Safety evaluation method for vehicles crossing bridges under strong winds based on TCN-BiLSTM-WOA

Through the TCN-BiLSTM-WOA method, a relationship model between non-uniform wind speed and vehicle safety indicators was established, which solved the problem of safety assessment of vehicles passing through bridges under strong winds in mountainous areas, achieved efficient and accurate safety prediction, and improved computing efficiency .

CN118761335BActive Publication Date: 2024-11-05XIHUA UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411252113.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-05
Estimated Expiration
2044-09-09

Smart Images

  • Figure CN118761335B_ABST
    Figure CN118761335B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of vehicle bridge safety evaluation under strong wind environment, and discloses a vehicle bridge safety evaluation method under strong wind based on TCN-BiLSTM-WOA. The spectral function of wind field data is simulated by wind tunnel test, and based on the spatial relationship of different points of the bridge, a linear interpolation method is used to obtain the average wind speed of each longitudinal wind speed simulation point, and the pulsating wind speed of each longitudinal wind speed simulation point is obtained by using wind spectrum and Davenport spatial coherence function; a multi-body dynamics model of the vehicle is established, and the joint simulation and calculation of the wind-vehicle-bridge system are completed to form a training data set for an intelligent prediction method; a TCN-BiLSTM deep learning network is constructed, and a WOA optimization algorithm is built for hyperparameter optimization of the TCN-BiLSTM network, and different vehicle safety index (lateral / vertical acceleration) time series samples are obtained through the TCN-BiLSTM-WOA model. A mapping relationship model between wind speed data and vehicle safety index is established, and prediction is performed efficiently to provide a data basis for the safety evaluation of vehicles crossing bridges under strong winds.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle bridge crossing safety evaluation under strong wind environment, and specifically to a vehicle bridge crossing safety evaluation method under strong wind environment based on TCN-BiLSTM-WOA. Background Art

[0002] The crosswind speed in mountainous areas is generally higher than that in plain areas, which intensifies the dynamic vibration of vehicles passing through bridges and poses a threat to the safety of railway vehicles.

[0003] Railway vehicles will be faster and lighter, which may increase the threat of wind excitation to windmill bridge systems. As railways extend into mountainous terrain, more bridge structures will be built over deep canyons, such as in southwest China. However, the inflow conditions in mountainous areas are more complex, which has an important impact on the dynamic response of bridges and vehicles, especially the safety assessment of vehicles passing through bridges. In general, the wind speed data along the axis of the bridge at the bridge site in coastal plains are basically the same, and the inflow conditions can be assumed to be uniform. In mountainous areas, the wind parameters are variable along the longitudinal direction of the bridge. Therefore, the characteristics of non-uniform wind should be considered in mountainous areas.

[0004] Field measurements, computational fluid dynamics (CFD) simulations, and wind tunnel tests are widely used to obtain wind characteristics. The ease of modeling and control of analytical methods, as well as the sustainability and repeatability of wind tunnel test conditions, make them the main sources for obtaining wind characteristics in mountainous terrain. Most existing studies only consider wind field characteristics or the corresponding response of bridges, and there is little research on the safety of vehicle driving on bridges with non-uniform wind characteristics in mountainous areas.

[0005] At this point, deep learning methods become a good alternative method, which can significantly reduce the computational cost and time cost while maintaining a certain degree of accuracy, and speed up the progress of research. At present, deep learning methods have been applied to non-stationary and non-uniform time series data such as earthquake excitation and wind excitation, and the application results are good. Deep learning methods have good application performance in the fields of earthquake excitation, wind excitation, and vehicle-bridge coupling systems. However, there is currently no attempt to use relevant deep learning methods for the prediction of relevant data on vehicle safety indicators when vehicles cross bridges under strong wind loads (wind loads in mountainous areas). Summary of the invention

[0006] In view of the above problems, the purpose of the present invention is to provide a safety evaluation method for vehicles crossing bridges under strong winds based on TCN-BiLSTM-WOA, which can efficiently predict the relationship between non-uniform wind speed data and vehicle safety index (lateral / vertical acceleration) time series, and provide a reference for safety evaluation of vehicles crossing bridges under strong winds. The technical solution is as follows:

[0007] The method for evaluating the safety of vehicles crossing a bridge under strong winds based on the TCN-BiLSTM-WOA method is characterized by comprising the following steps:

