results;

[0028] The system integration and communication module is responsible for coordinating the data flow and control flow between the modules within the system; it realizes the communication between the system and external devices (such as sensors and data storage), that is, it is used to communicate with the sensors deployed on the engineering structure to realize real-time data reading and updating.

[0029] A decision support module is used to provide suggestions for maintenance and management workflows based on the predictive analysis results. The decision support module integrates a rule engine to provide automated support for bridge maintenance decisions.

[0030] The system module design of the above-mentioned cable-stayed bridge structure displacement effect analysis method is intended to integrate the cable-stayed bridge structure displacement effect analysis method into a complete software and hardware platform to achieve full process automation from data acquisition to predictive analysis, explanatory reporting and decision support.

[0031] The present invention has at least the following beneficial effects:

[0032] 1. The present invention combines VMD with Bi-LSTM to perform noise reduction on the monitoring sequence data indexed by timestamps. While retaining the characteristics of the original data, the noise in the monitoring data is eliminated, ensuring the data quality and providing data support for establishing a high-performance structural displacement prediction model.

[0033] 2. The present invention realizes the unification of data scale by normalizing the experimental data set; at the same time, in view of the changes in data distribution after normalization, the distribution characteristics outside the original data scale are restored to the data through denormalization, thereby better retaining the key information in the data;

[0034] 3. In the process of time series data feature engineering, the sample diversity is expanded through feature construction, and the diversity at the feature level is enriched by combining spectrum analysis and frequency domain feature extraction. Through the fusion modeling of time and frequency features, the algorithm learning is more sufficient for the experimental data;

[0035] 4. By improving the traditional cross-validation, the time series data is prevented from being disrupted, the order of the time series is guaranteed as a whole, and the leakage of time series information is avoided;

[0036] 5. Using the improved Weighted SHAP method, different weights are assigned to features with different editing contributions, which better reflects the actual influence of each monitoring indicator on structural displacement and makes the analysis results closer to common sense. At the same time, an interpretable dynamic feedback framework is established, and the prediction model is dynamically updated in combination with the interpretable results, further improving the prediction accuracy.

[0037] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a spectrum analysis schematic diagram of the present invention;

[0039] Figure 2 This is the traditional cross validation graph;

[0040] Figure 3 A cross-validation diagram considering timestamps for the present invention (time series cross-validation);

[0041] Figure 4 A diagram showing the prediction results of the displacement prediction model of the present invention;

[0042] Figure 5 Schematic diagram of the interpretability feedback of the present invention. DETAILED DESCRIPTION

[0043] The present invention is described in detail and completely below in conjunction with the accompanying drawings. A person of ordinary skill in the art will be able to implement the present invention based on these descriptions. Before describing the present invention in conjunction with the accompanying drawings, it should be particularly pointed out that the technical solutions and technical features provided in each part of the present invention, including the following description, can be combined with each other without conflict.

[0044] In addition, the embodiments of the present invention involved in the following description are generally only a part of the embodiments of the present invention, rather than all the embodiments. Therefore, based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0045] The present invention is further described in detail below in conjunction with the accompanying drawings and implementations, and its specific implementation process is as follows:

[0046] The present invention provides a method for analyzing displacement effect of a cable-stayed bridge structure, comprising the following steps:

[0047] 1. Data collection and preprocessing:

[0048] Monitoring data related to the displacement response of cable-stayed bridges are collected, including influencing factors such as temperature, wind speed, load, and stress. The data is preprocessed, including noise reduction, normalization, and denormalization operations, to improve the quality and efficiency of model training. Among them, the present invention aims to address the problems of frequent dynamic changes in cable-stayed bridge monitoring data, large data jumps, and difficulty in removing noise. A cable-stayed bridge structure displacement monitoring data denoising algorithm combining variational mode decomposition and bidirectional upper short-term memory network is established, and normalization and denormalization operations are performed to achieve data scale consistency while ensuring the consistency of data distribution;

[0049] The noise reduction process of the cable-stayed bridge structure displacement monitoring data of the present invention is as follows:

[0050] 1) Process missing values ​​of original data;

[0051] 2) The pre-processed time series monitoring data of cable-stayed bridge structure displacement is used as the algorithm input, and the signal modal decomposition is performed using VMD to obtain n signal components; before the time series data is processed using VDM, the VDM is first optimized using the Grey Wolf Optimization Algorithm to determine the key parameters of the VMD decomposition sequence. The details are as follows:

