Bridge pier impact-resistant structure adaptive design method and system based on water flow simulation

Abstract

This application provides an adaptive design method and system for bridge pier impact-resistant structures based on flow simulation, relating to the field of bridge pier impact resistance technology. The method includes: acquiring flow velocity field, riverbed elevation, and bridge pier geometry and attitude data under extreme flood conditions; constructing a first time-series matrix and extracting second time-series data; inputting these into a CNN-BiLSTM prediction model to evaluate performance and obtain structural performance indicators; then fitting and generating a surrogate model response surface, and optimizing the bridge pier's geometric parameters with preset structural safety and material usage conditions as objectives; finally, inputting the optimization results into a fluid dynamics model based on historical hydrological data for bidirectional fluid-structure interaction simulation verification, and iteratively adjusting until convergence to obtain the final bridge pier structural parameters. This application enables automatic optimization and reliable verification of bridge pier structures under extreme flood conditions.

Description

Technical Field

[0001] This application relates to the field of bridge pier impact resistance technology, and in particular to an adaptive design method and system for bridge pier impact resistance structures based on water flow simulation. Background Technology

[0002] In the field of bridge engineering, improving the impact resistance of bridge piers in floods is the core task of ensuring structural safety, and its optimized design methods have clear engineering value for extending bridge life and reducing disaster risks.

[0003] Currently, in actual engineering design, the consideration of the impact resistance performance of bridge piers mainly relies on empirical formulas and simplified charts in the design specifications. Engineers usually base their decisions on a few static hydrological parameters such as the highest historical water level, combined with standard cross-sectional forms, and then perform static and stability checks by hand calculation or basic software. They also manually compare and select from multiple preset schemes based on experience.

[0004] However, these mainstream methods have obvious shortcomings: First, the static parameters and standard operating conditions they rely on cannot reflect the dynamic impact process of real floods and the continuous changes in riverbed topography, resulting in deviations between load assessment and reality; Second, they rely entirely on engineers' experience to select from a limited number of preset schemes, making it difficult to systematically explore better geometric shapes, and the impact resistance and economy of their design results often fail to achieve the best balance. Summary of the Invention

[0005] The purpose of this application is to provide an adaptive design method and system for bridge pier impact-resistant structures based on water flow simulation, so as to solve the problem in the prior art that it is difficult to efficiently obtain a safe, reliable and economically reasonable bridge pier design scheme under complex and dynamic actual water flow conditions.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides an adaptive design method for bridge pier impact-resistant structures based on water flow simulation, comprising:

[0007] Acquire flow velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as geometric parameters and structural attitude data of existing bridge piers;

[0008] The flow velocity field data, the riverbed elevation data, and the structural attitude data are combined to construct a first time series matrix. The first time series matrix is ​​then subjected to trend smoothing and feature extraction to obtain second time series data.

[0009] The second time series data is input into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators.

[0010] The surrogate model response surface is obtained by fitting the structural performance indices.

[0011] Based on the response surface of the surrogate model, the geometric parameters of the bridge pier are optimized with preset structural conditions and preset material usage conditions as optimization objectives, and the optimized geometric parameters are obtained.

[0012] The optimized geometric parameters are input into a fluid dynamics calculation model based on historical hydrological data for two-way fluid-structure interaction simulation to obtain verification results. If the verification results do not meet the preset convergence conditions, the parameter design of at least one condition in the optimization objective is modified according to the verification results, and the parameter optimization and simulation steps are re-executed until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

[0013] Optionally, fitting the structural performance index to obtain the surrogate model response surface includes:

[0014] An initial sample set is constructed based on the geometric parameters of the bridge piers and the aforementioned structural performance indicators;

[0015] The initial sample set is fitted using a Kriging surrogate model to obtain the initial surrogate model;

[0016] An uncertainty assessment is performed on the initial proxy model to obtain an uncertainty distribution. Based on the uncertainty distribution, an adaptive sampling algorithm is used to identify supplementary sample points.

[0017] The initial sample set is updated based on the supplementary sample points and the corresponding structural performance indicators, and the fitting, evaluation and identification steps are repeated until the preset iteration stopping condition is met, so as to obtain the target sample set and the target proxy model corresponding to the target sample set.

[0018] By combining the hydraulic characteristics corresponding to the target proxy model, the velocity field data, and the riverbed elevation data, the key geometric parameters are obtained.

[0019] Based on the physical constraint values ​​corresponding to the key geometric parameters, the target surrogate model is modified to obtain the surrogate model response surface.

[0020] Optionally, the step of performing uncertainty assessment on the initial proxy model to obtain an uncertainty distribution, and then using an adaptive sampling algorithm to identify supplementary sample points based on the uncertainty distribution, includes:

[0021] Based on the initial proxy model, Monte Carlo simulation prediction is performed on the unsampled space to obtain the uncertainty measure of each prediction point;

[0022] Using spatial interpolation methods, an uncertainty distribution map is constructed based on the uncertainty metric, and the uncertainty distribution map is subjected to trend fitting and identification processing to obtain the extreme value region;

[0023] In the extreme value region, the entropy weight method is applied to perform sensitivity analysis on the geometric parameters of the bridge piers to obtain the contribution of each geometric parameter;

[0024] Select the dominant parameters whose contribution exceeds a preset contribution threshold from all geometric parameters;

[0025] Latin hypercube sampling is performed within the sub-region defined by the dominant parameters to obtain supplementary sample points.

[0026] Optionally, based on the surrogate model response surface, and with preset structural conditions and preset material usage conditions as optimization objectives, the geometric parameters of the bridge pier are optimized to obtain the optimized geometric parameters, including:

[0027] Based on the preset structural conditions and preset material usage conditions, the improved NSGA-III algorithm is used to solve the response surface of the surrogate model and generate a Pareto solution set.

[0028] The Pareto solution set is analyzed to obtain sparse regions and clustered regions;

[0029] Supplementary sampling is performed in the sparse region to obtain the first solution, and the geometric parameters of the bridge pier are filtered by solution set processing in the clustered region to obtain the second solution.

[0030] By combining the first solution and the second solution, an equilibrium solution set is obtained;

[0031] A multi-criteria decision-making method based on the ideal point method is adopted to select the final solution from the equilibrium solution set, and the geometric parameters corresponding to the final solution are used as the optimized geometric parameters.

[0032] Optionally, the supplementary sampling process is performed in the sparse region to obtain a first solution, and the geometric parameters of the bridge pier are subjected to solution set filtering in the clustered region to obtain a second solution, including:

[0033] Obtain the local performance gradient corresponding to the response surface of the proxy model in the sparse region;

[0034] Based on the direction and magnitude of the local performance gradient, deterministic guided sampling is performed in the sparse region to obtain preliminary guided points;

[0035] The initial guiding point is validated based on the geometrically feasible region and spacing constraints to obtain the first solution;

[0036] In the clustered region, the geometric parameters of the bridge pier are subjected to Euclidean distance-based clustering partitioning to obtain multiple sub-clusters. Within each sub-cluster, the material property values ​​and front distance of each geometric parameter are calculated.

[0037] By combining the material performance values ​​and the frontal distance, a ranking index is obtained. Based on the ranking index, non-dominated solutions are selected from each of the subclusters to obtain a second solution.

[0038] Optionally, the step of inputting the second time-series data into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators includes:

[0039] The second time series data is subjected to spatiotemporal decomposition to obtain a time series feature sequence and a spatial feature map;

[0040] The temporal feature sequence is modeled temporally using a bidirectional recurrent network layer in the CNN-BiLSTM prediction model to obtain preliminary temporal features. These preliminary temporal features are then enhanced to generate the first fused features.

