WO2022099915A1 - 基于超网络模型的组织架构建模方法及空间探索算法 - Google Patents

基于超网络模型的组织架构建模方法及空间探索算法 Download PDF

Info

Publication number
WO2022099915A1
WO2022099915A1 PCT/CN2021/070648 CN2021070648W WO2022099915A1 WO 2022099915 A1 WO2022099915 A1 WO 2022099915A1 CN 2021070648 W CN2021070648 W CN 2021070648W WO 2022099915 A1 WO2022099915 A1 WO 2022099915A1
Authority
WO
WIPO (PCT)
Prior art keywords
architecture
combat
node
network
decision
Prior art date
Application number
PCT/CN2021/070648
Other languages
English (en)
French (fr)
Inventor
周鑫
王维平
井田
杨松
王彦锋
黄美根
王涛
李小波
林木
李童心
段婷
张�杰
王梦
Original Assignee
中国人民解放军国防科技大学
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 中国人民解放军国防科技大学 filed Critical 中国人民解放军国防科技大学
Priority to GB2200021.0A priority Critical patent/GB2618302A/en
Publication of WO2022099915A1 publication Critical patent/WO2022099915A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00Details relating to CAD techniques
    • G06F2111/04Constraint-based CAD
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00Details relating to CAD techniques
    • G06F2111/08Probabilistic or stochastic CAD
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02PCLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/02Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Definitions