[0008] Step 1: According to the mountainous wind environment, a terrain model is constructed and a D-type wind field is used as the boundary layer for simulation and wind tunnel testing;

[0009] Step 2: Use the experimental average wind speed profile and the D-type wind field to compare and verify the simulated wind field; calculate the wind spectrum of the simulated wind field wind speed data, compare and verify with the Simiu spectrum, and determine that the simulated wind field meets the requirements of the wind tunnel test; obtain the average wind speed distribution relationship of each key point in the longitudinal direction of the bridge through wind tunnel test;

[0010] Step 3: Based on the average wind speed distribution relationship of each key point in the longitudinal direction of the bridge, the linear interpolation method is used to calculate the average wind speed of each key point in the longitudinal direction of the bridge; the fluctuating wind speed of each key point in the longitudinal direction of the bridge is calculated using the wind spectrum and the Davenport spatial coherence function; the sum of the average wind speed and the fluctuating wind speed is the total wind speed, and the total wind speed is used as the wind speed data input in the joint simulation of the wind-vehicle-bridge system;

[0011] Step 4: Establish a multi-body dynamics model of the vehicle, conduct joint simulation and calculation of the wind-vehicle-bridge system, input wind speed data of the simulated wind field, output vehicle safety indicators, including lateral / vertical acceleration, and form a training data set for the intelligent prediction method;

[0012] Step 5: Construct a TCN-BiLSTM deep learning network. Use TCN to extract the time series features of the input wind speed data. BiLSTM predicts the lateral / vertical acceleration of the vehicle in the wind-vehicle-bridge system and the vertical and lateral displacement responses of the bridge based on the extracted time series features. Build a separate WOA optimization algorithm to optimize the hyperparameters of the TCN-BiLSTM network. First, define the hyperparameter space as a population, and each individual is a set of hyperparameter combinations. Then, WOA initializes the hyperparameter space, evaluates the fitness of each hyperparameter combination, updates the population, and finally obtains the optimal hyperparameter combination.

[0013] Step 6: According to the optimal hyperparameter combination found by the WOA optimization algorithm, adjust the hyperparameters of the TCN-BiLSTM network, re-perform prediction training, and save the trained TCN-BiLSTM-WOA model;

[0014] Step 7: Predict the vehicle safety index time series through the TCN-BiLSTM-WOA model.

[0015] Furthermore, the step 1 comprises:

[0016] Step 1.1: Construct a terrain model at a set scale; use the TFI Cobra probe to measure the time-varying wind speed history within the set test time, and use the mutual relationship of the average values ​​of each point in the experimental data as the relationship between the average wind speeds at each key point in the longitudinal direction; the terrain model is made of high-density foam boards stacked according to the actual terrain contours;

[0017] Step 1.2: According to the wind resistance specification, use the D-type wind field as the boundary layer.

[0018] Furthermore, the total wind speed in step 3 is the sum of the average wind speed and the pulsating wind speed, and the specific steps include:

[0019] Step 3.1: Use the linear interpolation method to calculate the average wind speed at each key point in the longitudinal direction of the bridge;

[0020] Step 3.2: The Davenport spatial coherence function is used to calculate the fluctuating wind speed at each key point in the longitudinal direction of the bridge. The Davenport spatial coherence function is expressed as:

[0021]

[0022] in, for j Point and m The spatial coherence function expression of the point; C y is the dimensionless reduction factor; y j and y m They are j Point and m The vertical position of the point; z for y The vertical height of the point; is the average wind speed; is the wind speed frequency.

[0023] Furthermore, the step 4 specifically includes:

[0024] Step 4.1: Build a multibody dynamics model of the vehicle

[0025] Determine the number of degrees of freedom of a single vehicle model including the vehicle body, each bogie and each wheelset, define the stiffness and damping of the suspension system, calculate the creep force based on the single point contact model using the simplified rolling contact theory, and establish the multi-body dynamics model of the vehicle using SIMPACK software; and use the maximum residual acceleration of the vehicle-bridge coupling system to determine the correction of the multi-body dynamics model of the vehicle;

[0026] Step 4.2: According to the safety index of the vehicle crossing the bridge, the lateral / vertical acceleration of the vehicle body is selected as the prediction target;

[0027] Step 4.3: Perform joint simulation and calculation of wind-vehicle-bridge system

[0028] A finite element model of the bridge is established to simulate the vehicle-bridge coupling system. Wind excitation is used as an external excitation, and SIMPACK and ANSYS software are used to simulate the entire time course of vehicles passing through the bridge in the wind-vehicle-bridge system, forming a training data set for safety prediction.