[0041] The spatial feature map is spatially modeled using the convolutional network layer in the CNN-BiLSTM prediction model to obtain preliminary spatial features, and the preliminary spatial features are then enhanced to generate a second fused feature.

[0042] By using the attention fusion layer in the CNN-BiLSTM prediction model, the first fusion feature and the second fusion feature are weighted and fused to obtain the spatiotemporal fusion feature;

[0043] The spatiotemporal fusion features are processed by nonlinear transformation using the fully connected network layer in the CNN-BiLSTM prediction model to obtain structural performance indicators.

[0044] Optionally, the step of inputting the optimized geometric parameters into a fluid dynamics calculation model established based on historical hydrological data to perform two-way fluid-structure interaction simulation and obtain verification results includes:

[0045] The simulation boundary conditions of the fluid dynamics calculation model are set based on historical hydrological data;

[0046] The optimized geometric parameters are input into the fluid dynamics calculation model, and a two-way fluid-structure interaction simulation is performed based on the simulation boundary conditions to obtain physical field data, which includes the pressure distribution on the structure surface, the internal stress distribution, and the eddy shedding frequency.

[0047] The physical characteristics corresponding to the physical field data are compared with the structural performance indicators corresponding to the response surface of the surrogate model to obtain the verification results.

[0048] Secondly, this application provides an adaptive design system for bridge pier impact-resistant structures based on water flow simulation, comprising:

[0049] The acquisition module is used to acquire flow velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as geometric parameters and structural attitude data of the existing bridge piers;

[0050] The combination module is used to combine the flow velocity field data, the riverbed elevation data, and the structural attitude data to construct a first time series matrix, and to perform trend smoothing and feature extraction processing on the first time series matrix to obtain second time series data.

[0051] The evaluation module is used to input the second time series data into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators.

[0052] The fitting module is used to fit the structural performance indicators to obtain the surrogate model response surface;

[0053] The optimization module is used to optimize the geometric parameters of the bridge pier based on the response surface of the proxy model, with preset structural conditions and preset material usage conditions as optimization targets, so as to obtain the optimized geometric parameters.

[0054] The verification module is used to input the optimized geometric parameters into a fluid dynamics calculation model based on historical hydrological data to perform two-way fluid-structure interaction simulation and obtain verification results. If the verification results do not meet the preset convergence conditions, the module modifies the parameter design of at least one condition in the optimization objective according to the verification results and re-executes the parameter optimization and simulation steps until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

[0055] Thirdly, this application provides an electronic device, comprising:

[0056] Memory, used to store computer programs;

[0057] A processor is configured to execute the computer program to implement the steps of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation as described in the first aspect above.

[0058] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation as described in the first aspect above.

[0059] The adaptive design method for bridge pier impact-resistant structures based on water flow simulation provided in this application has the following beneficial effects: By acquiring flow velocity, riverbed, and bridge pier data under extreme flood conditions, this application can provide a realistic dynamic environment basis for analysis; furthermore, by combining and extracting the features of these time-series data, it can effectively capture the key correlation between water flow impact and structural response, thus laying the foundation for accurate performance evaluation; subsequently, by using a CNN-BiLSTM model to perform spatiotemporal analysis on the features, it can achieve rapid and accurate prediction of impact resistance performance; then, by fitting performance indicators through a surrogate model to generate a response surface, it can construct an efficient computational model that can replace time-consuming simulations; based on this, the parameters are automatically optimized with the goal of structural safety and material economy, and a more optimal balanced solution can be found in massive designs; finally, the optimization results are verified by high-fidelity fluid-structure interaction simulation and feedback adjustment is performed to form a design closed loop, thereby obtaining a safe, reliable, and economically reasonable bridge pier design scheme according to complex and dynamic actual water flow conditions.

[0060] Furthermore, in constructing the surrogate model, this application enables the model to continuously improve itself in key areas through iterative fitting and adaptive sampling based on uncertainty assessment, thereby improving prediction accuracy. By combining hydraulic features to identify key geometric parameters and correcting the model based on physical constraints, a highly efficient surrogate model that reflects complex physical relationships and conforms to engineering practice is finally obtained, providing a reliable foundation for automated multi-objective optimization. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart illustrating an adaptive design method for bridge pier impact-resistant structures based on water flow simulation, provided in an embodiment of this application;

[0063] Figure 2 A schematic diagram illustrating a specific implementation of an adaptive design method for bridge pier impact-resistant structures based on water flow simulation, provided in this application embodiment;

[0064] Figure 3 A schematic diagram of a bridge pier impact-resistant adaptive design system based on water flow simulation provided in this application embodiment;

[0065] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0066] In the practice of bridge impact resistance design, the traditional method of relying on static parameters and human experience for selection is difficult to accurately simulate the dynamic impact of floods and the complex loads brought about by continuous changes in the riverbed. The design process is inefficient and it is difficult to systematically explore the global optimal balance between the geometry of the bridge piers and the amount of materials used. The safety and economy of the final solution in the real harsh hydrological environment are often uncertain.

[0067] To address this, this application proposes an adaptive design method for bridge pier impact-resistant structures based on water flow simulation. The core idea is to: perform intelligent feature analysis by integrating time-series data of dynamic water flow, riverbed changes, and structural posture to quickly and accurately predict the impact resistance performance of bridge piers; then construct an efficient surrogate model to replace complex simulations and drive a multi-objective optimization algorithm to automatically search for geometric parameter schemes that balance safety and economy; finally, verify the design through high-fidelity fluid-structure interaction simulation and perform closed-loop iterative adjustments. This method can achieve accurate response to extreme flood dynamic processes and autonomous optimization, effectively overcoming the limitations of existing technologies that rely on static assumptions, have low optimization efficiency, and insufficient reliability of results.

[0068] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] The core of this application is to provide an adaptive design method for bridge pier impact-resistant structures based on water flow simulation. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0070] S101. Obtain flow velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as geometric parameters and structural attitude data of existing bridge piers.

[0071] Among them, velocity field data refers to information reflecting the three-dimensional spatial distribution of the magnitude and direction of water flow velocity when floods flow through the river section where the bridge piers are located; riverbed elevation data refers to measurement information describing the topographic undulations and elevation changes of the bottom of the same river section.

[0072] Existing bridge piers refer to the existing bridge pier structures that serve as a design reference or starting point for optimization; structural attitude data refers to information obtained through measurement that reflects the possible offset, tilt, or vibration state of the bridge pier's spatial position compared to its original design position under actual water flow or other loads.

[0073] In step S101, the flow velocity field data of the waters surrounding the bridge pier location is first obtained for the target river through historical hydrological observation records, satellite remote sensing images, or dedicated radar current meters during simulated or actual extreme flood events. At the same time, underwater topographic scanning of the riverbed around the bridge pier foundation is performed using multibeam sonar, or the sedimentation and scouring changes of the riverbed before and after the flood are analyzed to obtain accurate riverbed elevation data.

[0074] Then, for the existing bridge piers used as a reference, their precise shape and dimensions are obtained through three-dimensional laser scanning technology as geometric parameters; at the same time, sensors such as inclinometers and accelerometers are installed at key locations on the bridge piers to record their dynamic response under daily water flow or specific tests, thereby obtaining structural attitude data that reflects their true working state.

[0075] S102. Combine the flow velocity field data, the riverbed elevation data, and the structural attitude data to construct a first time series matrix. Perform trend smoothing and feature extraction processing on the first time series matrix to obtain second time series data.

