Railway bridge tower area wind barrier optimization method and system based on deep learning

Abstract

The invention provides a railway bridge tower area wind barrier optimization method and system based on deep learning, and relates to the technical field of railway barriers, and the method comprises the steps: obtaining the structure parameters of a target main beam, a bridge tower and a train, and building a train-bridge three-dimensional model of a tower area according to the structure parameters of a target bridge tower-main beam intersection area; processing the train-bridge model by adopting a self-adaptive grid technology to obtain an overlapped grid model for simulating train movement; defining initial variables and boundary conditions of the model in computational fluid dynamics software, and performing iterative computation through an N-S equation and a turbulence equation to obtain an aerodynamic coefficient time history of the train; based on the CFD calculation result, processing the aerodynamic coefficient time history to obtain a data set for neural network training; and training and testing the data set to complete optimization of the wind barrier in the bridge tower area. The deep reinforcement learning technology is adopted, the wind load sudden change effect of the tower area is effectively controlled, and the calculation time and the calculation cost are reduced.

Description

Technical Field

[0001] The present invention relates to the field of railway barrier technology, and in particular to a railway bridge tower area wind barrier optimization method and system based on deep learning. Background Art

[0002] Long-span bridges are favored by designers because of their strong spanning capacity and convenient navigation. The spans of suspension bridges and cable-stayed bridges have been challenging new limits. The design of wind barriers in the tower area of ​​traditional railway bridges usually relies on the results of wind tunnel tests and numerical simulations. However, these methods can often only obtain local optimal design solutions, and their optimization experience is difficult to directly apply to other similar bridge projects. Among them, the size of the bridge tower also increases with the increase of the proportion, even reaching tens of meters along the longitudinal bridge direction. The existence of the bridge tower will change the wind environment around the bridge deck, and the real aerodynamic force of the vehicle is determined by the flow field around it. In this case, the sudden change of the lateral wind speed in the bridge tower area is more likely to cause lateral instability of the vehicle and driver's handling / controllability problems. This is because the aerodynamic force acting on the vehicle changes dramatically when the vehicle passes the bridge tower. When the vehicle passes the bridge tower, the vehicle will be tightly shielded, and the wind load acting on the vehicle will decrease when the vehicle enters the bridge tower area, and then increase when the vehicle leaves the area, changing sharply, and the risk of the vehicle being blown over is high. Therefore, it is necessary to take wind protection measures near the bridge tower to reduce the negative impact of this sudden effect. Summary of the invention

[0003] The purpose of the present invention is to provide a method, system, device and readable storage medium for optimizing wind barriers in railway bridge towers based on deep learning to improve the above problems. In order to achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0004] In a first aspect, the present application provides a railway bridge tower area wind barrier optimization method based on deep learning, comprising:

[0005] Obtain the target main beam, bridge tower, train and their structural parameters. According to the structural parameters of the target bridge, after the three-dimensional bridge model is constructed, a tower area train-bridge model including the bridge section, bridge deck pavement, bridge deck anti-collision railings, bridge tower and main components of the train is obtained;

[0006] The train-bridge model including the bridge tower is processed by adaptive mesh refinement technology to obtain an overlapping mesh model for simulating train movement. Based on the overlapping mesh model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated through the NS equation and turbulence equation to obtain the aerodynamic coefficient time history of the corresponding train under different parameters.

[0007] The aerodynamic coefficient time history of trains under different wind barrier parameters is processed to obtain the aerodynamic mutation amplitude of the train, and then the data set for neural network training is obtained;

[0008] The data set is trained and tested to determine whether the training and test results meet the termination conditions. If so, the optimal parameters are obtained to complete the optimization of the wind barrier in the bridge tower area. If not, training continues until the termination conditions are met. The training method adopts a deep deterministic policy gradient algorithm, where the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of changes in design variables.

[0009] Preferably, the adaptive mesh refinement technology is used to process the tower area train-bridge model including the bridge tower to obtain an overlapping mesh model for simulating train movement, which includes:

[0010] According to the tower area train-bridge model, after meshing in the software, the mesh models of the main area and the subsidiary area are obtained, where the main area includes the bridge part and the outer area, and the subsidiary area includes the moving train part;

[0011] The grid model is processed by adaptive grid refinement technology to obtain a dynamically updated overlapping boundary model;

[0012] Based on the overlapping grid model, the flow field data exchange model between the main area and the subsidiary area is obtained after flow field information interpolation processing.

