Bridge deflection hybrid intelligent prediction system and method
By integrating the vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model, the problem of vehicle-bridge nonlinear coupling in traditional bridge deflection prediction is solved, efficient and accurate bridge deflection prediction is achieved, and the credibility of the project is improved.
Patent Information
- Application Number
- CN202510792532.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-13
- Publication Date
- 2025-09-30
AI Technical Summary
Traditional bridge deflection prediction models have difficulty dealing with the nonlinear coupling between vehicles and bridges, and purely data-driven methods are prone to produce non-physical interpretations when data is sparse, resulting in insufficient engineering credibility.
The vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model are integrated. The deflection response caused by vehicle load is obtained through successive variational modal decomposition. An analytical model is established by combining the Euler-Bernoulli beam theory and Hertz contact theory. The Kepler planet optimization algorithm is used to optimize the model hyperparameters and construct a hybrid prediction system.
It improves the global accuracy and optimization efficiency of bridge deflection prediction, reduces the deviation of data-driven results, realizes efficient prediction of full-bridge deflection, and enhances engineering credibility.
Smart Images
Figure CN120724070A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bridge monitoring, and in particular to a hybrid intelligent prediction system and method for bridge deflection that integrates a vehicle-bridge coupled vibration chaotic response model and a KOA-BiLSTM model. Background Art
[0002] During the service life of bridges, due to the coupling of adverse factors such as environmental erosion, material aging, long-term effects of loads, fatigue effects, and mutation effects, various bridge safety accidents occur frequently, such as bridge collapse, pier subsidence, structural instability, etc., which seriously affect the safety of people's lives and property.
[0003] With the acceleration of urbanization, bridges, as crucial infrastructure for urban transportation, face significant challenges in safety and stability, impacting the lives and property of residents and the smooth operation of cities. However, traditional bridge deflection prediction methods, often based on physical models like finite element models, rely on precise structural parameters and struggle to account for nonlinear vehicle-bridge coupling. Furthermore, purely data-driven methods lack physical constraints and are prone to producing non-physical interpretations when data is sparse, leading to insufficient engineering credibility.
[0004] In recent years, with the rapid development of technologies such as the Internet of Things, big data, and cloud computing, bridge monitoring technology has also ushered in new changes. Among them, bridge deflection prediction models based on long short-term memory networks have received widespread attention. However, current deflection prediction models in the bridge monitoring field lack efficient optimization mechanisms and sometimes rely entirely on data-driven methods, resulting in biased results. To address these issues, this paper proposes a hybrid intelligent bridge deflection prediction system and method that integrates a vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model. Summary of the Invention
[0005] The purpose of the present invention is to provide a hybrid intelligent prediction system and method for bridge deflection that integrates a vehicle-bridge coupled vibration chaotic response model and a KOA-BiLSTM model to solve the problems raised in the background technology. The present invention can simultaneously consider the multi-point deflection prediction of the bridge and improve the global accuracy of the prediction through information sharing.
[0006] To achieve the above object, the present invention adopts the following technical solutions:
[0007] A hybrid intelligent prediction system for bridge deflection includes the following modules:
[0008] The first module is used to obtain the vehicle-induced deflection response, perform successive variational modal decomposition on the original acceleration time series signal, and deduce the deflection response caused by vehicle loads in the bridge structure;
[0009] The second module is used to establish a chaotic response model of vehicle-bridge coupled vibration using analytical methods, integrating bridge load information and structural parameters to obtain the predicted initial deflection;
[0010] The third module is used to construct a vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network hybrid model, optimize the model hyperparameters using the Kepler planetary optimization algorithm, and predict the deflection of small and medium-span bridges based on the results of successive variational modal decomposition.
