Distributed tasks-platforms scheduling method to holonic-C2 organization
WANG Xun1,*,YAO Peiyang1,ZHANG Jieyong1,WAN Lujun2,and JIA Fangchao3
1.Information and Navigation College,Air Force Engineering University,Xi’an 710077,China;
2.Air Traffic Control and Navigation College,Air Force Engineering University,Xi’an 710051,China;
3.Sergeant School,Air Force Communication,Dalian 116600,China
Abstract: To solve the problem of distributed tasks-platforms scheduling in holonic command and control(C2)organization,the basic elements of the organization are analyzed firstly and the formal description of organizational elements and structure is provided.Based on the improvement of task execution quality,a single task resource scheduling model is established and the solving method based on the m-best algorithm is proposed.For the problem of tactical decision-holon cannot handle tasks with low priority effectively,a distributed resource scheduling collaboration mechanism based on platform pricing and a platform exchange mechanism based on resource capacities are designed.Finally,a series of experiments are designed to prove the effectiveness of these methods.The results show that the proposed distributed scheduling methods can realize the effective balance of platform resources.
Keywords: command and control(C2),decision-holon,distributed task allocation,task execution quality,platform price,order optimization.
1.Introduction
In recent years,decision-making problems in military field are highly concerned.A military organization with a more flexible decision-making mode is more likely to win the victory of the modern warfare[1].The traditional command and control(C2)organizational structure keeps authority and information at the center because the centralized scheduling can achieve global planning.However,the rigid structure cannot adapt to the networked and fastpaced battlefield environment[2,3].
“Holonic”derives from the word“holon”,which was developed by Koestler[4]in the context of social organizations and living organisms.“Holon”means a combination of“wholes”and“part”.It is a unit with autonomy and cooperation[5].On the one hand,holon can handle circumstances and incidents based on its own information and knowledge.On the other hand,holon can receive instructions from or be controlled by a higher lever holon.These characteristics of holon ensure the effectiveness of complex military operations.The flattened structure and marginalized power created by holons can adapt to the high confrontational combat environment[6].This paper mainly studies the distributed tasks-platforms scheduling method under the holonic-C2 organizational framework.A comprehensive process involves those elements–who,what,why,how,where,when,with what,implying“who”makes the plan(decision makers),“what”needs to be planned(missions,tasks,actions to be executed by using resources),“why”makes the plan(objective function or desired goal),“how”to achieve the expected outcome(the assignment of platforms to tasks),“where”and“when”the plan is executed(task location and time),and“with what”facilities to make the plan(information about tasks,platforms,etc)[7].The problem we solve is“why”and“how”.We mainly research on the tactical level planning problem.In the holonic framework,there are multiple tactical decision holons(TDHs)and one operational decision holon(ODH).TDHs are key decision makers and ODH is a coordinator.The distributed coordination mechanism is studied to match platform capabilities with task requirements.
（2）教师要深入把控教学生成问题。“植物的生殖”一课中，教师采用了生物学中常用的对比学习法，对有性生殖和无性生殖两种生殖方式在形成过程、遗传特性、与母体的性状一致性和后代个体变异等多个方面进行对比的同时，让学生体会到两种生殖方式各有利弊，从而知道：人们应当趋利避害，取长补短，在植物不同的生命阶段，采用不同的生殖方式以繁衍后代。在课堂中，教师还需分配更多时间、列举更多例子来帮助学生分析和理解为了适应环境以实现种族的繁衍，两种生殖方式应当如何合理搭配应用，将理论联系生活实际，服务于农业生产。
Recently,the idea of distributed tasks planning is widely used in unmanned aerial vehicles(UAVs),the multi-robot system,the manufacturing system,etc.The distributed planning method is a hotspot in the field of distributed artificial intelligence[8–12].Chen et al.established the task allocation model for the problem of multiple UAVs decentralized cooperative air combat.The asynchronous consensus-based auction algorithm was improved to solve the planning problem in[13].Zhang et al.proposed a distributed blackboard decision-making framework for collaborative planning that dynamically allocates tasksplatforms scheduling to accomplish mission objectives.The methodology combines with the nest genetic algorithm to maximize the tasks execution accuracy and minimize the weighted total workload of the decision-maker[14].The above methods can deal with the distributed tasks allocation problem.However,on the one hand,they design the methods based on intelligent optimization algorithms.These algorithms cannot output a stable solution and result in poor practicality.On the other hand,they ignore the parallel characteristic of distributed planning.The process of distributed decision-making has not been analyzed well.
