WO2021097917A1 - 一种数值预报的集合耦合同化系统及方法 - Google Patents

一种数值预报的集合耦合同化系统及方法 Download PDF

Info

Publication number
WO2021097917A1
WO2021097917A1 PCT/CN2019/122765 CN2019122765W WO2021097917A1 WO 2021097917 A1 WO2021097917 A1 WO 2021097917A1 CN 2019122765 W CN2019122765 W CN 2019122765W WO 2021097917 A1 WO2021097917 A1 WO 2021097917A1
Authority
WO
WIPO (PCT)
Prior art keywords
assimilation
algorithm
coupling
collective
module
Prior art date
Application number
PCT/CN2019/122765
Other languages
English (en)
French (fr)
Inventor
刘利
李锐喆
张�诚
王斌
Original Assignee
清华大学
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 清华大学 filed Critical 清华大学
Priority to US17/778,822 priority Critical patent/US20220398505A1/en
Publication of WO2021097917A1 publication Critical patent/WO2021097917A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation

Definitions

  • the present disclosure relates to the technical field of numerical prediction. More specifically, the present disclosure relates to an ensemble coupled assimilation system and method for numerical prediction.
  • German AWI Helmholtz Polar and Ocean Research Center
  • PDAF Parallel Data Assimilation Framework
  • the basic idea is to implement the parallel operation of all set members of the model in the same MPI (ie message passing interface, used for programming parallel programs, and almost all models use its parallel version).
  • MPI message passing interface
  • Implementation model All group members directly call the same assimilation algorithm, and develop an MPI-based interaction module between the two, so as to realize online interaction without file reading and writing, and avoid the restart of the mode.
  • PDAF integrates multiple integrated algorithms and has been applied to the development of multiple modes of assimilation system
  • its online interaction method is difficult to support the coupling mode, that is, it is difficult to support the coupling assimilation.
  • PDAF requires the assimilation algorithm to only use the MPI process of the first set member of the pattern, which limits the running speed of the assimilation algorithm.
  • the embodiments of the present disclosure provide a numerical prediction ensemble coupling assimilation system and method, which can realize the coordinated operation of all ensemble members of the same coupling mode under the same MPI task, support different component modes flexibly using different assimilation algorithm examples, and support one
  • the assimilation algorithm instance uses all the MPI processes of all the collection members of the corresponding component mode to run in parallel.
  • the completion mode and the assimilation algorithm do not need to go through the efficient online interaction of the data file, and in the online interaction process, a variety of collection data operations can be completed in parallel. Data interpolation between different grids will ultimately improve the operating efficiency of the corresponding assimilation system and the flexibility of assimilation configuration.
  • the ensemble coupling assimilation system includes: a coupling mode integration and collaborative collective operation management module, an assimilation algorithm integration module, an ensemble coupling assimilation test configuration module, and an ensemble coupling Assimilation online interactive module;
  • the coupling mode integration and collaborative collective operation management module is used to integrate arbitrary coupling modes, realize the configuration and compilation operation of multiple collective members in the same coupling mode, and start all collective members to run in parallel in the same MPI task at the same time;
  • the assimilation algorithm integration module is used to integrate any assimilation algorithm or assimilation algorithm library, and can automatically compile each assimilation algorithm or assimilation algorithm library into a dynamic link library for indirect calling in the component mode;
  • the collective coupling assimilation test configuration module is used for the user to configure the collective coupling assimilation test, supporting the user's choice of coupling mode, selection of the assimilation algorithm used for each component mode, setting of the number of collective members, and each assimilation algorithm The setting of the processed assimilation variables, the setting of the assimilation frequency of each assimilation algorithm;
  • the collective coupling assimilation online interaction module is used to complete the online interaction between the component mode and the assimilation algorithm during the simultaneous parallel operation of all collective members in the same coupling mode in the same MPI task.
  • the collective coupling assimilation online interaction module includes: an assimilation configuration information interface sub-module for reading and analyzing coupling mode selection, assimilation algorithm selection, and assimilation from the collective coupling assimilation test configuration module Variable and assimilation frequency configuration information.
  • the collectively coupled assimilation online interaction module further includes: an assimilation algorithm operation management sub-module for managing all assimilation algorithm instances, wherein the dynamic link library and its contents are automatically completed during the startup process of the collectively coupled assimilation system Loading of assimilation algorithms.
  • the collectively coupled assimilation online interaction module further includes: the collective calculation operation sub-module for performing input/output variables in the component mode and the assimilation algorithm according to the assimilation test configuration information and the assimilation algorithm declared input/output variables When interacting, the collection operation for each variable is automatically completed.
  • the set coupling assimilation online interaction module further includes: a data online exchange sub-module for completing the parallelization of related variables between the MPI process of each set member of the same component mode and the MPI process of the corresponding assimilation algorithm instance Data transfer.
  • the collectively coupled assimilation online interaction module further includes: a data parallel interpolation sub-module, which is used to complete the parallelization of variable data between different grids when the component mode is different from the grid of the corresponding assimilation algorithm instance Interpolation.
  • the embodiments of the present disclosure provide an ensemble coupled assimilation method for numerical prediction, and the ensemble coupled assimilation system based on the numerical prediction is implemented, and the method includes:
  • Integrate assimilation algorithm or assimilation algorithm library integrate all codes and corresponding input data of assimilation algorithm or assimilation algorithm library into assimilation algorithm integration module, so that assimilation algorithm or assimilation algorithm library can be automatically compiled into dynamic link library;