[0011] A hybrid intelligent prediction method for bridge deflection includes the following steps:
[0012] S1. Obtain vehicle-induced deflection response, perform successive variational modal decomposition on the original acceleration time series signal, and deduce the deflection response caused by vehicle loads in the bridge structure;
[0013] S2. Use analytical methods to establish a chaotic response model of vehicle-bridge coupled vibration, integrate bridge load information and structural parameters, and obtain the predicted initial deflection;
[0014] S3. Construct a vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network hybrid model, optimize the model hyperparameters using the Kepler planetary optimization algorithm, and predict the deflection of small and medium-span bridges based on the results of successive variational modal decomposition.
[0015] Preferably, the S1 specifically includes the following contents:
[0016] Based on the time-varying characteristics of the original acceleration time series signal, variational modal decomposition is used to decompose and filter the original acceleration time series signal. The retained modal function is reconstructed and then subjected to variational modal decomposition again to obtain the vehicle-induced acceleration time series signal. The deflection response caused by vehicle loads in the bridge structure is deduced from the vehicle-induced acceleration time series signal using the numerical integration method.
[0017] Preferably, S2 specifically includes the following contents:
[0018] According to the actual bridge structure and its basic parameters, a simplified model is constructed using the Euler-Bernoulli beam theory, and the coupling relationship between the tire and the bridge deck is defined by the contact force.
[0019] Based on the changes in external dynamic loads, the free vibration equation and forced vibration equation are established and the boundary conditions of the corresponding environment are set to construct the vehicle vertical motion equation and the vehicle pitch motion equation;
[0020] The nonlinear factors of vehicle suspension system are introduced, and polynomial function is used to express the nonlinear characteristics of suspension system stiffness;
[0021] The nonlinear factors of tire-bridge contact are introduced, and the Hertz contact theory is used to describe the contact stiffness between the tire and the bridge. Meanwhile, the local slip and friction nonlinear factors are considered, and the contact force equation is improved by introducing the friction coefficient and slip condition.
[0022] The bridge dynamic response equation is solved by integrating the bridge load information and structural parameters to obtain the predicted initial deflection.
[0023] Preferably, the S3 specifically includes the following contents:
[0024] The Kepler Planet Optimization Algorithm is used to optimize the model hyperparameters. Based on the results of successive variational mode decomposition, initialization is performed and optimization variables (such as learning rate, number of hidden layer nodes, and number of training rounds) are selected.
[0025] Calculate the fitness value and mechanical parameters and then update the parameter values. When the maximum number of iterations is reached or the fitness value of the solution no longer changes significantly, output the optimal hyperparameter combination.
[0026] A hybrid model of a vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network was constructed. The predicted initial deflection obtained from the vehicle-bridge coupled vibration chaotic response model was used as an additional feature in the input layer of the KOA-BiLSTM model, along with other sensor data.
[0027] The deflection value output by the KOA-BiLSTM model and the predicted initial deflection of the vehicle-bridge coupled vibration chaotic response model are weightedly combined to generate the final predicted deflection.
[0028] Preferably, the prediction method further includes the following:
[0029] The deviation between the final predicted deflection and the deflection predicted by the multi-degree-of-freedom dynamic response equation is added to the loss function of the KOA-BiLSTM model. The final predicted deflection is then substituted into the multi-degree-of-freedom dynamic response equation to verify its conformity and add error constraints to obtain the final loss function.
[0030] When the state of the bridge structure changes (such as aging or load pattern change), the parameters of the vehicle-bridge coupled vibration chaotic response equation and the weights of the KOA-BiLSTM model are dynamically adjusted to achieve adaptive updating of the hybrid model.
[0031] The present invention further protects a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded and executed by the processor to implement the above-mentioned hybrid intelligent prediction method for bridge deflection.
[0032] The present invention further protects a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded and executed by a processor to implement the above-mentioned hybrid intelligent prediction method for bridge deflection.
