WO2021253666A1 - 一种流场数据可视化方法、装置、设备及存储介质 - Google Patents

一种流场数据可视化方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2021253666A1
WO2021253666A1 PCT/CN2020/117023 CN2020117023W WO2021253666A1 WO 2021253666 A1 WO2021253666 A1 WO 2021253666A1 CN 2020117023 W CN2020117023 W CN 2020117023W WO 2021253666 A1 WO2021253666 A1 WO 2021253666A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
source
field
feedback force
fluid velocity
Prior art date
Application number
PCT/CN2020/117023
Other languages
English (en)
French (fr)
Inventor
李茹杨
彭慧民
赵雅倩
李仁刚
Original Assignee
广东浪潮智慧计算技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 广东浪潮智慧计算技术有限公司 filed Critical 广东浪潮智慧计算技术有限公司
Publication of WO2021253666A1 publication Critical patent/WO2021253666A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00Manipulating 3D models or images for computer graphics
    • G06T19/006Mixed reality
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00Indexing scheme for image generation or computer graphics
    • G06T2210/24Fluid dynamics
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00Indexing scheme for image generation or computer graphics
    • G06T2210/28Force feedback

Definitions

  • This application relates to the field of computer application technology, in particular to a method, device, equipment and storage medium for visualization of flow field data.
  • VR Virtual Reality
  • VR technology has been introduced in the research and application of fluid mechanics, which can directly observe various flow structures in the flow field, such as the formation, development and fragmentation of vortex structures, as well as the motion behavior of a large number of solid particles entrapped in the fluid.
  • VR technology not only contributes to the scientific understanding of flow issues, but also helps guide related industrial production. For example, through VR technology, the introduction speed and concentration of raw materials can be visually observed to produce products with more stable performance.
  • the flow field data calculated by computer numerical simulation methods are obtained, such as fluid velocity information and VR feedback force (that is, the force information of the solid in the flow field), and the VR technology is used to further guide research and production.
  • the purpose of this application is to provide a flow field data visualization method, device, equipment, and storage medium, so as to provide accurate VR feedback force and facilitate the visualization of flow field data.
  • a method for visualizing flow field data including:
  • the data set of the source field includes at least a source fluid velocity sample and a source VR feedback force label, and a group of the source fluid velocity samples corresponds to one source VR feedback force label
  • the data set of the target field includes at least target fluid velocity samples
  • the source fluid velocity sample, and the target fluid velocity sample migrating the source VR feedback force tag to predict and obtain the target VR feedback force in the target field;
  • a VR visualization display of the flow field data of the target field is performed.
  • the source VR feedback force tag is migrated based on the optimization target, the source fluid velocity sample, and the target fluid velocity sample, and the target field is predicted to be obtained
  • the target VR feedback power includes:
  • the target VR feedback force of the target field is predicted to be obtained on the mapping result of the target fluid velocity sample.
  • mapping result based on the source fluid velocity sample and the source VR feedback force label are predicted to obtain the target field based on the mapping result of the target fluid velocity sample Target VR feedback capabilities, including:
  • the target VR feedback force in the target field is predicted to be obtained on the mapping result of the target fluid velocity sample.
  • the solving the optimization target and determining the transformation matrix includes:
  • At least one of regularized linear regression, support vector machine, and principal component analysis is used to solve the optimization target and determine the transformation matrix.
  • said performing joint distribution adaptation on the data set of the source field and the data set of the target field to obtain the optimization target includes:
  • the performing edge probability distribution adaptation on the data set of the source field and the data set of the target field to obtain the first distance includes:
  • the maximum mean variance is used to adapt the edge probability distributions of the data set of the source field and the data set of the target field to obtain the first distance.
  • the performing class-conditional probability distribution adaptation on the data set of the source field and the data set of the target field to obtain the second distance includes:
  • the class condition probability distribution is adapted for each category to obtain the second distance.
  • the classification processing of the source fluid velocity sample and the target fluid velocity sample includes:
  • the average values of the source fluid velocity samples and the target fluid velocity samples are sorted from large to small, and are evenly divided into multiple categories.
  • Also after the prediction to obtain the target VR feedback force of the target field, before the VR visualization display of the flow field information of the target field according to the target VR feedback force ,Also includes:
  • a visualization device for flow field data including:
  • the data set obtaining module is used to obtain the data set of the source field and the data set of the target field.
  • the data set of the source field includes at least a source fluid velocity sample and a source VR feedback force label.
  • a group of the source fluid velocity samples corresponds to one
  • the data set of the target field includes at least target fluid velocity samples;
  • the optimization target obtaining module is configured to perform joint distribution adaptation on the data set of the source field and the data set of the target field to obtain the optimization target;
  • the VR feedback force prediction module is used to migrate the source VR feedback force tag based on the optimization target, the source fluid velocity sample, and the target fluid velocity sample, and predict to obtain the target VR feedback force in the target field ;
  • the visual display module is used to perform VR visual display of the flow field data in the target field according to the target VR feedback force.
  • a flow field data visualization device including:
  • Memory used to store computer programs
  • the processor is configured to implement the steps of any one of the above-mentioned flow field data visualization methods when the computer program is executed.
  • a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, realizes the steps of any one of the above-mentioned flow field data visualization methods.
  • the data sets of the two fields are jointly distributed and adapted to obtain the optimization target, based on the optimization target and the source fluid velocity sample And the target fluid velocity sample, migrate the source VR feedback force label, predict and obtain the target VR feedback force in the target field, so as to visualize the flow field data in the target field according to the target VR feedback force. That is, by migrating the existing feedback force in the source field, the accurate VR feedback force in the target field can be obtained, so as to facilitate the corresponding VR visualization display.
  • FIG. 1 is an implementation flowchart of a method for visualizing flow field data in an embodiment of the application
  • FIG. 2 is a schematic diagram of the VR feedback force prediction process in an embodiment of the application
  • FIG. 3 is a schematic structural diagram of a flow field data visualization device in an embodiment of the application.
  • Fig. 4 is a schematic structural diagram of a flow field data visualization device in an embodiment of the application.
  • an implementation flowchart of a method for visualizing flow field data provided by an embodiment of this application, the method may include the following steps:
  • the data set of the source domain includes at least a source fluid velocity sample and a source VR feedback force label.
  • a set of source fluid velocity samples corresponds to a source VR feedback force label
  • the data set of the target domain includes at least a target fluid velocity sample.
  • the source domain refers to the solid-fluid two-phase flow problem in which all data spaces have been obtained by high-precision solution, that is, the two-phase flow problem has been solved.
  • the data space here can include fluid velocity samples and VR feedback forces.
  • the target field refers to the solid-fluid two-phase flow problem that needs to be solved, that is, the two-phase flow problem to be solved.
  • the source domain can also be called the existing knowledge domain, and the target domain can also be called the unsolved domain.
  • This application is to obtain the VR feedback force of the visualization technology applied to the solid-fluid two-phase flow problem by migrating the source field to the target field.
  • the source domain and target domain are concepts in transfer learning, and each domain includes samples and labels.
  • the data set of the source field and the data set of the target field can be obtained first, and the source field can be migrated to the target field. Specifically, the corresponding data set can be obtained through numerical simulation.
  • the data set in the source domain can be represented by D s , which corresponds to the entire data space of the solid-fluid two-phase flow problem that has been obtained through numerical simulation, including the source fluid velocity sample u s and the source VR feedback force label f s .