[0013] Preferably, in the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated through the NS equation and the turbulence equation, so as to obtain the aerodynamic coefficient time history of the corresponding train under different parameters, including:

[0014] According to the overlapping grid model, after the initial variable definition processing, the initial flow field model is obtained;

[0015] According to the boundary conditions, after the boundary condition definition processing, the boundary flow field model is obtained;

[0016] Based on the initial flow field model and the boundary flow field model, the flow field distribution model is obtained through iterative calculation of the NS equation and the turbulence equation.

[0017] Preferably, the aerodynamic time history is calculated and processed by using CFD calculation and neural network training to obtain a data set for optimizing wind barrier parameters, which includes:

[0018] Based on the time history of train drag, lift and lift moment coefficients under different wind barrier parameters, the drag coefficient data of the train under different wind barrier parameter settings are collected and processed to obtain the mutation amplitude of the aerodynamic coefficient when the train passes through the tower area. After data set construction and processing, the data set for neural network training is obtained, in which the height and porosity of the wind barrier are adjustable variable parameters.

[0019] Preferably, the data set is trained and tested, wherein the training includes using a deep deterministic policy gradient algorithm, wherein as the Actor network parameter θμ changes;

[0020] Use the same reward function r during training t Evaluate the quality of taking a specific action under a specific state.

[0021] In the second aspect, the present application also provides a railway bridge tower area wind barrier optimization system based on deep learning, comprising:

[0022] Acquisition module: used to obtain the target main beam, bridge tower, train and their structural parameters. According to the structural parameters of the target bridge, after the three-dimensional bridge model is constructed and processed, a tower area train-bridge model including the bridge section, bridge deck pavement, bridge deck anti-collision railings, bridge tower and main components of the train is obtained;

[0023] The first processing module is used to process the train-bridge model including the bridge tower by using the adaptive grid refinement technology to obtain an overlapping grid model for simulating the movement of the train; based on the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated by the NS equation and the turbulence equation, thereby obtaining the aerodynamic coefficient time history of the corresponding train under different parameters;

[0024] The second processing module is used to perform calculations based on the aerodynamic coefficient time history using CFD to obtain a data set for neural network training;

[0025] Judgment module: used to train and test the data set, and judge whether the training and test results of the iterative calculation meet the termination conditions. If so, the optimal parameters are obtained, and the optimization of the wind barrier in the bridge tower area is completed; if not, the training continues until the termination conditions are met; the training method adopts the deep deterministic policy gradient algorithm, and the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of the design variable changes.

[0026] In a third aspect, the present application also provides a railway bridge tower area wind barrier optimization device based on deep learning, comprising:

[0027] Memory for storing computer programs;

[0028] A processor is used to implement the steps of the railway bridge tower area wind barrier optimization method based on deep learning when executing the computer program.

[0029] In a fourth aspect, the present application also provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned railway bridge tower area wind barrier optimization method based on deep learning are implemented.

[0030] The beneficial effects of the present invention are:

[0031] The present invention adopts deep reinforcement learning technology to effectively control the sudden change effect of wind load in the tower area. The advantage of this method is that its optimization strategy can greatly reduce the calculation time and cost.

[0032] The present invention adopts deep reinforcement learning technology to effectively control the sudden change effect of wind load in the tower area. The optimization strategy can reduce the calculation time and cost, and carefully optimize the parameters of the wind barrier to seek the best design solution. Through this process, it can ensure that the wind barrier can play the greatest protective effect in practical applications.

[0033] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 It is a schematic flow chart of a railway bridge tower area wind barrier optimization method based on deep learning described in an embodiment of the present invention;

[0036] Figure 2 It is a schematic diagram of the structure of the railway bridge tower area wind barrier optimization system based on deep learning described in an embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of the structure of the railway bridge tower area wind barrier optimization equipment based on deep learning described in an embodiment of the present invention;

[0038] Figure 4A schematic diagram of overlapping grids described in an embodiment of the present invention;

[0039] Figure 5 A schematic diagram of the impact width and mutation amplitude described in an embodiment of the present invention;

[0040] Figure 6 It is a schematic diagram of the CFD calculation domain arrangement described in an embodiment of the present invention;

[0041] Figure 7 It is a schematic diagram of the three-force train described in an embodiment of the present invention.