[0033] Compared with the existing technology, the present invention provides a hybrid intelligent prediction system and method for bridge deflection that integrates the vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model, which has the following beneficial effects:
[0034] (1) Compared with traditional physical models, the hybrid intelligent prediction system for bridge deflection proposed in this invention integrates the vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model. The system uses a deep learning model (KOA-BiLSTM model) to increase the processing speed of sensor data and can predict the deflection of the entire bridge. Compared with a single deep learning model, the system can greatly reduce the deviation of the results caused by data-driven optimization by combining the vehicle-bridge coupled vibration chaotic response model. The Kepler planet optimization algorithm is used to optimize the model hyperparameters, which improves the optimization efficiency and has a strong global optimization capability.
[0035] (2) The solution of the present invention is simple and convenient to implement, and has strong practicality. It solves the problems of low practicality and inconvenience in actual application of related technologies, can improve user experience, and has important market value. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction to the drawings involved in the embodiments is now provided. It is obvious that the drawings described below are only schematic illustrations of some embodiments of the present invention. Those skilled in the art can construct other forms of drawings based on these drawings without inventive effort.
[0037] Figure 1 This is an overall flow chart of the hybrid intelligent prediction method for bridge deflection that integrates the vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model proposed in Example 1 of the present invention;
[0038] Figure 2 Schematic diagram of the vehicle-bridge coupled vibration chaotic response model proposed in Example 1 of the present invention;
[0039] Figure 3 This is a schematic diagram of the Kepler planet optimization algorithm proposed in Example 1 of the present invention;
[0040] Figure 4 Schematic diagram of the structure of the bidirectional long short-term memory network proposed in Example 1 of the present invention. DETAILED DESCRIPTION
[0041] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are merely exemplary and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solutions of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, and such modifications and replacements fall within the scope of protection of the present invention.
[0042] It should be emphasized that unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Although any methods, devices, and materials similar or equivalent to those herein can be used in the practice or testing of the present invention, the preferred methods, devices, and materials are now described.
[0043] The present invention proposes a hybrid intelligent prediction system and method for bridge deflection that integrates a vehicle-bridge coupled vibration chaotic response model and a KOA-BiLSTM model. First, the vehicle-induced deflection response is acquired, and the original acceleration time series signal is subjected to successive variational modal decomposition to deduce the deflection response caused by vehicle loads in the bridge structure. Second, an analytical method is used to establish a vehicle-bridge coupled vibration chaotic response model, and the bridge load information and structural parameters are integrated to obtain the predicted initial deflection. Finally, a hybrid model of the vehicle-bridge coupled vibration chaotic response model and a bidirectional long short-term memory network is constructed. The model hyperparameters are optimized using the Kepler planet optimization algorithm, and the deflection of small and medium-span bridges is predicted based on the results of successive variational modal decomposition.
[0044] During implementation, the data collected by the sensors can be divided into different training, validation, and test sets. A model can then be established to predict bridge deflection, and the accuracy of the prediction model can be assessed by calculating the root mean square error (RMSE), absolute error (AE), and mean absolute error (MAE). The following describes the hybrid intelligent prediction system and method for bridge deflection proposed in this invention, which integrates a vehicle-bridge coupled vibration chaotic response model and the KOA-BiLSTM model, with reference to the accompanying figures and specific examples. The details are as follows.
[0045] Example 1:
[0046] See also Figure 1 This embodiment proposes a hybrid intelligent prediction method for bridge deflection that integrates a vehicle-bridge coupled vibration chaotic response model and a KOA-BiLSTM model. The implementation process includes the following steps:
[0047] Step 1:
[0048] Obtain vehicle-induced deflection response, perform successive variational modal decomposition on the original acceleration time series signal, and deduce the deflection response caused by vehicle loads in the bridge structure;
[0049] The raw acceleration signal data of the bridge is obtained from acceleration sensors installed at key locations on the bridge, the DC offset or long-term trend in the acceleration signal is removed, and the data value is mapped to a specific range (such as [0,1]) for subsequent processing.