  • the data set of the target field can be represented by D t , which is the entire data space of the problem to be solved, including the target fluid velocity sample u t , and the VR feedback force f t data is missing, which is a label that needs to be predicted.
  • the method of accurately solving the force of an object requires a higher calculation grid resolution, and accordingly, it will bring higher calculation costs. Therefore, in the embodiment of the present application, the method of accurately solving the force of the object can be used only in the acquisition of the data set of the source domain to obtain the source fluid velocity sample u s and the source VR feedback force label f s near the surface of the object.
  • the obtained data set of the source field can be used as the benchmark data to be reused in different target problems.
  • a sparse grid can be used to calculate the target fluid velocity sample u t around the surface of the object.
  • a set of source fluid velocity samples u s and a source VR feedback force label f s can be obtained; in the target field, a set of target fluid velocity samples u t can be obtained, but there is no corresponding one VR feedback force label. That is, in each computer numerical simulation calculation step, a set of source fluid velocity samples u s and a source VR feedback force label f s are obtained by solving the two-phase flow problem corresponding to the source field, and the two-phase flow problem corresponding to the target field is solved A set of target fluid velocity samples u t . Select the numerical simulation results at different moments for the two fields, and accumulate to obtain the complete data sets of the source field and the target field, namely D s and D t .
  • step S120 After obtaining the data set of the source field and the data set of the target field, the operation of step S120 can be continued.
  • S120 Perform joint distribution adaptation on the data set of the source field and the data set of the target field to obtain an optimization target.
  • the data set of the source field and the data set of the target field can be jointly distributed and adapted, such as edge probability distribution adaptation, condition Probability distribution adaptation, etc., to obtain the optimization goal.
  • Joint distribution adaptation can be expressed as JDA (Joint Distribution Adaptation).
  • S130 Based on the optimization target, the source fluid velocity sample, and the target fluid velocity sample, migrate the source VR feedback force tag to predict and obtain the target VR feedback force in the target field.
  • the source fluid velocity sample and the target fluid velocity sample After the optimization target is obtained, based on the optimization target, the source fluid velocity sample and the target fluid velocity sample, the source VR feedback force tag is migrated, and the target VR feedback force in the target field is predicted to be obtained.
  • step S130 may include the following steps:
  • the first step Solve the optimization goal and determine the transformation matrix
  • the second step use the transformation matrix to obtain the mapping result of the source fluid velocity sample and the target fluid velocity sample;
  • the third step based on the mapping result of the source fluid velocity sample and the source VR feedback force label, predict the target VR feedback force in the target field based on the mapping result of the target fluid velocity sample.
  • the optimization goal can be solved. Specifically, methods such as regularized linear regression, support vector machines, and principal component analysis can be used to solve the optimization objective, so that the transformation matrix A can be obtained.
  • this application is to find a suitable transformation matrix A to minimize the joint distribution distance of the data set D s in the source field and the data set D t in the target field, thereby realizing migration and predicting the VR feedback force in the target field.
  • the distance between the two fields can be expressed as:
  • P * (u * ) is the marginal probability distribution
  • u * ) is the conditional probability distribution
  • the two distributions are collectively called the joint probability distribution.
  • mapping results of the fluid velocity samples in the source domain and the target domain can be obtained, that is, the mapping result u sA of the source fluid velocity sample u s and the mapping result u tA of the target fluid velocity sample u t can be obtained.
  • Mapping result obtained source fluid velocity sample u s after u sA can u sA and source VR feedback force labels f s based on the mapping result of source fluid velocity of the sample, the prediction to obtain the target VR target areas on a mapping result the target velocity of the fluid sample Feedback force.
  • a neural network (Neural Network, NN) can be used to train a regression model, and the model parameters can be updated through Back Propagation (BP).
  • BP Back Propagation
  • the target VR feedback force in the target field is predicted based on the mapping result of the target fluid velocity sample. That is, after the target regression model is trained, it can be directly predicted on the mapping result u tA of the target fluid velocity sample u t to accurately obtain the target VR feedback force
  • the mapping result u tA of the target fluid velocity sample u t can be used as input, and the target VR feedback force can be obtained through the calculation processing of the target regression model F which is
  • S140 Perform a VR visual display of the flow field data in the target field according to the target VR feedback force.
  • the flow field data in the target field can be visualized in VR based on the target VR feedback force and target fluid velocity samples.
  • the data sets of the two fields are jointly distributed and adapted to obtain the optimization target, which is based on the optimization target, the source fluid velocity sample and Target fluid velocity samples, migrate the source VR feedback force label, predict and obtain the target VR feedback force in the target field, so as to visualize the flow field data in the target field according to the target VR feedback force. That is, by migrating the existing feedback force in the source field, the accurate VR feedback force in the target field can be obtained, so as to facilitate the corresponding VR visualization display.
  • step S120 performs joint distribution adaptation on the data set of the source field and the data set of the target field to obtain the optimization target, which may include the following steps:
  • Step 1 Perform edge probability distribution adaptation on the data set of the source field and the data set of the target field to obtain the first distance;
  • Step 2 Perform class conditional probability distribution adaptation on the data set of the source field and the data set of the target field to obtain the second distance;
  • Step 3 Combine the first distance and the second distance to obtain the optimization target.
  • the target domain may be the data set D t D s of the source and target areas in the field of data sets probability distribution edge adapted to obtain the source field data sets, to obtain a first distance.
  • MMD Maximum Mean Discrepancy
  • RKHS Reproducing Kernel Hilbert Space
  • the first distance obtained can be expressed as:
  • n and m are the number of samples in the source field and target field, respectively, Indicates that the first distance is calculated in Hilbert space, and A represents the transformation matrix.
  • the target domain may be the data set D t D s of the source and target areas in the field of data sets for class conditional probability distribution further adapted to obtain the source field of data sets, to obtain a second distance.
  • the embodiment of the application adopts the class conditional distribution Q s (u s
  • the initial VR feedback force of the target field can be determined first Specifically, the initial VR feedback force can be set based on experience or historical data. The initial VR feedback force is determined as the adaptive VR feedback force.
  • the source fluid velocity samples and the target fluid velocity samples can be classified. Specifically, the average values of the source fluid velocity samples and the target fluid velocity samples can be sorted from large to small, and evenly divided into multiple categories. For example, the samples in the two fields are divided into C categories.
  • the class condition probability distribution can be adapted for each category to obtain the second distance. That is, for each category c, the class conditional probability Q s (u s
  • the obtained second distance that is, the MMD distance between classes, can be expressed as:
  • n c and m c are the number of samples from the c-th category in the source field and the target field, respectively.
  • the first distance and the second distance can be combined to obtain the optimization target.
  • the optimization target After the optimization target is obtained, it can be solved to obtain the transformation matrix. With the help of the transformation matrix, the mapping result of the source fluid velocity sample and the target fluid velocity sample can be obtained. Through the mapping result of the source fluid velocity sample and the source VR feedback force The trained target regression model can predict the target VR feedback force based on the mapping result of target fluid velocity samples in the target field.
  • the method may further include the following step:
  • the end condition can be set according to the actual situation. For example, when the number of iterations reaches the set threshold, the set end condition is considered to be reached, or the set end condition is considered to be reached when the deviation of the results of two adjacent iterations is less than the set percentage.