We design a kind of distributed tasks-platforms scheduling method based on the holonic-C2 organizational structure in this paper.TDHs make their own resource scheduling plan and ODH works as a coordinator when the task cannot be finished by one TDH.The single task platform scheduling model,taking the quality of task execution as the objective function,is established.We propose the solution optimization algorithm based on m-best to solve the model.In the distributed coordination part,for the problem of platform shortage for low priority tasks,we propose the platform pricing model,the price-based order optimization algorithm and the platform exchange method based on resource capability.Finally,the validity of the proposed method is proved by experiments.
Step 1 Calculate the prices of the platforms in task Ti’s plan.Set k=d,c=1.
2.C2 structure of holonic-C2 organization
The C2 structure of the holonic-C2 organization is constructed on the basis of holon units[15].Holon is a hierarchical structural unit with autonomy and collaboration.Each holon unit consists of multiple sub-holon units.Multiple holon units can form a higher level holon.Each holon has the characteristics of autonomy and collaboration.Autonomy means that holon itself has the ability to complete the given tasks autonomously.Collaboration is the ability of holon to accept the upper command and collaborate with other holons to complete tasks[16].Therefore,we regard the autonomous and cooperative entity in the holonic-C2 organization as the holon unit.Firstly,we define the concept of elements in the holonic-C2 organization.
行政合同是行政机关为了实现行政管理目的，与公民、法人或者其他组织之间，经双方意思表示一致所达成的协议。行政合同作为传统行政管理的替代性方式或者手段，在我国行政领域得到了越来越广泛的运用。一般来讲，只要符合两个条件，就可以采用行政合同的方式进行行政管理：一是法律法规没有相反的规定，即法律法规没有禁止性规定。二是行政管理的性质适宜采用行政合同的方式进行管理，即行政管理事务的性质具有合意或者讨价还价的空间，或者说行政管理事务的性质适宜通过合同建立与公民、法人或其他组织之间的行政法上的权利义务关系。
Definition 1 Resource:Resource is a kind of measureable capability provided by platforms used in the processing of tasks.Resource cannot be divided into smaller units.Each task is specified by resource requirements in each functional resource category and each platform provides resource capabilities in each resource category.The resource capacity vector is denoted by r=[r1,r2,...,rl,...,rL],where L is the number of resource capabilities.
Definition 2 Task:Task is the action that operational entity takes to achieve some certain operational purposes.The set of tasks is denoted by T={T1,...,Ti,...,Tn},where n is the number of tasks.Every task requires a set of relevant resources to be processed.In this paper,a task i(i=1,2,...,n)is defined by its properties including ts,i,tp,I,Ri and ρ.ts,i is the start time of the task i.tp,i is the processing time of the task i.Ri is the resource requirement vector denoted by[Ri1,...,Ril,...,RiL],where Ril is the number of units of resource type l(l=1,2,...,L)required for successful processing of the task i.ρ is the task priority.
Definition 3 Platform-holon (PH) or platform:Platform-holon is the physical entity with specified resource capabilities to execute the tasks.It is a direct participant in combat.The platform holon set is denoted by PH={PH1,...,PHj,...,PHm}.m is the number of the platform holon.Each platform holon j(j=1,...,m)is defined by its properties including rj,and nj.rj is the resource capability vector denoted by[rj1,...,rjl,...,rjL],where rjl is the number of units of resource type l possessed by the platform holon j.
Definition 4 Decision-holon(DH):Decision-holon is a basic unit with independent decision making,interactive collaboration and C2 capabilities in the holonic-C2 organization.According to different levels,DH can be divided into ODH and TDH.
These relations among tasks,resources and platformholons in the holonic-C2 organization are shown in Fig.1.
In the macroscopic sight,according to the different levels of C2,holonic-C2 organization elements can be divided into the operational holon and the tactical holon.Operational holon consists of one ODH and several tactical holons.A tactical-holon has one TDH and several platform-holons.In this paper,we only study the organization with one operational holon and several tactical holons.
患者如院后，若患者病情发生变化，则采用改良评估量表对患者进行常态化评估，并记录评估结果。当评分≥1分时，护理人员应给予患者护理措施干预，防止压疮发生。
Step 3 If QT new ＞QTthreshold,y new is the plan to the task Ti.Otherwise,go to Step 2.