[0076] The first time series matrix refers to the matrix formed by aligning and integrating the three types of data—velocity field, riverbed elevation, and structural attitude—in a unified time order. In this matrix, time is the core dimension, with each row corresponding to a specific sampling time and each column representing the specific values ​​of different spatial locations or different measurement types at that time.

[0077] In step S102, the acquired flow velocity field, riverbed elevation and structural attitude data are first aligned on the timeline. Since these data may come from sensors of different frequencies or observations at different times, the interpolation algorithm can ensure that the aligned data is valid at the same series of time points.

[0078] Subsequently, all spatial points and all types of measurement data at each aligned time node are arranged into a vector according to their spatial location, and the vectors of all time nodes are sorted according to their data type, thus forming a two-dimensional matrix in which rows represent time and columns represent different variables and spatial locations, namely the first time series matrix.

[0079] Next, for each column of data in the matrix, a moving average method is used to smooth the trend. This process aims to preserve the overall rising and falling trend and periodicity of the sequence, while weakening instantaneous spikes or noise that may obscure the main pattern.

[0080] Finally, feature extraction is performed on each smoothed data sequence, that is, by calculating the statistics of the sequence to summarize its distribution characteristics, or by extracting its energy characteristics in different frequency bands through time-frequency analysis, and then these features are reorganized in chronological order to obtain the second time series data.

[0081] This application not only eliminates irrelevant interference but also improves the information density and interpretability of the data, thereby laying a reliable data foundation for subsequent models to make accurate performance predictions.

[0082] S103. Input the second time series data into the pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators.

[0083] The CNN-BiLSTM prediction model can include a bidirectional recurrent network layer, a convolutional network layer, and an attention fusion layer connected in sequence. The training data of the entire CNN-BiLSTM model comes from historical high-fidelity computational fluid dynamics and finite element analysis simulation results, as well as some physical model experimental data. During training, the mean squared error is used as the loss function, the Adam optimizer is used to update the parameters, and the model learns the complex mapping relationship between the input multi-source time series data and the target performance index through a large amount of data iteration training.

[0084] It should be noted that this embodiment does not impose any limitations on the structural design of the CNN-BiLSTM prediction model, the structural design of each layer, the parameter design, the training process, etc., and can be set accordingly according to the actual situation.

[0085] In one specific implementation, step S103 includes:

[0086] Step 1031: Perform spatiotemporal decomposition processing on the second time series data to obtain a time series feature sequence and a spatial feature map.

[0087] Among them, the temporal feature sequence refers to the data sequence that reflects the dynamic changes of each measurement feature itself with time as the only dimension of change; the spatial feature map refers to the data structure that maps spatial positional relationships into a row and column grid similar to an image, and carries multiple feature values ​​at each grid point.

[0088] In step 1031, firstly, in order to extract the temporal feature sequence, the second temporal data is reshaped along the spatial dimension to generate a two-dimensional matrix. Specifically, while keeping the time dimension unchanged, all feature values ​​of all spatial points at each time point are arranged into a long vector to form a matrix with time as the row and the combination of "space and feature" as the column, which serves as the temporal feature sequence.

[0089] Simultaneously, a representative time segment is selected from the entire time feature sequence, and the statistical values ​​of each feature at each spatial point within that time segment are calculated, such as the mean, to obtain a two-dimensional matrix, where rows represent spatial points and columns represent feature types. Then, based on the predefined mapping relationship between spatial points and two-dimensional grids, the feature vector of each point is placed at the corresponding position on the grid, and for grid positions without point mapping, an interpolation method is used to fill in the grid, ultimately generating a spatial feature map.

[0090] For example, suppose the second time series data contains 100 time steps and monitors 20 points around the bridge pier, and extracts two features, flow velocity and scour depth, for each point; then, during the spatiotemporal decomposition preprocessing, the time series feature sequence is constructed into a 100-row, 40-column matrix. Subsequently, to construct the spatial feature map, the entire time series feature is selected, and the average of the two features for each point over all 100 time steps is calculated, resulting in a 20-row, 2-column matrix.

[0091] Then, based on the real geographic coordinates of the 20 points, they are mapped onto a regular grid of 5 rows and 4 columns. For cells in the grid without original points, bilinear interpolation is used to calculate and fill them based on the feature values ​​of the surrounding points, and finally a 5×4×2 spatial feature map is generated. It should be noted that the explanation and specific implementation process of bilinear interpolation can be referred to relevant technologies, and will not be repeated in the embodiments of this application.

[0092] Step 1032: Perform temporal modeling on the temporal feature sequence through the bidirectional recurrent network layer in the CNN-BiLSTM prediction model to obtain preliminary temporal features, and perform feature enhancement processing on the preliminary temporal features to generate the first fused features.

[0093] In step 1032, the temporal feature sequence is input into the bidirectional recurrent network layer of the CNN-BiLSTM model. This layer consists of two stacked BiLSTM units, each with 128 hidden units. The network then processes the temporal feature sequence simultaneously from both forward and backward directions. Forward modeling captures the dynamic evolution trend of the bridge pier's vibration response and stress accumulation under water flow impact. Backward modeling traces the historical impact and hysteresis effect of water flow impact load. After processing all time steps, the forward hidden state of the last time step is concatenated with the backward hidden state of the first time step to obtain the preliminary temporal feature that encodes the complete sequence context information. This feature reflects the evolution of the bridge pier's vibration response and surface pressure fluctuations over time under water flow impact.

[0094] Subsequently, feature enhancement based on scaled dot product attention is performed on the initial temporal features. Specifically, the initial temporal features are first passed through three independent linear projection layers to generate query vectors, key vectors, and value vectors, respectively. Then, the features are enhanced using the formula... Calculate attention weights Where Q is the query vector, K is the key vector, and V is the value vector. Let T be the dimension of the key vector, and let T represent the transpose operation of the matrix. The normalized exponential function can be expressed with reference to relevant technologies. The attention weight enables the model to adaptively focus on the key time points or flow stages that have the greatest impact on the impact resistance performance of the bridge pier. Finally, the attention weight is residually connected with the original preliminary temporal features and then normalized to generate the first fusion feature. This feature enhances the temporal patterns that are strongly correlated with the impact resistance performance in the dynamic response of the bridge pier.

[0095] Step 1033: Spatial modeling is performed on the spatial feature map through the convolutional network layer in the CNN-BiLSTM prediction model to obtain preliminary spatial features, and feature enhancement processing is performed on the preliminary spatial features to generate the second fused features.

[0096] In step 1033, the spatial feature map is input into the convolutional neural network layer, which contains two convolutional blocks. The first convolutional block uses 32 3×3 convolutional kernels and employs the ReLU activation function, followed by 2×2 max pooling to extract local hydraulic features in the flow field around the pier, such as the high-speed region near the wall and the core region of the separated vortex. The second convolutional block uses 64 3×3 convolutional kernels and also employs the ReLU activation function to further fuse multi-scale spatial information to characterize the overall structure of the water flow around the pier and the spatial distribution pattern of the impact load, thereby obtaining a one-dimensional preliminary spatial feature. This feature reflects the spatial load pattern formed by the interaction between the pier geometry and the surrounding flow field.

[0097] Next, the preliminary spatial features are fed into a feature enhancement layer, which is a two-layer feedforward neural network to form a bottleneck structure. The first layer reduces the dimension of the preliminary spatial features, and the second layer restores them to the original dimension. In the process, the ReLU activation function is used to introduce nonlinearity. Finally, the second fused feature is output. This feature enhances the spatial pattern related to key impact resistance phenomena such as pressure concentration on the bridge pier surface and vortex-induced vibration through nonlinear transformation.