[0052] a. First, initialize the gray wolf optimization algorithm parameters, set the number of gray wolves to m1, and the maximum number of iterations to n1. Set the two key parameters K and α of the VMD algorithm to gray wolf individuals, and their value ranges are [1,100] and [1500,2600] respectively;

[0053] b. Then, calculate the fitness. Under different gray wolf individual settings, use VMD to decompose the structural displacement sequence monitoring data, determine the fitness of each position based on its collocation entropy and inverse normalized mutual information entropy, and update the positions of the leader wolf γ, deputy leader wolf β and ordinary gray wolf δ at the same time;

[0054] c. Then, update the position of the gray wolf. Update the positions of other wolves according to the positions of γ, β, and δ. This process simulates the cooperation and competition relationship between gray wolves. The position update process is shown in the following formula:

[0055]

[0056] In the above formula, γ, β, and δ are the leader gray wolf, the deputy leader gray wolf, and the ordinary gray wolf respectively; x new is the updated position, x oldis the original position; A is a random number, generally ranging from [0,3]; distance γ 、distance β 、distance δ They are the distances from the current gray wolf to the leader gray wolf, deputy leader gray wolf and ordinary gray wolf respectively.

[0057] d. Finally, steps b and c are repeated continuously until the maximum number of times is reached and the optimal parameter combination k and α are output as the parameters of the current VMD decomposition sequence.

[0058] 3) Calculate the SE value of each signal component and set the corresponding threshold, then select some effective components from the signal components 1.,,n that meet the threshold for noise reduction. The SE value is used to measure the complexity of the structural displacement monitoring sequence number in the time series format and describe the randomness and irregularity of the sequence data;

[0059]

[0060] In the above formula, A(m,r) represents the proportion of similar sub-monitoring data sequences given the sub-monitoring data sequence length m and threshold r.

[0061] 4) The IMF components in step 3) are taken as input respectively, and then normalized and denormalized to improve the processing efficiency of Bi-LSTM.

[0062] 5) The processed monitoring data sequence components are synthesized to obtain the denoised structural displacement monitoring data sequence.

[0063] The normalization and denormalization process used in the present invention is as follows:

[0064] aFirst, the normalization process of the input monitoring data sequence is carried out. The mean and variance of the direction of the data row are calculated respectively, and each prediction window is normalized separately, as follows:

[0065]

[0066] In the above formula, E t , The mean and variance of each row of samples respectively; is the scaled sample obtained after processing; after the processing in this step, the impact of distribution changes on the model will be reduced, allowing the structural displacement prediction model to better focus on the time series correlation within the monitoring data sequence.

[0067] b Then there is the denormalization operation; in order to further avoid the change in data distribution caused by the normalization operation, the denormalization operation is further performed to restore the non-stationary information in the original input monitoring data sequence to the prediction output result of BiLSTM. The specific operations are as follows:

[0068]

[0069] 2. Feature Engineering:

[0070] The wind speed, wind direction, temperature, load, stress and other features are extracted from the original time series monitoring data of the cable-stayed bridge. The spectrum analysis is performed on these time series data with timestamp as index, and feature engineering and time-frequency feature extraction are performed to expand the existing feature number. First, the effect of spectrum analysis on the structural displacement monitoring data sequence is shown in the attached figure. Figure 1 As shown; the present invention also expands the vertical axis data obtained through spectrum analysis, that is, the amplitude spectrum, into the data in step 1 as a new data column.

[0071] Then, numerical calculations are performed on each column in the expanded structural displacement prediction modeling data set according to the corresponding physical meaning, that is, new feature data columns are obtained through feature construction; for example, the temperature monitoring data t of different main beam stages are 1 ,t 2 By making a difference, we can get the temperature difference characteristics dif between steel beams at different stages t1-t2 .

[0072] 3. Model construction:

[0073] A multi-objective output prediction algorithm wrapper suitable for real-time prediction of upstream, downstream, X, Y, and Z directions of cable-stayed bridge structures is established. Basic algorithms such as CatBoost, RandomForest, and decision trees are used as meta-learning algorithms to construct structural displacement prediction algorithm chains respectively, and the constructed integrated learning model is integrated at the decision layer.