Fig.1 Relations among tasks,resources and platform-holons in holonic-C2 organization
Fig.2 Holonic-C2 organization structure
3.Holonic-C2 organization distributed tasks-platforms scheduling model
3.1 Measurement model of tasks-platforms scheduling
The measure model of tasks-platforms scheduling is used to judge whether the plan is good or not.In this paper,the task execution quality model is designed as the measure[18,19].The higher the quality is,the better the plan is.The concepts about the task execution quality are defined as follows.
Definition 5 Tasks-platforms assignment: The platform-to-task assignment determines the types and number of platforms to execute each individual task.It is denoted by Y =[yi],where the ith task assignment is represented by a column vector given by
where yij indicates the number of the platform,type j is allocated to task i,and Z is the set of integers.
Definition 6 Resource satisfaction degree: The resource satisfaction degree is the ratio of the number of the lth resource provided by platforms to the number of resources needed by the task i and the maximum value is 1.The lth resource satisfaction degree for the task i is denoted as follows:
Definition 7 Task execution quality:The task execution quality is related with how well the platforms’capabilities match the task requirements.
第三，积极制定并落实相关配套措施，建立健全社会保障体系。借鉴国外在产假制度的完善措施方面的经验，在产假的基础上，建立陪产假和育儿假制度，使男女共同承担起生育和抚养孩子的责任，保障男女有平等的就业机会。在女性离职生育和其丈夫离职进行陪产期间，保障其工资、奖金、津贴的正常发放和职位升降的正常进行，提供更加完善的生育津贴制度，健全医疗保障体系，为女性职工提供津贴和补助。在企业内部及社区加强相关育婴设施建设，减少女性职工的后顾之忧，为其提供良好的哺育环境，保障女性职工的权益。
To ensure the task can be executed,any kind of resource needed by the task cannot be lacking.Therefore,the geometric average of all resource satisfaction degrees is used to measure the task execution quality.Without any kind of resource,it will be 0.The definition of the task i execution quality as follows:
where
is the resource needed to process the task i.
is the number of element in
3.2 Distributed tasks-platforms scheduling model and solving method
3.2.1 Distributed tasks-platforms scheduling model
Platform planning is proven to be a non-deterministic polynomial(NP)-hard problem[20].It is difficult to efficiently solve especially distributed planning.Therefore,we divide the distributed tasks-platforms scheduling problem into two stages.One is each TDH planning for its own tasks and platforms.The other is distributed cooperation over multiple TDHs when the task cannot be finished by one TDH.In this part,we mainly study the tasks-platforms planning problem for the TDH.Each TDH makes the plan for tasks one by one by using its own available platforms at that moment.All tasks are handled from high to low according to the tasks priority.The single task scheduling model is built first.
(i)Objective function
The task execution quality model is given in(3).If the maximized task execution quality(QT)is used as the objective function,the tasks-platforms planning will be a mixed integer nonlinear programming problem.It is hard to be solved because of complex calculations.It is inefficient compared with linear programming.Moreover,it is easy to appear the lack of resources because the TDH only matches the local tasks with local platforms.The nonlinear programming cannot provide a stable solution.This is not conducive to the cooperation stage.Compared to nonlinear programming,linear programming can overcome the above shortcomings.Although the solution may lack some resources,it can be remedied in the cooperation stage.
In this paper,we take the sum of resource satisfaction as the objective function and the task execution quality as the standard to judge whether the solution is good or not.The objective function for the task Ti is built as follows:
where
is the resource type set needed by Ti.(ii)Constraints
The available platforms vector of the kth TDH when handling the task Ti is denoted by 
where
is the number of platform PHj belonging to TDHk.The task set done by TDHk is denoted by TΦ(k).The following two constraints should be satisfied when dealing with the task Ti.
Constraint 1 The total number of platforms allocated to the task Ti is no greater than a certain value.
Constraint 2 The number of platforms allocated to the task Ti is no greater than the number of available platforms.
Above all,the single task resource scheduling model is designed as follows:
据了解，淄博全市共规划21处以医带养型养老服务设施、10处以医托养型养老服务设施、1处以医联养型养老服务设施、15处医养共建型养老服务设施。至2020年，基本实现医疗卫生和养老服务资源有序共享、覆盖城乡、规模适宜、功能合理的医养结合服务格局。
3.2.2 Solving method
The mathematical model built in(7)is a typical mixed integer linear programming model.We can solve the problem by using the branch and bound algorithm[21].By solving the single task model we can find the optimal solution of the task by using available platforms.However,the method of dealing with tasks one by one leads to resource allocation imbalance.The task with high priority occupies the strong platforms.It is not conducive to the resource satisfaction of low priority tasks.To alleviate this problem,we use the m-best algorithm[22]to optimize the solution.