[0098] For example, a 5×4×2 spatial feature map is input into a convolutional network layer, and after passing through the first convolutional layer and 2×2 max pooling, it becomes a 2×2×32 feature map. After passing through the second convolutional layer, the size remains unchanged, and then it is flattened to obtain a preliminary 256-dimensional spatial feature. This feature is then processed through a bottleneck MLP: first, it is linearly mapped from 256 dimensions to 128 dimensions, then activated by ReLU, and finally mapped back from 128 dimensions to 256 dimensions, thus outputting a second fused feature. This process highlights the flow field characteristics of key areas such as the upstream face of the bridge pier and the wake region.

[0099] Step 1034: Through the attention fusion layer in the CNN-BiLSTM prediction model, the first fusion feature and the second fusion feature are weighted and fused to obtain the spatiotemporal fusion feature.

[0100] In step 1034, the first fusion feature and the second fusion feature are input into the attention fusion layer. This layer first projects the two feature vectors to the same dimensional space through independent linear transformations. Then, the two projected feature vectors are concatenated and input into a small feedforward neural network. The network outputs a two-dimensional weight vector, which is then normalized by the softmax function to obtain the weights assigned to the first fusion feature and the second fusion feature. These weights adaptively quantify the relative contributions of temporal dynamic response and spatial load distribution in determining the overall impact resistance of the bridge pier. For example, in the early stage of flood impact, more attention may be paid to spatial pressure distribution, while in the sustained impact stage, more attention may be paid to cumulative vibration response. Finally, the first fusion feature and the second fusion feature are multiplied by their corresponding weights and then added together to obtain the spatiotemporal fusion feature. This feature integrates the dynamic behavior of the bridge pier under water flow impact and the spatial interaction of the surrounding flow field.

[0101] Step 1035: Perform nonlinear transformation on the spatiotemporal fusion features through the fully connected network layer in the CNN-BiLSTM prediction model to obtain structural performance indicators.

[0102] In step 1035, the spatiotemporal fusion features are fed into the output module consisting of three fully connected layers. The first layer contains 128 neurons, the second layer contains 64 neurons, and both use the ReLU activation function. The number of neurons in the last output layer is equal to the number of structural performance indicators to be predicted, such as 3, and they correspond to the maximum equivalent stress, maximum lateral displacement, and fatigue damage index, respectively.

[0103] This application enables rapid and accurate prediction of the complex mechanical response of bridge piers under dynamic water flow impact, thereby providing an efficient performance evaluation basis for optimized design.

[0104] S104. Fit the structural performance index to obtain the surrogate model response surface.

[0105] In one specific implementation, such as Figure 2 As shown, step S104 includes:

[0106] Step 1041: Based on the geometric parameters of the bridge piers and the structural performance indicators, construct an initial sample set.

[0107] In step 1041, the design variables to be optimized, such as the bottom diameter ratio and the curvature of the water-facing surface, need to be determined first based on the geometric parameters of the pier. Then, within the range of these variables, the Latin hypercube design method is used to generate a set of initial design points. For each generated design point, the corresponding structural performance index is calculated by calling the CNN-BiLSTM prediction model. Finally, the variable values ​​of each design point and their corresponding performance index values ​​are combined into a complete data pair to form the initial sample set.

[0108] For example, using the three parameters of the pier's bottom diameter ratio, the radius of curvature of the water-facing surface, and the pier's height-to-diameter ratio as design variables, and using the Latin hypercube design method to generate 50 initial design points within the design space, a CNN-BiLSTM prediction model is run for each design point to obtain its maximum equivalent stress and maximum lateral displacement. The 50 sets of geometric parameters and performance index data are then organized into a table, thus constructing an initial sample set containing 50 samples. It should be noted that the explanation and specific implementation process of the Latin hypercube design method can be referred to relevant technologies, and will not be elaborated in the embodiments of this application.

[0109] Step 1042: Fit the initial sample set using the Kriging surrogate model to obtain the initial surrogate model.

[0110] In step 1042, the geometric parameters in the initial sample set are used as the input matrix, and the structural performance index is used as the output vector. The model is then trained using the Kriging model framework. This model assumes that the output response follows a Gaussian process and uses a covariance function to characterize the correlation between the output responses of different design points. The optimal values ​​of all hyperparameters in the covariance function are obtained by maximizing the marginal likelihood function of the given samples. Once the hyperparameters are learned, an initial surrogate model that can predict any new design point and provide a confidence interval is obtained.

[0111] For example, an initial sample set containing 50 samples is input into the Kriging model, and the input is in the form of: The covariance function, where The covariance function defined in the Kriging proxy model. , Let i and j be the vectors of the geometric parameters of the bridge piers at the i-th and j-th design points. Let be a natural exponential function, and D be the dimension of the design space, i.e., the number of geometric parameters. For signal variance, and Let be the parameter values ​​of the i-th and j-th design points in the d-th dimension. Let d be the length scale of the d-th dimension. For noise variance, The Kronecker function has a value of 1 when i=j and 0 otherwise. The optimal values ​​of all hyperparameters in the covariance function are obtained through optimization. Once the hyperparameters are learned, the initial surrogate model is constructed.

[0112] Step 1043: Perform uncertainty assessment on the initial proxy model to obtain the uncertainty distribution. Based on the uncertainty distribution, use an adaptive sampling algorithm to identify supplementary sample points.

[0113] The process of identification using an adaptive sampling algorithm based on the uncertainty distribution is as described in steps a1 to a5. Therefore, step 1043 may specifically include the following steps:

[0114] Step a1: Based on the initial surrogate model, perform Monte Carlo simulation prediction on the unsampled space to obtain the uncertainty measure of each prediction point.

[0115] In step a1, in order to evaluate the prediction confidence of the initial surrogate model in the entire design space, a large number of uniformly distributed prediction points are first randomly generated in the entire geometric parameter design space. Then, the initial surrogate model is called to make predictions for these prediction points, and the model will output the prediction mean and prediction variance of the performance for each point. The prediction variance is used as a measure of uncertainty for that point. The larger the prediction variance, the more ambiguous the model's understanding of the region and the less reliable the prediction.

[0116] Step a2: Using spatial interpolation, construct an uncertainty distribution map based on the uncertainty metric, and perform trend fitting and identification processing on the uncertainty distribution map to obtain the extreme value region.

[0117] In step a2, all Monte Carlo prediction points and their corresponding uncertainty metrics are treated as discrete spatial data. Then, a spatial interpolation method is used to fit these discrete uncertainty values ​​into a continuous surface covering the entire design space, which serves as an uncertainty distribution map. By analyzing this surface, peak point groups with significantly higher uncertainty values ​​than the surrounding areas are identified. These areas are defined as extreme value regions, representing the weakest points in the model's understanding and the areas most in need of supplementary information. It should be noted that the explanation and specific implementation process of the spatial interpolation method can be found in relevant technologies, and will not be elaborated upon in the embodiments of this application.

[0118] Step a3: In the extreme value region, the entropy weight method is applied to perform sensitivity analysis on the geometric parameters of the bridge pier to obtain the contribution of each geometric parameter.

[0119] Among them, the entropy weight method refers to an objective weighting method based on information entropy; the contribution degree refers to the weight calculated by this method, which represents the magnitude of the influence of the corresponding parameter on the current high uncertainty state.

[0120] In step a3, several sample points are extracted from each extreme value region, and the entropy weight method is applied to analyze the geometric parameters of all sample points in the region. This method evaluates the degree of variation by calculating the information entropy of the value sequence of each parameter. The smaller the information entropy, the more concentrated the value of the parameter is in the region, and the stronger its contribution or distinguishing ability to the overall uncertainty pattern. Then, the information entropy of each parameter is normalized to obtain a set of weights as the contribution of each geometric parameter.