  • the edge probability distribution adaptation can be performed on the data set D s of the source field and the data set D t of the target field first to obtain the first distance, determine the initial VR feedback force of the target field, and determine the initial VR feedback force as the appropriate Equipped with VR feedback force
  • the source fluid velocity sample u s and the target fluid velocity sample u t are classified based on the source fluid velocity sample u s , the source VR feedback force label f s , the target fluid velocity sample u t and the adapted VR feedback force
  • the class condition probability distribution is adapted for each category respectively, and the second distance can be obtained.
  • the optimization target can be obtained.
  • the mapping velocity can be obtained, that is, the mapping result u sA of the source fluid velocity sample u s and the mapping result u tA of the target fluid velocity sample u t .
  • the target regression model F is obtained through training.
  • the target VR feedback force is predicted to be obtained on the mapping result utA of the target fluid velocity sample ut.
  • the first distance and the second distance are combined to obtain the optimization target, the optimization target is solved, and the transformation matrix is determined.
  • the mapping result of the source fluid velocity sample and the target fluid velocity sample can be obtained.
  • the target regression model is obtained through training. Through the target regression model, the target VR feedback force is predicted based on the mapping result of the target fluid velocity sample. The predicted target VR feedback force has been updated.
  • the target VR feedback force after iteration is obtained. After multiple iterations, a more accurate target VR feedback force can be predicted, making the target VR feedback force more accurate. According to the target VR feedback force finally obtained after the iteration, the VR visualization display of the flow field data in the target field can improve the VR visualization display effect.
  • the flow field data calculated by the computer simulation method is usually imported, and the VR technology is used to further guide the research and production.
  • the accuracy of the output flow field velocity and VR feedback force in the numerical simulation stage directly affects the subsequent VR visualization effect.
  • the feedback force is mostly introduced in the form of a force model in order to improve the efficiency of the numerical simulation.
  • the embodiments of this application use the transfer idea of the joint distributed adaptation (JDA) method to narrow the distance between the existing knowledge domain (source domain) and the unsolved domain (target domain).
  • JDA joint distributed adaptation
  • the embodiments of the present application also provide a flow field data visualization device, and the flow field data visualization device described below and the flow field data visualization method described above can be referenced correspondingly.
  • the device may include the following modules:
  • the data set obtaining module 310 is used to obtain the data set of the source field and the data set of the target field.
  • the data set of the source field includes at least a source fluid velocity sample and a source VR feedback force label.
  • a group of source fluid velocity samples corresponds to a source VR feedback Force label
  • the data set of the target field includes at least the target fluid velocity sample;
  • the optimization target obtaining module 320 is configured to perform joint distribution adaptation on the data set of the source field and the data set of the target field to obtain the optimization target;
  • the VR feedback force prediction module 330 is used to migrate the source VR feedback force tag based on the optimization target, the source fluid velocity sample, and the target fluid velocity sample, so as to predict and obtain the target VR feedback force in the target field;
  • the visual display module 340 is used to visually display the flow field data in the target field according to the target VR feedback force.
  • the data sets of the two fields are jointly distributed and adapted to obtain the optimization target, which is based on the optimization target, the source fluid velocity sample and Target fluid velocity samples, migrate the source VR feedback force label, predict and obtain the target VR feedback force in the target field, so as to visualize the flow field data in the target field according to the target VR feedback force. That is, by migrating the existing feedback force in the source field, the accurate VR feedback force in the target field can be obtained, so as to facilitate the corresponding VR visualization display.
  • the VR feedback force prediction module 330 is used to:
  • the target VR feedback force in the target field is predicted based on the mapping result of the target fluid velocity sample.
  • the VR feedback force prediction module 330 is used to:
  • the target VR feedback force in the target field is predicted based on the mapping result of the target fluid velocity sample.
  • the VR feedback force prediction module 330 is used to:
  • At least one of regularized linear regression, support vector machine, and principal component analysis is used to solve the optimization goal and determine the transformation matrix.
  • the optimization target obtaining module 320 is used for:
  • the optimization target obtaining module 320 is used for:
  • the optimization target obtaining module 320 is used for:
  • the class condition probability distribution is adapted for each category to obtain the second distance.
  • the optimization target obtaining module 320 is used for:
  • the average values of the source fluid velocity samples and the target fluid velocity samples are sorted from large to small, and are evenly divided into multiple categories.
  • the target VR feedback force After predicting the target VR feedback force in the target area, and before visualizing the flow field information in the target area according to the target VR feedback force, determine the target VR feedback force as the adaptive VR feedback force;
  • an embodiment of the present application also provides a flow field data visualization device, including:
  • Memory used to store computer programs
  • the processor is used to implement the steps of the flow field data visualization method when the computer program is executed.
  • the flow field data visualization device may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13.
  • the processor 10, the memory 11, and the communication interface 12 all communicate with each other through the communication bus 13.
  • the processor 10 may be a central processing unit (Central Processing Unit, CPU), an application-specific integrated circuit, a digital signal processor, a field programmable gate array, or other programmable logic devices.
  • CPU Central Processing Unit
  • application-specific integrated circuit e.g., an application-specific integrated circuit
  • digital signal processor e.g., a digital signal processor
  • field programmable gate array e.g., a field programmable gate array
  • the processor 10 can call a program stored in the memory 11, and specifically, the processor 10 can perform operations in the embodiment of the flow field data visualization method.
  • the memory 11 is used to store one or more programs, the programs may include program codes, and the program codes include computer operation instructions.
  • the memory 11 stores at least programs for implementing the following functions:
  • the data set of the source field includes at least the source fluid velocity sample and the source VR feedback force label.
  • a group of source fluid velocity samples corresponds to a source VR feedback force label, and the target field data set At least include the target fluid velocity sample;
  • the source fluid velocity sample and the target fluid velocity sample Based on the optimized target, the source fluid velocity sample and the target fluid velocity sample, the source VR feedback force label is migrated, and the target VR feedback force in the target field is predicted to be obtained;
  • VR visual display of the flow field data in the target field is carried out.
  • the memory 11 may include a program storage area and a data storage area.
  • the program storage area may store an operating system and an application program required for at least one function (such as a numerical simulation function and an image playback function). Etc.; the data storage area can store data created during use, such as data set data, forecast data, etc.
  • the memory 11 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device or other volatile solid-state storage devices.
  • the communication interface 13 may be an interface of a communication module for connecting with other devices or systems.
  • FIG. 4 does not constitute a limitation on the flow field data visualization device in the embodiment of the present application.
  • the flow field data visualization device may include more or more than that shown in FIG. 4 Few parts, or combine some parts.
  • the embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned flow field data visualization method are realized .
  • the steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of the two.
  • the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage media.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Graphics (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Indicating Or Recording The Presence, Absence, Or Direction Of Movement (AREA)
  • Aerodynamic Tests, Hydrodynamic Tests, Wind Tunnels, And Water Tanks (AREA)