The idea of the m-best algorithm is reserving more resources for the following tasks by reducing the quality of the current task appropriately.Finally,we can get a better global solution than before.The steps are as follows:
Step 1 Solve the single task scheduling model for Ti.Get the optimal solution as the 1-best solution 
.The objective function value is
Put the solution into the candidate set.Set p=2,q=1.
箫声流淌，四小姐进入乐音营造的世界。她仿佛看见，一只哀鸿在寒潭照影，在青霄孤鸣，心中突如其来涌上阵阵感动，再看面前的腊梅，仿佛也跟着颤动。
Step 2 Select nonzero elements’location v from(p-1)-best solution.The number of elements is V.The constraint condition 
is added to(7).Solve the new model and get p-best to(p+V)-best solutions.The objective function value set is
Put these solutions into the candidate set.Set p=p+V.
Step 3 If p ≥m,retain m best solutions and go to Step 4.Otherwise,go to Step 2.
智慧教室以学习者为中心，打破物理空间的束缚，为学习者打造一个开放的学习空间。其不再是一间封闭的教室。网络互联扩延了传统教室的空间，基于网络的互动教学突破了空间局限，调动了生生之间、师生之间、班级之间的互动积极性，让区域间的班级互动成为可能，学习也由个体的机械记忆转变为集体的、互动的过程。
Step 5 If q ＞m,go to Step 6.Otherwise,go to Step 4.
Step 6 Select the maximum value Sumd from SUM.The d-best solution is the plan for the task Ti.
1.2.4.6 含量测定。取不同产地的咖啡生豆和焙炒豆，按“1.2.2”方法制成供试品溶液，按照“1.2.3”色谱条件测定咖啡中绿原酸、葫芦巴碱、D-(-)-奎宁酸、咖啡酸的含量。
4.Holonic-C2 organization distributed cooperative mechanism
4.1 Distributed resource parallel scheduling process
The key of the distributed scheduling is parallel planning of multi-TDHs.All TDHs make their own plans synchronously as shown in Fig.3.Each TDH deals with its own tasks one by one.The processing time for each task is divided into two parts.One is the plan generation time.During this time,the C2 planning system generates the platforms scheduling scheme for the task.The other is the planning time of platforms’action.The former is insignificant compared with the latter.Every time a task is handled,TDHk updates the shared platform and price information to the coordinator.(The ODH works as a coordinator in this paper.)When the execution quality of the task is less than the specified minimum threshold(QTthreshold),TDHk will submit the task and platforms to the coordinator.The coordinator generates the platform collaboration plan by using the order optimization algorithm and the platform exchange algorithm.Finally,the coordinator sends the plan to TDHk and updates the available platform information to all TDHs.
Fig.3 Distributed parallel tasks process
Actually,the platform action planning time is obtained from the actual operation.We have to make approximations and assume that the planning time 
is proportional to the number of platforms assigned to Ti,as shown below
where ω is a constant,ω=1.
4.2 Platform pricing method
In the distributed coordination phase,the coordinator finishes the task collaboration based on the information of sharable platforms provided by all TDHs.In fact,these platforms can be classified into two categories.One is assigned platforms;the other is unassigned platforms.The price of these platforms is an important basis for the coordinator to coordinate tasks.Therefore,how to price the two kinds of platforms reasonably is the key to achieve the distributed task coordination.
4.2.1 Price model of assigned platform
These assigned platforms include sharable platforms and unshared platforms.We price the platform mainly according to its impact on the task execution quality.If the quality is lower than the threshold by the original plan without PHj′,the platform PHj′cannot be shared and its price is infinite.If not,the platform PHj′can be shared and we price it based on the utilization rate of resource capacities as shown in(9).The price of PHj′is the ratio of its contribution to the task Ti to the sum of its resource capacities.
where min(a,b)is taking the minimum from a and b.
yi is the plan for task Ti without PHj′.
4.2.2 Price model of unassigned platform
When price the unassigned platforms,we need to calculate its demand degree for following tasks.As shown in(10),the price is the ratio of PHj’s maximum resource utilization amount for unprocessed tasks to the sum of its resource capacities.