[0121] For example, if m sample points are drawn from an extreme region, and each point has n geometric parameters, the parameter values ​​are first normalized: ,in, It is the j-th parameter value of the i-th sample. It is its normalized value; then calculate the information entropy of the j-th parameter: ,in, It refers to the information entropy of the j-th parameter. The smaller the entropy value, the more concentrated the value of the parameter is and the smaller the variation in the extreme value region. c=1 / ln(m);

[0122] Next, calculate the coefficient of variation: ,in, This refers to the difference coefficient of the j-th parameter, which ultimately yields the contribution: ,in, This refers to the contribution of the j-th geometric parameter, where n is the total number of geometric parameters. For example, the calculated contributions of the bottom diameter ratio, radius of curvature, and height-to-diameter ratio are 0.50, 0.45, and 0.05, respectively.

[0123] Step a4: Select the dominant parameters whose contribution exceeds the preset contribution threshold from all geometric parameters.

[0124] In step a4, a contribution threshold is preset, such as θ=0.3. The contribution of each geometric parameter is then compared with 0.3. Parameters with a contribution exceeding 0.3, such as the bottom diameter ratio and radius of curvature, are selected as the dominant parameters because they are key factors that play a leading role in the high uncertainty of the current extreme value region.

[0125] Step a5: Perform Latin hypercube sampling within the sub-region formed by the dominant parameters to obtain supplementary sample points.

[0126] In step a5, based on the screening results of step a4, the sampling attention is focused on the dominant parameters, and new design points, i.e. supplementary sample points, are generated in the subspace spanned by these dominant parameters using the Latin hypercube sampling strategy. During sampling, it is necessary to ensure that the samples are evenly distributed in the dimension of the dominant parameters. For non-critical geometric parameters, their values ​​are fixed as the average value of the samples in the extreme value region.

[0127] For example, assuming that according to the results of step a3, the contribution of the bottom diameter ratio is 0.50>θ, the radius of curvature is 0.45>θ, and the height-to-diameter ratio is 0.05<θ, the bottom diameter ratio and the radius of curvature are selected as the dominant parameters. A two-dimensional subspace is formed by the two dominant parameters as coordinate axes, and Latin hypercube sampling is performed in this two-dimensional subspace to generate 5 supplementary sample points. The height-to-diameter ratio of the non-dominant parameter is set to the average value of the samples in the extreme value region, such as 1.2.

[0128] Step 1044: Update the initial sample set based on the supplementary sample points and the corresponding structural performance indicators, and repeat the fitting, evaluation and identification steps until the preset iteration stopping condition is met, to obtain the target sample set and the target surrogate model corresponding to the target sample set.

[0129] In step 1044, the structural performance indices corresponding to the supplementary sample points are first obtained, and then these structural performance indices are added to the initial sample set. Subsequently, the updated sample set is used to re-perform the fitting process to update the surrogate model. Then, uncertainty assessment and adaptive sampling identification are performed on the new model, and this process is iterated until the preset stopping condition is met, such as the decrease of the maximum prediction variance of the surrogate model tends to be stable or the total sample size reaches the upper limit, such as 120. When the iteration terminates, the final sample set used is the target sample set, and the last surrogate model fitted is the high-precision target surrogate model.

[0130] Step 1045: Combine the hydraulic characteristics corresponding to the target proxy model, the velocity field data, and the riverbed elevation data to obtain the key geometric parameters.

[0131] Among them, hydraulic characteristics refer to the features that can be extracted from velocity field data, including Froude number, velocity gradient, etc., and features that can be extracted from riverbed elevation data, including average scour depth, scour and deposition rate, etc.

[0132] Key geometric parameters refer to the contour and cross-sectional shape parameters that play a dominant role in the impact resistance performance of bridge piers, identified through global performance sensitivity analysis. These include, but are not limited to, the bottom diameter ratio, the radius of curvature of the water-facing surface, the pier height-to-diameter ratio, the pier cross-sectional shape coefficient, or the pier side inclination angle.

[0133] In step 1045, a set of representative hydraulic characteristic values ​​are first calculated from the velocity field data and riverbed elevation data to quantify the environmental load of the current flood scenario. Then, based on the target surrogate model, a global sensitivity analysis is performed. Specifically, the hydraulic characteristic values ​​are used as fixed conditions in the target surrogate model to perturb each geometric parameter, and the average impact of each geometric parameter changing individually on the structural performance index is quantified. Then, by calculating the corresponding sensitivity index, one or more geometric parameters that play a dominant role in performance under a given hydraulic environment are identified. These are determined as key geometric parameters strongly correlated with the scenario.

[0134] For example, assuming the calculated hydraulic characteristics are: Froude number 0.8, riverbed scour rate 0.05 m / h, and fixing the Froude number and riverbed scour rate in the target surrogate model, the effects of ±10% variation in the bottom diameter ratio R and the upstream curvature C on the maximum stress are analyzed. The influence of the bottom diameter ratio was then considered, and the sensitivity index of the water-facing surface curvature was calculated: ,in, Represents the geometric parameters of the i-th pier. This represents the maximum equivalent stress on the bridge pier. Conditional expectation: The i-th parameter is fixed. After taking the value, predict The average value, Let S represent the variance of the conditional expectation; for example, we obtain S. R =0.6, S C =0.3. Since 0.6 > 0.3, the bottom expansion ratio is determined to be the key geometric parameter under the current hydraulic conditions.

[0135] Step 1046: Based on the physical constraint values ​​corresponding to the key geometric parameters, modify the target surrogate model to obtain the surrogate model response surface.

[0136] In step 1046, the physical constraint values ​​of the key geometric parameters are first determined. These constraints may originate from construction technology, material properties, or design specifications. Then, this physical constraint is used as a mandatory boundary condition and input into the prediction logic of the target proxy model. In terms of technical implementation, when the target proxy model receives a new design point for prediction, it will first check whether its key geometric parameters exceed the range of physical constraints. If they do, the model will directly return a penalty value indicating poor performance.

[0137] If the error does not exceed the limit, the model is invoked normally for prediction. After this rule enhancement, the model not only retains the performance mapping relationship learned from a large amount of data, but also embeds the engineer's experience and standard knowledge. Finally, by querying the predicted value of this corrected model in the entire design space and visualizing or storing it as a data structure, the surrogate model response surface is obtained.

[0138] For example, assuming the physical constraint upper limit of the bottom diameter ratio R of the key geometric parameter is Rmax=2.5, the target surrogate model M is then modified to obtain the modified model M′. The logic is as follows: if the input design R>Rmax, then the modified target surrogate model M′(x)=P, where P is a very large penalty stress value, such as 106MPa, and x represents the complete geometric parameter vector of the pier, which usually includes key parameters and non-key parameters; otherwise, M′(x)=M(x), where M(x) represents the target surrogate model before modification. Subsequently, on the design space grid points composed of R, water-facing curvature C, and height-to-diameter ratio H, M′ is called in batches to perform predictions to obtain the performance value corresponding to each grid point. The data cube composed of these performance values ​​is the final surrogate model response surface.

[0139] This application generates a proxy model response surface that balances prediction accuracy and engineering feasibility, providing a core basis for efficient and reliable multi-objective optimization design.

[0140] S105. Based on the response surface of the surrogate model, with preset structural conditions and preset material usage conditions as optimization objectives, the geometric parameters of the bridge pier are optimized to obtain the optimized geometric parameters.