[0029] Furthermore, in step 5, the WOA optimization algorithm is built separately for optimizing the hyperparameters of the TCN-BiLSTM network, specifically including:

[0030] Step 5.1: Define the hyperparameter space of the TCN-BiLSTM model, that is, the upper and lower limits of the hyperparameter values;

[0031] Step 5.2: Initialize the whale group: randomly generate num_whales whales, each of which corresponds to a set of hyperparameter configurations of the TCN-BiLSTM model;

[0032] Step 5.3: Evaluate whales: Build a TCN-BiLSTM model for each whale's hyperparameter configuration, train the model on the training set, and train each TCN-BiLSTM model for 10 epochs; use the root mean square error as the fitness function to calculate the fitness of each whale, that is, the root mean square error calculated by each trained TCN-BiLSTM model on the validation set; record the current best whale and its fitness;

[0033] Step 5.4: Update the position of the whale: According to the update formula of the WOA algorithm, update the position of each whale, that is, update the new hyperparameter configuration;

[0034] Step 5.5: Repeat steps 5.3 and 5.4 until the fitness reaches convergence, or until the maximum number of iterations limited by the code is reached;

[0035] Step 5.6: Use the hyperparameter configuration with optimal fitness to build the TCN-BiLSTM model, and train the model until iterative convergence. The training result at this time is the optimal result of the training.

[0036] Compared with the prior art, the beneficial effects of the present invention include:

[0037] (1) This paper first used the TCN-BiLSTM-WOA method to predict the safety index of vehicles crossing bridges in a strong wind environment in mountainous areas, and verified the applicability of this method in this field.

[0038] (2) Compared with the traditional SIMPACK software simulation and finite element analysis method, the computational efficiency of the proposed method can be improved by nearly 6 times and 10 times, respectively.

[0039] (3) The present invention adds a WOA optimization algorithm to further improve the prediction accuracy of the TCN-BiLSTM model. By establishing a TCN-BiLSTM deep learning neural network model based on the whale optimization algorithm (WOA), a relationship model between non-uniform wind speed data and vehicle safety index (lateral / vertical acceleration) time series is established, and efficient prediction is performed, providing a reference for the safety assessment of vehicles crossing bridges under strong winds. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is the wind speed profile of type D wind field.

[0041] Figure 2 is the turbulence profile of the D-type wind field.

[0042] Figure 3 The multibody dynamics model of the vehicle.

[0043] Figure 4 Loss curve for TCN-BiLSTM model and TCN-BiLSTM-WOA training.

[0044] Figure 5 (a) shows the comparison between the predicted and actual values ​​of the vehicle acceleration time history - lateral acceleration.

[0045] Figure 5 (b) shows the comparison between the predicted and actual values ​​of the vehicle acceleration time history - vertical acceleration. DETAILED DESCRIPTION

[0046] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] A method for evaluating the safety of vehicles crossing a bridge under strong winds based on a TCN-BiLSTM-WOA algorithm comprises the following steps:

[0048] Step 1: According to the wind environment in the mountainous area under study, a terrain model with a scale of 1:2000 is constructed. According to the wind resistance specification for bridges, a D-type wind field is used as the boundary layer for simulation and wind tunnel tests are conducted.

[0049] Step 1.1: The wind environment at the bridge site was studied using wind tunnel tests. Considering the availability of numerical data and test sections, a simulation model was constructed at a scale of 1:2000.

[0050] Assume that the river is at the bottom of the model. The time-varying wind speed time history is measured with a TFI Cobra probe over a test period of 120 seconds, and the average wind speed is the average of the experimental data. The terrain model is made of high-density foam boards with thicknesses of 10 mm and 5 mm (corresponding to the actual height difference of 20 m and 10 m) stacked according to the actual terrain contours. In this embodiment, the core purpose of the terrain model wind tunnel test is to obtain the non-uniform characteristics of the wind at the bridge site, which is only related to the terrain.