[0050] Synchronization constraints are imposed on the original acceleration time series signal to minimize the synchronization optimization objective. The synchronization optimization strategy is combined with the bandwidth constraint to construct the Lagrangian objective function of successive variational mode decomposition.
[0051] Successive variational mode decomposition uses an alternating optimization method to iteratively solve each mode function and the corresponding center frequency;
[0052] Fix the center frequency, solve the modal function through Fourier transform, fix the modal function again, optimize the center frequency, adjust the Lagrange multiplier to constrain the synchronization of the modal components and reconstruct the error;
[0053] When the update amplitude of the modal component is less than the preset threshold or the maximum number of iterations is reached, the optimization is stopped;
[0054] Among the K modal functions obtained by decomposition, the validity of the mode is judged based on the center frequency: high-frequency noise modes (modes with higher frequencies) are removed, and low-frequency modes related to bridge load vibration are retained;
[0055] The acceleration signal is reconstructed using the effective mode function, and finally a signal denoised by successive variational mode decomposition is obtained for subsequent deflection prediction using the BiLSTM model.
[0056] The deflection response of the bridge structure caused by vehicle loads is deduced from the vehicle-induced acceleration time series signal using the numerical integration method.
[0057] Step 2:
[0058] An analytical method is used to establish a chaotic response model of vehicle-bridge coupled vibration, integrating bridge load information and structural parameters to obtain the predicted initial deflection;
[0059] See also Figure 2 The specific steps of the vehicle-bridge coupled vibration chaotic response model are as follows:
[0060] According to the actual bridge structure and its basic parameters, a simplified model is constructed using the Euler-Bernoulli beam theory, and the coupling relationship between the tire and the bridge deck is defined by the contact force.
[0061] Assuming that the tire and the bridge deck are always in contact, the contact force is related to the deformation of the tire and the displacement of the bridge deck, and can be represented by a nonlinear spring-damper model, taking into account the nonlinear changes of the tire's stiffness and damping characteristics with tire deformation;
[0062] Based on the changes in external dynamic loads, the free vibration equation and forced vibration equation are established and the boundary conditions of the corresponding environment are set to construct the vehicle vertical motion equation and the vehicle pitch motion equation;
[0063] According to Newton's second law, for each degree of freedom of the vehicle, a corresponding motion equation can be established.
[0064] Taking the four-degree-of-freedom vehicle model as an example, let the vehicle body mass be m b The front and rear axle masses are m f and m r , the vertical displacement of the vehicle body is z b , the pitch angle is θ, and the vertical displacement of the front and rear axes is z f and z r , the vehicle front suspension stiffness and damping are k sf and c sf , the rear suspension stiffness and damping are k sr and c sr ; Then the vertical motion equation of the vehicle is:
[0065]
[0066] The vehicle pitch motion equation is:
[0067]
[0068] Assume that the bridge is discretized into n beam elements, each beam element has two nodes, and each node has two degrees of freedom: vertical displacement and rotation. Let the mass matrix of the bridge be M b , the stiffness matrix is K b , the damping matrix is C b , the node displacement vector is {u}, the external load vector is {F}, then the dynamic equation of the bridge is:
[0069]
[0070] The nonlinear factors of vehicle suspension system are introduced, and polynomial function is used to express the nonlinear characteristics of suspension system stiffness;
[0071] The nonlinear factors of tire-bridge contact are introduced, and the Hertz contact theory is used to describe the contact stiffness between the tire and the bridge. Meanwhile, the local slip and friction nonlinear factors are considered, and the contact force equation is improved by introducing the friction coefficient and slip condition.
[0072] The bridge dynamic response equation is solved by integrating the bridge load information and structural parameters to obtain the predicted initial deflection.
[0073] Step 3:
[0074] A vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network hybrid model were constructed. The model hyperparameters were optimized using the Kepler planetary optimization algorithm. Based on the results of successive variational modal decomposition, the deflection of small and medium-span bridges was predicted.