[0074] The data structure used for model construction in the present invention is shown in the following table:

[0075] Table 1 Data structure table

[0076]

[0077] As shown in the table above, each sample of the data set constructed in the present invention contains multiple target output values ​​Y, and each specific sample corresponds to 7 output values. Therefore, according to the traditional method, at least 7 models need to be constructed each time to meet the real-time prediction output of multiple target values.

[0078] a. Create a multi-target output regressor chain; Specifically, the present invention takes the decision tree algorithm as an example, defines a wrapper, creates a linear model sequence, and takes the algorithm model that does not support multi-target output as the basic parameter, which is sequentially passed into the wrapper to form a regressor chain, so that multiple outputs can be predicted at one time. The implementation code example is as follows:

[0079] #Code Example

[0080] from sklearn.datasets import make_regression

[0081] from sklearn.multioutput import MultiOutputRegressor

[0082] from sklearn.tree import DecisionTreeRegressor

[0083] #Create dataset

[0084] X,y=make_regression(n_samples=1000,n_features=10,n_informative=5,n_targets=7,random_state=1)

[0085] #Define model

[0086] model=DecisionTreeRegressor()

[0087] # Pass the created model object as a parameter

[0088] wrapper=MultiOutputRegressor(model)

[0089] #Train model

[0090] wrapper.fit(X,y)

[0091] # Use the wrapper model to make predictions

[0092] data_test=[[-2.02220122,0.31563495,0.82797464,-0.30620401,0.16003707,

[0093] -1.44411381,0.87616892,-0.50446586,0.23009474,0.76201118]]

[0094] yhat=wrapper.predict(data_test)

[0095] #Summarize the prediction results and train a separate model for each output based on MultiOutputRegressor

[0096] print(yhat)

[0097] In the above code example, a decision tree regression algorithm is created and used as a parameter of the wrapper, a regressor chain is constructed, and the expansion of the decision tree algorithm suitable for single-target prediction tasks to a multi-target output algorithm is realized. Among them, n_targets = 7, corresponding to the number of output targets in the present invention is 7; n_features = 10, corresponding to the number of monitoring indicators used to predict structural displacement in the present invention is 10, n_samples = 1000, corresponding to the structural displacement prediction data set example constructed in the present invention has 1000 samples; similarly, regressor chains based on CatBoost and RandomForest are constructed respectively, and predictions are performed.

[0098] b. Fusing the created regressor chain into the decision-making model, and improving the reliability of the final prediction effect through group decision-making;

[0099] The process of re-fusing the model of the regressor chain in the present invention is as follows:

[0100] The prediction results of each structural displacement prediction regressor chain are averaged using the averaging method, and the formula is as follows:

[0101]

[0102] In the above formula, N is the number of regressor chains, is the prediction result of each regressor chain, The final fusion prediction result.

[0103] 4. Model Validation:

[0104] The structural displacement prediction model is trained using the training data set, and the improved cross-validation is used to evaluate the model performance to obtain comprehensive and stable prediction results.

[0105] Among them, as attached Figures 2-3 As shown, the difference between the present invention and the traditional cross-validation is that when dividing the data, the order of timestamps is restricted, ensuring that the data at the later moment always appears after the data at the previous moment, avoiding the problem of data leakage when the data at the future moment is scattered in disorder and put into the training set, resulting in model overfitting.

[0106] 5. Hyperparameter Tuning:

[0107] The hyperparameters of the structural displacement prediction model are tuned by the grid search method to determine the optimal model parameter configuration. The tuned hyperparameters include learning rate, number of trees, maximum tree depth, minimum child weight, sample sampling rate, and feature sampling rate. At the same time, for decision layer fusion, the objects of hyperparameter tuning also include basic model weights.

[0108] 6. Performance evaluation:

[0109] Use statistical performance indicators (such as R 2 , MSE) to evaluate the prediction accuracy and generalization ability of the model; the above indicators are defined as follows:

[0110]

[0111] In the above formula, y p (i) and y t (i) are the predicted displacement result and the real displacement result of the ith displacement respectively. The predicted results are shown in the attached Figure 4 shown.

[0112] 7. Model explanation and update:

[0113] Apply Weighted SHAP to interpret model predictions and identify the contribution of each feature to the prediction results. According to the relative size of each feature's contribution, adjust the contribution weight of the input feature, and dynamically update the prediction algorithm. That is, while using Weighted SHAP analysis to provide individual and global explanations of each feature's output of the model, it enhances the transparency of the model's prediction results, and further utilizes the explanation results to achieve dynamic updates of the prediction model, realizing a dynamic cycle of data, modeling prediction, explanation analysis, and model update.