[0051] Step 1.2: The bridge is located at the bottom of a mountain ridge, and the wind tunnel test uses a D-type wind field as the boundary layer. The D-type wind field is derived from the bridge wind resistance specification.

[0052] Step 2: Use the experimental average wind speed profile and theoretical fitting results to verify the simulated wind field. Calculate the wind spectrum of the simulated wind field wind speed data and compare and verify it with the Simiu spectrum to ensure that the simulated wind field meets the requirements of the wind tunnel test.

[0053] Step 2.1: In the wind tunnel test, use the experimental average wind speed profile and fitting results to verify the simulated wind field. Figure 1 As shown in the figure, the comparison shows that the experimental data is consistent with the theoretical fitting results. The fitting value of the profile coefficient at the entire height is 0.29, and the theoretical fitting value is 0.30. The experimental data and the required turbulence profile are shown in Figure 2 As shown in the figure, the measured turbulence profile at the bridge deck height meets the requirements of the theoretical data. The average turbulence degree at the bridge deck height is 17.9%, and the data in the specification is 18%. Therefore, the simulation of the D-type wind field is verified.

[0054] Step 2.2: Compare and verify the simulated wind spectrum with the lateral Simiu spectrum to ensure that the experimental data is basically consistent with the wind spectrum. Therefore, the simulated wind field meets the requirements of the wind tunnel test. When the simulation point is close to the ground, the average wind speed and turbulence intensity change with the increase of height. When the height between the simulation point and the ground reaches a certain level, the speed and intensity almost become a constant value, which means that the average wind speed and turbulence intensity can reach the uniform flow field condition. The Simiu spectrum is expressed as:

[0055] ;

[0056] ;

[0057] in, S u ( n ) is the function of the Simiu spectrum, n is the frequency, f is the price reduction frequency, is the friction speed, U ( z ) is the height ZThe average wind speed at the location.

[0058] Step 2.3: Through wind tunnel test, obtain the average wind speed distribution relationship of each key point in the longitudinal direction of the bridge.

[0059] Step 3: Before the wind-vehicle-bridge system joint simulation analysis, a wind field simulation is required. Based on the above-mentioned average wind speed distribution relationship of each key point in the longitudinal direction of the bridge, the average wind speed of each wind speed simulation point in the longitudinal direction of the bridge is calculated by linear interpolation method; the fluctuating wind speed of each wind speed simulation point in the longitudinal direction of the bridge is calculated by wind spectrum and Davenport spatial coherence function. The sum of the average wind speed and the fluctuating wind speed is the total wind speed, which is used as the wind speed data input in the wind-vehicle-bridge joint simulation.

[0060] Step 3.1: Calculate the average wind speed at each key point in the longitudinal direction of the bridge based on the linear interpolation method.

[0061] Step 3.2: Use the Davenport spatial coherence function to calculate the fluctuating wind speed at each key point in the longitudinal direction of the bridge. The Davenport spatial coherence function can be expressed as:

[0062] ;

[0063] in, for j and m The spatial coherence function expression of the point; C y is the dimensionless reduction factor; y j and y m They are j Point and m The vertical position of the point; z for y The vertical height of the point, in this case the beam height, is the average wind speed; is the wind speed frequency.

[0064] The wind speed at 69 equally spaced points on the bridge deck was studied by fast Fourier transform, simulating the wind field of the wind-vehicle-bridge system. The average wind speed was 20m / s, the upper frequency limit was 2Hz, and the time step was 0.25s. The angle of 145.9° clockwise from the north direction (perpendicular to the local ridge) in the mountainous area was selected for wind field simulation because the wind speed of the southward airflow in this embodiment has a large non-uniform characteristic.

[0065] Step 4: Use SIMPACK and ANSYS software to build a multi-body dynamics model of the vehicle, perform joint simulation and calculation of the wind-vehicle-bridge system, input simulated wind speed data, output the lateral and vertical acceleration of the vehicle, and form a training data set for the intelligent prediction method.

[0066] Traditional wind-vehicle-bridge system analysis is mostly achieved by deriving the system motion equations and compiling programs. During the modeling process, researchers usually simplify the model according to their own focus, which makes it more difficult for other researchers to repeat or verify the calculation results. The vehicle-bridge interaction is a coupled dynamic process, which includes a flexible bridge system and a multi-body vehicle system. In view of the rapid development of commercial software and the substantial improvement of computer performance, it is possible to combine multi-body system dynamics software and finite element software to perform wind-vehicle-bridge system analysis.