Abstract

一种流场数据可视化方法、装置、设备及存储介质。该方法包括以下步骤:获得源领域和目标领域的数据集(S110);对源领域和目标领域的数据集进行联合分布适配,获得优化目标(S120);基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力(S130);根据目标VR反馈力对目标领域的流场数据进行VR可视化展示(S140)。该方法通过对源领域中已有的反馈力迁移,可以得到目标领域中准确的VR反馈力,从而方便进行相应的VR可视化展示。

Description

一种流场数据可视化方法、装置、设备及存储介质
本申请要求于2020年06月19日提交至中国专利局、申请号为202010568718.6、发明名称为“一种流场数据可视化方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机应用技术领域,特别是涉及一种流场数据可视化方法、装置、设备及存储介质。
背景技术
随着计算机技术的快速发展、终端设备的轻量化发展,VR(Virtual Reality,虚拟现实)技术也快速发展起来,逐渐应用于沉浸式游戏、制造维修、辅助医疗、科学问题可视化和教育教学演示等领域。
近年来,流体力学研究和应用中引入VR技术,能够直接观测流场中各种流动结构,如涡结构形成、发展和破碎现象,以及流体中裹挟的大量固体颗粒的运动行为。VR技术不仅有助于对流动问题的科学理解,也可以很好地辅助指导相关工业生产,如通过VR技术直观观测原材料的导入速度、浓度等状态,制造出性能更稳定的产品。VR技术应用于流体力学研究和相关工业生产时,本质上面对的是固体-流体两相流问题。在具体实施时,获得通过计算机数值模拟方法计算得到的流场数据,如流体速度信息和VR反馈力(即流场中固体的受力信息),通过VR技术来进一步指导研究和生产。
综上所述,如何准确获得VR反馈力,以便利用VR反馈力进行流场数据的可视化,是目前本领域技术人员急需解决的技术问题。
发明内容
本申请的目的是提供一种流场数据可视化方法、装置、设备及存储介质,以准确VR反馈力,方便进行流场数据的可视化。
为解决上述技术问题,本申请提供如下技术方案:
一种流场数据可视化方法,包括:
获得源领域的数据集和目标领域的数据集,所述源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组所述源流体速度样本对应一个所述源VR反馈力标签,所述目标领域的数据集至少包括目标流体速度样本;
对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标;
基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力;
根据所述目标VR反馈力对所述目标领域的流场数据进行VR可视化展示。
在本申请的一种具体实施方式中,所述基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力,包括:
对所述优化目标进行求解,确定变换矩阵;
利用所述变换矩阵,获得所述源流体速度样本和所述目标流体速度样本的映射结果;
基于所述源流体速度样本的映射结果和所述源VR反馈力标签,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力。
在本申请的一种具体实施方式中,所述基于所述源流体速度样本的映射结果和所述源VR反馈力标签,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力,包括:
基于所述源流体速度样本的映射结果和所述源VR反馈力标签,训练获得目标回归模型;
通过所述目标回归模型,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力。
在本申请的一种具体实施方式中,所述对所述优化目标进行求解,确定变换矩阵,包括:
使用正则化线性回归、支持向量机、主成分分析中的至少一种方式对 所述优化目标进行求解,确定变换矩阵。
在本申请的一种具体实施方式中,所述对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标,包括:
对所述源领域的数据集和所述目标领域的数据集进行边缘概率分布适配,获得第一距离;
对所述源领域的数据集和所述目标领域的数据集进行类条件概率分布适配,获得第二距离;
结合所述第一距离和所述第二距离,获得优化目标。
在本申请的一种具体实施方式中,所述对所述源领域的数据集和所述目标领域的数据集进行边缘概率分布适配,获得第一距离,包括:
使用最大均值方差适配所述源领域的数据集和所述目标领域的数据集的边缘概率分布,获得第一距离。
在本申请的一种具体实施方式中,所述对所述源领域的数据集和所述目标领域的数据集进行类条件概率分布适配,获得第二距离,包括:
确定所述目标领域的初始VR反馈力;
将所述初始VR反馈力确定为适配VR反馈力;
对所述源流体速度样本和所述目标流体速度样本进行分类处理;
基于所述源流体速度样本、所述源VR反馈力标签、所述目标流体速度样本和所述适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离。
在本申请的一种具体实施方式中,所述对所述源流体速度样本和所述目标流体速度样本进行分类处理,包括:
分别将所述源流体速度样本和所述目标流体速度样本的均值按照从大到小排序,均匀分为多个类别。
在本申请的一种具体实施方式中,在所述预测获得所述目标领域的目标VR反馈力之后、所述根据所述目标VR反馈力对所述目标领域的流场信息进行VR可视化展示之前,还包括:
将所述目标VR反馈力确定为所述适配VR反馈力;
迭代执行所述基于所述源流体速度样本、所述源VR反馈力标签、所述目标流体速度样本和所述适配VR反馈力,分别针对每个类别进行类条件概 率分布适配,获得第二距离的步骤,直至达到设定结束条件,获得迭代后的所述目标VR反馈力。
一种流场数据可视化装置,包括:
数据集获得模块,用于获得源领域的数据集和目标领域的数据集,所述源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组所述源流体速度样本对应一个所述源VR反馈力标签,所述目标领域的数据集至少包括目标流体速度样本;
优化目标获得模块,用于对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标;
VR反馈力预测模块,用于基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力;
可视化展示模块,用于根据所述目标VR反馈力对所述目标领域的流场数据进行VR可视化展示。
一种流场数据可视化设备,包括:
存储器,用于存储计算机程序;
处理器,用于执行所述计算机程序时实现上述任一项所述流场数据可视化方法的步骤。
一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机程序,所述计算机程序被处理器执行时实现上述任一项所述流场数据可视化方法的步骤。
应用本申请实施例所提供的技术方案,获得源领域的数据集和目标领域的数据集之后,对两个领域的数据集进行联合分布适配,获得优化目标,基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力,从而根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。即通过对源领域中已有的反馈力迁移,可以得到目标领域中准确的VR反馈力,从而方便进行相应的VR可视化展示。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为本申请实施例中一种流场数据可视化方法的实施流程图;
图2为本申请实施例中VR反馈力预测过程示意图;
图3为本申请实施例中一种流场数据可视化装置的结构示意图;
图4为本申请实施例中一种流场数据可视化设备的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案,下面结合附图和具体实施方式对本申请作进一步的详细说明。显然,所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
参见图1所示,为本申请实施例所提供的一种流场数据可视化方法的实施流程图,该方法可以包括以下步骤:
S110:获得源领域的数据集和目标领域的数据集。
源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组源流体速度样本对应一个源VR反馈力标签,目标领域的数据集至少包括目标流体速度样本。
在本申请实施例中,源领域是指通过高精度求解已经获得全部数据空间的固体-流体两相流问题,即已解决两相流问题,这里的数据空间可以包括流体速度样本和VR反馈力。目标领域是指需要求解的固体-流体两相流问题,即待解决两相流问题。源领域也可称为已有知识领域,目标领域也可称为待解决领域。
本申请是通过将源领域向目标领域迁移,获得可视化技术应用于固体-流体两相流问题的VR反馈力。源领域和目标领域是迁移学习中的概念,每个领域中包括样本和标签。
在需要对目标领域的流场数据进行可视化处理时,可以先获得源领域的数据集和目标领域的数据集,将源领域向目标领域迁移。具体的,可以通过数值模拟的方式获得相应数据集。
源领域的数据集可使用D s表示,对应已经通过数值模拟获得的固体-流体两相流问题的全部数据空间,其中包括源流体速度样本u s,源VR反馈力标签f s。目标领域的数据集可使用D t表示,为待解决问题的全部数据空间,其中包括目标流体速度样本u t,其VR反馈力f t数据缺失,是需要预测的标签。
在实际应用中,精确求解物体受力的方式对计算网格分辨率要求较高,相应地,会带来较高的计算成本。所以,在本申请实施例中,可以只在源领域的数据集的获取中采用精确求解物体受力的方式,获得物体表面附近的源流体速度样本u s和源VR反馈力标签f s。获得的源领域的数据集可作为基准数据在不同目标问题中重复使用。