[0141] In one specific implementation, step S105 includes:

[0142] Step 1051: Based on the preset structural conditions and preset material usage conditions, the improved NSGA-III algorithm is used to solve the response surface of the surrogate model to generate a Pareto solution set.

[0143] Among them, the improved NSGA-III algorithm refers to an enhanced algorithm based on the standard NSGA-III algorithm, which introduces an adaptive reference point generation and constraint domination mechanism to address the characteristics of high-dimensionality, nonlinearity and complex constraints of bridge pier geometric parameters.

[0144] In step 1051, the surrogate model response surface is first established as the core engine for optimization calculation to evaluate the performance of any design scheme. Then, minimizing the critical structural response and minimizing the material usage are defined as two competing optimization objectives. Subsequently, the population of the improved NSGA-III algorithm is initialized, with each individual encoding a specific set of pier geometry parameters.

[0145] The improvement of the algorithm lies in the fact that its reference point set can be dynamically adjusted according to the current population distribution to better guide the search direction. It also uses a penalty function and special selection pressure to deal with individuals that violate the constraints. Then, during the iteration process, the algorithm uses the response surface to efficiently evaluate the population, generates new individuals through crossover and mutation, and uses an improved environmental selection mechanism to retain high-quality and diverse solutions. When the evolution reaches the preset number of generations, the algorithm outputs all non-dominated individuals in the final population. These individuals together constitute the Pareto solution set.

[0146] For example, the optimization objectives are to minimize the maximum equivalent stress and minimize the concrete volume. The design variables are the bottom diameter ratio R and the pier height ratio β, with R ∈ [1.5, 2.5] and β ∈ [3.0, 6.0], respectively. Then, the improved NSGA-III algorithm is adopted, and it is assumed that the population size N = 100 and the maximum number of generations G = 250. After iterative optimization, a Pareto solution set {x1, x2, ..., x40} containing 40 non-dominated solutions is obtained, where xi = (Ri, βi). Each solution xi corresponds to a set of objective values ​​(I(xi), V(xi)), where I(xi) represents the specific value of the maximum equivalent stress corresponding to the i-th solution, and V(xi) represents the specific value of the concrete volume corresponding to the i-th solution.

[0147] Step 1052: Analyze the Pareto solution set to obtain sparse regions and clustered regions.

[0148] Among them, the sparse region refers to the region in the space of objective function values ​​in which the Pareto solution set is distributed that the distribution density of solutions is lower than the average level. Its physical meaning is the performance range that has not yet been fully explored by current optimization, and that may be better or have special trade-off value.

[0149] Clustered regions refer to areas in the objective function value space where the distribution density of solutions is higher than the average level. Physically, they represent a large number of redundant design schemes with similar performance but slightly different geometric parameters.

[0150] In step 1052, firstly, each solution in the Pareto solution set is located in the two-dimensional target space based on its two objective function values; then, a distance-based density estimation method is used to calculate the Euclidean distance between each solution and its z-th nearest neighbor solution, and the reciprocal of this distance is calculated as the local density around the solution; then, the average local density of the entire solution set is calculated, and a first density threshold is set based on the average local density. Solutions with local densities lower than the first density threshold are classified as being located in sparse regions, which indicate parts of the Pareto front that are under-covered or can be further explored; conversely, solutions with local densities higher than a preset second density threshold are classified as being located in clustered regions, where the first density threshold is less than the second density threshold.

[0151] Step 1053: Perform supplementary sampling processing in the sparse region to obtain the first solution; perform solution set filtering processing on the geometric parameters of the bridge pier in the clustered region to obtain the second solution.

[0152] Step 1053 may specifically include the following steps:

[0153] Step b1: Obtain the local performance gradient corresponding to the response surface of the proxy model in the sparse region.

[0154] In step b1, firstly, from the Pareto solution set that has been determined to be a sparse region, all solutions belonging to that region are selected as the starting point for analysis; then, for each such solution, the local performance gradient of the surrogate model response surface at that point is calculated, for example, by calling the function of calculating the first-order partial derivative of the surrogate model response surface at that point, a local gradient vector characterizing the rate of performance change is obtained, where the calculation of the gradient reveals how, at this point, a slight adjustment of any geometric parameter will lead to changes in structural performance indicators and material usage.

[0155] Step b2: Based on the direction and magnitude of the local performance gradient, perform deterministic guided sampling in the sparse region to obtain preliminary guided points.

[0156] In step b2, for each point in the sparse region, a small step is taken in the opposite direction of its local performance gradient. The step size is proportional to the magnitude of the gradient, with larger gradients resulting in larger adjustments. In this way, a new initial guide point that may have greater performance potential is generated from each original sparse point.

[0157] For example, assuming the local performance gradient is (-3.2, 1.5), to explore the direction of stress reduction, we move along the opposite direction of this gradient (3.2, -1.5) and set the step size factor to 0.05 to generate new parameter points: , This generates an initial guiding point. .

[0158] Step b3: Verify the preliminary guiding point based on the geometrically feasible region and spacing constraints to obtain the first solution.

[0159] In step b3, all preliminary guide points need to be screened to ensure their quality and usability. The screening is based on two criteria: First, check whether the geometric parameters of each preliminary guide point are still within the predefined design feasible region and remove points that exceed the limits. For example, the feasible region requires the bottom diameter ratio to be within [1.5, 2.5] and the pier height-to-diameter ratio to be within [3.5, 5.0]. Second, to avoid the newly introduced solution being too similar to the existing Pareto solution, calculate the minimum Euclidean distance between each preliminary guide point and all solutions in the existing Pareto solution set. If the distance is less than a preset distance threshold, such as 0.1, the point is considered too close to the existing solution and lacks new information, so it is removed, thus obtaining the first solution.

[0160] Step b4: Perform Euclidean distance-based clustering partitioning on the geometric parameters of the pier in the clustered region to obtain multiple sub-clusters, and calculate the material property values ​​and front distance of each geometric parameter in each sub-cluster.

[0161] In step b4, firstly, the geometric parameters corresponding to all Pareto solutions belonging to the clustered region are used as features, and K-means clustering algorithm based on Euclidean distance is used to divide them into several sub-clusters with similar internal features. Then, after clustering, within each sub-cluster, two core indicators are calculated for each solution in the cluster: one is its material performance value, that is, the specific value of each solution in terms of material usage target, and the other is its frontier distance. The frontier distance is usually defined as the distance from the solution to the theoretical optimal point in the sub-cluster. The theoretical optimal point is the minimum value of all solutions in the corresponding sub-cluster in terms of the target.

[0162] Step b5: Combine the material performance value with the front distance to obtain a ranking index. Based on the ranking index, select non-dominated solutions from each of the subclusters to obtain a second solution.

[0163] In step b5, the material performance values ​​and the front distance are normalized and then fused into a comprehensive ranking index according to preset weights, such as a weight of 0.7 for the material performance values ​​and a weight of 0.3 for the front distance. Next, within each sub-cluster, all solutions are ranked according to this ranking index. Finally, according to preset quotas, such as retaining the top two in each sub-cluster, the second-ranked solution is selected.

[0164] Step 1054: Merge the first solution and the second solution to obtain an equilibrium solution set.

[0165] In step 1054, the first solution and the second solution are merged to form a temporary merged set. Then, a non-dominated sorting operation is performed on the merged set. This operation identifies and removes individuals in the set that are dominated by other solutions, that is, those individuals that are inferior to at least one other solution in all optimization objectives. After the non-dominated sorting purification, all the solutions that are finally retained constitute the equilibrium solution set.