[0075] Based on the results of successive variational mode decomposition, initialization processing is performed and optimization variables (such as learning rate, number of hidden layer nodes, and number of training rounds) are selected;
[0076] See also Figure 3 , the specific steps of Kepler planet optimization algorithm are as follows:
[0077] Randomly initialize the initial positions of P candidate planets (i.e., hyperparameter combinations), each position represents a hyperparameter combination:
[0078] x i =[N i ,η i ,E i ]
[0079] Among them, N i is the number of hidden layer nodes, η i is the learning rate, E i is the number of training sessions;
[0080] The value range of each hyperparameter is set based on experience or prior knowledge and distributed in the search space through random generation or Latin hypercube sampling methods;
[0081] For each candidate planet, its corresponding hyperparameter combination x i Apply it to the BiLSTM model for model training.
[0082] The preprocessed raw response data is input into the model, features are extracted through the bidirectional LSTM layer, hidden layer output is obtained, and then the prediction value is generated through the fully connected layer;
[0083] The mean square error (MSE) is used as the loss function to calculate the training error:
[0084]
[0085] Use an optimizer (such as Adam) to adjust model parameters until training is completed or the maximum number of iterations is reached;
[0086] After training is complete, the trained model is applied to the validation dataset and the validation error is calculated:
[0087]
[0088] The inverse of the validation error is used as the fitness value, and the fitness values of all candidate planets are recorded and put into a set.
[0089] Based on the current fitness value, adjust the orbit of the planet (hyperparameter combination). Calculate the gravitational effect based on the relative position of the planet and the star (optimal solution):
[0090]
[0091] Where G is the gravitational constant, M i is the planet mass (positively correlated with the fitness value), r i* is the distance between the planet and the star, e i* is the direction vector.
[0092] Under the influence of gravity, the position of each planet is adjusted according to the following formula:
[0093] x i,t+1 =x i,t +v i,t +αr
[0094] Among them, v i,t is the current speed, and αr is a random perturbation term used to increase the global exploration ability.
[0095] After adjusting the position, the new hyperparameter combination x i Retrain the BiLSTM model, repeat the above training and validation process, and calculate the updated fitness value.
[0096] Check whether the current optimization status meets the following convergence conditions:
[0097] 1. The maximum number of iterations reaches the preset value;
[0098] 2. The fitness value change of all planets is less than a certain threshold;
[0099] If the above convergence conditions are met, it means that the search has become stable.
[0100] If it does not converge, continue to adjust the trajectory and update the fitness; if it converges, terminate the optimization and output the hyperparameter combination with the highest current fitness value.
[0101] The optimal hyperparameter combination x Ж =[N Ж ,η Ж ,E Ж ] was applied to the BiLSTM model, the full data was used for final training, and the prediction performance of the model was verified to obtain the final results.
[0102] The original response data after stepwise variational mode decomposition and denoising is organized into a time series format suitable for BiLSTM model input X input ={x t|t=1, 2, …, T}, of dimension T×F, where T is the number of time steps and F is the feature dimension (data at multiple sensor locations);
[0103] Divide the data into training set, validation set and test set, ensuring that the data in the test set does not participate in the training process to verify the generalization ability of the model. Normalize or standardize the data;
[0104] Perform data inspection on the denoised original response data. If there are missing values, they can be filled by interpolation. If there are outliers, they can be detected and processed by the interquartile range method.
[0105] The Min-Max normalization method is used to calculate the x value of the data feature at each sensor position. min and x max , according to the formula, the original corresponding data of each time step is mapped to [0,1], and only the statistics of the training set are used to normalize the validation set and test set.