  • the integrated coupling mode integrates all the codes and corresponding input data of the coupling mode into the coupling mode integration and collaborative collective operation management module, so that all the members of the coupling mode can be configured, compiled and run at the same time;
  • the coupling mode integration and collaborative collective operation management module is adopted, and after the compilation of all codes is completed, the collective coupling assimilation system is run.
  • the step of integrating the assimilation algorithm or the assimilation algorithm library further includes: generating the assimilation algorithm or the framework interface program of the assimilation algorithm library based on the API in the application program interface submodule, and connecting the assimilation algorithm or the assimilation algorithm Start the program, run the program and end the program of the library, complete the declaration of the input/output variables of the assimilation algorithm, and switch on the input/output variables of the assimilation algorithm.
  • the step of integrating the coupling mode further includes: the corresponding input data of the coupling mode includes startup data and external force data.
  • the step of setting the set coupling assimilation test configuration further includes: setting the assimilation variable and assimilation frequency of each assimilation calculation instance.
  • Figure 1 is a schematic diagram of an assimilation system based on reading and writing data files in some cases
  • Figure 2 is a schematic diagram of an assimilation system based on the framework of reconstruction mode in some cases
  • Fig. 3 is a schematic structural diagram of an ensemble coupled assimilation system for numerical prediction provided by an embodiment of the present disclosure.
  • the embodiments of the present disclosure provide a numerical prediction ensemble coupling assimilation system, which can realize the coordinated operation of all ensemble members of the same coupling mode under the same MPI task, supports the flexible use of different assimilation algorithm instances in different component modes, and supports one assimilation algorithm instance Use all the MPI processes of all set members of the corresponding component mode to run in parallel.
  • the completion mode and the assimilation algorithm do not need to go through the efficient online interaction of the data file, and in the online interaction process, it can complete a variety of collective data operations and different grids in parallel. Interpolation between data, etc., ultimately improve the operating efficiency of the corresponding assimilation system and the flexibility of assimilation configuration.
  • FIG. 3 is a schematic diagram of a framework of an ensemble coupled assimilation system for numerical prediction in an embodiment of the present disclosure.
  • the ensemble coupled assimilation system for numerical prediction mainly includes four modules: a coupled mode integration and collaborative collective operation management module, an assimilation algorithm integration module, an ensemble coupled assimilation test configuration module, and an ensemble coupled assimilation online interaction module.
  • Coupling mode integration and collaborative collective operation management module used to integrate arbitrary coupling modes, realize one-click configuration and one-click compilation of multiple set members in the same coupling mode, and start all set members to run in parallel in the same MPI task at the same time, Manage the working directories of all collection members.
  • Coupling mode integration and the successful integration of the collaborative collective operation management module is usually a parallel program running in multiple MPI processes, and the framework interface program has been implemented using the application program interface API provided by the collective coupling assimilation online interaction module.
  • the assimilation algorithm integration module is used to integrate any assimilation algorithm or assimilation algorithm library, which can automatically compile each assimilation algorithm or assimilation algorithm library into a dynamic link library for indirect call of the model.
  • the assimilation algorithm or assimilation algorithm library successfully integrated by the assimilation algorithm integration module uses the application program interface API provided by the collection coupling assimilation online interaction module to realize the framework interface program.
  • the collective coupling assimilation test configuration module is used for the user to configure the collective coupling assimilation test. It supports the user's choice of coupling mode, selection of the assimilation algorithm used in each component mode, setting of the number of collective members, and processing of each assimilation algorithm The setting of assimilation variables and the setting of assimilation frequency for each assimilation algorithm.
  • the collective coupling assimilation online interaction module is used to complete the online interaction between the component mode and the assimilation algorithm when all the members of the same coupling mode run in parallel in the same MPI task.
  • the above-mentioned collective coupling assimilation online interaction module includes application program interface sub-module, assimilation configuration information interface sub-module, assimilation algorithm operation management sub-module, collective computing operation sub-module, data online exchange sub-module, data parallel interpolation sub-module, and interactive program flow management Sub-module.
  • the application program interface sub-module is used to provide different application program interfaces for the ensemble coupled assimilation system of numerical forecasting, including the application program interface API for all ensemble members of the coupling mode to operate cooperatively, and the application program interface for the mode start/run/stop assimilation algorithm instance API, the application program interface API that specifies the input/output variable information of the mode, the application program interface API that the assimilation algorithm obtains input parameter information from the component mode, the application program interface API that the assimilation algorithm declares input/output variables, and the application program interface API for obtaining the assimilation test configuration information Application programming interface API.
  • the mode-specified input/output variable information includes: variable name, grid, parallel division, and storage space information.
  • the assimilation configuration information interface sub-module is used to read in and analyze the coupling mode selection, assimilation algorithm selection, assimilation variable and assimilation frequency configuration information from the collective coupling assimilation test configuration module.