对于待解决两相流问题对应的目标领域,可以采用较稀疏网格计算得到物体表面周围的目标流体速度样本u t
每一时刻,在源领域中,可以获得一组源流体速度样本u s和一个源VR反馈力标签f s;在目标领域中,可以获得一组目标流体速度样本u t,但没有相应的一个VR反馈力标签。即在每一个计算机数值模拟计算步骤中,源领域对应的两相流问题求解得到的一组源流体速度样本u s和一个源VR反馈力标签f s,目标领域对应的两相流问题求解得到的一组目标流体速度样本u t。针对两个领域分别选取不同时刻的数值模拟结果,积累可得到源领域和目标领域各自完整的数据集,即D s和D t
获得源领域的数据集和目标领域的数据集之后,可以继续执行步骤S120的操作。
S120:对源领域的数据集和目标领域的数据集进行联合分布适配,获得优化目标。
在本申请实施例中,在获得源领域的数据集和目标领域的数据集之后,可以对源领域的数据集和目标领域的数据集进行联合分布适配,如进行边缘概率分布适配、条件概率分布适配等,获得优化目标。
联合分布适配可表示为JDA(Joint Distribution Adaptation)。
S130:基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力。
获得优化目标之后,基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力。
在本申请的一种具体实施方式中,步骤S130可以包括以下步骤:
第一个步骤:对优化目标进行求解,确定变换矩阵;
第二个步骤:利用变换矩阵,获得源流体速度样本和目标流体速度样本的映射结果;
第三个步骤:基于源流体速度样本的映射结果和源VR反馈力标签,在目标流体速度样本的映射结果上预测获得目标领域的目标VR反馈力。
为便于描述,将上述三个步骤结合起来进行说明。
在获得优化目标后,可以对优化目标进行求解。具体的,可以使用正则化线性回归、支持向量机、主成分分析等方法对优化目标进行求解,从而可以得到变换矩阵A。
即本申请是要寻找一个合适的变换矩阵A,使得源领域的数据集D s和目标领域的数据集D t的联合分布距离最小,从而实现迁移,预测出目标领域中的VR反馈力。依据联合分布适配JDA算法的思想,两个领域之间的距离可表示为:
D(D s,D t)≈||P s(u s)-P t(u t)|| 2+||Q s(f s|u s)-Q t(f t|u t)|| 2
其中,P *(u *)为边缘概率分布,Q *(f *|u *)为条件概率分布,两个分布合称联合概率分布。
利用变换矩阵A,可以得到源领域和目标领域的流体速度样本的映射结果,即可获得源流体速度样本u s的映射结果u sA和目标流体速度样本u t的映射结果u tA
具体的,u sA=Au s,u tA=Au t
获得源流体速度样本u s的映射结果u sA之后,可以基于源流体速度样本的映射结果u sA和源VR反馈力标签f s,在目标流体速度样本的映射结果上预测获得目标领域的目标VR反馈力。
具体的,可以先基于源流体速度样本的映射结果和源VR反馈力标签,训练获得目标回归模型。即基于源领域的(u sA,f s)训练出一个简单的目标回归模型F,如多项式回归F(x)=ax 4+bx 3+cx 2+dx+e,其中,a、b、c、d、e为参数,x为输入量。如可以使用神经网络(Neural Network,NN)训练回归模型,通过误差反向传播(Back Propagation,BP)更新模型参数。
然后通过目标回归模型,在目标流体速度样本的映射结果上预测获得目标领域的目标VR反馈力。即在训练获得目标回归模型之后,可以在目标流体速度样本u t的映射结果u tA上直接进行预测,准确获得目标VR反馈力
Figure PCTCN2020117023-appb-000001
具体的,在训练获得目标回归模型F后,可以将目标流体速度样本u t的映射结果u tA作为输入,经过目标回归模型F的计算处理,得到目标VR反馈力
Figure PCTCN2020117023-appb-000002
Figure PCTCN2020117023-appb-000003
S140:根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。
预测获得目标领域的目标VR反馈力之后,可以根据目标VR反馈力及目标流体速度样本,对目标领域的流场数据进行VR可视化展示。
应用本申请实施例所提供的方法,获得源领域的数据集和目标领域的数据集之后,对两个领域的数据集进行联合分布适配,获得优化目标,基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力,从而根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。即通过对源领域中已有的反馈力迁移,可以得到目标领域中准确的VR反馈力,从而方便进行相应的VR可视化展示。
在本申请的一个实施例中,步骤S120对源领域的数据集和目标领域的数据集进行联合分布适配,获得优化目标,可以包括以下步骤:
步骤一:对源领域的数据集和目标领域的数据集进行边缘概率分布适配,获得第一距离;
步骤二:对源领域的数据集和目标领域的数据集进行类条件概率分布适配,获得第二距离;
步骤三:结合第一距离和第二距离,获得优化目标。
为便于描述,将上述三个步骤结合起来进行说明。
获得源领域的数据集D s和目标领域的数据集D t之后,可以对源领域的数据集D s和目标领域的数据集D t进行边缘概率分布适配,获得第一距离。
具体的,可以使用最大均值方差(Maximum Mean Discrepancy,MMD)来适配源领域与目标领域的边缘概率分布,即P s(u s)和P t(u t),在再生核希尔伯特空间(Reproducing Kernel Hilbert Space,RKHS)中最小化两个领域的距离。
得到的第一距离可以表示为:
Figure PCTCN2020117023-appb-000004
其中,n和m分别为源领域和目标领域中的样本数量,
Figure PCTCN2020117023-appb-000005
表示第一距离在希尔伯特(Hilbert)空间中计算,A表示变换矩阵。
获得源领域的数据集D s和目标领域的数据集D t之后,还可以对源领域的数据集D s和目标领域的数据集D t进行类条件概率分布适配,获得第二距离。
由于目标领域中没有VR反馈力标签f t,无法求出目标领域的条件概率,也就无法直接采用MMD对两个领域的条件概率进行适配。所以,本申请实施例采用类条件分布Q s(u s|f s)和
Figure PCTCN2020117023-appb-000006
替代原始的条件概率分布,其中
Figure PCTCN2020117023-appb-000007
是在目标领域中预测的VR反馈力。
在本申请的一种具体实施方式中,可以先确定目标领域的初始VR反馈力
Figure PCTCN2020117023-appb-000008
具体的,可以根据经验或者历史数据等设定初始VR反馈力。将初始VR反馈力确定为适配VR反馈力。同时,可以对源流体速度样本和目标流体速度样本进行分类处理。具体的,可以分别将源流体速度样本和目标流体速度样本的均值按照从大到小排序,均匀分为多个类别。如将两个领域的样本分为C个类别。基于源流体速度样本、源VR反馈力标签、目标流体速度样本和适配VR反馈力,可以分别针对每个类别进行类条件概率分布适配,获得第二距离。即对每个类别c分别进行类条件概率Q s(u s|f s∈c)和
Figure PCTCN2020117023-appb-000009
的适配。获得的第二距离即类与类之间的MMD距离可表示为:
Figure PCTCN2020117023-appb-000010
其中,n c和m c分别为源领域和目标领域中的来自第c个类的样本数量。
获得第一距离和第二距离后,可以结合第一距离和第二距离,获得优化目标。
将两个距离结合起来,得到的总的优化目标可表示为:
Figure PCTCN2020117023-appb-000011
得到优化目标后,对其进行求解,可得到变换矩阵,从而借助于变换矩阵,可以得到源流体速度样本和目标流体速度样本的映射结果,通过针对源流体速度样本的映射结果和源VR反馈力训练得到的目标回归模型,可以在目标领域的目标流体速度样本的映射结果上预测得到目标VR反馈力。
在本申请的一个实施例中,在步骤S130预测获得目标领域的目标VR反馈力之后,在步骤S140根据目标VR反馈力对目标领域的流场信息进行VR可视化展示之前,该方法还可以包括以下步骤:
将目标VR反馈力确定为适配VR反馈力;
迭代执行基于源流体速度样本、源VR反馈力标签、目标流体速度样本和适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离的步骤,直至达到设定结束条件,获得迭代后的目标VR反馈力。然后根据最终得到的目标VR反馈力对目标领域的流场数据进行VR可视化展示。
在实际应用中,可以根据实际情况进行结束条件的设定。如在迭代次数达到设定次数阈值时认为达到设定结束条件,或者,在相邻两次迭代结果偏差小于设定百分比时认为达到设定结束条件。
结合图2,对本申请实施例进行说明。获得源领域的数据集D s和目标领域的数据集D t后,联合分布适配目标
Figure PCTCN2020117023-appb-000012
即对源领域的数据集D s和目标领域的数据集D t进行联合分布适配,获得优化目标。
具体的,可以先对源领域的数据集D s和目标领域的数据集D t进行边缘 概率分布适配,获得第一距离,确定目标领域的初始VR反馈力,将初始VR反馈力确定为适配VR反馈力
Figure PCTCN2020117023-appb-000013
对源流体速度样本u s和目标流体速度样本u t进行分类处理,基于源流体速度样本u s、源VR反馈力标签f s、目标流体速度样本u t和适配VR反馈力
Figure PCTCN2020117023-appb-000014
分别针对每个类别进行类条件概率分布适配,可以获得第二距离。