[0042] In the figure: 701, acquisition module; 702, first processing module; 703, second processing module; 704, judgment module; 800, railway bridge tower area wind barrier optimization equipment based on deep learning; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0044] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0045] Embodiment 1:

[0046] In the prior art, key parameters such as the height and air permeability of the wind barrier have a significant impact on the flow field around it, which in turn determines the quality of its windproof effect. In view of the differences in windproof effects under different parameter configurations, it is particularly important to carefully optimize the parameters of the wind barrier to seek the best design solution. Through this process, it can be ensured that the wind barrier can play the greatest protective effect in practical applications. In this embodiment, with the development of artificial intelligence and computer science, machine learning is increasingly widely used in the fields of fluid dynamics and aerodynamics, such as for reduced-order modeling, experimental data processing, shape optimization, turbulent closure modeling and control. The use of machine learning to optimize the aerodynamic shape of the wind barrier reduces the cost of numerical simulation, and the use of deep learning models instead of computational fluid dynamics (CFD) models can accelerate the optimization process.

[0047] This embodiment provides a railway bridge tower area wind barrier optimization method based on deep learning.

[0048] See also Figure 1 , the figure shows that the method includes step S100, step S200, step S300 and step S400.

[0049] S100, obtaining target main beams, bridge towers, trains and their structural parameters, and obtaining a tower area train-bridge model including a bridge section, bridge deck pavement, bridge deck anti-collision railings, bridge towers and main components of the train through a three-dimensional bridge model construction process according to the structural parameters of the target bridge.

[0050] It is understandable that in this step, it should be noted that fluid mechanics simulation is a technology that uses computers to solve various conservation control partial differential equations of fluid flow to simulate the actual fluid flow. Its basic principle is to numerically solve the differential equations that control fluid flow, and obtain the discrete distribution of the flow field of fluid flow in a continuous area, thereby approximately simulating the fluid flow. In this step, the target bridge tower is determined and a three-dimensional bridge model is constructed according to the structural parameters of the bridge to which the target bridge tower belongs. Among them, the structural parameters include the structural parameters of the bridge section (which can fully express the geometric characteristics of the main beam such as length, width, height, angle, spatial position of the rod, etc.), the bridge deck pavement (thickness, length, height), the bridge deck anti-collision railing (component size, spatial position, connection method), the bridge tower (which can fully express the geometric characteristics of the bridge tower such as length, width, height, angle, etc.), and the main components such as the train.

[0051] S200. The train-bridge model including the bridge tower is processed by using the adaptive grid refinement technology to obtain an overlapping grid model for simulating the movement of the train. Based on the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated through the NS equation and the turbulence equation, thereby obtaining the aerodynamic coefficient time history of the corresponding train under different parameters.

[0052] It can be understood that step S200 includes S201, S202 and S203, wherein:

[0053] S201, according to the tower area train-bridge model, through meshing processing in the software, a mesh model of the main area and the subsidiary area is obtained, wherein the main area includes the bridge part and the outer area, and the subsidiary area includes the moving train part;

[0054] S202, processing the grid model by using an adaptive grid refinement technology to obtain a dynamically updated overlapping boundary model;

[0055] S203, based on the overlapping grid model, after flow field information interpolation processing, a flow field data exchange model between the main area and the subsidiary area is obtained.

[0056] It should be noted that when the above model is meshed in the software, overlapping meshes can also be used to simulate the movement of the train. The core of overlapping meshes is to divide the calculation domain into multiple sub-areas, each sub-area generates a separate mesh, and there is an overlapping relationship between these sub-areas. The flow field information is realized by interpolating the overlapping boundaries to ensure the exchange of flow field data between the main area and the subsidiary area. In this study, the entire calculation domain is divided into the main area and the subsidiary area. The main area includes the bridge part and the external domain, while the subsidiary area contains the moving train part. During the movement of the train, the overlapping boundaries will be continuously updated to ensure continuity with the main area, so as to calculate the aerodynamic force when the train passes through the tower area. Finally, it is exported to the computational fluid dynamics software, such as Figure 4 shown.

[0057] In this step, first, according to the geometry and complexity of the model, the initial mesh is generated using the meshing tool provided by the software. The initial mesh can be uniform or non-uniform, depending on the characteristics of the problem being studied. Secondly, by evaluating the difference between the initial solution and the expected solution, it is concluded that the mesh may need to be refined or coarsened according to different areas. Error evaluation can be performed according to different criteria, such as node displacement, strain, stress, etc. Then, based on the results of the error evaluation, the software will automatically refine or coarsen the mesh in the required area. Mesh refinement improves accuracy by adding more nodes and units to the problem area, while mesh coarsening reduces unnecessary details and improves computational efficiency; during the calculation process, the standards for refinement and coarsening can be checked regularly to perform mesh adaptation. For example, in Fluent, dynamic adaptation can be set to perform mesh refinement and coarsening operations regularly during the calculation process; finally, the standard for mesh refinement can be set, such as through cell register settings or expression settings, that is, when a physical quantity in a certain area reaches the set standard, the mesh in that area will be refined and encrypted. These steps ensure that the overlay mesh is dynamically updated as the train moves to ensure continuity with the main area, thereby accurately calculating the aerodynamic forces when the train passes through the tower area.