[0067] Step 4.1: Build a multibody dynamics model of the vehicle

[0068] The vehicle is a complex multi-degree-of-freedom spatial vibration system, and its vibration is divided into lateral motion, longitudinal motion, vertical motion, yaw motion, pitch motion and roll motion. High-speed vehicles are basically composed of a car body, a bogie and a wheelset. The wheelset is connected to the bogie through the primary suspension system, and the bogie is connected to the car body through the secondary suspension system, which includes air springs, lateral and vertical shock absorbers, anti-snaking shock absorbers and lateral stops. The springs and dampers that provide stiffness and damping are defined as a spring-damper system. In addition, the car body, bogie and wheelset are regarded as rigid bodies, and the center of mass of each object is assumed to be bilaterally symmetrical. According to the basic theory of multi-body dynamics, a multi-body model of the vehicle is established. The process is more efficient and can avoid setting up and solving the complex D'Alembert equations of the vehicle.

[0069] The vehicle body and each bogie established in this example have 6 degrees of freedom, and each wheel set has 4 degrees of freedom. Therefore, a single vehicle model has a total of 34 degrees of freedom. According to the actual situation, the stiffness and damping of the suspension system are defined as linear or nonlinear. According to the single point contact model, the simplified rolling contact theory (FASTSIM) is used to calculate the creep force. The multi-body dynamics model of the vehicle established by SIMPACK software is as follows Figure 3 shown.

[0070] The maximum residual acceleration (MRA) of the system is used to determine the rationality of the multi-body vehicle dynamics model. If the MRA of the vehicle is less than 0.01 m / s 2 , the system indicates that it is stable in the initial state. Otherwise, the established model has errors and needs to be rechecked and modified. The calculated MRA of the currently established vehicle model is only 1.52×10 -3m / s 2 , indicating the effectiveness of the model.

[0071] Step 4.2: According to the high-speed railway design specifications and the safety indicators of vehicles crossing bridges in domestic and foreign automobile papers, the lateral / vertical acceleration of the vehicle body is selected as the prediction target.

[0072] Step 4.3: Perform joint simulation and calculation of wind-vehicle-bridge system

[0073] The finite element model of the bridge was established and interpreted as a multi-body dynamic analysis framework of a flexible body, and the vehicle-bridge coupling system was simulated. The wind load was used as an external excitation to simulate the full time course of the vehicle passing through the bridge. According to the actual marshaling situation, the vehicle is arranged in the form of (trailer + 6×motor vehicle + trailer). The German low interference spectrum is used as the track irregularity. The buffeting force is determined based on the quasi-steady-state assumption. The span of the bridge is arranged as (81+162+432+162+81) m, with a total length of 918 m.

[0074] Step 5: Build a TCN-BiLSTM deep learning network and build a separate WOA optimization algorithm to optimize the hyperparameters of the TCN-BiLSTM network and obtain the optimal hyperparameter combination; adjust the hyperparameters of TCN-BiLSTM according to the optimal hyperparameter combination found by the WOA optimization algorithm, re-perform prediction training, and save the trained model; predict the lateral / vertical acceleration time series samples through the above TCN-BiLSTM-WOA model, verify the accuracy of the model, and calculate the mean and standard deviation of the difference with the real data.

[0075] Temporal Convolutional Network (TCN) is used to process time series data. Its core idea is to capture long-term dependencies in time series through convolution operations, and to process time steps in parallel. Long short-term memory neural network (LSTM) is a variant RNN structure that aims to solve the problem of gradient explosion or gradient disappearance in long sequence training. BiLSTM is a long short-term memory network with forward and reverse connections. One processes the input in chronological order, and the other processes the input in reverse chronological order, capturing the features of the input sequence from the forward and reverse directions respectively. The forward LSTM inputs the sequence from left to right according to the time step to calculate the hidden state of each time step. h t and cell status c t The reverse LSTM inputs the sequence from right to left in time steps, and similarly calculates the hidden state of each time step. and cell status Finally, the output of BiLSTM is the concatenation of the hidden states in two directions.

[0076] The Whale Optimization Algorithm (WOA) is a meta-heuristic algorithm that simulates the intelligent hunting process of humpback whales. WOA mainly includes three steps: surrounding prey, attacking with bubble nets, and searching for prey.