  • the assimilation algorithm operation management sub-module is used to manage all assimilation algorithm instances. For each assimilation algorithm instance, the assimilation algorithm operation management sub-module uses the MPI process of all set members of the corresponding component mode to start, run, and end the assimilation algorithm instance. During the process, the dynamic link library and its assimilation algorithm are loaded automatically.
  • the collective calculation operation sub-module is used to automatically complete the collective operation for each variable when the component mode interacts with the assimilation algorithm according to the assimilation test configuration information and the input/output variables declared by the assimilation algorithm.
  • the collective operations include collection aggregation and collection Average, largest collection, smallest collection, distribution; and when the time average is used for assimilation, the time average of the model variables is automatically calculated.
  • the data online exchange sub-module is used to complete the parallel data transfer of related variables between the MPI process of each set member of the same component mode and the MPI process of the corresponding assimilation algorithm instance, involving MPI data communication.
  • the data parallel interpolation sub-module is used to complete the parallel interpolation of variable data between different grids when the component mode is different from the grid of the corresponding assimilation algorithm instance.
  • the interactive program flow management sub-module is used to call the corresponding functions of the assimilation algorithm operation management sub-module, the collective computing operation sub-module, the data online exchange sub-module and the data parallel interpolation sub-module, according to the information derived from the application program interface and the assimilation configuration,
  • the online interactive program flow between the component mode and the assimilation algorithm is automatically constructed.
  • the online interactive program flow is automatically run, and when the component mode ends the assimilation algorithm instance, it automatically ends Online interactive program flow.
  • the embodiment of the present disclosure also provides a numerical prediction ensemble coupling assimilation method, which can realize the coordinated operation of all ensemble members of the same coupling mode under the same MPI task, supports the flexible use of different assimilation algorithm instances in different component modes, and supports one assimilation algorithm
  • the instance uses all the MPI processes of all set members of the corresponding component mode to run in parallel.
  • the completion mode and the assimilation algorithm do not need to go through the efficient online interaction of the data file, and in the online interaction process, it can complete a variety of collective data operations and different networks in parallel. Data interpolation between grids, etc., ultimately improve the operating efficiency of the corresponding assimilation system and the flexibility of assimilation configuration.
  • the ensemble coupled assimilation method of numerical prediction is realized by the ensemble coupled assimilation system based on numerical prediction, including the following steps:
  • Step 1 Integrate the assimilation algorithm or the assimilation algorithm library, and integrate all the codes of the assimilation algorithm or the assimilation algorithm library and the corresponding input data into the assimilation algorithm integration module, where the input data includes observation data.
  • the assimilation algorithm or assimilation algorithm library can be automatically compiled into a dynamic link library.
  • develop the framework interface program of the assimilation algorithm or the assimilation algorithm library connect the start program, run the program and the end program of the assimilation algorithm or the assimilation algorithm library, complete the declaration of the input/output variables of the assimilation algorithm, Turn on the input/output variables of the assimilation algorithm.
  • Step 2 Integrate the coupling mode, integrate all the codes of the coupling mode and the corresponding input data into the coupling mode integration and collaborative collective operation management module, and the corresponding input data includes start-up data and external force data.
  • All set members in the coupling mode can be configured with one key, compiled with one key, and run at the same time with one key.
  • the framework interface program of the pattern is developed, which realizes the cooperative operation of all set members of the pattern, specifies the information of the input/output variables of the pattern, and calls the API to start/run/end the assimilation algorithm instance.
  • Step 3 Set up the set coupling assimilation test configuration. After selecting the coupling mode, set the number of set members in the assimilation test, select the assimilation algorithm used by each component mode, and set the assimilation calculation examples for some component modes without using the assimilation algorithm.
  • the coupling mode includes several component modes.
  • Step 4 Run the collective coupling assimilation system, using the coupling mode integration and collaborative collective operation management module. After compiling all the codes, run the collective coupling assimilation system to realize the efficient online interaction between the component mode and the assimilation algorithm.
  • the ensemble coupled assimilation system and method for numerical prediction provided by the embodiments of the present disclosure can support all the MPI processes of all ensemble members of the corresponding component mode by supporting an instance of an assimilation algorithm for parallel operation, and the completion mode and the assimilation algorithm do not need to go through data files Efficient online interaction improves the operating efficiency of the corresponding assimilation system and the flexibility of assimilation configuration.
  • the present disclosure can realize the coordinated operation of all set members of the same coupling mode under the same MPI task, support the flexible use of different assimilation algorithm instances in different component modes, and support one assimilation algorithm instance to use all MPI processes of all set members in the corresponding component mode for parallelism Operation, and ultimately improve the operating efficiency of the corresponding assimilation system and the flexibility of assimilation configuration.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Environmental & Geological Engineering (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Environmental Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Atmospheric Sciences (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Ecology (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Debugging And Monitoring (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种数值预报的集合耦合同化系统及方法。所述集合耦合同化系统包括：耦合模式集成与协同集合运行管理模块、同化算法集成模块、集合耦合同化试验配置模块和集合耦合同化在线交互模块。