[0106] The hyperparameters of the BiLSTM model are optimized using the Kepler Optimization Algorithm (KOA) with the input being the time step sequence X input , the number of hidden layer nodes is N * (Optimized parameters) are output as bidirectional features h t =[h t前 ,h t后 ], with dimensions of T×2N * ;
[0107] See also Figure 4 , the main structure of the BiLSTM model is as follows:
[0108] The input layer receives the original response data of the time series. After the above preprocessing steps, the input data directly enters the BiLSTM network;
[0109] Bidirectional LSTM (BiLSTM) consists of two LSTMs: a forward LSTM and a backward LSTM. The forward LSTM processes data from time step t = 1 to t = T, while the backward LSTM does the opposite. At the same time, the data of each time step is passed to the forward and backward LSTM units in parallel to produce a bidirectional feature representation.
[0110] There are three gating mechanisms inside the LSTM unit, including:
[0111] Input layer, which decides whether new information needs to be added to the cell state:
[0112] i t =σ(W i [h t-1 ,x t ]+b i )
[0113] The forget gate determines which information in the current state needs to be forgotten:
[0114] f t =σ(W f [h t-1 ,x t ]+b f )
[0115] Output gate, which determines what information of the current cell state needs to be output:
[0116] o t =σ(W o [h t-1 ,x t ]+b o )
[0117] In addition, the unit state and hidden state are updated in a timely manner, and the output features of the forward and reverse LSTM are spliced. By stacking multiple bidirectional LSTM layers, the model's ability to extract time series features is enhanced.
[0118] The predicted initial deflection obtained from the vehicle-bridge coupled vibration chaotic response model is used as an additional feature of the input layer of the KOA-BiLSTM model and input into the BiLSTM model together with other sensor data. The sequence is divided using the sliding window technique, with each window being a time step of length L (e.g., 60 time steps).
[0119] The sliding step size can be set to s to generate adjacent windows. The original response data is input into the BiLSTM through each window for prediction. The input tensor dimension is B×L×F, where B is the batch size.
[0120] According to the formula h t ,c t =LSTM(x t ,h t-1 ,c t-1 ) Update the hidden state h at each time step t in the BiLSTM model t and cell state c t ;
[0121] BiLSTM output h t The predicted deflection at each time step is obtained through the fully connected layer:
[0122] y t =Wh t +b
[0123] Where W and b are the weights and biases of the linear layer;
[0124] Integrate deflection predictions at different bridge locations or time periods into a shared layer and use their correlation to improve global prediction accuracy;
[0125] Multi-task learning is achieved through shared layers to capture the correlation between data at different locations or periods of the bridge, and the output is the predicted value;
[0126] The deflection value output by the KOA-BiLSTM model and the predicted initial deflection of the vehicle-bridge coupled vibration chaotic response model are weightedly combined to generate the final predicted deflection;
[0127] The task sharing layer calculates the final deflection by combining the output features of different tasks:
[0128]
[0129] in, is the predicted deflection value of the mth task.
[0130] Combine the predicted values at each time step into a complete curve of bridge deflection versus time;
[0131] Input the original response data of the test set into the model, compare the output deflection prediction value with the actual value, and calculate the evaluation index:
[0132] Mean Squared Error (MSE):
[0133]
[0134] Coefficient of determination:
[0135]
[0136] The deviation between the final predicted deflection and the deflection predicted by the multi-degree-of-freedom dynamic response equation is added to the loss function of the KOA-BiLSTM model. The final predicted deflection is then substituted into the multi-degree-of-freedom dynamic response equation to verify its conformity and add error constraints to obtain the final loss function.
[0137] When the state of the bridge structure changes (such as aging or load pattern change), the parameters of the vehicle-bridge coupled vibration chaotic response equation and the weights of the KOA-BiLSTM model are dynamically adjusted to achieve adaptive updating of the hybrid model.
[0138] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.