近年来，关于学生学业负担过重的问题已经引起社会相关部门和机构的高度重视，教育行政部门也明确提出给学生减负的号召。但面对升学的压力，作为教育的主战场，需要结合实际教学的需要，通过在学生作业的布置方式和批改方式等方面，研究出具体的优化措施。
Step 4 Select q-best solution from the candidate set. Solve the models of the task Ti+1 to Tn and get their 1-best solution and objective function value set
Calculate the sum of the values Put Sumq into set SUM.q=q+1.
where T unf is the set of unprocessed tasks.
4.3 Order optimization algorithm based on price
The order optimization is an optimization mechanism based on the greedy strategy.It includes two stages.One is the accumulation stage.In this stage,we construct a feasible solution by adding platforms constantly.The other is the reduction stage during which we delete redundant platforms.The simplest platform scheduling scheme is obtained in the end.
4.3.1 Platforms accumulation method
The basic flow of the platforms accumulation method is as follows.
Step 1 All platforms in the shareable platforms set are sorted by price from low to high.The number of platforms is num share PH;
Step 2 If num share PH ＞0,go to Step 3.Otherwise,stop the algorithm,output“Lack of resource and the task cannot be completed”.
Step 3 Select the lowest price platform and add it to the task plan.Calculate the task execution quality QT.
The above introduces the organization elements.The relations between elements are also important for the organization.Holonic-C2 organization relations include the C2 relation between ODH and the tactical holon,the cooperation relation among tactical holons,the execution relation between the tactical holon and the task,the C2 relation between TDH and the platform,the allocation relation between the platform and the task,tasks sequential relation and so on[17].
Step 4 If QT ＞QTthreshold,go to Step 5.Otherwise,go to Step 3.
Step 5 Output the plan to the task Ti.
4.3.2 Platforms reduction method
在文本当中，有一些内容是不符合常理的，而这些内容也正是作者所强调的内容所在。如果学生能理解这些内容，学生对于文本的理解也就更加的深入了。例如教学《台阶》时，文中提到父亲花费巨大精力把九级台阶的房子造好了，但是他却并没有高兴，反而感到很难过，按照常理，父亲在新房子建好之后应当感到高兴，而在父亲脸上流露出来的只有惆怅。为什么会这样呢？因为父亲的腰闪了，对于一个劳力来说，腰受伤了，就表示劳动能力的下降，这对于父亲来说是致命的打击。引导学生通过关注文本当中的矛盾，加深了对文本的理解。
The basic flow of the platforms reduction method is as follows.
Step 3 Judge whether the execution qualities of the task Tiand
original task are both greater than the threshold QTthreshold.If yes,go to Step 5.Otherwise,restore the exchanged platforms and go to Step 4.
Step 2 Delete the combination PH Cm from the original plan.Generate the new plan y new and calculate its task execution quality QT new,m=m+1.
The structure of the holonic-C2 organization is shown in Fig.2.Different from the traditional C2 organization,the mission planner of the holonic-C2 organization is not just ODH.TDHs also have the ability to make their own plan.In the holonic-C2 organization,the decision mode is not only centralized decision making in which ODH is the decision maker,but also distributed decision making in which TDH is the decision maker.In this paper,we study the distributed negotiation mechanism among TDHs to build the optimal allocation relationship between tasks and platforms.
Step 4 Output the plan to the task Ti.
4.4 Platform exchange algorithm
To solve the problem of“Lack of resource and the task cannot be completed”appearing in the order optimization algorithm,we design the platform exchange algorithm basing on platform’s resource capacity.According to the resource capacity lacked in the task’s plan,select the lowestpriced platform from the plan and exchange it to a platform with the resource capacity.
从以上列举的译文中可以看出，对于“道”的第一种释义，三者的译法各不相同，分别选用了“the Way”、“truth”、“doctrine”作为对应词。
We assume that TDHd’s task Ti for lack of resource capacity rl needs to be cooperated.The platforms with rl controlled by TDHdis
The number of platforms is num PHr.
The basic flow of the algorithm is as follows.