[0166] Step 1055: Using a multi-criteria decision-making method based on the ideal point method, select the final solution from the equilibrium solution set, and use the geometric parameters corresponding to the final solution as the optimized geometric parameters.

[0167] In step 1055, a multi-criteria decision-making method based on the ideal point method is used to assign decision weights to the structural performance target and the material performance value according to the actual engineering requirements, so as to quantify the relative importance between the two. This embodiment does not limit the implementation process of the multi-criteria decision-making method. Subsequently, an ideal point is constructed in the space composed of the two targets, and its coordinates are composed of the minimum value of each target in the equilibrium solution set to represent the theoretical global optimum.

[0168] Next, for each solution in the equilibrium solution set, the weighted Euclidean distance from its weighted objective value to the ideal point is calculated; for example, the ideal point is (Imin, Vmin), where Imin is the minimum equivalent stress and Vmin is the minimum concrete volume. Finally, the solution with the smallest distance to the ideal point is selected as the final solution. This final solution encodes the specific shape of the pier, and the values ​​of the geometric parameters it contains are the final and optimal geometric parameters output by this adaptive design optimization process. For example, in this application, the values ​​of the geometric parameters are the values ​​of the bottom diameter ratio R and the pier height-to-diameter ratio β.

[0169] This application uses intelligent algorithms to automatically find and select Pareto solutions, and finally decides on a single optimal design scheme based on engineering preferences, which can achieve efficient and accurate adaptive optimization of bridge pier geometric parameters under multi-objective constraints.

[0170] S106. Input the optimized geometric parameters into the fluid dynamics calculation model established based on historical hydrological data to perform two-way fluid-structure interaction simulation and obtain the verification results. If the verification results do not meet the preset convergence conditions, modify the parameter design of at least one condition in the optimization objective according to the verification results, and re-execute the parameter optimization and simulation steps until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

[0171] In one specific implementation, step S106 includes:

[0172] Step 1061: Set the simulation boundary conditions for the fluid dynamics calculation model based on historical hydrological data.

[0173] Historical hydrological data refers to data obtained through long-term observation or statistics that reflects the characteristics of the target river's flow.

[0174] The fluid dynamics calculation model can be an OpenFOAM-based model, and this application does not limit the structure of the model or the process of setting simulation boundary conditions.

[0175] In step 1061, in order to construct a simulation environment that reflects the actual flood conditions, historical hydrological data is required. First, key parameters corresponding to the target return period flood, such as peak flow and corresponding velocity profile, are extracted from the historical hydrological data. Then, simulation boundary conditions are set in the fluid dynamics calculation model based on these parameters: the inlet of the calculation domain is set as the velocity inlet boundary condition according to the extracted velocity profile, the outlet of the calculation domain is set as the pressure outlet boundary condition, and the bottom of the basin and the bank are set as no-slip wall boundary conditions with specific roughness based on the topographic data.

[0176] Step 1062: Input the optimized geometric parameters into the fluid dynamics calculation model, and perform a two-way fluid-structure interaction simulation based on the simulation boundary conditions to obtain physical field data, which includes the pressure distribution on the structural surface, the internal stress distribution, and the eddy shedding frequency.

[0177] In step 1062, firstly, the optimized geometric parameters are used to create a three-dimensional geometric model, which is then imported into a fluid dynamics calculation model with pre-defined boundary conditions for mesh generation. Next, a two-way fluid-structure interaction (FSI) calculation is configured in the solver. The specific settings include: using the RANS equations and the SST k-omega turbulence model in the fluid domain, and assigning concrete material properties to the pier in the solid domain. Then, the two-way FSI method is used for iterative solution. After the calculation is started, the solver iteratively solves the coupling problem of fluid motion and structural deformation under the given simulation boundary conditions through fluid-structure data exchange. After the calculation is completed, the transient and time-averaged surface pressure distribution of the pier, the internal stress distribution of key sections inside the structure, and the vortex shedding frequency identified after spectral analysis of the pier wake velocity signal are extracted and output from the result file.

[0178] Step 1063: Compare the physical characteristics corresponding to the physical field data with the structural performance indicators corresponding to the response surface of the surrogate model to obtain the verification results.

[0179] In step 1063, firstly, physical features related to the optimization objective are extracted from the physical field data, such as maximum lateral displacement, maximum equivalent stress, and vortex shedding frequency. Next, the optimized geometric parameters are input into the surrogate model response surface to query and obtain the predicted maximum stress and maximum displacement, among other structural performance indicators. Then, the extracted maximum lateral displacement is numerically compared with the queried maximum displacement, and the corresponding relative error is calculated. Similarly, the maximum equivalent stress is numerically compared with the queried maximum stress, and the corresponding relative error is calculated. Finally, the vortex shedding frequency is compared with a preset resonance hazard range, which is the hazard range obtained through structural dynamic modal analysis based on the bridge pier's current optimized geometric parameters and material properties. Finally, the error is evaluated according to preset convergence conditions, specifically quantified as follows: if the relative error of the maximum stress is less than 5%, the relative error of the maximum displacement is less than 3%, and the vortex shedding frequency is not within the preset resonance hazard range, then the convergence conditions are met, thus generating the final verification result.

[0180] This application constitutes a closed loop of design and verification. The optimization results are rigorously verified through high-fidelity simulation, and the optimization objectives are adjusted based on the quantification error feedback, ultimately ensuring that the obtained bridge pier structural parameters have reliable impact resistance in the real physical world.

[0181] Figure 3 This application provides a schematic diagram of a specific implementation of an adaptive design system for bridge pier impact-resistant structures based on water flow simulation, as shown in the following embodiment. Figure 3 The system may include:

[0182] The acquisition module 31 is used to acquire the velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as the geometric parameters and structural attitude data of the existing bridge piers.

[0183] The combination module 32 is used to combine the flow velocity field data, the riverbed elevation data and the structural attitude data to construct a first time series matrix, and to perform trend smoothing and feature extraction processing on the first time series matrix to obtain second time series data.

[0184] Evaluation module 33 is used to input the second time series data into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators.

[0185] The fitting module 34 is used to fit the structural performance index to obtain the surrogate model response surface.

[0186] The optimization module 35 is used to optimize the geometric parameters of the bridge pier based on the response surface of the surrogate model, with preset structural conditions and preset material usage conditions as optimization targets, so as to obtain the optimized geometric parameters.

[0187] The verification module 36 is used to input the optimized geometric parameters into a fluid dynamics calculation model based on historical hydrological data to perform two-way fluid-structure interaction simulation and obtain verification results. If the verification results do not meet the preset convergence conditions, the parameter design of at least one condition in the optimization objective is modified according to the verification results, and the parameter optimization and simulation steps are re-executed until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

[0188] The adaptive design system for bridge pier impact-resistant structures based on water flow simulation in this application is used to implement the aforementioned adaptive design method for bridge pier impact-resistant structures based on water flow simulation. Therefore, the specific implementation of the adaptive design system for bridge pier impact-resistant structures based on water flow simulation can be found in the embodiment section of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0189] like Figure 4 As shown, this application also provides an electronic device, including: a memory 41 for storing a computer program; and a processor 42 for executing the computer program to implement the steps of any of the above-described adaptive design methods for bridge pier impact-resistant structures based on water flow simulation.

[0190] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described adaptive design methods for bridge pier impact-resistant structures based on water flow simulation.

[0191] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0192] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation.