Description

一种数值预报的集合耦合同化系统及方法
本公开要求享有2019年11月22日提交的名称为“一种数值预报的集合耦合同化系统及方法”的中国专利申请CN201911154881.1的优先权，其全部内容通过引用并入本文中。
技术领域
本公开涉及数值预报技术领域。更具体的，本公开涉及一种数值预报的集合耦合同化系统及方法。
背景技术
应对延长预报时效与无缝隙预报业务发展的需求，传统的基于大气模式的数值天气预报正朝着基于耦合模式和耦合同化技术的方向发展。与耦合模式相比，耦合同化正处在发展的初期，不仅同化方法面临着诸多挑战性问题，如何构建通用耦合同化框架是其与计算机科学交叉中面临的重大挑战。
基于最优控制的四维变分和基于集合预报的集合卡曼滤波是业务中应用较广泛的先进同化方法，两者互有优缺点；而有机结合集合和变分的混合同化方法(如集合三维变分、集合四维变分等)可有效避免两者的缺点，逐步成为未来数值预报中数据同化技术的主流。在采用集合卡曼滤波或混合同化等基于同一模式集合预报的同化算法来构建耦合同化系统(简称为集合耦合同化系统)时，需要实现相应耦合模式所有集合成员与同化算法之间的交互。
基于文件读写的离线交互是最早采用的实现方式，其基本思路(如图1所示)是：继续保持模式各集合成员和同化模块之间运行的独立性(模式各集合成员、同化模块均独立执行)，约定好数据文件接口(数据文件格式、变量名等)，并利用脚本控制来实现模式各集合成员与同化模块之间的交替重启动、运行与交互。这一实现方式可尽量保持模式及同化模块的代码不变，因而容易实现，较多集合耦合同化系统采用了这一实现方式，如美国NCAR(国家大气研究中心)开发的DART、美国GFDL(国家地球物理流体动力学实验室)开发的CDA、美国 DTC(Developmental Testbed Center)开发的GSI_EnKF等。尽管如此，由于数据文件读写操作和软件(模式和同化模块)重启动过程的开销大、并行可扩展性差(例如图2)，这一实现方式存在着计算效率低这一明显缺点。随着模式分辨率和同化频率的提高，软件重启动和数据文件读写的开销会不断增加，而同化系统的运行速度会不断降低，这将会限制预报结果的及时性，甚至使得在墙钟时间内无法完成规定时长的预报。
为了避免离线交互的上述问题，德国AWI(亥姆霍兹极地与海洋研究中心)研发了支持在线交互的PDAF(并行数据同化框架)。其基本思路(如图2所示)是：实现模式所有集合成员在同一MPI(即消息传递接口，用于编程并行程序，几乎所有模式都采用其实现并行版本)任务内的并行运行，实现模式所有集合成员对同一同化算法的直接调用，研发两者之间基于MPI的交互模块，从而实现无需文件读写的在线交互，并避免了模式的重启动。尽管PDAF集成了多个集合同化算法并已应用于多个模式的同化系统研制，但其在线交互方式难以支持耦合模式，即难以支持耦合同化，这是因为：1)PDAF要求被同化的分量模式使用所有MPI进程，而在绝大部分耦合模式中，一个分量模式仅使用耦合模式的部分MPI进程；2)在耦合同化试验中，可能需要采用不同的同化算法对多个分量模式进行同化，但PDAF只允许使用一个同化算法，这给耦合同化带来了限制。此外，PDAF要求同化算法只使用模式第一个集合成员的MPI进程，这给同化算法的运行速度带来了限制。
综上所述，当前集合耦合同化的发展面临着严峻技术挑战，国内外均没有基于高效在线交互方式实现的集合耦合同化框架，只能采用基于文件读写的低效离线交互方式来研制集合耦合同化系统。
发明内容
本公开实施例通过提供一种数值预报的集合耦合同化系统及方法，其能够实现同一耦合模式所有集合成员在同一MPI任务下的协同运行，支持不同分量模式灵活使用不同的同化算法实例，支持一个同化算法实例使用相应分量模式所有集合成员的所有MPI进程以进行并行运行，完成模式与同化算法间无需经过数据文件的高效在线交互，并在在线交互过程中，能并行完成多种集合数据操作、不同网格间的数据插值等，最终提高相应同化系统的运行效率和同化配置的灵活性。
本公开实施例提供了一种数值预报的集合耦合同化系统，其中：所述集合耦合同化系统包括：耦合模式集成与协同集合运行管理模块、同化算法集成模块、集合耦合同化试验配置模块和集合耦合同化在线交互模块；
所述耦合模式集成与协同集合运行管理模块，用于集成任意耦合模式，实现同一耦合模式多个集合成员的配置和编译操作，启动所有集合成员在同一MPI任务内的同时并行运行；
所述同化算法集成模块，用于集成任意同化算法或同化算法库，能将各同化算法或同化算法库自动编译为动态链接库，以供分量模式间接调用；
所述集合耦合同化试验配置模块，用于用户对集合耦合同化试验的配置，支持用户对耦合模式的选择、对各分量模式所采用同化算法的选择、对集合成员数量的设置、对各同化算法所处理同化变量的设置、对各同化算法同化频率的设置；