2.1 脂蛋白定义与分类 脂蛋白由蛋白质结合脂类形成，它们通过淋巴和循环系统输送脂质，是诱发认知障碍甚至痴呆的危险因素[20]。2016年修订版《中国成人血脂异常防治指南》将脂蛋白分为乳糜微粒、HDL、LDL、中密度脂蛋白、极低密度脂蛋白5大类，此外还有一种脂蛋白称为脂蛋白a[21]。脂蛋白内的蛋白质组分称为载脂蛋白(apolipoprotein，apo)，如apoA、apoB等。
Step 2 Select the lowest-priced platform and exchange it with 
《绸缪》是一首新婚诗歌。 新婚的晚上，看到新人的美丽，如在梦境之中，不知如何是好。 值得注意的是，这是正规的婚姻，不是其他国风里大量反映的野合，所以乐而不放荡。
Step 1 The full combinations of all platforms assigned to the task Ti are put in the set PH C.The number of platforms is M and the total number of combinations is 2M-1.Calculate the price of each combination.Ranking these combinations according to their prices from high to low and build the new set PH C={PH C1,...,PH C2M-1},m=1.
Step 4 If c ≤num PHr,go to Step 2.Otherwise,set
and go to Step 2.
Step 5 Output the task collaboration plan.
4.5 Distributed tasks-platforms scheduling process
The flow chart of the distributed tasks-platforms scheduling is shown in Fig.4.
Fig.4 Flow chart of distributed tasks-platforms scheduling
5.Simulation results
5.1 Mission scenario
In this section,the maritime operations center(MOC)experiment information[19]is used for our experimental simulation example.The mission needs to be done in 30 h and three TDHs need to build an effective plan for allocating platforms to tasks distributed over the combat area.The mission is divided into 11 tasks,namely n=11.These tasks include“AEW Area A,TAMD GREEN,TAMD BLUE in A,SURF SURV Area A,MIW in Strait A,CVG Penetrate into A,DEF vs.CDCM Attack,Attack Air Bases,Attack C2 Nodes,Attack IADS,Attack MSL Bases”which are labeled“T1”to“T11”.There are 12 kinds of available platforms,namely m=12.They are“CVN,CG,DDG,SSN,P3,MH53,AWACS,JSTAR,U2,RJ,UAV,AEF”which are labeled“PH1”to“PH12”.The resource capabilities include“C2,STRK,AW,BMD,CMD,SUW,USW,MIW,ISR(A),ISR(S),ISR(G)and BDA”which are labeled“r1”to“r12”.The number of resource capabilities,L,is 12.The task information is given in Table 1.Resource capabilities of platforms are given in Table 2.We can find that most of those tasks are parallel because they need to be done at the same time.Only the task T6 follows tasks T5,T7,T8 and T10.
Table 1 Resource requirements for tasks
Table 2 Resource capabilities of platforms
We assume that the time platforms taken to move to task location are considered when planners make the time programming.Based on the assumption,the platforms assigned to the above four tasks also can do the task T6.There are three TDHs in this organization.The attribution information of tasks and platforms is shown in Fig.5.
Fig.5 TDH’s resources and responsibilities
5.2 Results and analysis
In this paper,we design seven experiments to verify the effectiveness of the method.The first experiment,as shown in Table 3,is three TDHs making the decision by themselves.Each TDH make the tasks-platforms plan only according to its own resources.From Table 4 to Table 6,TDHs make tasks-platforms plans by using the distributed coordination.We set different values of QTthreshold,which are 0.55,0.6,0.65,0.7,0.75 and 0.8,in the following six experiments.
Table 3 Tasks-platforms plan before coordination
The result in Table 3 is the tasks-platforms plan without coordination.Only TDH2 has enough resources and all tasks’resource requirements can be satisfied.The resources of TDH1 and TDH3 are not enough.These two TDHs can only handle tasks with high priority.Qualities of low priority tasks(T3 and T10)are 0 because of the lack of some necessary resources.There are two reasons for this situation.One is that platforms with strong resources are used by high priority tasks and there is no resource for following tasks.The other is that TDH does not have enough platforms at the beginning.These resources are not balanced among TDHs.
In Table 4,QTthreshold is 0.55 and 0.6.Compared to the plan before coordination,tasks T3 and T10 are completed after different TDHs’cooperation.All tasks are handled well and their qualities are over 60%.
Table 4 Tasks-platforms plan after coordination(QTthreshold=0.55 and 0.6)
In Table 5,QTthreshold is 0.65 and 0.7.The task T2 is handled better after cooperation.Its quality is improved from 64.94%to 81.82%.Although all tasks execution qualities are over 70%,the quality of the task T11 is reduced from 100%to 71.21%.This is because PH11 is sharable in assigned platforms to task T11.The task T11 is not planned again since its quality still satisfy threshold when moving PH11 away.Table 6 indicates the results after coordination when QTthreshold is 0.75 and 0.8.In order to reach the threshold,more tasks and platforms are involved in cooperation than before.All qualities are over 80%.In our experiments,we also try greater threshold such as 0.85 and 0.9.The tasks-platforms plan is not generated successfully because the assigned platforms are all unshared.The high priority task occupies platforms with strong resources to ensure its quality is always over the threshold.There is not enough resources for low priority tasks.