[0193] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0194] The above provides a detailed description of the adaptive design method and system for bridge pier impact-resistant structures based on water flow simulation provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An adaptive design method for bridge pier impact-resistant structures based on water flow simulation, characterized in that, include: Acquire flow velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as geometric parameters and structural attitude data of existing bridge piers; The flow velocity field data, the riverbed elevation data, and the structural attitude data are combined to construct a first time series matrix. The first time series matrix is ​​then subjected to trend smoothing and feature extraction processing to obtain the second time series data. The second time series data is input into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators. The surrogate model response surface is obtained by fitting the structural performance indices. Based on the response surface of the surrogate model, the geometric parameters of the bridge pier are optimized with preset structural conditions and preset material usage conditions as optimization objectives, and the optimized geometric parameters are obtained. The optimized geometric parameters are input into a fluid dynamics calculation model based on historical hydrological data for two-way fluid-structure interaction simulation to obtain verification results. If the verification results do not meet the preset convergence conditions, the parameter design of at least one condition in the optimization objective is modified according to the verification results, and the parameter optimization and simulation steps are re-executed until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

2. The method according to claim 1, characterized in that, The fitting of the structural performance indicators to obtain the surrogate model response surface includes: An initial sample set is constructed based on the geometric parameters of the bridge piers and the aforementioned structural performance indicators; The initial sample set is fitted using a Kriging surrogate model to obtain the initial surrogate model; An uncertainty assessment is performed on the initial proxy model to obtain an uncertainty distribution. Based on the uncertainty distribution, an adaptive sampling algorithm is used to identify supplementary sample points. The initial sample set is updated based on the supplementary sample points and the corresponding structural performance indicators, and the fitting, evaluation and identification steps are repeated until the preset iteration stopping condition is met, so as to obtain the target sample set and the target proxy model corresponding to the target sample set. By combining the hydraulic characteristics corresponding to the target proxy model, the velocity field data, and the riverbed elevation data, the key geometric parameters are obtained. Based on the physical constraint values ​​corresponding to the key geometric parameters, the target surrogate model is modified to obtain the surrogate model response surface.

3. The method according to claim 2, characterized in that, The initial proxy model is subjected to uncertainty assessment to obtain an uncertainty distribution. Based on the uncertainty distribution, an adaptive sampling algorithm is used to identify supplementary sample points, including: Based on the initial proxy model, Monte Carlo simulation prediction is performed on the unsampled space to obtain the uncertainty measure of each prediction point; Using spatial interpolation methods, an uncertainty distribution map is constructed based on the uncertainty metric, and the uncertainty distribution map is subjected to trend fitting and identification processing to obtain the extreme value region; In the extreme value region, the entropy weight method is applied to perform sensitivity analysis on the geometric parameters of the bridge piers to obtain the contribution of each geometric parameter; Select the dominant parameters whose contribution exceeds a preset contribution threshold from all geometric parameters; Latin hypercube sampling is performed within the sub-region defined by the dominant parameters to obtain supplementary sample points.

4. The method according to claim 1, characterized in that, Based on the surrogate model response surface, and with preset structural and material usage conditions as optimization objectives, the geometric parameters of the bridge pier are optimized to obtain the optimized geometric parameters, including: Based on the preset structural conditions and preset material usage conditions, the improved NSGA-III algorithm is used to solve the response surface of the surrogate model and generate a Pareto solution set. The Pareto solution set is analyzed to obtain sparse regions and clustered regions; Supplementary sampling is performed in the sparse region to obtain the first solution, and the geometric parameters of the bridge pier are filtered by solution set processing in the clustered region to obtain the second solution. By combining the first solution and the second solution, an equilibrium solution set is obtained; A multi-criteria decision-making method based on the ideal point method is adopted to select the final solution from the equilibrium solution set, and the geometric parameters corresponding to the final solution are used as the optimized geometric parameters.

5. The method according to claim 4, characterized in that, The process involves supplementing sampling in the sparse region to obtain a first solution, and then performing solution set filtering on the geometric parameters of the bridge pier in the clustered region to obtain a second solution, including: Obtain the local performance gradient corresponding to the response surface of the proxy model in the sparse region; Based on the direction and magnitude of the local performance gradient, deterministic guided sampling is performed in the sparse region to obtain preliminary guided points; The initial guiding point is validated based on the geometrically feasible region and spacing constraints to obtain the first solution; In the clustered region, the geometric parameters of the bridge pier are subjected to Euclidean distance-based clustering partitioning to obtain multiple sub-clusters. Within each sub-cluster, the material property values ​​and front distance of each geometric parameter are calculated. By combining the material performance values ​​and the frontal distance, a ranking index is obtained. Based on the ranking index, non-dominated solutions are selected from each of the subclusters to obtain a second solution.

6. The method according to claim 1, characterized in that, The second time-series data is input into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators, including: The second time series data is subjected to spatiotemporal decomposition to obtain a time series feature sequence and a spatial feature map; The temporal feature sequence is modeled temporally using a bidirectional recurrent network layer in the CNN-BiLSTM prediction model to obtain preliminary temporal features. These preliminary temporal features are then enhanced to generate the first fused features. The spatial feature map is spatially modeled using the convolutional network layer in the CNN-BiLSTM prediction model to obtain preliminary spatial features, and the preliminary spatial features are then enhanced to generate a second fused feature. By using the attention fusion layer in the CNN-BiLSTM prediction model, the first fusion feature and the second fusion feature are weighted and fused to obtain the spatiotemporal fusion feature; The spatiotemporal fusion features are processed by nonlinear transformation using the fully connected network layer in the CNN-BiLSTM prediction model to obtain structural performance indicators.

7. The method according to claim 1, characterized in that, The optimized geometric parameters are input into a fluid dynamics calculation model based on historical hydrological data for two-way fluid-structure interaction simulation to obtain verification results, including: The simulation boundary conditions of the fluid dynamics calculation model are set based on historical hydrological data; The optimized geometric parameters are input into the fluid dynamics calculation model, and a two-way fluid-structure interaction simulation is performed based on the simulation boundary conditions to obtain physical field data, which includes the pressure distribution on the structure surface, the internal stress distribution, and the eddy shedding frequency. The physical characteristics corresponding to the physical field data are compared with the structural performance indicators corresponding to the response surface of the surrogate model to obtain the verification results.

8. An adaptive design system for bridge pier impact-resistant structures based on water flow simulation, characterized in that, include: The acquisition module is used to acquire flow velocity field data and riverbed elevation data of the target river under extreme flood scenarios, as well as geometric parameters and structural attitude data of the existing bridge piers; The combination module is used to combine the flow velocity field data, the riverbed elevation data, and the structural attitude data to construct a first time series matrix, and to perform trend smoothing and feature extraction processing on the first time series matrix to obtain second time series data. The evaluation module is used to input the second time series data into a pre-trained CNN-BiLSTM prediction model for spatiotemporal feature analysis and shock resistance performance evaluation to obtain structural performance indicators. The fitting module is used to fit the structural performance indicators to obtain the surrogate model response surface; The optimization module is used to optimize the geometric parameters of the bridge pier based on the response surface of the proxy model, with preset structural conditions and preset material usage conditions as optimization targets, so as to obtain the optimized geometric parameters. The verification module is used to input the optimized geometric parameters into a fluid dynamics calculation model based on historical hydrological data to perform two-way fluid-structure interaction simulation and obtain verification results. If the verification results do not meet the preset convergence conditions, the module modifies the parameter design of at least one condition in the optimization objective according to the verification results and re-executes the parameter optimization and simulation steps until the verification results meet the preset convergence conditions to obtain the pier structure parameters.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the adaptive design method for bridge pier impact-resistant structures based on water flow simulation as described in any one of claims 1 to 7.

Images