[0058] It can be understood that step S200 also includes S204, S205 and S206, wherein:

[0059] S204, obtaining an initial flow field model through initial variable definition processing according to the overlapping grid model;

[0060] S205, according to the boundary conditions, after boundary condition definition processing, a boundary flow field model is obtained;

[0061] It should be noted that in this step, the initial state of the flow field is first required, which includes basic fluid parameters such as velocity, pressure, and temperature. This step involves setting up a reasonable initial flow field model, which provides a starting point for numerical solution. In actual operation, the initial conditions are usually based on experimental data, theoretical predictions, or simplified assumptions. The establishment of the initial flow field model is completed through the physical understanding of the flow field and the grasp of the characteristics of the problem. For example, when simulating a train passing through a bridge, the initial flow field may be assumed to be a stationary state or a uniform flow state, which provides a benchmark for subsequent dynamic simulations; after the initial variable definition process, the initial flow field model is obtained. In this step, by defining the initial state of the flow field, the necessary starting conditions are provided for solving the approximate solution of the control equation, and these initial conditions must be physically true to ensure the reliability of the simulation results. The next step is to define the boundary conditions, where the boundary conditions are an important part of the numerical method, which tells the solver the information of the region boundary. In CFD simulation, boundary conditions can be the value of the physical variable to be determined on a given boundary, the normal gradient of the physical quantity on a given boundary, or the Robin type (a mixture of the two). The choice depends on geometric considerations, the characteristics of the physical problem, the physical information that the boundary position can provide, and the consideration of numerical calculations. The resulting boundary flow field model ensures the correct behavior of the flow field at the boundary and provides the necessary constraints for the simulation.

[0062] S206. Based on the initial flow field model and the boundary flow field model, the flow field distribution model is obtained through iterative calculation of the NS equation and the turbulence equation. The calculation formula of the turbulence equation is as follows:

[0063]

[0064] In the formula, G k is the kinetic energy of turbulence, G w is the w equation, Γ K , Γ w are the effective diffusion terms of k and w, respectively, K , Y w are the divergent terms of k and w, respectively, D w is the orthogonal divergence term, S K , S w Can be customized by the user.

[0065] It should be noted that in this step, initial variables and boundary conditions are defined in the computational fluid dynamics software, and iterative calculations are performed by solving the NS equations and turbulence equations.

[0066] That is, in the rectangular coordinate system, the unsteady incompressible NS equations are:

[0067]

[0068] Where ρ is the air density, u, v, and w are the three components of the velocity in the rectangular coordinate system O-xyz, respectively, which is a time variable, p is the air static pressure, and μ is the aerodynamic viscosity coefficient.

[0069] S300, processing the time history of aerodynamic coefficients of trains with different wind barrier parameters to obtain the aerodynamic mutation amplitude of the train, and then obtaining a data set for neural network training.

[0070] It can be understood that step S300 includes step S301, wherein:

[0071] S301. Based on the time history of train resistance, lift and lift moment coefficients under different wind barrier parameters, the drag coefficient data of the train under different wind barrier parameter settings are collected and processed to obtain the mutation amplitude of the aerodynamic coefficient when the train passes through the tower area. After data set construction and processing, a data set for neural network training is obtained, in which the height and porosity of the wind barrier are adjustable variable parameters.

[0072] It can be understood that the generated wind barrier geometry model with different parameters is added to the tower area train-bridge model for calculation. The results of CFD calculation constitute the data set for neural network training, which is input into a deep learning framework. Using the powerful learning ability of neural network, the model is trained to identify and predict the impact of different wind barrier parameters on the tower area train-bridge system, such as Figure 6 shown.