  • the invention relates to the technical field of combat systems, in particular to a method for modeling a combat system architecture and a space exploration algorithm based on a super network model.
  • the organizational structure reflects the configuration of the components in the system and the interaction between the components and the external environment.
  • the organizational structure focuses on physical entities, information structures and system functions, and is the core framework of the system.
  • the organizational structure runs through the whole process of design, requirement demonstration, prototype development, application testing and field testing. Therefore, it is the best configuration to realize the core elements of the organization by studying the organization through the organization structure and defining a reasonable formal organization structure.
  • An OA is a collection of equipment connected by a charge network that has certain functions to support the completion of a specific mission. OA is used to guide the construction of a specific organization.
  • an architectural model, an architectural solution space exploration problem model and a solution algorithm are constructed.
  • the following problems need to be solved in architecture modeling and selection: First, the potential capabilities of the architecture are uncertain. In previous studies, organizational capabilities were determined after the architecture was established. In fact, the uncertainty of the potential capabilities of the architecture is reflected in the uncertainty of tasks and the diversity of resource combinations on the one hand; on the other hand, the influence of secondary factors, because only the main factors that affect the capabilities of the system are often considered when designing the architecture.
  • the purpose of the present invention is to provide a combat system architecture modeling method and a space exploration algorithm based on a super network model, the modeling method can construct a combat system architecture model selected by multi-architecture schemes, and is convenient for decision makers to make the best choice;
  • the algorithm is a parallel search algorithm based on decision indicators, and the algorithm is a polynomial time algorithm. Its return value is significantly better than other benchmark algorithms, and it is optimal under the assumption that the combat system architecture scheme space is independent.
  • the present invention provides the following scheme:
  • the combat system architecture modeling method based on the super network model takes the mission, equipment system and command and command structure as the core elements of the combat system architecture. It consists of the following steps:
  • the reporter is Agent n; among them, Agent n represents the plan space of the nth decision maker, each plan space includes a variety of combat system architectures, and the combat system architecture is used to describe the combat system; N represents the decision maker space,
  • the decision maker problem in the space exploration problem model of the combat architecture is converted into a formalized dynamic programming problem.
  • the edge set of G TS represents the correspondence between task nodes and system nodes.
  • the development cost of the combat architecture is defined as c, and c ⁇ C, C represents the cost space,
  • the potential capability of the combat architecture is defined as w, and the combat architecture has a certain ability to complete the mission w ⁇ W, W represents the capability of the combat architecture to complete the mission;
  • the combat architecture model consists of the topology model of the combat architecture, development costs and capabilities, denoted as ⁇ GA,C,W>.
  • the present invention also provides a space exploration algorithm for a combat system architecture solution.
  • the exploration algorithm is based on the dynamic planning of the combat system architecture space, and the algorithm selects actions by judging defined indicators, and includes the following steps:
  • Judgment indicators based on the classic Pandora's rules, define the decision indicators for each decision maker to perform different actions:
  • each plan represents a combat architecture
  • K n represents the set of solutions of the nth decision maker
  • m represents the action
  • M k represents the kth plan.
  • action set represents the decision index of the nth action of the kth plan of the nth decision maker
  • x k represents the reward value of program k
  • step 2) a search algorithm, according to the search rule in step 1), the calculation of the optimal solution is simplified to the judgment of the index, and the search algorithm includes a single decision maker search algorithm and a cooperation algorithm between multiple decision makers;
  • step 2) the calculation of each of the indicators in step 2) is independent, and the indicators are not affected by the probability distribution of the return value of other combat architectures.
  • the present invention also provides a space exploration method for a combat system architecture scheme, including:
  • the corresponding relationship between task nodes and system nodes is given, and the system network is constructed according to the task network and the corresponding relationship between task nodes and system nodes.
  • the system nodes are equipment with specific functions and can independently complete specific tasks. is SY; the task node is the activity process performed by the equipment system, denoted as TA;
  • the corresponding relationship between the system node and the command node is given, and the command network is established according to the relationship, and the combat system architecture model is constructed in combination with the task network and the system network; among them, the command node is used for information processing, management organization, decision-making and planning.
  • the accusation node is denoted as C2
  • the exploration algorithm is used for dynamic planning of the combat system architecture space;
  • the exploration algorithm includes a single decision-maker search algorithm and a cooperation algorithm between multiple decision-makers;
  • a single decision-maker search algorithm includes three stages: index sorting, index judgment, and plan selection. ;
  • Dynamic planning of the combat architecture space using a single decision maker search algorithm including:
  • the set of decision indicators is k represents a plan, each plan represents a combat system architecture, K n represents the set of solutions of the nth decision maker, m represents the action, and M k represents the set of actions of the kth plan, represents the decision index of the nth action of the kth plan of the nth decision maker, represents the development cost corresponding to the nth action of the kth plan of the nth decision maker, Indicates that the reward value x of each program obeys a probability distribution, and x k represents the reward value of program k;
  • the present invention discloses the following technical effects:
  • the method for modeling the combat system architecture based on the super network model of the present invention is based on the generation elements of the system architecture capability, according to the formal definition of the combat system architecture, the multi-Agent dynamic planning problem, and the combat system architecture solution space exploration problem framework, and can construct a multi-tasking system.
  • the combat system architecture model selected by the architecture plan is convenient for decision makers to make the best choice;
  • the space exploration algorithm of the combat system architecture plan is a parallel search algorithm based on decision indicators.
  • the algorithm is polynomial time, and its return value is significantly better than other Baseline algorithm, which is optimal under the assumption of space independence of combat architecture solutions.
  • FIG. 1 is a schematic diagram of a task network in an embodiment of the present invention.
  • FIG. 2 is a schematic diagram of a system network in an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of a charging network in an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of three types of network mapping relationships in an embodiment of the present invention, FIG. 4 a is a “task-system” bipartite graph, and FIG. 4b is a “system-command” bipartite graph;
  • FIG. 5 is a schematic diagram of a transition relationship between possible states of an architectural effect in an embodiment of the present invention.
  • FIG. 6 is an experimental data analysis diagram of an analysis experiment of a sequential search algorithm based on a decision index in an embodiment of the present invention.
  • the combat system architecture modeling method based on the super network model, according to the capability generation mechanism, takes the mission, equipment system and command and command structure as the core elements of the combat system architecture.
  • the combat system architecture consists of three heterogeneous networks: mission network, system network and command and command network.
  • the composition includes the following steps:
  • a task node is an activity process that can be performed by an equipment system, denoted as TA.
  • the decision maker problem in the combat system search problem model is converted into a formalized dynamic programming problem.
  • the potential capability of the combat system architecture is the capability of the combat system developed according to the combat system architecture to complete a specific mission, denoted as W.
  • the uncertainty of the potential capability of an architecture can be represented by a probability distribution, that is, W obeys a certain probability distribution.
  • a hypernetwork as a heterogeneous network connecting multiple types of nodes.
  • the operational architecture based on hypernetwork is composed of three heterogeneous networks: mission network, system network and command and command network.
  • mission network In order to complete the mission of the combat system, the mission is decomposed into a series of executable tasks, called a task network, as shown in Figure 1, which is an example of a task network.
  • Figure 2 is a An example of a system network, which represents the logical relationship between system functions. As shown in Figure 4, it shows the corresponding relationship between the task network and the system network.
  • Figure 4a is the "task-system” bipartite graph
  • Figure 4b is the "system-command" bipartite graph.
  • the accusation node is used to process the instruction information of the superior and inferior.
  • the task information is received from the upper-level accusing node, and the sub-task information is transmitted to the lower-level accusing node after being processed by this node.
  • an accusation network refers to an organizational network that connects all accusing nodes through an order relationship, as shown in Figure 3, which is an example of an accusation network.
  • the combat system capability in this embodiment is measured by the return value, which is a comprehensive measure of the development cost of the combat system architecture and the benefits obtained by the cluster executing the reconnaissance strategy.
  • the reward value x of each scheme obeys the probability distribution W(x), and the reward values of different schemes are independent of each other.
  • k n ⁇ K n , K n ⁇ 1,2,...,
  • ⁇ , K n is the number of solutions in the solution space of Agent n.
  • some agents have an intersection in the solution space, at this time It is also possible that there is no intersection, at this time
  • the reward value of each architecture is not determined in advance, but can be obtained through different actions.
  • Agent n can take action to develop. Continue to explore options in the undeveloped options space, and finally select one option among all the developed options space as the final option.