所述集合耦合同化在线交互模块，用于同一耦合模式所有集合成员在同一MPI任务内同时并行运行过程中，完成分量模式与同化算法之间的在线交互。
在一个实施例中，所述集合耦合同化在线交互模块，包括：同化配置信息接口子模块，用于读入并解析来自于所述集合耦合同化试验配置模块的耦合模式选择、同化算法选择、同化变量和同化频率配置信息。
在一个实施例中，所述集合耦合同化在线交互模块，还包括：同化算法运行管理子模块，用于管理所有同化算法实例，其中在集合耦合同化系统启动过程中，自动完成动态链接库及其中同化算法的加载。
在一个实施例中，所述集合耦合同化在线交互模块，还包括：所述集合计算操作子模块，用于根据同化试验配置信息和同化算法声明的输入/输出变量，在分量模式与同化算法进行交互时，自动完成针对各变量的集合操作。
在一个实施例中，所述集合耦合同化在线交互模块，还包括：数据在线交换子模块，用于完成同一分量模式各集合成员的MPI进程与相应同化算法实例的MPI进程之间相关变量的并行数据传递。
在一个实施例中，所述集合耦合同化在线交互模块，还包括：数据并行插值子模块，用于完成当分量模式与相应同化算法实例的网格不同时，变量数据在不同网格间的并行插值。
本公开实施例提供了一种数值预报的集合耦合同化方法，基于所述数值预报的集合耦合同化系统实现，所述方法包括：
集成同化算法或同化算法库，将同化算法或同化算法库的所有代码及相应输入数据集成至同化算法集成模块，使同化算法或同化算法库能被自动编译为动态链接库；
集成耦合模式，将耦合模式的所有代码及相应输入数据集成至耦合模式集成与协同集合运行管理模块，使耦合模式所有集合成员能被配置、编译且同时运行；
设置集合耦合同化试验配置，在选定耦合模式后，设定同化试验的集合成员数量，选定各分量模式使用的同化算法；
运行所述集合耦合同化系统，采用耦合模式集成与协同集合运行管理模块，在完成所有代码的编译后，运行集合耦合同化系统。
在一个实施例中，所述集成同化算法或同化算法库的步骤，还包括：基于应用程序接口子模块中的API，生成同化算法或同化算法库的框架接口程序，接通同化算法或同化算法库的启动程序、运行程序和结束程序、完成同化算法输入/输出变量的声明、接通同化算法的输入/输出变量。
在一个实施例中，所述集成耦合模式的步骤，还包括：耦合模式的相应输入数据包括启动数据、外强迫数据。
在一个实施例中，所述设置集合耦合同化试验配置的步骤，还包括：设定各同化算实例的同化变量和同化频率。
附图说明
图1是一些情况中的基于读写数据文件方式实现同化系统示意图；
图2是一些情况中的基于重构模式框架方式实现的同化系统示意图；
图3是本公开实施方式提供的数值预报的集合耦合同化系统的结构示意图。
具体实施方式
下面结合说明书附图针对本公开的实施方式作进一步详细的说明。
本公开实施方式提供一种数值预报的集合耦合同化系统，其能够实现同一耦合模式所有集合成员在同一MPI任务下的协同运行，支持不同分量模式灵活使用不同的同化算法实例，支持一个同化算法实例使用相应分量模式所有集合成员的所有MPI进程以进行并行运行，完成模式与同化算法间无需经过数据文件的高效在线交互，并在在线交互过程中，能并行完成多种集合数据操作、不同网格间的 数据插值等，最终提高相应同化系统的运行效率和同化配置的灵活性。
请参阅图3，所示为本公开一实施方式中数值预报的集合耦合同化系统的框架示意图。在本实施方式中，数值预报的集合耦合同化系统主要包括四个模块：耦合模式集成与协同集合运行管理模块、同化算法集成模块、集合耦合同化试验配置模块和集合耦合同化在线交互模块。
耦合模式集成与协同集合运行管理模块，用于集成任意耦合模式，实现同一耦合模式多个集合成员的一键配置和一键编译等操作，启动所有集合成员在同一MPI任务内的同时并行运行，管理所有集合成员的工作目录。耦合模式集成与协同集合运行管理模块成功集成的耦合模式，通常是采用多个MPI进程运行的并行程序，并已利用集合耦合同化在线交互模块提供的应用程序接口API，实现了框架接口程序。
同化算法集成模块，用于集成任意同化算法或同化算法库，能将各同化算法或同化算法库自动编译为动态链接库，以供模式间接调用。同化算法集成模块成功集成的同化算法或同化算法库，利用集合耦合同化在线交互模块提供的应用程序接口API，实现了框架接口程序。
集合耦合同化试验配置模块，用于用户对集合耦合同化试验的配置，支持用户对耦合模式的选择、对各分量模式所采用同化算法的选择、对集合成员数量的设置、对各同化算法所处理同化变量的设置、对各同化算法同化频率的设置。
集合耦合同化在线交互模块，用于同一耦合模式所有集合成员在同一MPI任务内同时并行运行过程中，完成分量模式与同化算法之间的在线交互。