[0077] In the prey encirclement phase, the first step is to find the location of the prey. Then, WOA will select the best individual (optimal solution) after calculating the fitness of each search agent, and the other individuals will update their positions according to the given steps to pursue the position of the best individual:

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] in, is the distance between the current position and the best individual position during the process of encircling the prey, t For the current iteration, is the position of the best individual, is the position vector, is the vector that decreases linearly during the iteration, is a random vector in [0,1], and is the coefficient vector, and • represents element-wise multiplication. By updating and The value of the vector can be found at the best individual position obtain different new positions in the neighborhood of X , Y ).

[0083] The bubble net attack phase mainly includes two behaviors, shrinking the encirclement and updating the position in a spiral. When hunting, these two behaviors occur simultaneously because the whale moves toward the surface of the water in a spiral trajectory while shrinking the diameter of the encirclement. In order to simulate this, the algorithm assumes a 50% probability of updating the position through the first or second behavior, which can be expressed as follows:

[0084] ;

[0085] Among them, the simulation formula of spiral motion is:

[0086] ;

[0087] ;

[0088] in, is the update probability, b is a constant that defines the shape of the logarithmic spiral, l is a random number in [-1,1]. is the distance between the current position and the current best individual position during the spiral update process.

[0089] In the prey search phase, the main focus is to explore the search space outside the current best individual position in order to find a better individual position. Therefore, WOA has global search capabilities. This is done by updating the position through a random vector ( X , Y ) to achieve:

[0090] ;

[0091] ;

[0092] in, is the current random position vector.

[0093] In the present invention, the TCN-BiLSTM model is first built. TCN is considered to be better than RNN or even GRU in the ability to extract time series features. Therefore, this part is mainly responsible for extracting the time series features of the input wind speed data, that is, inputting wind speed data into TCN and outputting the corresponding time series features. BiLSTM predicts the lateral / vertical acceleration of the vehicle in the wind-vehicle-bridge system based on the extracted time series features, which is used to judge the safety indicators of the vehicle. Finally, the WOA algorithm is written to seek the optimal hyperparameters in TCN and BiLSTM to avoid the subjectivity of the hyperparameter setting so that the model cannot reach the global optimal result.

[0094] In the model of the present invention, each whale in WOA refers to the model configuration of the candidate TCN-BiLSTM, which is also the hyperparameter combination to be optimized. The main hyperparameters include: TCN convolution kernel size (kernel), TCN expansion factor, TCN number of channels (channel), BiLSTM number of hidden units and dropout rate. The prey in WOA represents the TCN-BiLSTM model configuration with the best performance, that is, the optimal hyperparameter combination. The whale group will update its position according to the relative position with the prey (optimal model) in order to find a better model configuration. The specific steps of WOA's TCN-BiLSTM hyperparameter optimization process are as follows:

[0095] Step (1) defines the hyperparameter space of the TCN-BiLSTM model, that is, the upper and lower limits of the values ​​of the above hyperparameters.

[0096] Step (2) Initialize the whale group: randomly generate num_whales whales, each of which corresponds to a set of hyperparameter configurations of the TCN-BiLSTM model.

[0097] Step (3) Evaluate whales: A TCN-BiLSTM model is constructed for each whale's hyperparameter configuration, and the model is trained on the training set. Each TCN-BiLSTM model is trained for 10 epochs. RMSE (Root Mean Square Error) is used as the fitness function to calculate the fitness of each whale, that is, the RMSE calculated on the validation set for each trained TCN-BiLSTM model. The current best whale and its fitness are recorded.

[0098] Step (4) Update the position of the whale: According to the update formula of the WOA algorithm, update the position of each whale, that is, update the new hyperparameter configuration.

[0099] Step (5) Repeat steps (3) and (4) until the fitness reaches convergence, or until the maximum number of iterations limited by the code is reached.

[0100] Step (6) constructs the TCN-BiLSTM model using the hyperparameter configuration with the best fitness, and trains the model until iterative convergence. The training result at this time is the optimal result of the training.