  • the goal of the agent is to choose an architecture with the highest expected reward value and the least accumulated search cost.
  • the combat architecture state is the state of the combat architecture in the development process, including an undeveloped state and an developed state. As shown in Figure 5, it represents an operational architecture state transition relationship. Among them, the unknown state indicates that the program has not been developed and its reward value is unknown; the known state indicates that the program has been developed and its reward value is known.
  • constraint (a) ensures that for any agent's solution space, solutions are either developed or not.
  • Constraint (b) says that if the agent finally chooses a scheme, it must have already developed that scheme.
  • Constraint (c) means that each agent finally chooses only one scheme.
  • Constraint (d) represents the value space of the two classes of decision variables.
  • Constraint (e) represents the cost for each agent to perform different actions for each scheme.
  • the problem is searched according to the combat architecture, and the problem is converted into a formalized dynamic programming problem.
  • Agent n as an example to illustrate, in dynamic programming, its scheme space is first Divided into two mutually exclusive sets: one is a growing set of developed architectures The other is a decreasing set of unexplored architectures
  • the operational architecture solution space is known in advance, and is obtained by decision makers for a specific mission and specific field, and its focus is to find a solution that maximizes the objective function.
  • Agent n can choose whether to choose from the set Select and develop unknown solutions from , or stop the search and select a final solution from the set D n . If the agent chooses to continue searching, then it has M k types of action development architecture k. If Agent n stops searching, choose the solution with the highest reward value in the developed solution space:
  • the state evaluation function after executing action m, c m,k represent the cost of all Agents executing action m k to develop architecture k n , Further, the agent needs to compare the expected value generated by different actions, and select and execute the action with the largest expected reward value. Taking action m as an example, if the reward value x k ⁇ y, then the current highest reward value will not change, and the expected state evaluation value is If x k > y, then the current highest reward value will be updated to x, and the expected state evaluation value will be
  • Each task network has a start task node, an end task node, and an intermediate node.
  • the task network includes two types of logical relationships, causality and juxtaposition.
  • the development cost of the combat architecture is defined as c, and c ⁇ C
  • the potential capability of the combat architecture is defined as w
  • the combat architecture has a certain ability to complete the mission w ⁇ W.
  • the combat architecture model consists of the topology model, development cost and capability of the combat architecture, denoted as ⁇ GA,C,W>.
  • Embodiment 2 The present invention also provides a space exploration algorithm for a combat system architecture solution.
  • the search algorithm is based on the dynamic planning of the combat system architecture space, and the algorithm selects actions by judging the defined indicators, including the following steps:
  • Judgment indicators based on the classic Pandora's rules, define the decision indicators for each decision maker to perform different actions:
  • step 2) a search algorithm, according to the search rule in step 1), the calculation of the optimal solution is simplified to the judgment of the index, and the search algorithm includes a single decision maker search algorithm and a cooperation algorithm among multiple decision makers.
  • each index in step 2) is independent, and the index is not affected by the probability distribution of the return value of other combat architectures.
  • the optimal architecture scheme can be calculated after at most K iterations.
  • the current maximum sampling value is compared with the maximum index in each iteration. If the maximum sampled value is not less than the maximum index, the search is stopped and the architecture m with the current maximum sampled reward value is used as the selected architecture. On the contrary, according to the corresponding architecture index i and action a, the execution program continues to search. If the sampled reward value for architecture i is obtained, then update the variable, which represents the removal of architecture i from the set.
  • means sampling, which means sampling the probability distribution.
  • the sequential search algorithm based on the decision index is a polynomial time algorithm.
  • the time complexity of this algorithm depends on the time complexity of the sorting algorithm.
  • the agent performs corresponding actions in order based on the size of the architectural index value, and this order will not change during the entire search process. Therefore, the complexity of the algorithm proposed by the present invention is equal to that of the sorting algorithm, so the algorithm is a polynomial time algorithm.
  • each scheme selected by the sequential search algorithm based on decision index is conditionally optimal, and the algorithm has local optimality.
  • the selection of each plan can be mapped to the classic Pandora problem.
  • the Pandora problem the return value of each project obeys a probability distribution, and the actual return value of the project is not known before running the project. The actual return value needs to be obtained by sampling.
  • each existing architecture can be regarded as a project k, which has a sampling reward value rk. Once the sampled reward value for option k is obtained, the three items are moved into the explored set D.
  • the combat architecture search problem model uses an index-based search strategy, that is, if the agent wants to explore a new solution, it selects the unexplored solution with the highest index, otherwise it selects the explored solution with the largest sampling reward value. It is proved that this search strategy can effectively solve Pandora's problem and get the best expected return value.
  • the present invention also provides a space exploration method for a combat system architecture scheme, including:
  • the corresponding relationship between task nodes and system nodes is given, and a system network is constructed according to the task network and the corresponding relationship between task nodes and system nodes.
  • the system nodes are equipment with specific functions and can independently complete specific tasks. is SY; the task node is the activity process performed by the equipment system, denoted as TA;
  • the corresponding relationship between the system node and the command node is given, and the command network is established according to the relationship, and the combat system architecture model is constructed in combination with the task network and the system network; among them, the command node is used for information processing, management organization, decision-making and planning.
  • the accusation node is denoted as C2
  • the exploration algorithm is used for dynamic planning of the combat system architecture space;
  • the exploration algorithm includes a single decision-maker search algorithm and a cooperation algorithm between multiple decision-makers;
  • a single decision-maker search algorithm includes three stages: index sorting, index judgment, and plan selection. ;
  • Dynamic planning of the combat architecture space using a single decision maker search algorithm including:
  • the set of decision indicators is k represents a plan, each plan represents a combat system architecture, K n represents the set of solutions of the nth decision maker, m represents the action, and M k represents the set of actions of the kth plan, represents the decision index of the nth action of the kth plan of the nth decision maker, represents the development cost corresponding to the nth action of the kth plan of the nth decision maker, Indicates that the reward value x of each program obeys a probability distribution, and x k represents the reward value of program k;
  • the technical solution for mission capability is generally a top-down approach, and finally turns to a solution method based on a multi-Agent system: first, the mission is decomposed into a task network; secondly, UAVs with certain functions can complete specific tasks, thereby constructing The mapping relationship between UAVs in the task domain; in addition, there is an accusation relationship between UAVs, so as to build an accusation network; finally, a multi-Agent system model is established, each agent has a task list, specific functions, and accusation relationships, so that The multi-agent system is a kind of equipment architecture scheme. In order to maximize the operational effectiveness of the system, it is necessary to select the optimal architecture scheme.
  • the objective function described in which the performance of a simulation is the difference between the return value of the architecture and the cumulative cost value; (2) Number of known architecture times, the average number of times the architecture has been developed; (3) Running time Time, the time the program runs during recording .
  • the average performance is used to evaluate the performance of the algorithm, and the number of consultations and the number of known architectures are used to analyze the search process.
  • Random Algorithm that is, the Agent randomly selects an action at each moment. Specifically, an architecture k is randomly selected from the set K first. If k ⁇ D, i.e. the architecture has been developed, end the search and get the reward value for that architecture; if That is, if the architecture k is undeveloped and the reward value is unknown, an action is randomly selected, then returns and continues to perform the random action until the end of the search.
  • TDA Traversal Development Algorithm
  • the agent chooses the least expensive action to develop, When the Agent has completed the development of all architectures, the architecture with the highest return value among all the developed architectures is selected as the final solution.
  • GSA General Development Algorithm
  • the evaluation index of the local exploration algorithm is the difference between the highest expected value and the development cost, namely When the highest reward value among the developed architectures has exceeded this metric, the search is stopped and the architecture with the highest reward value is selected.
  • NoS is the number of selected solutions, that is, how many solutions are finally selected by the multi-Agent system
  • FIG. 6 is an experimental data analysis diagram of an analysis experiment performed by a sequential search algorithm based on a decision index.
  • the method for modeling a combat system architecture based on a super network model provided by the present invention is based on the capability generation elements of the system architecture, according to the formal definition of the combat system architecture, the multi-agent dynamic planning problem, and the space exploration of the combat system architecture solution.
  • the problem framework can build a combat system architecture model with multi-architecture options, which is convenient for decision makers to make the best choice;
  • the combat system architecture solution space exploration algorithm is a parallel search algorithm based on decision indicators, which is polynomial time, Its return value is significantly better than other benchmark algorithms, and it is optimal under the assumption of space independence of combat architecture solutions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于超网络模型的作战体系架构建模方法,该建模方法包括以下步骤:将作战体系使命分解为可由装备系统执行的任务网络;根据任务网络和任务节点与系统节点之间的对应关系构建系统网络;建立指控网络,结合任务网络、系统网络构建成作战体系架构模型;构建作战体系架构的空间探索问题模型;将作战体系搜索问题模型中的决策者问题转换为一种形式化的动态规划问题。该建模方法构建的作战体系架构模型,便于决策者做出最佳选择。