结合第一距离和第二距离,可以获得优化目标。
对优化目标进行求解,确定变换矩阵A。利用变换矩阵A,可以获得映射速度,即源流体速度样本u s的映射结果u sA和目标流体速度样本u t的映射结果u tA
基于源流体速度样本u s的映射结果u sA和源VR反馈力标签f s,训练获得目标回归模型F。
通过目标回归模型F,预测VR反馈力
Figure PCTCN2020117023-appb-000015
即在目标流体速度样本u t的映射结果u tA上预测获得目标VR反馈力。
将此时得到的VR反馈力确定为适配VR反馈力
Figure PCTCN2020117023-appb-000016
可以迭代执行基于源流体速度样本u s、源VR反馈力标签f s、目标流体速度样本u t和适配VR反馈力
Figure PCTCN2020117023-appb-000017
分别针对每个类别进行类条件概率分布适配,获得第二距离及其以下的步骤。对类条件概率分布进行更新,可以提高联合分布适配质量。
得到新的第二距离后,结合第一距离和第二距离,可以获得优化目标,对优化目标进行求解,确定变换矩阵。利用变换矩阵,可以获得源流体速度样本和目标流体速度样本的映射结果。基于源流体速度样本的映射结果和源VR反馈力标签,训练获得目标回归模型。通过目标回归模型,在目标流体速度样本的映射结果上预测获得目标VR反馈力。预测得到的目标VR反馈力有更新。
直至迭代次数达到设定次数阈值,获得迭代后的目标VR反馈力。经过多次迭代后,可以预测得到更为准确的目标VR反馈力,使得目标VR反馈力精度更高。根据迭代后最终得到的目标VR反馈力对目标领域的流场数据进行VR可视化展示,可以提升VR可视化展示效果。
在当前技术中,在对固体-流体两相流问题进行具体实施时,通常是导入通过计算机模拟方法计算得到的流场数据,通过VR技术进一步指导研 究和生产。数值模拟阶段输出流场速度和VR反馈力的精度,直接影响后续VR可视化的效果。现有数值模拟中,反馈力多采用力模型的形式引入,以便提高数值模拟效率。
而随着机器学习的发展和广泛应用,已有一些两相流研究工作使用监督学习方法针对大量数据进行训练,回归得到一个复杂模型,再面向不同问题预测得到VR反馈力。然而,这些反馈力模型都没有考虑待解决问题中复杂流动和固体群体行为的影响,容易造成VR反馈力与实际情况偏差较大,甚至导致一些关键问题上出现错误。
本申请实施例针对现有VR反馈力模型的不足,借助联合分布适配(JDA)方法的迁移思想,通过缩小已有知识领域(源领域)和待解决领域(目标领域)之间的距离,完成VR反馈力的迁移,准确预测目标领域中VR反馈力,可以实现相关科学研究和工业应用的高精度可视化。
相应于上面的方法实施例,本申请实施例还提供了一种流场数据可视化装置,下文描述的流场数据可视化装置与上文描述的流场数据可视化方法可相互对应参照。
参见图3所示,该装置可以包括以下模块:
数据集获得模块310,用于获得源领域的数据集和目标领域的数据集,源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组源流体速度样本对应一个源VR反馈力标签,目标领域的数据集至少包括目标流体速度样本;
优化目标获得模块320,用于对源领域的数据集和目标领域的数据集进行联合分布适配,获得优化目标;
VR反馈力预测模块330,用于基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力;
可视化展示模块340,用于根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。
应用本申请实施例所提供的装置,获得源领域的数据集和目标领域的数据集之后,对两个领域的数据集进行联合分布适配,获得优化目标,基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签 进行迁移,预测获得目标领域的目标VR反馈力,从而根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。即通过对源领域中已有的反馈力迁移,可以得到目标领域中准确的VR反馈力,从而方便进行相应的VR可视化展示。
在本申请的一种具体实施方式中,VR反馈力预测模块330,用于:
对优化目标进行求解,确定变换矩阵;
利用变换矩阵,获得源流体速度样本和目标流体速度样本的映射结果;
基于源流体速度样本的映射结果和源VR反馈力标签,在目标流体速度样本的映射结果上预测获得目标领域的目标VR反馈力。
在本申请的一种具体实施方式中,VR反馈力预测模块330,用于:
基于源流体速度样本的映射结果和源VR反馈力标签,训练获得目标回归模型;
通过目标回归模型,在目标流体速度样本的映射结果上预测获得目标领域的目标VR反馈力。
在本申请的一种具体实施方式中,VR反馈力预测模块330,用于:
使用正则化线性回归、支持向量机、主成分分析中的至少一种方式对优化目标进行求解,确定变换矩阵。
在本申请的一种具体实施方式中,优化目标获得模块320,用于:
对源领域的数据集和目标领域的数据集进行边缘概率分布适配,获得第一距离;
对源领域的数据集和目标领域的数据集进行类条件概率分布适配,获得第二距离;
结合第一距离和第二距离,获得优化目标。
在本申请的一种具体实施方式中,优化目标获得模块320,用于:
使用最大均值方差适配源领域的数据集和目标领域的数据集的边缘概率分布,获得第一距离。
在本申请的一种具体实施方式中,优化目标获得模块320,用于:
确定目标领域的初始VR反馈力;
将初始VR反馈力确定为适配VR反馈力;
对源流体速度样本和目标流体速度样本进行分类处理;
基于源流体速度样本、源VR反馈力标签、目标流体速度样本和适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离。
在本申请的一种具体实施方式中,优化目标获得模块320,用于:
分别将源流体速度样本和目标流体速度样本的均值按照从大到小排序,均匀分为多个类别。
在本申请的一种具体实施方式中,还包括迭代处理模块,用于:
在预测获得目标领域的目标VR反馈力之后、根据目标VR反馈力对目标领域的流场信息进行VR可视化展示之前,将目标VR反馈力确定为适配VR反馈力;
迭代执行基于源流体速度样本、源VR反馈力标签、目标流体速度样本和适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离的步骤,直至达到设定结束条件,获得迭代后的目标VR反馈力。
相应于上面的方法实施例,本申请实施例还提供了一种流场数据可视化设备,包括:
存储器,用于存储计算机程序;
处理器,用于执行计算机程序时实现上述流场数据可视化方法的步骤。
如图4所示,为流场数据可视化设备的组成结构示意图,流场数据可视化设备可以包括:处理器10、存储器11、通信接口12和通信总线13。处理器10、存储器11、通信接口12均通过通信总线13完成相互间的通信。
在本申请实施例中,处理器10可以为中央处理器(Central Processing Unit,CPU)、特定应用集成电路、数字信号处理器、现场可编程门阵列或者其他可编程逻辑器件等。
处理器10可以调用存储器11中存储的程序,具体的,处理器10可以执行流场数据可视化方法的实施例中的操作。
存储器11中用于存放一个或者一个以上程序,程序可以包括程序代码,程序代码包括计算机操作指令,在本申请实施例中,存储器11中至少存储有用于实现以下功能的程序:
获得源领域的数据集和目标领域的数据集,源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组源流体速度样本对应一个源VR反馈力标签,目标领域的数据集至少包括目标流体速度样本;
对源领域的数据集和目标领域的数据集进行联合分布适配,获得优化目标;
基于优化目标、源流体速度样本和目标流体速度样本,对源VR反馈力标签进行迁移,预测获得目标领域的目标VR反馈力;
根据目标VR反馈力对目标领域的流场数据进行VR可视化展示。
在一种可能的实现方式中,存储器11可包括存储程序区和存储数据区,其中,存储程序区可存储操作系统,以及至少一个功能(比如数值模拟功能、图像播放功能)所需的应用程序等;存储数据区可存储使用过程中所创建的数据,如数据集数据、预测数据等。
此外,存储器11可以包括高速随机存取存储器,还可以包括非易失性存储器,例如至少一个磁盘存储器件或其他易失性固态存储器件。
通信接口13可以为通信模块的接口,用于与其他设备或者系统连接。
当然,需要说明的是,图4所示的结构并不构成对本申请实施例中流场数据可视化设备的限定,在实际应用中流场数据可视化设备可以包括比图4所示的更多或更少的部件,或者组合某些部件。
相应于上面的方法实施例,本申请实施例还提供了一种计算机可读存储介质,计算机可读存储介质上存储有计算机程序,计算机程序被处理器执行时实现上述流场数据可视化方法的步骤。
本说明书中各个实施例采用递进的方式描述,每个实施例重点说明的都是与其它实施例的不同之处,各个实施例之间相同或相似部分互相参见即可。
专业人员还可以进一步意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、计算机软件或者二者的结合来实现,为了清楚地说明硬件和软件的可互换性,在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本申请的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块,或者二者的结合来实施。软件模块可以置于 随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
本文中应用了具体个例对本申请的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本申请的技术方案及其核心思想。应当指出,对于本技术领域的普通技术人员来说,在不脱离本申请原理的前提下,还可以对本申请进行若干改进和修饰,这些改进和修饰也落入本申请权利要求的保护范围内。