[0114] aFirst, the model interpretability analysis is carried out; in Weighted SHAP, a weighted average method is added to realize the assignment of different weights to the marginal contribution of each monitoring table, so as to more flexibly reflect the influence degree and influence trend of each detection indicator on the actual structural displacement; the specific algorithm is as follows:

[0115]

[0116] b. Update the structural displacement prediction model using the explanatory analysis results. Specifically, according to the SHAPELY Value of each monitoring index calculated by WeightedSHAP, the factors that have a greater impact on the structural displacement can be distinguished. If these indicators are given greater weights when the structural displacement prediction model is trained, the model prediction results will be more targeted. Therefore, according to the explanatory results obtained by Weighted SHAP, the present invention further updates the input weights of each influencing factor to obtain a new structural displacement prediction model.

[0117] As attached Figure 5 As shown in the figure, the first one is the traditional modeling and prediction process, which is a typical black box prediction process; the second one is some existing methods, which increase the interpretability of model prediction results through SHAP analysis on the basis of traditional methods; the third one is the interpretability feedback method proposed in the present invention, that is, on the basis of some existing methods, further combined with the explanatory analysis results, weights are assigned to each monitoring indicator input into the prediction model, and the prediction model is updated again, thereby achieving a trade-off between prediction accuracy and reliability.

[0118] 8. Result verification and application:

[0119] The trained model is applied to a test data set that was not involved in the training to verify the model's prediction performance. The practicality and effectiveness of the model are verified through actual bridge monitoring data to provide decision support for bridge maintenance and management.

[0120] In another embodiment, the present invention also provides a method and system for analyzing displacement effects of a cable-stayed bridge structure. The following is the corresponding system module design:

[0121] 1. Data acquisition module:

[0122] Responsible for real-time collection of environmental and operating condition data of cable-stayed bridges, including temperature, wind speed, humidity, load, etc. Specifically, this module integrates with the sensor data interface deployed on the bridge structure, dynamically receives input signals from various sensors in real time, obtains dynamic monitoring data indexed by timestamps, and provides data support for the construction of cable-stayed bridge structure displacement prediction model;

[0123] 2. Data preprocessing module:

[0124] The data processing module is mainly used to pre-process the collected raw data, including time stamp conversion, timestamp and time series monitoring data denoising, normalization and denormalization, etc., to improve data quality;

[0125] 3. Feature extraction and selection module:

[0126] The feature extraction and selection module is mainly used to automatically identify and extract features related to the displacement response of cable-stayed bridges. This includes performing spectrum analysis on the input time series monitoring data and extracting the frequency domain features of the monitoring data; at the same time, feature construction is performed on the input time series monitoring data, such as calculating the temperature difference and dimensional gradient of different main beam segments, to achieve feature construction and expansion and expand sample diversity.

[0127] 4. Model training and management module:

[0128] The model training and management module is mainly used for the training configuration management of the cable-stayed bridge structure displacement prediction model, including the initialization and adjustment of hyperparameters, real-time feedback of model training loss and prediction accuracy, and the preservation and serialization export of the optimal model checkpoint at the stage.

[0129] 5. Interpretive Analysis Module:

[0130] Combined with the Weighted SHAP method, the qualitative and quantitative relationship between the model prediction results and each input feature time is analyzed to ensure that users can intuitively know the impact trend of each monitoring feature on the structural displacement, as well as the specific impact degree.

[0131] At the same time, combined with module 4 (model training and management module), according to the evaluation results provided by the explanatory analysis, the influence weights of the input features are dynamically adjusted, and the cable-stayed bridge structure displacement prediction model is updated, thereby further improving the model prediction accuracy.

[0132] 6. Other modules

[0133] Other modules also include relevant sub-function modules related to the present invention, including:

[0134] (1) User Interface (UI): Provides a user interface that allows users to input parameters, start analysis, and view results and interpretive analysis reports. It also includes dynamic charts and dashboards to display real-time data and forecast results.

[0135] (2) System integration and communication module: responsible for coordinating the data flow and control flow between modules within the system. It enables communication between the system and external devices (such as sensors and data storage), that is, it is used to connect with sensors deployed on the engineering structure to achieve real-time data reading and updating.

[0136] (3) Decision support module: The decision support module is used to provide suggestions for maintenance and management workflows based on the predictive analysis results.