Description

基于超网络模型的组织架构建模方法及空间探索算法
本申请要求于2020年11月13日提交中国专利局、申请号为202011267527.2、发明名称为“基于超网络模型的作战体系架构建模方法及空间探索算法”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
本发明涉及作战系统技术领域,特别是涉及一种基于超网络模型的作战体系架构建模方法及空间探索算法。
背景技术
随着系统信息化和智能化的发展,系统之间的相互联系变得越来越多样化。组织是一种体系,是有限数量的组分系统的集成,这些系统是独立且可运行的,并且在一段时间内相互连接以实现某个更高的目标。由于组织十分复杂,如何研究组织是当前研究者迫切需要解决的问题。幸运的是,组织架构为解决该问题提供了一种行之有效的思路。组织架构反映了体系中组件的配置以及组件与外部环境之间的交互。组织架构关注物理实体、信息结构和体系功能,是体系的核心框架。组织架构贯穿设计、需求演示、原型开发、应用程序测试和外场试验全过程。因此,通过组织架构研究组织,定义一种合理的形式化组织架构,是实现组织核心要素的最佳配置。
OA是由指控网络连接的装备集合,这些装备具有某些功能以支持特定任务的完成。OA用于指导特定组织的构建。针对组织架构潜在能力不确定问题,构建架构模型、架构方案空间探索问题模型及求解算法。架构建模与选取中有如下问题需要解决:首先,架构潜在能力具有不确定性。在以往的研究中,组织能力在架构建立之后就确定了。事实上,架构潜在能力的不确定性一方面体现在任务不确定和资源组合多样性;另一方面次要因素的影响,因为在设计架构时往往只考虑影响体系能力的主要因素。其次,如果选择继续开发架构,决策者有多种策略来获得架构潜在能力。因而决策者应该评估这些策略的期望回报值,以便做出最佳选择。第三, 从多个架构方案空间中选择若干个最优架构,以往研究往往只选择一个架构方案,缺乏对多架构方案选取的研究。鉴于此,亟需构建一种新颖的架构模型和组织架构方案空间探索问题,并提出了一种架构方案空间动态探索算法,以解决上述问题。
发明内容
本发明的目的是提供一种基于超网络模型的作战体系架构建模方法及空间探索算法,该建模方法能够构建出具多架构方案选取的作战体系架构模型,便于决策者做出最佳选择;该算法为基于决策指标的并行搜索算法,且该算法为多项式时间算法,其回报值明显优于其他基准算法,其在作战体系架构方案空间独立情况的假设下是最优的。
为实现上述目的,本发明提供了如下方案:
基于超网络模型的作战体系架构建模方法,根据能力生成机理,将任务、装备系统和指控结构作为作战体系架构的核心要素,所述作战体系架构由任务网络、系统网络和指控网络三种异构网络组成,具体包括以下步骤:
S1、将作战体系使命分解为由装备系统执行的任务网络;
S2、给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY;任务节点为可由装备系统执行的活动过程,记为TA;
S3、给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型;其中, 指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中,G C2表示指控网络,V C2表示指控节点,E C2表示V C2节点之间的边集合,G SC表示系统节点与指控节点之间的对应关系,V SY表示系统节点,E SC表示节点V SY和节点V C2之间的边集合;
S4、根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n;其中,Agent n表示第n个决策者的方案空间,每个方案空间包括多种作战体系架构,作战体系架构用于对作战体系进行描述;N表示决策者空间,|N|表示决策者的数量;
S5、根据作战体系架构的空间探索问题模型,将作战体系架构的空间探索问题模型中的决策者问题转换为一种形式化的动态规划问题。
进一步地,所述任务网络可抽象为有向图,记为G TA=<V TA,E TA>,其中,V TA表示任务节点集合,E TA表示任务节点之间的边集合,G TA表示任务网络;每个所述任务网络均具有起始任务节点、结束任务节点和中间节点;所述任务网络包括因果关系和并列关系两类逻辑关系。
进一步地,所述系统网络表示系统节点之间的功能关系,记为G SY=<V SY,E SY>,其中V SY表示系统节点集合,E SY表示系统节点之间的边集合,G SY表示系统网络;所述任务节点和系统节点之间的对应关系定义为二分图,记为G TS=<V TA,V SY,E TS>,其中E TS表示节点V TA和节点V SY之间的边集合,G TS表示任务节点和系统节点之间的对应关系。
进一步地,所述指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>;所述系统节点和指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中E SC表示节点V SY和节点V C2之间的边集合,G SC表示系统节点和指控节点之间的对应关系。
进一步地,所述作战体系架构的拓扑模型由三类节点与五类关系构成的异构网络GA,记为GA=<V TA,V SY,V C2,E TA,E SY,E C2,E TS,E SC>;所述作战体系架构的开发花费代价定义为c,且c∈C,C表示代价空间,所述作战体系架构的潜在能力定义为w,作战体系架构具有一定完成使命的能力w∈W,W表示作战体系架构具有的完成使命的能力;所述作战体系架构模型由作战体系架构的拓扑模型、开发代价和能力组成,记为<GA,C,W>。
本发明还提供一种作战体系架构方案空间探索算法,所述探索算法基于作战体系架构空间的动态规划,且该算法通过判断定义的指标来进行动作选择,具备包括以下步骤:
1)、判定指标,基于经典潘多拉规则,定义每个决策者执行不同行动的决策指标:
Figure PCTCN2021070648-appb-000001
推导出:
Figure PCTCN2021070648-appb-000002
根据状态
Figure PCTCN2021070648-appb-000003
和指标集合
Figure PCTCN2021070648-appb-000004
设计一种简单且最优的搜索规则;其中,k表示方案,每一个方案代表一个作战体系架构,K n表示第n个决策者的方案集合,m表示行动,M k表示第k个方案的行动集合,
Figure PCTCN2021070648-appb-000005
表示第n个决策者第k个方案的第n个行动的决策指标,
Figure PCTCN2021070648-appb-000006
表示第n个决策者第k个方案的第n个行动对应的开发花费代价,
Figure PCTCN2021070648-appb-000007
表示每个方案的回报值x服从概率分布,x k表示方案k的回报值;
2)、搜索算法,根据步骤1)中的搜索规则,将最优方案的计算简化为指标的判断,所述搜索算法包括单个决策者搜索算法和多个决策者之间的合作算法;
3)、将单决策者搜索算法分为指标排序、指标判断、方案选择三个阶段,并在Sorting程序,根据公式
Figure PCTCN2021070648-appb-000008
计算所有架构中所有动作的指标,然后对该指标进行排序,并将排序结果保存至向量,然后调用Developing程序求得最优架构方案;其中,y n表示第n个决策者的最大回报值,D n表示已开发架构集合。
进一步地,步骤2)中每个所述指标的计算均是独立的,且所述指标不受其他作战体系架构回报值概率分布的影响。
本发明还提供一种作战体系架构方案空间探索方法,包括:
将作战体系使命分解为由装备系统执行的任务网络;
给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY;任务节点为由装备系统执行的活动过程,记为TA;
给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型;其中,指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,指控节点表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络, 记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中,G C2表示指控网络,V C2表示指控节点,E C2表示V C2节点之间的边集合,G SC表示系统节点与指控节点之间的对应关系,V SY表示系统节点,E SC表示节点V SY和节点V C2之间的边集合;
根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n;其中,Agent n表示第n个决策者的方案空间,每个方案空间包括多种作战体系架构,作战体系架构用于对作战体系进行描述;N表示决策者空间,|N|表示决策者的数量;
根据作战体系架构的空间探索问题模型,将作战体系架构的空间探索问题模型中的决策者问题转换为动态规划问题;
采用探索算法进行作战体系架构空间的动态规划;其中,探索算法包括单个决策者搜索算法和多个决策者之间的合作算法;单个决策者搜索算法包括指标排序、指标判断、方案选择三个阶段;
采用单个决策者搜索算法进行作战体系架构空间的动态规划,具体包括:
基于潘多拉规则,确定每个决策者执行不同行动的决策指标:
Figure PCTCN2021070648-appb-000009
Figure PCTCN2021070648-appb-000010
式中,决策指标集合为
Figure PCTCN2021070648-appb-000011
k表示方案,每一个方案代表一个作战体系架构,K n表示第n个决策者的方案集合,m表示 行动,M k表示第k个方案的行动集合,
Figure PCTCN2021070648-appb-000012
表示第n个决策者第k个方案的第n个行动的决策指标,
Figure PCTCN2021070648-appb-000013
表示第n个决策者第k个方案的第n个行动对应的开发花费代价,
Figure PCTCN2021070648-appb-000014
表示每个方案的回报值x服从概率分布,x k表示方案k的回报值;
对每个决策者执行不同行动的决策指标进行排序;
将最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合;已开发架构集合与未开发架构集合的并集为决策者方案空间;
确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值;
判断回报值是否大于或等于未开发架构集合中作战体系架构对应的最大决策指标,得到第一判断结果;
若第一判断结果为是,则根据公式
Figure PCTCN2021070648-appb-000015
确定已开发架构集合中最大回报值对应的作战体系架构为最终方案;其中,y n表示第n个决策者的最大回报值,D n表示已开发架构集合;
若第一判断结果为否,则将未开发架构集合中的最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合,然后返回步骤“确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值”。
根据本发明提供的具体实施例,本发明公开了以下技术效果:
本发明的基于超网络模型的作战体系架构建模方法,基于体系架构 能力生成要素,根据作战体系架构的形式化定义、多Agent动态规划问题以及作战体系架构方案空间探索问题框架,能够构建出具多架构方案选取的作战体系架构模型,便于决策者做出最佳选择;该作战体系架构方案空间探索算法,为基于决策指标的并行搜索算法,该算法为多项式时间的,其回报值明显优于其他基准算法,其在作战体系架构方案空间独立情况的假设下是最优的。