[0101] SIMPACK and ANSYS software were used to perform joint simulation and calculation of the wind-vehicle-bridge system, and a data set for the intelligent prediction method was formed. The data set was split in a ratio of 0.8:0.1:0.1, and 80% of the data was randomly selected as the training set, 10% as the test set, and 10% as the validation set. RMSE was selected as the loss function. Figure 4The figure shows the loss curves of the TCN-BiLSTM model and TCN-BiLSTM-WOA training. Obviously, after finding the optimal hyperparameters through WOA, the TCN-BiLSTM model can converge faster, and the loss value reached after the final convergence is significantly smaller. Figure 5 (a), Figure 5 (b) and Table 1 show the comparison between the predicted values ​​and the true values ​​of the two algorithms. After adding the WOA algorithm to optimize the hyperparameters of the TCN-BiLSTM model, the mean and variance of the mean absolute error (MAE Mean Absolute Error) and the mean square error (MSE Mean Square Error) are significantly smaller. Therefore, this also means that after adding the WOA algorithm, better accuracy can be achieved without a large number of iterations, and the computational efficiency can be greatly improved. In addition, the computer used for calculation is configured with i7-8750H CPU and 64G running memory. The calculation time of this method is about 1-2 minutes per case, the finite element analysis of Newmark-β is about 12-14 minutes per case, and the calculation time of SIMPACK and ANSYS software joint simulation is about 6-7 minutes per case. Therefore, compared with the traditional SIMPACK software simulation and finite element analysis method, the calculation efficiency of the method of the present invention can be improved by nearly 6 times and 10 times respectively.

[0102] Table 1 Acceleration loss values

[0103] .

[0104] In summary, the TCN-BiLSTM-WOA model proposed in the present invention is intended to predict the safety indicators of vehicle systems in strong wind environments. The model uses the TCN module to effectively extract features from the input data, and uses the BiLSTM module to predict and output the lateral / vertical acceleration of the vehicle, and uses the WOA algorithm to optimize the hyperparameters of the TCN-BiLSTM model to obtain the optimal hyperparameters. Through actual calculation examples, it is found that the results predicted by the TCN-BiLSTM model have reached the engineering allowable range, but the addition of the WOA algorithm enables the prediction model to achieve better accuracy while greatly improving the computational efficiency, so that it spends fewer iterations and saves training time.

Claims

1. A vehicle bridge safety evaluation method based on TCN-BiLSTM-WOA under strong winds is characterized by: The following steps are involved: Step 1: According to the mountainous wind environment, a terrain model is constructed and a D-type wind field is used as the boundary layer for simulation and wind tunnel testing; Step 2: Use the experimental average wind speed profile and the D-type wind field to compare and verify the simulated wind field; calculate the wind spectrum of the simulated wind field wind speed data, compare and verify with the Simiu spectrum, and determine that the simulated wind field meets the requirements of the wind tunnel test; obtain the average wind speed distribution relationship of each key point in the longitudinal direction of the bridge through wind tunnel test; Step 3: Based on the average wind speed distribution relationship of each key point in the longitudinal direction of the bridge, the linear interpolation method is used to calculate the average wind speed of each key point in the longitudinal direction of the bridge; the fluctuating wind speed of each key point in the longitudinal direction of the bridge is calculated using the wind spectrum and the Davenport spatial coherence function; the sum of the average wind speed and the fluctuating wind speed is the total wind speed, and the total wind speed is used as the wind speed data input in the joint simulation of the wind-vehicle-bridge system; Step 4: Establish a multi-body dynamics model of the vehicle, conduct joint simulation and calculation of the wind-vehicle-bridge system, input wind speed data of the simulated wind field, output vehicle safety indicators, including lateral / vertical acceleration, and form a training data set for the intelligent prediction method; Step 5: Construct a TCN-BiLSTM deep learning network. Use TCN to extract the time series features of the input wind speed data. BiLSTM predicts the lateral / vertical acceleration of the vehicle in the wind-vehicle-bridge system and the vertical and lateral displacement responses of the bridge based on the extracted time series features. Build a separate WOA optimization algorithm to optimize the hyperparameters of the TCN-BiLSTM network. First, define the hyperparameter space as a population, and each individual is a set of hyperparameter combinations. Then, WOA initializes the hyperparameter space, evaluates the fitness of each hyperparameter combination, updates the population, and finally obtains the optimal hyperparameter combination. Step 6: According to the optimal hyperparameter combination found by the WOA optimization algorithm, adjust the hyperparameters of the TCN-BiLSTM network, re-perform prediction training, and save the trained TCN-BiLSTM-WOA model; Step 7: Predict the vehicle safety index time series through the TCN-BiLSTM-WOA model.