Claims (12)

  1. 一种流场数据可视化方法,其特征在于,包括:
    获得源领域的数据集和目标领域的数据集,所述源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组所述源流体速度样本对应一个所述源VR反馈力标签,所述目标领域的数据集至少包括目标流体速度样本;
    对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标;
    基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力;
    根据所述目标VR反馈力对所述目标领域的流场数据进行VR可视化展示。
  2. 根据权利要求1所述的方法,其特征在于,所述基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力,包括:
    对所述优化目标进行求解,确定变换矩阵;
    利用所述变换矩阵,获得所述源流体速度样本和所述目标流体速度样本的映射结果;
    基于所述源流体速度样本的映射结果和所述源VR反馈力标签,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力。
  3. 根据权利要求2所述的方法,其特征在于,所述基于所述源流体速度样本的映射结果和所述源VR反馈力标签,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力,包括:
    基于所述源流体速度样本的映射结果和所述源VR反馈力标签,训练获得目标回归模型;
    通过所述目标回归模型,在所述目标流体速度样本的映射结果上预测获得所述目标领域的目标VR反馈力。
  4. 根据权利要求2所述的方法,其特征在于,所述对所述优化目标进 行求解,确定变换矩阵,包括:
    使用正则化线性回归、支持向量机、主成分分析中的至少一种方式对所述优化目标进行求解,确定变换矩阵。
  5. 根据权利要求1至4之中任一项所述的方法,其特征在于,所述对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标,包括:
    对所述源领域的数据集和所述目标领域的数据集进行边缘概率分布适配,获得第一距离;
    对所述源领域的数据集和所述目标领域的数据集进行类条件概率分布适配,获得第二距离;
    结合所述第一距离和所述第二距离,获得优化目标。
  6. 根据权利要求5所述的方法,其特征在于,所述对所述源领域的数据集和所述目标领域的数据集进行边缘概率分布适配,获得第一距离,包括:
    使用最大均值方差适配所述源领域的数据集和所述目标领域的数据集的边缘概率分布,获得第一距离。
  7. 根据权利要求5所述的方法,其特征在于,所述对所述源领域的数据集和所述目标领域的数据集进行类条件概率分布适配,获得第二距离,包括:
    确定所述目标领域的初始VR反馈力;
    将所述初始VR反馈力确定为适配VR反馈力;
    对所述源流体速度样本和所述目标流体速度样本进行分类处理;
    基于所述源流体速度样本、所述源VR反馈力标签、所述目标流体速度样本和所述适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离。
  8. 根据权利要求7所述的方法,其特征在于,所述对所述源流体速度样本和所述目标流体速度样本进行分类处理,包括:
    分别将所述源流体速度样本和所述目标流体速度样本的均值按照从大到小排序,均匀分为多个类别。
  9. 根据权利要求7所述的方法,其特征在于,在所述预测获得所述目 标领域的目标VR反馈力之后、所述根据所述目标VR反馈力对所述目标领域的流场信息进行VR可视化展示之前,还包括:
    将所述目标VR反馈力确定为所述适配VR反馈力;
    迭代执行所述基于所述源流体速度样本、所述源VR反馈力标签、所述目标流体速度样本和所述适配VR反馈力,分别针对每个类别进行类条件概率分布适配,获得第二距离的步骤,直至达到设定结束条件,获得迭代后的所述目标VR反馈力。
  10. 一种流场数据可视化装置,其特征在于,包括:
    数据集获得模块,用于获得源领域的数据集和目标领域的数据集,所述源领域的数据集至少包括源流体速度样本和源VR反馈力标签,一组所述源流体速度样本对应一个所述源VR反馈力标签,所述目标领域的数据集至少包括目标流体速度样本;
    优化目标获得模块,用于对所述源领域的数据集和所述目标领域的数据集进行联合分布适配,获得优化目标;
    VR反馈力预测模块,用于基于所述优化目标、所述源流体速度样本和所述目标流体速度样本,对所述源VR反馈力标签进行迁移,预测获得所述目标领域的目标VR反馈力;
    可视化展示模块,用于根据所述目标VR反馈力对所述目标领域的流场数据进行VR可视化展示。
  11. 一种流场数据可视化设备,其特征在于,包括:
    存储器,用于存储计算机程序;
    处理器,用于执行所述计算机程序时实现如权利要求1至9任一项所述流场数据可视化方法的步骤。
  12. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有计算机程序,所述计算机程序被处理器执行时实现如权利要求1至9任一项所述流场数据可视化方法的步骤。
PCT/CN2020/117023 2020-06-19 2020-09-23 一种流场数据可视化方法、装置、设备及存储介质 WO2021253666A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010568718.6 2020-06-19
CN202010568718.6A CN111724487B (zh) 2020-06-19 2020-06-19 一种流场数据可视化方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021253666A1 true WO2021253666A1 (zh) 2021-12-23