Claims
1. A hybrid intelligent prediction system for bridge deflection, characterized in that: Includes the following modules: The first module is used to obtain the vehicle-induced deflection response, perform successive variational modal decomposition on the original acceleration time series signal, and deduce the deflection response caused by vehicle loads in the bridge structure; The second module is used to establish a chaotic response model of vehicle-bridge coupled vibration using analytical methods, integrating bridge load information and structural parameters to obtain the predicted initial deflection; The third module is used to construct a vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network hybrid model, optimize the model hyperparameters using the Kepler planetary optimization algorithm, and predict the deflection of small and medium-span bridges based on the results of successive variational modal decomposition.
2. A hybrid intelligent prediction method for bridge deflection implemented using the system of claim 1, characterized in that: The steps include: S1. Obtain vehicle-induced deflection response, perform successive variational modal decomposition on the original acceleration time series signal, and deduce the deflection response caused by vehicle loads in the bridge structure; S2. Use analytical methods to establish a chaotic response model of vehicle-bridge coupled vibration, integrate bridge load information and structural parameters, and obtain the predicted initial deflection; S3. Construct a vehicle-bridge coupled vibration chaotic response model and a bidirectional long-short-term memory network hybrid model, optimize the model hyperparameters using the Kepler planetary optimization algorithm, and predict the deflection of small and medium-span bridges based on the results of successive variational modal decomposition.
3. The method according to claim 2, characterized in that The S1 specifically includes the following contents: Based on the time-varying characteristics of the original acceleration time series signal, variational modal decomposition is used to decompose and filter the original acceleration time series signal. The retained modal function is reconstructed and then subjected to variational modal decomposition again to obtain the vehicle-induced acceleration time series signal. The deflection response caused by vehicle loads in the bridge structure is deduced from the vehicle-induced acceleration time series signal using the numerical integration method.
4. The method according to claim 3, characterized in that The S2 specifically includes the following contents: Based on the actual bridge structure and its basic parameters, a simplified model was constructed using the Euler-Bernoulli beam theory. The free vibration equation and forced vibration equation were established based on the changes in external dynamic loads, and the boundary conditions of the corresponding environment were set. The bridge dynamic response equation was solved by integrating the bridge load information and structural parameters to obtain the predicted initial deflection.
5. The method according to claim 4, characterized in that The S3 specifically includes the following contents: A vehicle-bridge coupled vibration chaotic response model and a bidirectional long short-term memory network hybrid model were constructed. The model hyperparameters were optimized using the Kepler planetary optimization algorithm. Based on the results of successive variational modal decomposition, the predicted initial deflection output by the vehicle-bridge coupled vibration chaotic response model was used as the feature input into the KOA-BiLSTM model. The output of the KOA-BiLSTM model was then weightedly combined with the output of the bridge's multi-degree-of-freedom dynamic response equation to generate the final predicted deflection.
6. The method according to claim 5, characterized in that The Kepler planet optimization algorithm is used to optimize the model hyperparameters, specifically including the following: Based on the results of successive variational modal decomposition, initialization processing is performed and optimization variables (such as learning rate, number of hidden layer nodes, and number of training rounds) are selected. The fitness value and mechanical parameters are calculated and then the parameter values are updated. When the maximum number of iterations is reached or the fitness value of the solution no longer changes significantly, the optimal hyperparameter combination is output.
7. The method according to claim 6, characterized in that The prediction method further includes the following: After obtaining the final predicted deflection, the deviation between the final predicted deflection and the deflection predicted by the multi-degree-of-freedom dynamic response equation is added to the loss function of the KOA-BiLSTM model. The final predicted deflection is then substituted into the multi-degree-of-freedom dynamic response equation to verify its conformity and add error constraints to obtain the final loss function. When the state of the bridge structure changes, the parameters of the vehicle-bridge coupled vibration chaotic response equation and the weight of the KOA-BiLSTM model are dynamically adjusted to achieve adaptive updating of the hybrid model.
8. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded and executed by the processor to implement the hybrid intelligent prediction method for bridge deflection as described in any one of claims 2-7.
9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded and executed by the processor to implement the hybrid intelligent prediction method for bridge deflection as described in any one of claims 2-7.