说明书附图
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例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发明的一些实施例,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是本发明实施例中任务网络示意图;
图2是本发明实施例中系统网络示意图;
图3是本发明实施例中指控网络示意图;
图4是本发明实施例中三类网络映射关系示意图,图4a为“任务-系统”二分图,图4b为“系统-指控”二分图;
图5是本发明实施例中架构效果的可能状态之间的转移关系示意图;
图6为本发明实施例中对基于决策指标的顺次搜索算法进行分析实验的实验数据分析图。
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,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,都属于本发明保护的范围。
为使本发明的所述目的、特征和优点能够更加明显易懂,下面结合附图和具体实施方式对本发明作进一步详细的说明。
实施例
基于超网络模型的作战体系架构建模方法,根据能力生成机理,将任务、装备系统和指控结构作为作战体系架构的核心要素,作战体系架构由任务网络、系统网络和指控网络三种异构网络组成,具体包括以下步骤:
S1、将作战体系使命分解为可由装备系统执行的任务网络。
S2、给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY。任务节点为可由装备系统执行的活动过程,记为TA。
S3、给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型。其中,指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中E SC表示节点V SY和节点V C2之间的边集合。
S4、根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n。
S5、根据作战体系架构搜索问题模型,将作战体系搜索问题模型中的决策者问题转换为一种形式化的动态规划问题。
在本实施例中,作战体系架构潜在能力为根据作战体系架构开发的作 战体系所具备的完成特定使命的能力,记为W。
在数学上,架构潜在能力的不确定性可以用概率分布表示,即W服从某种概率分布。此外,本文认为超网络是连接多种类型节点的异构网络。基于超网络的作战体系架构是由任务网络、系统网络和指控网络三种异构网络组成的。为了完成作战体系使命,将使命分解为一系列可执行的任务,称为任务网络,如图1所示,为一个任务网络的示例。在作战体系架构中,具有特定功能的装备系统用于完成特定任务,因而系统节点(例如无人机、坦克和舰艇)之间的关系受任务节点的影响,如图2所示,其为一个系统网络的例子,表示了系统功能之间的逻辑关系。如图4所示,表示任务网络与系统网络的对应关系,图4a为“任务-系统”二分图,图4b为“系统-指控”二分图。
指控节点用于处理上级和下级的指令信息。例如,从上级指控节点接收任务信息,经过本节点处理后将子任务信息传递给下级指控节点。另一方面,保持与同级指控节点之间的信息交互。因此,指控网络是指通过指令关系连接所有指控节点的组织网络,如图3所示,其为指控网络的一个例子。
在本实施例中,根据作战体系架构的拓扑模型定性可以得到,若节点越多、连接关系越复杂,作战体系架构开发的代价将会越大,而架构潜在能力可能越大。
本实施例中的作战系统能力由回报值度量,是对作战体系架构开发代价与集群执行侦察策略所得收益的综合度量。每个方案的回报值x服从概率分布W(x),不同方案的回报值相互独立。其中k n∈K n,K n={1,2,…,|K n|}, K n为Agent n的方案空间中的方案数量。另外,有些Agent的方案空间存在交集,此时
Figure PCTCN2021070648-appb-000016
也有可能不存在交集,此时
Figure PCTCN2021070648-appb-000017
每个架构的回报值事先是不确定的,但可以通过不同行动获得。对于同一种方案k,Agent n可以采用行动
Figure PCTCN2021070648-appb-000018
进行开发。不断探索未开发方案空间中的方案,最后在所有已开发的方案空间中选择一个方案作为最终选项。Agent的目标是选择一个具有最高预期回报值和最少累积搜索成本的架构。
在本实施例中,作战体系架构状态为作战体系架构在开发过程中的形态,包括未开发状态和已开发状。如图5所示,其表示一个作战体系架构状态转移关系。其中,未知状态表明该方案尚未开发,其回报值是未知的;已知状态表明该方案已经被开发,并且其回报值是已知的。基于此,基于作战体系架构空间探索问题模型的形式化描述,具体地,定义两类二元决策变量的集合:
Figure PCTCN2021070648-appb-000019
当Agent n采用行动a开发架构k时
Figure PCTCN2021070648-appb-000020
开发代价为
Figure PCTCN2021070648-appb-000021
否则
Figure PCTCN2021070648-appb-000022
表示第n个Agent是否采取行动a开发架构k,当Agent n最终采用行动m开发架构k为最终架构方案时
Figure PCTCN2021070648-appb-000023
表示是否最终选取第n个Agent以行动m开发架构k,否则
Figure PCTCN2021070648-appb-000024
w表示能力值。作战体系架构空间探索问题模型的形式化描述如下:
CSoSAS:
Figure PCTCN2021070648-appb-000025
S.t.
Figure PCTCN2021070648-appb-000026
Figure PCTCN2021070648-appb-000027
Figure PCTCN2021070648-appb-000028
Figure PCTCN2021070648-appb-000029
Figure PCTCN2021070648-appb-000030
目标函数为最大化已开发架构的回报值与最小化累积搜索成本的加和。具体而言,约束(a)确保对于任意Agent的方案空间,方案要么已经开发,要么未开发。约束(b)表示如果Agent最终选择了一个方案,那么必须已经开发了该方案。约束(c)表示每个Agent最终只选择一个方案。约束(d)表示两类决策变量的值空间。约束(e)表示对于每个方案,每个Agent执行不同动作的成本。
在本实施例中,根据作战体系架构搜索问题,将该问题转换为一种形式化的动态规划问题。以Agent n为例进行说明,在动态规划中,首先将其方案空间
Figure PCTCN2021070648-appb-000031
划分为两个互斥的集合:一个是不断增长的已开发架构集合
Figure PCTCN2021070648-appb-000032
另一个是一组递减的未开发架构
Figure PCTCN2021070648-appb-000033
作战体系架构方案空间是预先获知的,是决策者面向特定使命特定领域得到的,其重点是寻找使目标函数最大化的方案。对于每次决策,Agent n可以选择是否从集合
Figure PCTCN2021070648-appb-000034
中选择和开发未知方案,或者停止搜索并从集合D n中选择一个最终方案。如果Agent选择继续搜索,那么它有M k种类型的行动开发架构k。如果Agent n停止搜索,则在已开发方案空间中选择具有最高回报值 的方案:
Figure PCTCN2021070648-appb-000035
在任意时刻,Agent n的状态被定义为统计
Figure PCTCN2021070648-appb-000036
所有Agent未开发架构空间定义为
Figure PCTCN2021070648-appb-000037
所有Agent的已开发方案空间的最高回报值为y={y 1,y 2,...,y N},那么系统的状态被定义为
Figure PCTCN2021070648-appb-000038
此外,定义状态评估函数
Figure PCTCN2021070648-appb-000039
为当最大已知回报值为y且未开发架构集合为
Figure PCTCN2021070648-appb-000040
时,从该时刻按照最优策略能够获得的期望折扣值。对于每个子集
Figure PCTCN2021070648-appb-000041
和最大已知回报值y,状态评估函数
Figure PCTCN2021070648-appb-000042
需要满足基本的迭代关系。
OSoSAS:
Figure PCTCN2021070648-appb-000043
where
Figure PCTCN2021070648-appb-000044
变量
Figure PCTCN2021070648-appb-000045
表示在状态
Figure PCTCN2021070648-appb-000046
下执行行动m后的状态评估函数,c m,k表示所有Agent执行行动m k开发架构k n的代价,
Figure PCTCN2021070648-appb-000047
进一步,Agent需要比较不同动作产生的期望值,选择和执行具有最大期望回报值的动作。以行动m为例,如果回报值x k≤y,那么当前最高回报值将不改变,期望状态评估值为
Figure PCTCN2021070648-appb-000048
如果x k>y,那么当前的最高回报值将更新为x,期望状态评估值为
Figure PCTCN2021070648-appb-000049
任务网络可抽象为有向图,记为G TA=<V TA,E TA>,其中,V TA表示任务节点集合,E TA表示节点之间的边集合。每个任务网络均具有起始任务节点、结束任务节点和中间节点。任务网络包括因果关系和并列关系两类逻辑关系。
系统网络表示系统节点之间的功能关系,记为G SY=<V SY,E SY>,其中V SY表示系统节点集合,E SY表示系统节点之间的边集合。任务节点和系 统节点之间的对应关系定义为二分图,记为G TS=<V TA,V SY,E TS>,其中E TS表示节点V TA和节点V SY之间的边集合。
指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>。系统节点和指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中E SC表示节点V SY和节点V C2之间的边集合。
作战体系架构的拓扑模型由三类节点与五类关系构成的异构网络GA,记为GA=<V TA,V SY,V C2,E TA,E SY,E C2,E TS,E SC>。作战体系架构的开发花费代价定义为c,且c∈C,作战体系架构的潜在能力定义为w,作战体系架构具有一定完成使命的能力w∈W。作战体系架构模型由作战体系架构的拓扑模型、开发代价和能力组成,记为<GA,C,W>。
实施例2:本发明还提供一种作战体系架构方案空间探索算法,搜索算法基于作战体系架构空间的动态规划,且该算法通过判断定义的指标来进行动作选择,具备包括以下步骤:
1)、判定指标,基于经典潘多拉规则,定义每个决策者执行不同行动的决策指标:
Figure PCTCN2021070648-appb-000050
推导出:
Figure PCTCN2021070648-appb-000051