[0073] It should be noted that the tower area train-bridge-wind barrier model in the CFD batch calculation data set is used to obtain the resistance lift and lift moment of the corresponding train under different wind barrier parameters. The mutation amplitude of the three forces of the train is the target of model optimization, where the mutation amplitude is defined as the impact width and mutation amplitude, such as Figure 5 As shown, the goal of this application is to optimize the design of wind barriers to reduce the drastic changes in the aerodynamic force of the train. Specifically, the height and porosity of the wind barrier are adjustable variable parameters, which have a significant impact on the sudden change of wind load in the tower area. This sudden change is characterized by the impact width (d) and sudden change amplitude (Y), and the optimization goal is to reduce the maximum sudden change amplitude Y of the drag coefficient. max , which is the higher value of Y1 and Y2, and optimizes the model performance by adjusting the wind barrier parameters to minimize Ymax.

[0074] In this step, the static wind force of the train is only related to the average wind speed of the incoming flow. In the body axis coordinate system, the unit static wind load acting on the bridge and the train is expressed as:

[0075]

[0076] Where D0, L0, and M0 are the static wind resistance, lift, and lift moment per unit length in the body axis coordinate system, respectively; ρ is the air density, which is generally taken as 1.225 kg / m 3 ; U is the average wind speed, in m / s; α0 is the wind attack angle; H and B are the height and width of the bridge or train section; C D (a0), C L (a0), C M (a0) are the three-force coefficients of the main beam or train in the body axis coordinate system under the wind attack angle of α0, which are obtained based on numerical simulation, such as Figure 7 shown.

[0077] S400, training and testing the data set to determine whether the training and test results meet the termination conditions. If so, the optimal parameters are obtained to complete the optimization of the wind barrier in the bridge tower area. If not, the training is continued until the termination conditions are met. The training method adopts a deep deterministic policy gradient algorithm, wherein the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of the design variable changes.

[0078] It can be understood that step S400 includes S401 and S402, wherein:

[0079] S401, the data set is trained and tested, wherein the training includes using a deep deterministic policy gradient algorithm, wherein as the Actor network parameter θμ changes, the change calculation formula of the design variable is as follows:

[0080] α=π(sθ μ )

[0081] In the formula, α is the deterministic action output under the current state s, and π is the strategy function, which is implemented by the Actor network;

[0082] S402 uses the same reward function r during training t To evaluate the quality of taking a specific action in a specific state, the calculation formula is as follows:

[0083]

[0084] Among them, Δf(s t )-Δf(s0) is the current design configuration s t Performance improvement relative to the baseline design configuration s0, f(s) is the mutation amplitude of the train resistance coefficient, and Δf(s) is the change in this indicator.

[0085] It should be noted that this setting encourages the algorithm to find a design configuration that can reduce the sudden change in the train drag coefficient. The baseline refers to the sudden change in the train drag coefficient without a wind barrier. In order to avoid local optimal solutions, the long-term reward is defined as follows:

[0086]

[0087] Where γ is the reward discount factor used to balance the importance of immediate rewards and future rewards, N is the time horizon or the total number of time steps, and r t+n is the reward obtained at time step t+n.

[0088] The value function of the design is denoted as V θ , which represents the expected long-term reward according to the current policy θ in state s.

[0089] V θ (s) = E θ [G t |S t =s]

[0090] The value function of the design variable change is the Q function (Qθ), which represents the expected long-term reward after taking action a in state s.

[0091] Q θ (s) = E θ [G t |S t =s,A t =α]

[0092] The “Actor” network is parameterized by μ As an optimization experience, the "Actor" network parameters are updated by gradient ascent method to maximize the expected long-term reward, which is calculated as follows:

[0093]

[0094] The “Critic” network function evaluates the selection and the network parameters are updated via a loss (LOSS) function, which calculates the difference between the reward value predicted by the Critic network and the actual reward value (plus discounted future reward predictions).

[0095]

[0096] Check whether the termination condition is met and output the result. In summary, the goal of DDPG algorithm training is to maximize the reward function rt while minimizing the loss of the value network Q.

[0097] Embodiment 2:

[0098] like Figure 2 As shown, this embodiment provides a railway bridge tower area wind barrier optimization system based on deep learning, see Figure 2 The system comprises:

[0099] Acquisition module 701: used to acquire target main beams, bridge towers, trains and their structural parameters, and obtain a tower area train-bridge model including bridge sections, bridge deck pavement, bridge deck anti-collision railings, bridge towers and main components of the train through three-dimensional bridge model construction processing according to the structural parameters of the target bridge;

[0100] The first processing module 702 is used to process the train-bridge model including the bridge tower by using the adaptive grid refinement technology to obtain an overlapping grid model for simulating the movement of the train; based on the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated by the NS equation and the turbulence equation, so as to obtain the aerodynamic coefficient time history of the corresponding train under different parameters;

[0101] The second processing module 703 is used to process the aerodynamic coefficient time history of the train under different wind barrier parameters to obtain the aerodynamic mutation amplitude of the train, and then obtain a data set for neural network training;

[0102] Judgment module 704: used to train and test the data set, and judge whether the training and test results of the iterative calculation meet the termination conditions. If so, the optimal parameters are obtained, and then the optimization of the wind barrier in the bridge tower area is completed; if not, the training is continued until the termination conditions are met; the training method adopts a deep deterministic policy gradient algorithm, and the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of the design variable changes.