上述集合耦合同化在线交互模块包括应用程序接口子模块、同化配置信息接口子模块、同化算法运行管理子模块、集合计算操作子模块、数据在线交换子模块、数据并行插值子模块和交互程序流程管理子模块。
应用程序接口子模块，用于为数值预报的集合耦合同化系统提供不同应用程序接口，包括耦合模式所有集合成员进行协同运行的应用程序接口API、模式启动/运行/结束同化算法实例的应用程序接口API、模式指定输入/输出变量信息的应用程序接口API、同化算法从分量模式获取到输入参数信息的应用程序接口API、同化算法声明输入/输出变量的应用程序接口API、获取同化试验配置信息的应用程序接口API。其中，模式指定输入/输出变量信息包括：变量名、所在网格、并行剖分、存储空间信息。
同化配置信息接口子模块，用于读入并解析来自于集合耦合同化试验配置模块的耦合模式选择、同化算法选择、同化变量和同化频率配置信息。
同化算法运行管理子模块，用于管理所有同化算法实例，对于各同化算法实例，同化算法运行管理子模块利用相应分量模式所有集合成员的MPI进程，启动、运行并结束同化算法实例，其中在启动过程中，自动完成动态链接库及其中同化算法的加载。
集合计算操作子模块，用于根据同化试验配置信息和同化算法声明的输入/输出变量，在分量模式与同化算法进行交互时，自动完成针对各变量的集合操作，其中集合操作包括集合聚合、集合平均、集合最大、集合最小、分发；且当采用时间平均量进行同化时，自动求取模式变量的时间平均值。
数据在线交换子模块，用于完成同一分量模式各集合成员的MPI进程与相应同化算法实例的MPI进程之间相关变量的并行数据传递，涉及到MPI数据通信。
数据并行插值子模块，用于完成当分量模式与相应同化算法实例的网格不同时，变量数据在不同网格间的并行插值。
交互程序流程管理子模块，用于调用同化算法运行管理子模块、集合计算操作子模块、数据在线交换子模块和数据并行插值子模块的相应功能，根据来源于应用程序接口和同化配置的信息，在分量模式启动同化算法实例时，自动构建分量模式与同化算法间的在线交互程序流程，在分量模式运行同化算法时，自动运行在线交互程序流程，并在分量模式结束同化算法实例时，自动结束在线交互程序流程。
本公开实施方式还提供一种数值预报的集合耦合同化方法，其能够实现同一耦合模式所有集合成员在同一MPI任务下的协同运行，支持不同分量模式灵活使用不同的同化算法实例，支持一个同化算法实例使用相应分量模式所有集合成员的所有MPI进程以进行并行运行，完成模式与同化算法间无需经过数据文件的高效在线交互，并在在线交互过程中，能并行完成多种集合数据操作、不同网格间的数据插值等，最终提高相应同化系统的运行效率和同化配置的灵活性。
该数值预报的集合耦合同化方法，基于数值预报的集合耦合同化系统实现，包括以下步骤：
步骤1：集成同化算法或同化算法库，将同化算法或同化算法库的所有代码及相应输入数据集成至同化算法集成模块，其中输入数据包括观测数据。使同化 算法或同化算法库能被自动编译为动态链接库。基于应用程序接口子模块中的API，研制同化算法或同化算法库的框架接口程序，接通同化算法或同化算法库的启动程序、运行程序和结束程序、完成同化算法输入/输出变量的声明、接通同化算法的输入/输出变量。
步骤2：集成耦合模式，将耦合模式的所有代码及相应输入数据集成至耦合模式集成与协同集合运行管理模块，相应输入数据包括启动数据、外强迫数据。使耦合模式所有集合成员能被一键配置、一键编译且一键同时运行。基于应用程序接口子模块中的API，研制模式的框架接口程序，其中实现模式所有集合成员的协同运行、指定模式输入/输出变量的信息、调用启动/运行/结束同化算法实例的API。
步骤3：设置集合耦合同化试验配置，在选定耦合模式后，设定同化试验的集合成员数量，选定各分量模式使用的同化算法，部分分量模式可不使用同化算法，设定各同化算实例的同化变量和同化频率。所述耦合模式包括若干个分量模式。
步骤4：运行集合耦合同化系统，采用耦合模式集成与协同集合运行管理模块，在完成所有代码的编译后，运行集合耦合同化系统，实现分量模式与同化算法的高效在线交互。
本公开实施方式提供的数值预报的集合耦合同化系统及方法，能够实现支持一个同化算法实例使用相应分量模式所有集合成员的所有MPI进程以进行并行运行，完成模式与同化算法间无需经过数据文件的高效在线交互，提高相应同化系统的运行效率和同化配置的灵活性。
本公开能够实现同一耦合模式所有集合成员在同一MPI任务下的协同运行，支持不同分量模式灵活使用不同的同化算法实例，支持一个同化算法实例使用相应分量模式所有集合成员的所有MPI进程以进行并行运行，最终提高相应同化系统的运行效率和同化配置的灵活性。
本领域的技术人员应当理解，可以对本公开的技术方案进行修改或等同替换，而不脱离本公开技术方案的精神和范围。