根据状态
Figure PCTCN2021070648-appb-000052
和指标集合
Figure PCTCN2021070648-appb-000053
设计一种简单且最优的搜索规则。
2)、搜索算法,根据步骤1)中的搜索规则,将最优方案的计算简化为指标的判断,搜索算法包括单个决策者搜索算法和多个决策者之间的合作算法。
3)、将单决策者搜索算法分为指标排序、指标判断、方案选择三个阶段,并在Sorting程序,根据公式
Figure PCTCN2021070648-appb-000054
计算所有架构中所有动作的指标,然后对改指标进行排序,并将排序结果保存至向量,然后调用Developing程序求得最优架构方案。
步骤2)中每个指标的计算均是独立的,且指标不受其他作战体系架构回报值概率分布的影响。
在SequenceSearching程序中,最多经过K次迭代就可以计算得到最优架构方案。根据设定的规则,每轮迭代中将当前最大采样值与最大指标进行对比。如果最大采样值不小于最大指标,则停止搜索,并将具有当前最大采样回报值的架构m作为选择的架构。反之,则根据对应的架构索引i和动作a,执行程序继续搜索。如果获得了架构i的采样回报值,则更新变量,其中表示去除集合中的架构i。
在Executing程序中,如果采取的动作是咨询,则判断是否能够通过相关机构获得架构i的回报值,即判断是否为true。其中“~”表示采样,表示对概率分布进行采样。
在本实施例中,基于决策指标的顺次搜索算法是一种多项式时间算法。该算法的时间复杂度取决于排序算法的时间复杂度。在算法中,Agent是基于架构指标值的大小顺序执行相应动作,而这个顺序的在整个搜索过程中是不会发生改变的。因此,本发明提出的算法复杂度等于排序算法的复杂度,故该算法是一种多项式时间算法成立。
基于决策指标的顺次搜索算法选取的每个方案都是条件最优的,且该算法具有局部最优性。在作战体系架构问题中每个方案的选取能够映射至 经典的潘多拉问题,在潘多拉问题中,每个项目的回报值都服从一个概率分布,在运行项目之前是不知道项目的实际回报值的,需要通过采样得到实际回报值。在作战体系中,每个已经架构可以看成是一个项目k,该项目具有采样回报值rk。一旦获得了方案k的采样回报值,这三个项目就被移入已探索的集合D。作战体系架构搜索问题模型使用基于指标的搜索策略,即如果Agent要探索新方案,则选择指标最高的未探索方案,否则选择具有最大采样回报值的已探索方案。则证明这种搜索策略可以有效地解决潘多拉问题并得到最好的期望回报值。
本发明还提供一种作战体系架构方案空间探索方法,包括:
将作战体系使命分解为由装备系统执行的任务网络;
给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY;任务节点为由装备系统执行的活动过程,记为TA;
给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型;其中,指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,指控节点表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中,G C2表示指控网络,V C2表示指控节点,E C2表示V C2节点之间的边集合,G SC表示系统节点与指控节点之间的对应关系,V SY表示系统节点,E SC表示节点V SY和节点V C2之间的边集合;
根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n;其中,Agent n表示第n个决策者的方案空间,每个方案空间包括多种作战体系架构,作战体系架构用于对作战体系进行描述;N表示决策者空间,|N|表示决策者的数量;
根据作战体系架构的空间探索问题模型,将作战体系架构的空间探索问题模型中的决策者问题转换为动态规划问题;
采用探索算法进行作战体系架构空间的动态规划;其中,探索算法包括单个决策者搜索算法和多个决策者之间的合作算法;单个决策者搜索算法包括指标排序、指标判断、方案选择三个阶段;
采用单个决策者搜索算法进行作战体系架构空间的动态规划,具体包括:
基于潘多拉规则,确定每个决策者执行不同行动的决策指标:
Figure PCTCN2021070648-appb-000055
Figure PCTCN2021070648-appb-000056
式中,决策指标集合为
Figure PCTCN2021070648-appb-000057
k表示方案,每一个方案代表一个作战体系架构,K n表示第n个决策者的方案集合,m表示行动,M k表示第k个方案的行动集合,
Figure PCTCN2021070648-appb-000058
表示第n个决策者第k个方案的第n个行动的决策指标,
Figure PCTCN2021070648-appb-000059
表示第n个决策者第k个方案的第n个行动对应的开发花费代价,
Figure PCTCN2021070648-appb-000060
表示每个方案的回报值x服从概率分布,x k表示方案k的回报值;
对每个决策者执行不同行动的决策指标进行排序;
将最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合;已开发架构集合与未开发架构集合的并集为决策者方案空间;
确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值;
判断回报值是否大于或等于未开发架构集合中作战体系架构对应的最大决策指标,得到第一判断结果;
若第一判断结果为是,则根据公式
Figure PCTCN2021070648-appb-000061
确定已开发架构集合中最大回报值对应的作战体系架构为最终方案;其中,y n表示第n个决策者的最大回报值,D n表示已开发架构集合;
若第一判断结果为否,则将未开发架构集合中的最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合,然后返回步骤“确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值”。
以下为基于仿真实验对基于决策指标的顺次搜索算法进行分析的实验:
实验设置:假设为了完成某一使命,如边境巡逻、持续侦察、电磁干扰,需要派遣一个无人机集群前往目标区域执行任务。对于这样的新使命,如何建设无人机集群,如何规划集群任务序列,是指挥人员需要解决的问题。面向使命能力的技术方案一般是自顶向下的思路,最后转为基于多Agent系统的求解方法:首先将使命分解为任务网络;其次具有某些功能 的无人机能够完成特定任务,从而构建任务域无人机之间的映射关系;再者无人机之间是具有指控关系的,从而构建指控网络;最后建立多Agent系统模型,每个Agent具有任务列表、特定功能、指控关系,这样的多Agent系统就是一种装备体系架构方案。为使体系作战效能达到最高,需要选择最优的架构方案。
为了评估GSDP(Greedy Search based Dynamic Planning,基于贪婪搜索的动态规划)算法的性能,定义如下统计指标:(1)平均效能Reward,即公式:
CSoSAS:
Figure PCTCN2021070648-appb-000062
S.t.
Figure PCTCN2021070648-appb-000063
Figure PCTCN2021070648-appb-000064
Figure PCTCN2021070648-appb-000065
Figure PCTCN2021070648-appb-000066
Figure PCTCN2021070648-appb-000067
描述的目标函数,其中一次仿真的效能为架构的回报值与累积代价值之差;(2)已知架构次数Number,已开发架构的平均次数;(3)运行时间Time,记录程序运行的时间。其中,平均效能用于评估算法的性能,咨询次数和已知架构次数用于分析搜索过程。
为了对比GSDP算法的性能,设计在OAS问题框架下的四种基准算法。(1)随机算法(Random Algorithm,RA),即Agent在每个时刻随机选择一个动作。具体地,先在集合K中随机选择一个架构k。如果k∈D,即架构已被开发,则结束搜索,并获得该架构的回报值;如果
Figure PCTCN2021070648-appb-000068
即 架构k的未开发且回报值未知,随机选择一个行动,之后返回并继续执行随机动作,直至搜索结束。(2)遍历开发算法(Traversal Development Algorithm,TDA),即Agent将开发所有的架构,获得每个架构的采样回报值。对于一个未开发的架构k,Agent选择代价最小的动作进行开发,
Figure PCTCN2021070648-appb-000069
当Agent完成了所有架构的开发,选择所有已开发架构中回报值最高的架构作为最终方案。(3)通用开发算法(General Development Algorithm,GEA),类似于采用指标判断的算法。局部探索算法的评判指标是最高期望值与开发代价之差,即
Figure PCTCN2021070648-appb-000070
当已开发架构中的最高回报值已经超过该指标,则停止搜索并选择具有最高回报值的架构。
实验结果:
设计四个场景用以评估架构空间的可扩展性,架构方案数量分别为K={20,100,1000,10000},开发每个架构有3种行动,每种行动的代价分别服从三种均匀分布U 1(1,3),U2(0.5,4),U3(1.5,2.5),每种架构的回报值服从概率分布W k(w k)~U(a k,b k),其中,a k~U(50,60),b k~U(90,100)。考察Agent的数量,选择的方案数NoS(NoS为选择的方案数量,即多Agent系统最终选择多少个方案),为1到10时,各个算法的性能指标。图6为基于决策指标的顺次搜索算法进行分析实验的实验数据分析图。
场景B1:在包含K=20个架构的方案空间中探索,选择最优架构,如下表p1所示,LEA为局部遍历算法。
表p1
Figure PCTCN2021070648-appb-000071
Figure PCTCN2021070648-appb-000072
场景B2:在包含K=10 4个架构的方案空间中探索,选择最优架构,如下表p2所示。
表p2
Figure PCTCN2021070648-appb-000073
场景B3:在包含K=10 6个架构的方案空间中探索,选择最优架构,如下表p3所示。
表p3
Figure PCTCN2021070648-appb-000074
Figure PCTCN2021070648-appb-000075
在本实施例中,本发明提供的基于超网络模型的作战体系架构建模方法,基于体系架构能力生成要素,根据作战体系架构的形式化定义、多Agent动态规划问题以及作战体系架构方案空间探索问题框架,能够构建出具多架构方案选取的作战体系架构模型,便于决策者做出最佳选择;该作战体系架构方案空间探索算法,为基于决策指标的并行搜索算法,该算法为多项式时间的,其回报值明显优于其他基准算法,其在作战体系架构方案空间独立情况的假设下是最优的。
本文中应用了具体个例对本发明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发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处。综上所述,本说明书内容不应理解为对本发明的限制。
提供以上实施例仅仅是为了描述本发明的目的,而并非要限制本发明的范围。本发明的范围由所附权利要求限定。不脱离本发明的精神和原理而做出的各种等同替换和修改,均应涵盖在本发明的范围之内。