[0103] Specifically, the first processing module 702 includes:

[0104] The first processing unit is used to obtain the grid models of the main area and the subsidiary area according to the tower area train-bridge model through the grid division processing in the software, wherein the main area includes the bridge part and the outer area, and the subsidiary area includes the moving train part;

[0105] The second processing unit is used to process the grid model through the adaptive grid refinement technology to obtain a dynamically updated overlapping boundary model;

[0106] The third processing unit is used to obtain the flow field data exchange model between the main area and the subsidiary area through flow field information interpolation processing based on the overlapping grid model.

[0107] Specifically, the first processing module 702 includes:

[0108] The fourth processing unit is used for obtaining an initial flow field model according to the overlapping grid model after initial variable definition processing;

[0109] The fifth processing unit is used for obtaining a boundary flow field model according to the boundary conditions through boundary condition definition processing;

[0110] The sixth processing unit is used to obtain the flow field distribution model based on the initial flow field model and the boundary flow field model through iterative calculation processing of the NS equation and the turbulence equation. The calculation formula of the turbulence equation is as follows:

[0111]

[0112] In the formula, G k is the kinetic energy of turbulence, G w is the w equation, Γ K , Γ w are the effective diffusion terms of k and w, respectively, K , Y w are the divergent terms of k and w, respectively, D w is the orthogonal divergence term, S K , S w Can be customized by the user.

[0113] Specifically, the second processing module 703 includes:

[0114] Acquisition unit: used for the time history of train resistance, lift and lift moment coefficients based on different wind barrier parameters, wherein the drag coefficient data of the train under different wind barrier parameter settings are collected and processed to obtain the mutation amplitude of the aerodynamic coefficient when the train passes through the tower area. After data set construction and processing, a data set for neural network training is obtained, wherein the height and porosity of the wind barrier are adjustable variable parameters.

[0115] Specifically, the judgment module 704, wherein the training includes adopting a deep deterministic policy gradient algorithm, wherein as the Actor network parameter θμ changes, the change calculation formula of the design variable is as follows:

[0116] α=π(sθ μ )

[0117] In the formula, α is the deterministic action output under the current state s, and π is the strategy function, which is implemented by the Actor network;

[0118] Evaluation unit: used to use the same reward function r during training t To evaluate the quality of taking a specific action in a specific state, the calculation formula is as follows:

[0119]

[0120] Among them, Δf(s t )-Δf(s0) is the current design configuration s t Performance improvement relative to the baseline design configuration s0, f(s) is the mutation amplitude of the train resistance coefficient, and Δf(s) is the change in this indicator.

[0121] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0122] Embodiment 3:

[0123] Corresponding to the above method embodiment, a railway bridge tower area wind barrier optimization device based on deep learning is also provided in this embodiment. The railway bridge tower area wind barrier optimization device based on deep learning described below and the railway bridge tower area wind barrier optimization method based on deep learning described above can refer to each other.

[0124] Figure 3 8 is a block diagram of a railway bridge tower area wind barrier optimization device 800 based on deep learning according to an exemplary embodiment. Figure 3 As shown, the railway bridge tower area wind barrier optimization device 800 based on deep learning includes: a processor 801 and a memory 802. The railway bridge tower area wind barrier optimization device 800 based on deep learning also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0125] The processor 801 is used to control the overall operation of the railway bridge tower area wind barrier optimization device 800 based on deep learning to complete all or part of the steps in the above-mentioned railway bridge tower area wind barrier optimization method based on deep learning. The memory 802 is used to store various types of data to support the operation of the railway bridge tower area wind barrier optimization device 800 based on deep learning, and these data may include, for example, instructions for any application or method operating on the railway bridge tower area wind barrier optimization device 800 based on deep learning, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse or a button, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the deep learning-based railway bridge tower area wind barrier optimization device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: a Wi-Fi module, a Bluetooth module or an NFC module.