Claims (10)

  1. 一种数值预报的集合耦合同化系统，其中：所述集合耦合同化系统包括：耦合模式集成与协同集合运行管理模块、同化算法集成模块、集合耦合同化试验配置模块和集合耦合同化在线交互模块；
    所述耦合模式集成与协同集合运行管理模块，用于集成任意耦合模式，实现同一耦合模式多个集合成员的配置和编译操作，启动所有集合成员在同一MPI任务内的同时并行运行；
    所述同化算法集成模块，用于集成任意同化算法或同化算法库，能将各同化算法或同化算法库自动编译为动态链接库，以供分量模式间接调用；
    所述集合耦合同化试验配置模块，用于用户对集合耦合同化试验的配置，支持用户对耦合模式的选择、对各分量模式所采用同化算法的选择、对集合成员数量的设置、对各同化算法所处理同化变量的设置、对各同化算法同化频率的设置；
    所述集合耦合同化在线交互模块，用于同一耦合模式所有集合成员在同一MPI任务内同时并行运行过程中，完成分量模式与同化算法之间的在线交互。
  2. 根据权利要求1所述的集合耦合同化系统，其中：所述集合耦合同化在线交互模块，包括：
    同化配置信息接口子模块，用于读入并解析来自于所述集合耦合同化试验配置模块的耦合模式选择、同化算法选择、同化变量和同化频率配置信息。
  3. 根据权利要求2所述的集合耦合同化系统，其中：所述集合耦合同化在线交互模块，还包括：
    同化算法运行管理子模块，用于管理所有同化算法运行实例，其中在启动同化算法运行实例过程中，自动完成动态链接库及其中同化算法的加载。
  4. 根据权利要求3所述的集合耦合同化系统，其中：所述集合耦合同化在线交互模块，还包括：
    所述集合计算操作子模块，用于根据同化试验配置信息和同化算法声明的输入/输出变量，在分量模式与同化算法进行交互时，自动完成针对各变量的集合操作。
  5. 根据权利要求4所述的集合耦合同化系统，其中：所述集合耦合同化在线交互模块，还包括：
    数据在线交换子模块，用于完成同一分量模式各集合成员的MPI进程与相应 同化算法实例的MPI进程之间相关变量的并行数据传递。
  6. 根据权利要求5所述的集合耦合同化系统，其中：所述集合耦合同化在线交互模块，还包括：数据并行插值子模块，用于完成当分量模式与相应同化算法实例的网格不同时，变量数据在不同网格间的并行插值。
  7. 一种数值预报的集合耦合同化方法，其中，基于权利要求1至6中任一项所述的数值预报的集合耦合同化系统实现，所述方法包括：
    集成同化算法或同化算法库，将同化算法或同化算法库的所有代码及相应输入数据集成至同化算法集成模块，使同化算法或同化算法库能被自动编译为动态链接库；
    集成耦合模式，将耦合模式的所有代码及相应输入数据集成至耦合模式集成与协同集合运行管理模块，使耦合模式所有集合成员能被配置、编译且同时运行；
    设置集合耦合同化试验配置，在选定耦合模式后，设定同化试验的集合成员数量，选定各分量模式使用的同化算法；
    运行所述集合耦合同化系统，采用耦合模式集成与协同集合运行管理模块，在完成所有代码的编译后，运行集合耦合同化系统。
  8. 根据权利要求7所述的集合耦合同化方法，其中：所述集成同化算法或同化算法库的步骤，还包括：基于应用程序接口子模块中的API，生成同化算法或同化算法库的框架接口程序，接通同化算法或同化算法库的启动程序、运行程序和结束程序、完成同化算法输入/输出变量的声明、接通同化算法的输入/输出变量。
  9. 根据权利要求8所述的集合耦合同化方法，其中：所述集成耦合模式的步骤中，耦合模式的相应输入数据包括启动数据和外强迫数据。
  10. 根据权利要求9所述的集合耦合同化方法，其中：所述设置集合耦合同化试验配置的步骤，还包括：设定各同化算实例的同化变量和同化频率。
PCT/CN2019/122765 2019-11-22 2019-12-03 一种数值预报的集合耦合同化系统及方法 WO2021097917A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/778,822 US20220398505A1 (en) 2019-11-22 2019-12-03 Ensemble Coupled Assimilation System for Numerical Prediction and Method thereof

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911154881.1 2019-11-22
CN201911154881.1A CN110909297B (zh) 2019-11-22 2019-11-22 一种数值预报的集合耦合同化系统及方法

Publications (1)

Publication Number Publication Date
WO2021097917A1 true WO2021097917A1 (zh) 2021-05-27

Family

ID=69818858

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/122765 WO2021097917A1 (zh) 2019-11-22 2019-12-03 一种数值预报的集合耦合同化系统及方法

Country Status (3)

Country Link
US (1) US20220398505A1 (zh)
CN (1) CN110909297B (zh)
WO (1) WO2021097917A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117217027A (zh) * 2023-11-02 2023-12-12 中国人民解放军国防科技大学 基于四维变分同化的污染物点源廓线排放估算方法和装置