Claims (8)

  1. 基于超网络模型的作战体系架构建模方法,根据能力生成机理,将任务、装备系统和指控结构作为作战体系架构的核心要素,所述作战体系架构由任务网络、系统网络和指控网络三种异构网络组成,其特征是:具体包括以下步骤:
    S1、将作战体系使命分解为可由装备系统执行的任务网络;
    S2、给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY;任务节点为可由装备系统执行的活动过程,记为TA;
    S3、给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型;其中,指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中E SC表示节点V SY和节点V C2之间的边集合;
    S4、根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n;
    S5、根据作战体系架构搜索问题模型,将作战体系搜索问题模型中的决策者问题转换为一种形式化的动态规划问题。
  2. 根据权利要求1所述的基于超网络模型的作战体系架构建模方法,其特征是,所述任务网络可抽象为有向图,记为G TA=<V TA,E TA>,其 中,V TA表示任务节点集合,E TA表示节点之间的边集合;每个所述任务网络均具有起始任务节点、结束任务节点和中间节点;所述任务网络包括因果关系和并列关系两类逻辑关系。
  3. 根据权利要求1所述的基于超网络模型的作战体系架构建模方法,其特征是,所述系统网络表示系统节点之间的功能关系,记为G SY=<V SY,E SY>,其中V SY表示系统节点集合,E SY表示系统节点之间的边集合;所述任务节点和系统节点之间的对应关系定义为二分图,记为G TS=<V TA,V SY,E TS>,其中E TS表示节点V TA和节点V SY之间的边集合。
  4. 根据权利要求1所述的基于超网络模型的作战体系架构建模方法,其特征是,所述指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>;所述系统节点和指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中E SC表示节点V SY和节点V C2之间的边集合。
  5. 根据权利要求1所述的基于超网络模型的作战体系架构建模方法,其特征是,所述作战体系架构的拓扑模型由三类节点与五类关系构成的异构网络GA,记为GA=<V TA,V SY,V C2,E TA,E SY,E C2,E TS,E SC>;所述作战体系架构的开发花费代价定义为c,且c∈C,所述作战体系架构的潜在能力定义为w,作战体系架构具有一定完成使命的能力w∈W;所述作战体系架构模型由作战体系架构的拓扑模型、开发代价和能力组成,记为<GA,C,W>。
  6. 一种作战体系架构方案空间探索算法,基于权利要求1至5任意 一项所述的基于超网络模型的作战体系架构建模方法,其特征是:所述搜索算法基于作战体系架构空间的动态规划,且该算法通过判断定义的指标来进行动作选择,具备包括以下步骤:
    1)、判定指标,基于经典潘多拉规则,定义每个决策者执行不同行动的决策指标:
    Figure PCTCN2021070648-appb-100001
    推导出:
    Figure PCTCN2021070648-appb-100002
    根据状态
    Figure PCTCN2021070648-appb-100003
    和指标集合
    Figure PCTCN2021070648-appb-100004
    设计一种简单且最优的搜索规则;
    2)、搜索算法,根据步骤1)中的搜索规则,将最优方案的计算简化为指标的判断,所述搜索算法包括单个决策者搜索算法和多个决策者之间的合作算法;
    3)、将单决策者搜索算法分为指标排序、指标判断、方案选择三个阶段,并在Sorting程序,根据公式
    Figure PCTCN2021070648-appb-100005
    计算所有架构中所有动作的指标,然后对改指标进行排序,并将排序结果保存至向量,然后调用Developing程序求得最优架构方案。
  7. 根据权利要求6所述的一种作战体系架构方案空间探索算法,其特征是:步骤2)中每个所述指标的计算均是独立的,且所述指标不受其他作战体系架构回报值概率分布的影响。
  8. 一种作战体系架构方案空间探索方法,其特征是:方法包括:
    将作战体系使命分解为由装备系统执行的任务网络;
    给出任务节点与系统节点之间的对应关系,并根据任务网络和任务节点与系统节点之间的对应关系构建系统网络,其中,系统节点为具有特定功能并能够独立完成特定任务的装备,记为SY;任务节点为由装备系统执行的活动过程,记为TA;
    给出系统节点和指控节点之间的对应关系,并根据该关系建立指控网络,并结合任务网络、系统网络构建成作战体系架构模型;其中,指控节点为用于处理信息、管理组织、决策规划和控制反馈的逻辑节点,指控节点表示为C2,指控网络为通过指令关系连接所有指控节点的组织网络,记为G C2=<V C2,E C2>,系统节点与指控节点之间的对应关系定义为二分图,记为G SC=<V SY,V C2,E SC>,其中,G C2表示指控网络,V C2表示指控节点,E C2表示V C2节点之间的边集合,G SC表示系统节点与指控节点之间的对应关系,V SY表示系统节点,E SC表示节点V SY和节点V C2之间的边集合;
    根据作战体系架构模型和决策者n,构建作战体系架构的空间探索问题模型,其中,决策者n∈N,N={1,2,...,|N|},第n个决策者记为Agent n;其中,Agent n表示第n个决策者的方案空间,每个方案空间包括多种作战体系架构,作战体系架构用于对作战体系进行描述;N表示决策者空间,|N|表示决策者的数量;
    根据作战体系架构的空间探索问题模型,将作战体系架构的空间探索问题模型中的决策者问题转换为动态规划问题;
    采用探索算法进行作战体系架构空间的动态规划;其中,探索算法包括单个决策者搜索算法和多个决策者之间的合作算法;单个决策者搜索算法包括指标排序、指标判断、方案选择三个阶段;
    采用单个决策者搜索算法进行作战体系架构空间的动态规划,具体包括:
    基于潘多拉规则,确定每个决策者执行不同行动的决策指标:
    Figure PCTCN2021070648-appb-100006
    Figure PCTCN2021070648-appb-100007
    式中,决策指标集合为
    Figure PCTCN2021070648-appb-100008
    k表示方案,每一个方案代表一个作战体系架构,K n表示第n个决策者的方案集合,m表示行动,M k表示第k个方案的行动集合,
    Figure PCTCN2021070648-appb-100009
    表示第n个决策者第k个方案的第n个行动的决策指标,
    Figure PCTCN2021070648-appb-100010
    表示第n个决策者第k个方案的第n个行动对应的开发花费代价,
    Figure PCTCN2021070648-appb-100011
    表示每个方案的回报值x服从概率分布,x k表示方案k的回报值;
    对每个决策者执行不同行动的决策指标进行排序;
    将最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合;已开发架构集合与未开发架构集合的并集为决策者方案空间;
    确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值;
    判断回报值是否大于或等于未开发架构集合中作战体系架构对应的最大决策指标,得到第一判断结果;
    若第一判断结果为是,则根据公式
    Figure PCTCN2021070648-appb-100012
    确定已开发架构集合中最大回报值对应的作战体系架构为最终方案;其中,y n表示第n个决策者 的最大回报值,D n表示已开发架构集合;
    若第一判断结果为否,则将未开发架构集合中的最大决策指标对应的作战体系架构添加至已开发架构集合,更新未开发架构集合,然后返回步骤“确定已开发架构集合中的作战体系架构的空间探索问题模型对应的回报值”。
PCT/CN2021/070648 2020-11-13 2021-01-07 基于超网络模型的组织架构建模方法及空间探索算法 WO2022099915A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
GB2200021.0A GB2618302A (en) 2020-11-13 2021-01-07 Hypernetwork model-based organization architecture modeling method and space exploration algorithm