Table 5 Tasks-platforms plan after coordination(QTthreshold=0.65 and 0.7)
Table 6 Tasks-platforms plan after coordination(QTthreshold=0.75 and 0.8)
The comparison of average task execution qualities is shown in Fig.6.
Fig.6 Comparison of average task execution qualities
Although the threshold increases constantly,the average of tasks execution qualities does not always increase.The best threshold is 75%–80%.The average task execution qualities after coordination are much better than before.It indicates that the proposed method is very suitable to dealing with the distributed tasks-platforms planning problem.
Fig.7 indicates the comparison between task execution qualities before and after coordination.The qualities of tasks T2,T3 and T10 are improved obviously.Although the qualities of the tasks T8,T10,and T11 decline,their shareable platforms make sure all tasks satisfy the requirements.These platforms are used more efficiently.In conclusion,the experiment proves the correctness of the distributed tasks-platforms scheduling method.
Fig.7 Comparison of task execution qualities before and after coordination
6.Conclusions
In this paper,we study the distributed tasks-platforms scheduling problem of the holonic-C2 organization.There are mainly three aspects as follows.Firstly,we give a formal description about the C2 structure of the holonic-C2 organization.Secondly,the single task resources scheduling model is built and the solving method based on the mbest algorithm is proposed.Thirdly,under the framework of the distributed parallel resources scheduling,we design the distributed cooperative mechanism based on platform price and resource capacity for the holonic-C2 organization.Finally,the validity of the proposed method is proved by experiments.
The shortcoming of this paper is that we only consider the satisfaction of resources between tasks and platforms and ignore more detailed information about platforms’location and moving speed.We assume that all the other conditions can be satisfied at the beginning of the task.The next work in the future is building a more perfect model and studying the distributed resource adjustment method for the holonic-C2 organization when some emergencies occur.
References
[1] SUN Y,YAO P Y,SHUI D D,et al.Uncertain optimal model and solving method to platform scheduling problem in battlefield.Journal of Systems Engineering and Electronics,2016,27(1):157–165.
[2] LEVCHUK G M,LEVCHUK Y N,LUO J,et al.Normative design of organizations-Part I:mission planning.IEEE Trans.on Systems,Man,and Cybernetics-Part A:Systems and Humans,2002,32(3):346–359.
[3] LEVCHUK G M,LEVCHUK Y N,MEIRINA C,et al.Normative design of project-based organizations-Part III:modeling congruent,robust,and adaptive organizations.IEEE Trans.on Systems,Man,and Cybernetics-Part A:Systems and Humans,2004,34(3):337–350.
[4] KOESTLER A.The ghost in the machine.London:Arkana Books,1967.
[5] PARK C,PATTIPATI K R,KLEINMAN D L.Multi-level operational C2 holonic reference architecture modeling for MHQ with MOC.Mansfield:Faculty Publications,2009.
[6] YU F,TU F,PATTIPATI K R.Integration of a holonic organizational control architecture and multiobjective evolutionary algorithm for flexible distributed scheduling.IEEE Trans.on Systems,Man,and Cybernetics-Part A:Systems and Humans,2008,38(5):1001–1017.
[7] HAN X.Optimization-based decision support algorithms for network identification and dynamic resource management.Connecticut:University of Connecticut,2016.
[8] TANG S Y,ZHU Y F,QUN L I,et al.Survey of task allocation in multi agent systems.Systems Engineering and Electronics,2010,32(10):2155–2161.(in Chinese)
[9] DE MENDONC¸A R M, NEDJAH N, DE MACEDO MOURELLE L.Efficient distributed algorithm of dynamic task assignment for swarm robotics.Neurocomputing,2016,172:345–355.
[10] NEDJAH N, DE MENDONC¸A R M, DE MACEDO MOURELLE L.PSO-based distributed algorithm for dynamic task allocation in a robotic swarm.Procedia Computer Science,2015,51:326–335.
[11] KIA S S.An augmented Lagrangian distributed algorithm for an in-network optimal resource allocation problem.Proc.of the IEEE American Control Conference,2017:3312–3317.