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112597669B (zh) * 2021-03-02 2021-08-27 北京卡普拉科技有限公司 一种模拟试验平台及其工作方法
CN113156395B (zh) * 2021-04-07 2022-08-12 中国科学院大气物理研究所 气溶胶激光雷达数据融合方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109426886A (zh) * 2017-08-29 2019-03-05 北京思湃德信息技术有限公司 一种气候预测系统
US20190107638A1 (en) * 2014-02-02 2019-04-11 Ertha Space Technologies Earthquake Forecast Device
CN109783932A (zh) * 2019-01-14 2019-05-21 哈尔滨工程大学 一种结合最优观测时间窗口的强耦合数据同化方法
CN110309481A (zh) * 2019-04-24 2019-10-08 武汉大学 农田土壤反应动力学过程模型建模方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102707932B (zh) * 2012-05-16 2013-07-24 清华大学 一种用于地球系统模式的并行耦合方法
CN108764515A (zh) * 2018-04-04 2018-11-06 河海大学 一种耦合数值气象水文集合预报的水库调度风险决策方法
CN109754125A (zh) * 2019-01-18 2019-05-14 中国农业大学 基于作物模型、历史和气象预报数据的作物产量预报方法
CN109976723B (zh) * 2019-03-12 2022-08-12 北京国电智深控制技术有限公司 一种算法开发平台、算法开发方法及计算机可读存储介质
CN110209353B (zh) * 2019-05-17 2022-10-21 青岛海洋科学与技术国家实验室发展中心 区域耦合预报系统中roms模式的i/o并行加速方法、装置及介质
CN110472281B (zh) * 2019-07-11 2021-09-21 北京师范大学 一种估算时空连续地表水热通量的数据同化方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190107638A1 (en) * 2014-02-02 2019-04-11 Ertha Space Technologies Earthquake Forecast Device
CN109426886A (zh) * 2017-08-29 2019-03-05 北京思湃德信息技术有限公司 一种气候预测系统
CN109783932A (zh) * 2019-01-14 2019-05-21 哈尔滨工程大学 一种结合最优观测时间窗口的强耦合数据同化方法
CN110309481A (zh) * 2019-04-24 2019-10-08 武汉大学 农田土壤反应动力学过程模型建模方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WEI ZEXUN, XU TENG-FEI; WANG YONG-GANG; LI HONG-PING; YANG XUE-LIAN; FENG WEI-ZHONG; ZHANG ZHI-YUAN: "Introduction to the Fine-Resolution Wind-Wave-Current Coupled Forecasting and Information Service System for the South China Sea and Its Adjacent Sea Areas", OCEAN TECHNOLOGY, GUOJIA HAIYANG JISHU ZHONGXIN, CN, vol. 34, no. 3, 1 June 2015 (2015-06-01), CN, pages 86 - 90, XP055814381, ISSN: 1003-2029 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117217027A (zh) * 2023-11-02 2023-12-12 中国人民解放军国防科技大学 基于四维变分同化的污染物点源廓线排放估算方法和装置
CN117217027B (zh) * 2023-11-02 2024-01-30 中国人民解放军国防科技大学 基于四维变分同化的污染物点源廓线排放估算方法和装置

Also Published As

Publication number Publication date
CN110909297A (zh) 2020-03-24
CN110909297B (zh) 2020-11-27
US20220398505A1 (en) 2022-12-15

Similar Documents

Publication Publication Date Title
WO2021097917A1 (zh) 一种数值预报的集合耦合同化系统及方法
US10095657B2 (en) Processor, accelerator, and direct memory access controller within a core reading/writing local synchronization flag area for parallel
US20220188086A1 (en) Off-load servers software optimal placement method and program
WO2020248511A1 (zh) 一种基于运行时模型实现对终端应用控制的方法
CN103593192A (zh) 一种基于slurm调度的算法集成与评测平台及方法
CN112286521A (zh) 一种任务流程处理方法、存储介质及处理器
CN110532072A (zh) 基于微内核操作系统的分布式流式数据处理方法及系统
Wang et al. A framework for distributed data-parallel execution in the Kepler scientific workflow system
US10496423B2 (en) Method for opening up data and functions of terminal application based on reconstruction technology
WO2023124543A1 (zh) 用于大数据的数据处理方法和数据处理装置
Vocke et al. Extending halide to improve software development for imaging dsps
Weber et al. MATOG: array layout auto-tuning for CUDA
CN103299277B (zh) Gpu系统及其处理方法
CN111679859B (zh) 一种面向i/o密集型高性能应用的自动化并行mpi-i/o加速方法
CN112068942B (zh) 一种基于单节点模拟的大规模并行系统模拟方法
CN111580826B (zh) 机器学习模型的编译优化方法和装置
CN112558977A (zh) 面向异构众核后端基于代价模型的多面体优化方法
CN1316359C (zh) 用户指导的程序半自动并行化方法
Aarnio Parallel data processing with MapReduce
CN113867714A (zh) 一种适配多语言的自动代码生成方法
CN111580827A (zh) 机器学习模型的编译优化方法和装置
TWI455030B (zh) 決定程式任務之平行度的方法與計算裝置及其機器可讀取媒體
US11714614B2 (en) Code generation tool for cloud-native high-performance computing
CN112992270A (zh) 一种基因测序方法和装置
Sheng et al. Layered Models for General Parallel Computation Based on Heterogeneous System

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: 19953189

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: 19953189

Country of ref document: EP

Kind code of ref document: A1