[0126] In an exemplary embodiment, the railway bridge tower area wind barrier optimization device 800 based on deep learning can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned railway bridge tower area wind barrier optimization method based on deep learning.

[0127] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned railway bridge tower area wind barrier optimization method based on deep learning are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the railway bridge tower area wind barrier optimization device 800 based on deep learning to complete the above-mentioned railway bridge tower area wind barrier optimization method based on deep learning.

[0128] Embodiment 4:

[0129] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the deep learning-based railway bridge tower area wind barrier optimization method described above can refer to each other.

[0130] A computer program is stored on the readable storage medium, and when the computer program is executed by the processor, the steps of the railway bridge tower area wind barrier optimization method based on deep learning in the above-mentioned method embodiment are implemented.

[0131] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0133] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A railway bridge tower area wind barrier optimization method based on deep learning, characterized in that: include: Obtain the target main beam, bridge tower, train and their structural parameters. According to the structural parameters of the target bridge, after the three-dimensional bridge model is constructed, a tower area train-bridge model including the bridge section, bridge deck pavement, bridge deck anti-collision railings, bridge tower and main components of the train is obtained; The train-bridge model including the bridge tower is processed by adaptive mesh refinement technology to obtain an overlapping mesh model for simulating train movement. Based on the overlapping mesh model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated through the NS equation and turbulence equation to obtain the aerodynamic coefficient time history of the corresponding train under different parameters. The aerodynamic coefficient time history of trains under different wind barrier parameters is processed to obtain the aerodynamic mutation amplitude of the train, and then the data set for neural network training is obtained; Train and test the data set to determine whether the training and test results meet the termination conditions. If so, the optimal parameters are obtained to complete the optimization of the wind barrier in the bridge tower area. If not, continue training until the termination conditions are met. The training method adopts a deep deterministic policy gradient algorithm, where the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of changes in design variables.

2. The railway bridge tower area wind barrier optimization method based on deep learning according to claim 1 is characterized in that: The train-bridge model including the bridge tower is processed by using the adaptive mesh refinement technology to obtain an overlapping mesh model for simulating train movement, which includes: According to the tower area train-bridge model, after meshing in the software, the mesh models of the main area and the subsidiary area are obtained, where the main area includes the bridge part and the outer area, and the subsidiary area includes the moving train part; The grid model is processed by adaptive grid refinement technology to obtain a dynamically updated overlapping boundary model; Based on the overlapping grid model, the flow field data exchange model between the main area and the subsidiary area is obtained after flow field information interpolation processing.

3. The railway bridge tower area wind barrier optimization method based on deep learning according to claim 1 is characterized in that: In the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated through the NS equation and the turbulence equation, thereby obtaining the aerodynamic coefficient time history of the corresponding train under different parameters, including: According to the overlapping grid model, after the initial variable definition processing, the initial flow field model is obtained; According to the boundary conditions, after the boundary condition definition processing, the boundary flow field model is obtained; Based on the initial flow field model and the boundary flow field model, the flow field distribution model is obtained through iterative calculation of the NS equation and the turbulence equation. The calculation formula of the turbulence equation is as follows: In the formula, G k is the kinetic energy of turbulence, G w is the w equation, Γ K , Γ w are the effective diffusion terms of k and w, respectively, K , Y w are the divergent terms of k and w, respectively, D w is the orthogonal divergence term, S K , S w Can be customized by the user.

4. The railway bridge tower area wind barrier optimization method based on deep learning according to claim 1 is characterized in that: The aerodynamic coefficient time history of trains under different wind barrier parameters is processed to obtain the aerodynamic mutation amplitude of the train, and then obtain a data set for neural network training, which includes: Based on the time history of train drag, lift and lift moment coefficients under different wind barrier parameters, the drag coefficient data of the train under different wind barrier parameter settings are collected and processed to obtain the mutation amplitude of the aerodynamic coefficient when the train passes through the tower area. After data set construction and processing, the data set for neural network training is obtained, in which the height and porosity of the wind barrier are adjustable variable parameters.

5. The railway bridge tower area wind barrier optimization method based on deep learning according to claim 1 is characterized in that: The data set is trained and tested, wherein the training includes using a deep deterministic policy gradient algorithm, wherein as the Actor network parameter θμ changes, the change calculation formula of the design variable is as follows: α=π(sθ μ ) In the formula, α is the deterministic action output under the current state s, and π is the strategy function, which is implemented by the Actor network; Use the same reward function r during training t To evaluate the quality of taking a specific action in a specific state, the calculation formula is as follows: Among them, Δf(s t )-Δf(s0) is the current design configuration s t Performance improvement relative to the baseline design configuration s0, f(s) is the mutation amplitude of the train resistance coefficient, and Δf(s) is the change in this indicator.