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011267527.2 2020-11-13
CN202011267527.2A CN112632744B (zh) 2020-11-13 2020-11-13 基于超网络模型的作战体系架构建模方法及空间探索方法

Publications (1)

Publication Number Publication Date
WO2022099915A1 true WO2022099915A1 (zh) 2022-05-19

Family

ID=75303252

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/070648 WO2022099915A1 (zh) 2020-11-13 2021-01-07 基于超网络模型的组织架构建模方法及空间探索算法

Country Status (2)

Country Link
CN (1) CN112632744B (zh)
WO (1) WO2022099915A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115906673A (zh) * 2023-01-10 2023-04-04 中国人民解放军陆军工程大学 作战实体行为模型一体化建模方法及系统
CN116108613A (zh) * 2022-11-07 2023-05-12 中山大学 一种可解释的装备组合快速构建方法、系统、设备及介质
CN116720296A (zh) * 2023-06-19 2023-09-08 湖南大学 一种边缘作战体系网络结构优化方法、装置及设备
CN118242935A (zh) * 2024-03-22 2024-06-25 哈尔滨工业大学 一种基于复杂网络的装备体系杀伤链生成方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113111440B (zh) * 2021-04-26 2023-02-03 河北交通职业技术学院 基于逻辑关系的集群无人机任务模型构建方法
CN113742998A (zh) * 2021-08-03 2021-12-03 中国人民解放军国防科技大学 一种基于超网络模型的作战体系架构建模方法
CN113566831B (zh) * 2021-09-26 2021-12-07 中国人民解放军国防科技大学 基于人机交互的无人机集群导航方法、装置和设备
CN114239275B (zh) * 2021-12-17 2024-09-10 中国人民解放军国防科技大学 基于适变网络的无人机群多目标优化与多属性决策方法
CN116489193B (zh) * 2023-05-04 2024-01-23 中国人民解放军陆军工程大学 一种作战网络自适应组合方法、装置、设备及介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7930196B2 (en) * 2001-04-11 2011-04-19 Fair Isaac Corporation Model-based and data-driven analytic support for strategy development
CN110401564A (zh) * 2019-07-04 2019-11-01 大连交通大学 基于相对混合择优的指挥控制超网络模型构建方法
CN110688754A (zh) * 2019-09-25 2020-01-14 中国人民解放军国防科技大学 一种作战体系架构建模与最优搜索方法
CN110929394A (zh) * 2019-11-14 2020-03-27 北京华如科技股份有限公司 基于超网络理论的联合作战体系建模方法以及存储介质
CN111784135A (zh) * 2020-06-22 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 基于超网络和ooda环理论的体系作战能力量化分析方法
CN111783291A (zh) * 2020-06-22 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 基于ooda环理论的作战体系超网络建模方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10488068B2 (en) * 2017-07-25 2019-11-26 Mitsubishi Electric Research Laboratories, Inc. System and method for generating a controller for multi-zone heating, ventilation, and air conditioning system
CN110580404B (zh) * 2019-07-26 2023-05-12 东南大学 一种基于城市多模式交通网络的网络运能确定方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7930196B2 (en) * 2001-04-11 2011-04-19 Fair Isaac Corporation Model-based and data-driven analytic support for strategy development
CN110401564A (zh) * 2019-07-04 2019-11-01 大连交通大学 基于相对混合择优的指挥控制超网络模型构建方法
CN110688754A (zh) * 2019-09-25 2020-01-14 中国人民解放军国防科技大学 一种作战体系架构建模与最优搜索方法
CN110929394A (zh) * 2019-11-14 2020-03-27 北京华如科技股份有限公司 基于超网络理论的联合作战体系建模方法以及存储介质
CN111784135A (zh) * 2020-06-22 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 基于超网络和ooda环理论的体系作战能力量化分析方法
CN111783291A (zh) * 2020-06-22 2020-10-16 中国人民解放军军事科学院国防科技创新研究院 基于ooda环理论的作战体系超网络建模方法

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116108613A (zh) * 2022-11-07 2023-05-12 中山大学 一种可解释的装备组合快速构建方法、系统、设备及介质
CN116108613B (zh) * 2022-11-07 2023-12-01 中山大学 一种可解释的装备组合快速构建方法、系统、设备及介质
CN115906673A (zh) * 2023-01-10 2023-04-04 中国人民解放军陆军工程大学 作战实体行为模型一体化建模方法及系统
CN115906673B (zh) * 2023-01-10 2023-11-03 中国人民解放军陆军工程大学 作战实体行为模型一体化建模方法及系统
CN116720296A (zh) * 2023-06-19 2023-09-08 湖南大学 一种边缘作战体系网络结构优化方法、装置及设备
CN118242935A (zh) * 2024-03-22 2024-06-25 哈尔滨工业大学 一种基于复杂网络的装备体系杀伤链生成方法

Also Published As

Publication number Publication date
CN112632744A (zh) 2021-04-09
CN112632744B (zh) 2023-05-16

Similar Documents

Publication Publication Date Title
WO2022099915A1 (zh) 基于超网络模型的组织架构建模方法及空间探索算法
WO2021057329A1 (zh) 一种作战体系架构建模与最优搜索方法
Ren et al. A reinforcement learning method for constraint-satisfied services composition
Isaila et al. Collective I/O tuning using analytical and machine learning models
Saeedvand et al. Novel hybrid algorithm for Team Orienteering Problem with Time Windows for rescue applications
Schillinger et al. Hierarchical ltl-task mdps for multi-agent coordination through auctioning and learning
Hoang et al. A large neighboring search schema for multi-agent optimization
Le et al. ER-DCOPs: A framework for distributed constraint optimization with uncertainty in constraint utilities
Apers et al. Simulation of quantum walks and fast mixing with classical processes
Nagm-Aldeen et al. A literature review of business process modeling techniques
CN112700099A (zh) 基于强化学习和运筹学的资源调度规划方法
Zhou et al. On the opportunities of green computing: A survey
Dorier et al. Hpc storage service autotuning using variational-autoencoder-guided asynchronous bayesian optimization
Cao et al. TASS: Transaction assurance in service selection
Srighakollapu et al. Optimizing network topology for average controllability
de Almeida et al. Bigdatanetsim: A simulator for data and process placement in large big data platforms
CN113312890B (zh) 多表拼接方法、装置、电子设备及存储介质
Siryani et al. Framework using Bayesian belief networks for utility effective management and operations
Bettker et al. Understanding sample generation strategies for learning heuristic functions in classical planning
Ghoshal et al. Two approaches for the min-degree constrained minimum spanning tree problem
Quan et al. Multi-objective evolutionary scheduling based on collaborative virtual workflow model and adaptive rules for flexible production process with operation reworking
Huang et al. A novel mission planning model and method for combat system-of-systems architecture design
Li et al. A PTS-PGATS based approach for data-intensive scheduling in data grids
Atwal et al. DeepSDN: connecting the dots towards self-driving networks
CN116718198B (zh) 基于时序知识图谱的无人机集群的路径规划方法及系统

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 202200021

Country of ref document: GB

Kind code of ref document: A

Free format text: PCT FILING DATE = 20210107

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

Ref document number: 21890453

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: 21890453

Country of ref document: EP

Kind code of ref document: A1