2. The method for evaluating the safety of vehicles crossing a bridge under strong winds based on TCN-BiLSTM-WOA according to claim 1 is characterized in that: The step 1 comprises: Step 1.1: Construct a terrain model at a set scale; use the TFI Cobra probe to measure the time-varying wind speed history within the set test time, and use the mutual relationship of the average values ​​of each point in the experimental data as the relationship between the average wind speeds at each key point in the longitudinal direction; the terrain model is made of high-density foam boards stacked according to the actual terrain contours; Step 1.2: According to the wind resistance specification, use the D-type wind field as the boundary layer.

3. The method for evaluating the safety of vehicles crossing a bridge under strong winds based on TCN-BiLSTM-WOA according to claim 1 is characterized in that: The sum of the average wind speed and the pulsating wind speed in step 3 is the total wind speed, and the specific steps include: Step 3.1: Use the linear interpolation method to calculate the average wind speed at each key point in the longitudinal direction of the bridge; Step 3.2: The Davenport spatial coherence function is used to calculate the fluctuating wind speed at each key point in the longitudinal direction of the bridge. The Davenport spatial coherence function is expressed as: ; in, for j Dot and m The spatial coherence function expression of the point; C y is the dimensionless reduction factor; y j and y m They are j Dot and m The vertical position of the point; z for y The vertical height of the point; is the average wind speed; is the wind speed frequency.

4. The method for evaluating the safety of vehicles crossing a bridge under strong winds based on TCN-BiLSTM-WOA according to claim 1 is characterized in that: The step 4 specifically includes: Step 4.1: Build a multibody dynamics model of the vehicle Determine the number of degrees of freedom of a single vehicle model including the vehicle body, each bogie and each wheelset, define the stiffness and damping of the suspension system, calculate the creep force based on the single point contact model using the simplified rolling contact theory, and establish the multi-body dynamics model of the vehicle using SIMPACK software; and use the maximum residual acceleration of the vehicle-bridge coupling system to determine the correction of the multi-body dynamics model of the vehicle; Step 4.2: According to the safety index of the vehicle crossing the bridge, the lateral / vertical acceleration of the vehicle body is selected as the prediction target; Step 4.3: Perform joint simulation and calculation of wind-vehicle-bridge system A finite element model of the bridge is established to simulate the vehicle-bridge coupling system. Wind excitation is used as an external excitation, and SIMPACK and ANSYS software are used to simulate the entire time course of vehicles passing through the bridge in the wind-vehicle-bridge system, forming a training data set for safety prediction.

5. The method for evaluating the safety of vehicles crossing a bridge under strong winds based on TCN-BiLSTM-WOA according to claim 1 is characterized in that: In step 5, the WOA optimization algorithm is built separately for optimizing the hyperparameters of the TCN-BiLSTM network, specifically including: Step 5.1: Define the hyperparameter space of the TCN-BiLSTM model, that is, the upper and lower limits of the hyperparameter values; Step 5.2: Initialize the whale group: randomly generate num_whales whales, each of which corresponds to a set of hyperparameter configurations of the TCN-BiLSTM model; Step 5.3: Evaluate whales: Build a TCN-BiLSTM model for each whale's hyperparameter configuration, train the model on the training set, and train each TCN-BiLSTM model for 10 epochs; use the root mean square error as the fitness function to calculate the fitness of each whale, that is, the root mean square error calculated by each trained TCN-BiLSTM model on the validation set; record the current best whale and its fitness; Step 5.4: Update the position of the whale: According to the update formula of the WOA algorithm, update the position of each whale, that is, update the new hyperparameter configuration; Step 5.5: Repeat steps 5.3 and 5.4 until the fitness reaches convergence, or until the maximum number of iterations limited by the code is reached; Step 5.6: Use the hyperparameter configuration with the optimal fitness to build the TCN-BiLSTM model, and train the model until iterative convergence. The training result at this time is the optimal result of the training.

Citation Information

Patent Citations

  • Combined deep learning method for solving wind-induced vibration response of uncertain structure

    CN115577436A

  • Bridge damage identification method considering uncertainty

    US11709979B1