6. A railway bridge tower area wind barrier optimization system based on deep learning, based on the railway bridge tower area wind barrier optimization method based on deep learning according to claim 1, characterized in that: include: Acquisition module: used to obtain the target main beam, bridge tower, train and their structural parameters. According to the structural parameters of the target bridge, after the three-dimensional bridge model is constructed and processed, a tower area train-bridge model including the bridge section, bridge deck pavement, bridge deck anti-collision railings, bridge tower and main components of the train is obtained; The first processing module is used to process the train-bridge model including the bridge tower by using the adaptive grid refinement technology to obtain an overlapping grid model for simulating the movement of the train; based on the overlapping grid model, the initial variables and boundary conditions of the model are defined in the computational fluid dynamics software, and the model is iteratively calculated by the NS equation and the turbulence equation, thereby obtaining the aerodynamic coefficient time history of the corresponding train under different parameters; The second processing module is used to process the aerodynamic coefficient time history of trains under different wind barrier parameters to obtain the aerodynamic mutation amplitude of the train, and then obtain a data set for neural network training; Judgment module: used to train and test the data set, and judge whether the training and test results of the iterative calculation meet the termination conditions. If so, the optimal parameters are obtained, and then the optimization of the wind barrier in the bridge tower area is completed; if not, the training continues until the termination conditions are met; The training method adopts a deep deterministic policy gradient algorithm, where the network training goal of the deep deterministic policy gradient algorithm is to extract optimization experience and map the design configuration to the probability distribution of changes in design variables.

7. The railway bridge tower area wind barrier optimization system based on deep learning according to claim 6 is characterized in that: The first processing module includes: The first processing unit is used to obtain the grid models of the main area and the subsidiary area according to the tower area train-bridge model through the grid division processing in the software, wherein the main area includes the bridge part and the outer area, and the subsidiary area includes the moving train part; The second processing unit is used to process the grid model through the adaptive grid refinement technology to obtain a dynamically updated overlapping boundary model; The third processing unit is used to obtain the flow field data exchange model between the main area and the subsidiary area through flow field information interpolation processing based on the overlapping grid model.

8. The railway bridge tower area wind barrier optimization system based on deep learning according to claim 6 is characterized in that: The first processing module includes: The fourth processing unit is used for obtaining an initial flow field model according to the overlapping grid model after initial variable definition processing; The fifth processing unit is used for obtaining a boundary flow field model according to the boundary conditions through boundary condition definition processing; The sixth processing unit is used to obtain the flow field distribution model based on the initial flow field model and the boundary flow field model through iterative calculation processing of the NS equation and the turbulence equation. The calculation formula of the turbulence equation is as follows: In the formula, G k is the kinetic energy of turbulence, G w is the w equation, Γ K , Γ w are the effective diffusion terms of k and w, respectively, K , Y w are the divergent terms of k and w, respectively, D w is the orthogonal divergence term, S K , S w Can be customized by the user.

9. The railway bridge tower area wind barrier optimization system based on deep learning according to claim 6, characterized in that: The second processing module includes: Acquisition unit: used for the time history of train resistance, lift and lift moment coefficients based on different wind barrier parameters, wherein the drag coefficient data of the train under different wind barrier parameter settings are collected and processed to obtain the mutation amplitude of the aerodynamic coefficient when the train passes through the tower area. After data set construction and processing, a data set for neural network training is obtained, wherein the height and porosity of the wind barrier are adjustable variable parameters.

10. The railway bridge tower area wind barrier optimization system based on deep learning according to claim 6, characterized in that: The judgment module, wherein the training includes adopting a deep deterministic policy gradient algorithm, wherein as the Actor network parameter θμ changes, the change calculation formula of the design variable is as follows: α=π(sθ μ ) In the formula, α is the deterministic action output under the current state s, and π is the strategy function, which is implemented by the Actor network; Evaluation unit: used to use the same reward function r during training t To evaluate the quality of taking a specific action in a specific state, the calculation formula is as follows: Among them, Δf(s t )-Δf(s0) is the current design configuration s t Performance improvement relative to the baseline design configuration s0, f(s) is the mutation amplitude of the train resistance coefficient, and Δf(s) is the change in this indicator.

Images