Family

ID=72568663

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/117023 WO2021253666A1 (zh) 2020-06-19 2020-09-23 一种流场数据可视化方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN111724487B (zh)
WO (1) WO2021253666A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113010132B (zh) * 2021-03-24 2023-04-18 成都维泰油气能源技术有限公司 一种用于智能井控的辅助系统和方法

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107704926A (zh) * 2017-11-23 2018-02-16 清华大学 一种大数据跨领域分析的深度迁移学习方法
US20180253627A1 (en) * 2017-03-06 2018-09-06 Xerox Corporation Conditional adaptation network for image classification
CN109657693A (zh) * 2018-10-22 2019-04-19 中国科学院软件研究所 一种基于相关熵和迁移学习的分类方法
CN109741429A (zh) * 2018-12-21 2019-05-10 浙江远算云计算有限公司 一种用于非结构化流体仿真模型的vr动画可视化方法
CN110533193A (zh) * 2019-08-20 2019-12-03 武汉理工大学 半监督场景下特征和实例联合迁移学习方法
CN110705029A (zh) * 2019-09-05 2020-01-17 西安交通大学 一种基于迁移学习的振荡扑翼能量采集系统流场预测方法
CN111027626A (zh) * 2019-12-11 2020-04-17 西安电子科技大学 基于可变形卷积网络的流场识别方法
CN111177882A (zh) * 2019-12-05 2020-05-19 西安交通大学 一种基于深度迁移学习的piv流场恢复方法

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9218365B2 (en) * 2011-12-15 2015-12-22 Yeda Research And Development Co. Ltd. Device, system, and method of visual inference by collaborative composition
CN103970920A (zh) * 2013-02-05 2014-08-06 周柏贾 地震应急演练虚拟仿真系统
EP3234742A4 (en) * 2014-12-16 2018-08-08 Quan Xiao Methods and apparatus for high intuitive human-computer interface
US10296846B2 (en) * 2015-11-24 2019-05-21 Xerox Corporation Adapted domain specific class means classifier
US20180024968A1 (en) * 2016-07-22 2018-01-25 Xerox Corporation System and method for domain adaptation using marginalized stacked denoising autoencoders with domain prediction regularization
CN110892408A (zh) * 2017-02-07 2020-03-17 迈恩德玛泽控股股份有限公司 用于立体视觉和跟踪的系统、方法和装置
CN109657159A (zh) * 2018-12-18 2019-04-19 哈尔滨工业大学 舆情数据角色识别中异构关系数据的迁移学习界限的确定方法
CN109828659A (zh) * 2018-12-25 2019-05-31 北京牡丹视源电子有限责任公司 一种基于虚拟现实的用户体验标注方法
CN109885833A (zh) * 2019-02-18 2019-06-14 山东科技大学 一种基于多域数据集联合嵌入的性感极性检测方法
CN110059744B (zh) * 2019-04-16 2022-10-25 腾讯科技(深圳)有限公司 训练神经网络的方法、图像处理的方法、设备及存储介质
CN110781970B (zh) * 2019-10-30 2024-04-26 腾讯科技(深圳)有限公司 分类器的生成方法、装置、设备及存储介质
CN110815258B (zh) * 2019-10-30 2023-03-31 华南理工大学 基于电磁力反馈和增强现实的机器人遥操作系统和方法

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180253627A1 (en) * 2017-03-06 2018-09-06 Xerox Corporation Conditional adaptation network for image classification
CN107704926A (zh) * 2017-11-23 2018-02-16 清华大学 一种大数据跨领域分析的深度迁移学习方法
CN109657693A (zh) * 2018-10-22 2019-04-19 中国科学院软件研究所 一种基于相关熵和迁移学习的分类方法
CN109741429A (zh) * 2018-12-21 2019-05-10 浙江远算云计算有限公司 一种用于非结构化流体仿真模型的vr动画可视化方法
CN110533193A (zh) * 2019-08-20 2019-12-03 武汉理工大学 半监督场景下特征和实例联合迁移学习方法
CN110705029A (zh) * 2019-09-05 2020-01-17 西安交通大学 一种基于迁移学习的振荡扑翼能量采集系统流场预测方法
CN111177882A (zh) * 2019-12-05 2020-05-19 西安交通大学 一种基于深度迁移学习的piv流场恢复方法
CN111027626A (zh) * 2019-12-11 2020-04-17 西安电子科技大学 基于可变形卷积网络的流场识别方法

Also Published As

Publication number Publication date
CN111724487A (zh) 2020-09-29
CN111724487B (zh) 2023-05-16

Similar Documents

Publication Publication Date Title
US20200364389A1 (en) Generating integrated circuit floorplans using neural networks
JP7087079B2 (ja) 深層学習アプリケーションのための堅牢な勾配重み圧縮方式
CN108491928A (zh) 模型参数训练方法、装置、服务器及存储介质
Singh et al. PI-LSTM: Physics-infused long short-term memory network
JP7483005B2 (ja) データ・ラベル検証
KR20200001466A (ko) 객체들의 엔트로피적 군집화
WO2022179533A1 (zh) 一种量子卷积操作器
CN112241478B (zh) 一种基于图神经网络的大规模数据可视化降维方法
US11373117B1 (en) Artificial intelligence service for scalable classification using features of unlabeled data and class descriptors
CN108830385A (zh) 深度学习模型训练方法和装置及计算机可读存储介质
US20230325675A1 (en) Data valuation using reinforcement learning
CN115239638A (zh) 一种工业缺陷检测方法、装置、设备及可读存储介质
CN114981800A (zh) 个性化自动化机器学习
US20220318591A1 (en) Inference method and information processing apparatus
CN111488925A (zh) 一种数据标注方法、装置、电子设备及存储介质
WO2022069740A1 (en) Simulating physical environments using mesh representations and graph neural networks
WO2021253666A1 (zh) 一种流场数据可视化方法、装置、设备及存储介质
US11593695B2 (en) Systems and methods for hybrid analog and digital processing of a computational problem using mean fields
CN113240113A (zh) 一种增强网络预测鲁棒性的方法
WO2019092868A1 (ja) 情報処理装置、情報処理方法及びコンピュータ読み取り可能な記録媒体
WO2022115656A1 (en) Data source correlation techniques for machine learning and convolutional neural models
CN117540096A (zh) 改进用于推荐系统的数据子采样
WO2023174189A1 (zh) 图网络模型节点分类方法、装置、设备及存储介质
US20220138554A1 (en) Systems and methods utilizing machine learning techniques for training neural networks to generate distributions
WO2020162294A1 (ja) 変換方法、訓練装置及び推論装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20940541

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20940541

Country of ref document: EP

Kind code of ref document: A1