[0137] A rule engine is integrated to provide automated support for bridge maintenance decisions.

[0138] The above system module design aims to integrate the displacement effect analysis method of cable-stayed bridge structure into a complete software and hardware platform to realize the automation of the whole process from data acquisition to predictive analysis, explanatory reporting and decision support.

[0139] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the embodiments shown and described herein.

Claims

1. A method for analyzing displacement effect of a cable-stayed bridge structure, characterized in that: The following steps are involved: S1. Data collection and processing: Collect monitoring data related to the displacement response of the cable-stayed bridge and pre-process the data to obtain a data set; S2, feature engineering: perform spectrum analysis on the time series data of the cable-stayed bridge with timestamp as index, perform feature engineering and time-frequency feature extraction, expand the existing feature number, expand the expanded data into the data set, and randomly divide the data set into training set and test set; S3, model construction: establish a multi-objective output prediction algorithm wrapper suitable for real-time prediction of upstream, downstream, X, Y, and Z directions of cable-stayed bridge structures, use the training set and adopt the basic algorithm as the meta-learning algorithm to construct the structural displacement prediction algorithm chain respectively, and perform decision-making layer fusion on the constructed integrated learning model to obtain the structural displacement prediction model; S4. Model testing: The structural displacement prediction model is tested using the test set, and cross-validation is used to evaluate the model performance; S5. Hyperparameter tuning: The hyperparameters of the structural displacement prediction model are tuned by the grid search method to determine the optimal model parameter configuration; S6. Performance evaluation: Use statistical performance indicators to evaluate the prediction accuracy and generalization ability of the model; S7, Model interpretation and update: Apply Weighted SHAP to interpret the model predictions and identify the contribution of each feature to the prediction results; adjust the contribution weights of the input features according to the relative size of the contribution of each feature, and dynamically update the prediction algorithm; S8. Result verification and application: Apply the trained model to the test set that has not participated in the training to verify the predictive performance of the model.

2. The method for analyzing displacement effect of a cable-stayed bridge structure according to claim 1, characterized in that: The monitoring data include temperature, wind speed, load, and stress; the data preprocessing specifically includes noise reduction, normalization, and denormalization operations.

3. The method for analyzing displacement effect of a cable-stayed bridge structure according to claim 1 or 2, characterized in that: A cable-stayed bridge structure displacement monitoring data denoising algorithm based on joint variational mode decomposition and bidirectional upper short-term memory network is established to denoise the monitoring data.

4. The method for analyzing displacement effect of a cable-stayed bridge structure according to claim 1, characterized in that: The hyperparameters to be tuned include learning rate, number of trees, maximum tree depth, minimum child weight, sample sampling rate, and feature sampling rate. At the same time, for decision layer fusion, the objects of hyperparameter tuning also include basic model weights.

5. A system for analyzing the displacement effect of a cable-stayed bridge structure according to any one of claims 1 to 4, characterized in that: include: Data acquisition module, used to collect real-time data on the cable-stayed bridge’s environmental and operating conditions; A data preprocessing module is used to preprocess the collected raw data; The feature extraction and selection module is used to automatically identify and extract features related to the displacement response of the cable-stayed bridge; including: performing spectrum analysis on the input time series monitoring data, extracting the frequency domain features of the monitoring data, and performing feature construction on the input time series monitoring data; The model training and management module is used for the training configuration management of the cable-stayed bridge structure displacement prediction model, including the initialization and adjustment of hyperparameters, real-time feedback of model training loss and prediction accuracy, and the saving and serialization export of the optimal model checkpoint at the stage; The explanatory analysis module analyzes the qualitative and quantitative relationship between the model prediction results and each input feature time.

6. The system of the cable-stayed bridge structure displacement effect analysis method according to claim 5, characterized in that: The explanatory analysis module is combined with the model training and management module to dynamically adjust the impact weights of input features according to the evaluation results provided by the explanatory analysis, and to update the cable-stayed bridge structure displacement prediction model.

7. The system according to claim 5, characterized in that Also includes: A user interface module provides a user interaction interface, allowing the user to input parameters, start analysis, view results and interpretative analysis reports; the user interface module includes dynamic charts and dashboards for displaying real-time data and forecast results; System integration and communication module, responsible for the coordination of data flow and control flow between modules within the system; Decision support module for providing recommendations for maintenance and management workflows based on predictive analysis results.

Images