Distributed Decision Making In Dynamic Uncertain Environment

The concept of decision making includes a wide scope of application,ranging from personal life to the development of the society.The emergence and development of decision science has contributed considerably to the optimal distribution of resources and has therefore prompted social productivity and the living standard.However,the framework of decision science,which originated from management science and conventional decision theory,has changed significantly in modern time.The changes lie especially in the following two aspects:Firstly,the assumptions of decision environment have been converted from a stable and certain situation to a dynamic and uncertain circumstance;and,secondly,the solution space of decision making has been generalized from numerical to functional and even more abstruse level.To perfect the methodology of decision making science,we will be needing to widen our scope incorporating inter-discipline knowledge.In addition,modern decision problems are often in large scale,which effectively prevents single decision unit to complete missions in acceptable time.Plus,for some decision tasks,taking a distributed decision scheme into utilization will be necessary because of their born natural distributive properties.Obviously,distributed decision making has become crucial in advancing to an effective and efficient decision system.Research and industrial practice of decision system under distributed framework have been promoted over the past years and related theories and applications have been growingly robust.However,it is undeniable that since the current development of distributed decision systems have originated from various disciplines,literatures in the area has indefinite and even conflicting descriptions towards the concept and failed in concluding the theories and notions to a certain and universal definition or penetrating the deeper insights of the problem.Moreover,most research outcomes in the area only focus on a certain portion of the problem and tend to exclude other distributed decision situations.To extract universal principles from previous researches and offer theroems for the development of distributed decision from a methedological level,the thesis discussed the following issues from a conceptual level:(1)In integrated consideration of system theory,information theory and control theory,the thesis fistly maintains that actually the concept and essence of management,decision-making and control are equivalent to each other.Under this premise,the analytic definitions pertaining to decision making is then introduced,including the concept,convergence,real-time,and stability of decision making.Besides,for time-realated sequential decision making system that has clear analytical dynamics,the concept of functional decision and related theories are put forward.In the end,in order to validate the engineering practicability of decision making theory stretching over,or derived from,cybernetics,a fire control method based on trajectory backstepping(inverse ballistic trajectory derivation)is proposed based on the theory of terminal value problem of differential equations and its solving algorithm.(2)A clear explanation regarding the basic properties of dynamic uncertain of environment decision making is firstly given;second,the thesis analysizes the decision making problems under dynamic uncertain environment,and further points out that the decision making problems under dynamic uncertain environment should be grouped into four categories: 1)classical stochastic decision making theory represented by multi-stage stochastic decision making;2)decision making problems partially known the analytic dynamics;3)decision making problems having statistical dynamics;and 4)decision making problem that the dynamics is absolutely unknown.Third,in view of the characteristics of all kinds of problems mentioned above,the solving models are established and the realated solutions of those models are given.Forth,the thesis expounds the basic ideas of decision making in the dynamic uncertain environment.(3)The five connotations of distributed decision making is defined: 1)constructing a decision model based on distributed sensing system;2)decision making system based on parallel computing technics;3)multi-model fusion system based on distributed architecture;4)collaborative assembly line system based on the distributed architecture;5)self-organized system.Moreover,to analysize,model and solve the five classes of distributed decision problems the distributed Max-min algorithm,matrix elimination method,fast matrix elimination method,distributed Benders algorithm,distributed reinforcement learning algorithm,wolf pack algorithms,et al,are proposed and applied.(4)Typical engineering examples of distributed decision problem in dynamic uncertain environment are introduced.Specifically: 1)with parallel computing theories,a distributed reforcement learning algorithm is proposed and applied to robots with which the effectiveness and efficiency to percept the environment information ans states and path planning is improved;2)taking distributed sensing system as the framework,the paper proposed a collaborative passive location method for multi-bearers(airplanes)based on time difference of arrival of radiated signal from target,with technics of dynamic variance estimation and IIR filtering;3)the wolf pack algorithm,as a particular illustration of self-organized system,used to seiege a prey by wolves,is designed.(5)Using Ethernet switch technology as the basis,the thesis makes birth a distributed simulation system architecture and running scheme.In detail,the toponogical structure and system composition are firstly introduced.Then we gives the concrete functions of each module.Finally,the information stream controlling scheme among sub-systems and the commucition protocol to interactive with other modules based on descriptive information stream concept are introduced.