[12] SUN Y.Research on multi-agent intelligent decision support system based on blackboard.Electronic Design Engineering,2012,562–564(21):1638–1641.(in Chinese)
[13] CHEN X,WEI X M,XU G Y.Multiple unmanned aerial vehicle decentralized cooperative air combat decision making with fuzzy situation.Journal of Shanghai Jiaotong University,2014,48(7):907–913,921.(in Chinese)
[14] ZHANG Y Z,ZHANG L,DU Z.Distributed blackboard decision-making framework for collaborative planning based on nested genetic algorithm.Journal of Systems Engineering and Electronics,2015,26(6):1236–1243.
[15] GUO F J.Research on C2S of manned vehicle/UAV formation air-to-ground attack based on Holon.Xi’an:Northwestern Polytechnical University,2012.(in Chinese)
[16] ZHANG Y X.Research on C2S of manned vehicle/UAV formation air-to-sea attack based on Holon-Agent.Xi’an:Northwestern Polytechnical University,2016.(in Chinese)
[17] SUN Y,YAO P Y,WAN L J,et al.A quantitative method to design command and control structure of distributed military organization.Proc.of the IEEE International Conference on Information System and Artificial Intelligence(ISAI),2016:451–458.
[18] HAN X,MANDAL S,PATTIPATI K R,et al.An optimizationbased distributed planning algorithm:a blackboard-based collaborative framework.IEEE Trans.on Systems,Man,and Cybernetics:Systems,2014,44(6):673–686.
[19] HAN X,BUI H,MANDAL S,et al.Optimization-based decision support software for a team-in-the-loop experiment:asset package selection and planning.IEEE Trans.on Systems,Man,and Cybernetics:Systems,2013,43(2):237–251.
[20] GUHA S,MUNAGALA K,SHI P.Approximation algorithms for restless bandit problems.Journal of the Association for Computing Machinery(ACM),2007,58(1):1–50.
[21] GUPTA O K,RAVINDRAN A.Branch and bound experiments in convex nonlinear integer programming.Management Science,1985,31(12):1533–1546.
[22] WU R J,SUN P,SUN Y.Distributed dynamic task plan adjustment model and algorithm.Systems Engineering and Electronics,2017,39(2):322–328.(in Chinese)
DOI: 10.21629/JSEE.2019.01.11
Manuscript received November 01,2017.
*Corresponding author .
This work was supported by the National Natural Science Foundation of China (61573017; 61703425), the Aeronautical Science Fund(20175796014), and Shaanxi Province Natural Science Foundation(2016JQ6062;2017JM6062).
Biographies
WANG Xun was born in 1990.He is currently a Ph.D.candidate of Air Force Engineering University.He received his B.S.degree in communication engineering from Shandong University in 2013,and his M.S.degree in command information system from Air Force Engineering University in 2013.His research interests include command information system and mission planning.E-mail:wxkgdxy@163.com
YAO Peiyang was born in 1960.Currently he is a professor in Information and Navigation College,Air Force Engineering University.He received his B.S.degree in 1982 and his M.S.degree in 1991 from Xidian University.His research interests include command and control theory and command automation system.
E-mail:ypy 664@163.com
ZHANG Jieyong was born in 1983.Currently he is a lecturer in Information and Navigation College,Air Force Engineering University.He received his B.S.degree in 2006,his M.S.degree in 2008 and his Ph.D.degree in 2012 from Air Force Engineering University.His research interests include mission planning technique and military organizational analysis.
E-mail:dumu3110728@126.com
WAN Lujun was born in 1986.Currently he is a lecturer in Air Traffic Control and Navigation College,Air Force Engineering University.He received his B.S.degree in 2007,his M.S.degree in 2010 and his Ph.D.degree in 2014 from Air Force Engineering University.His research interest includes combat agent modeling and simulation.
E-mail:pandawlj@126.com
JIA Fangchao was born in 1989.Currently he is a lecturer in Dalian Sergeant School of Air Force Communication.He received his B.S.degree in 2011 from Taiyuan University of Technology.He received his M.S.degree in 2013 from Air Force Engineering University.His research interest includes command information system.
E-mail:hijiafc@163.com
标签：COMMAND论文; and论文; control(C2)论文; decision-holon论文; distributed论文; TASK论文; ALLOCATION论文; EXECUTION论文; QUALITY论文; platform论文; PRICE论文; order论文; optimization论文; Information and Navigation College论文; Air Force Engineering University论文; Air Traffic Control and Navigation College论文; Sergeant School论文; Air Force Communication论文;