Research On Qualitative Preference Representation And Argumentative Reasoning Technology For Service-planning Of Responsive Satellites

The responsive satellite system has proposed a new requirement in building its ground supporting system,summarized as planning of responsive satellites.As a new pattern for utilizing satellites,this new type of planning demands human users to present their requirements via preferences,and demands the planner to achieve plans which satisfy the users best.The challenge of solving this problem lies in two aspects.In the representation part,it is needed to describe the sequencial relations and time-constrainted relations in an unified frame.In the planning part,one should make a fair and effective selection of preferences when it is impossible to satisfy all the preferences.On handling these those chanlleges,this thesis estalishes an extensive temporal planning theory,a language for qualitative preference,and also the preference-based temporal planning problem(PBTP)so as to accomplish the problem representation,ending with an argumentation framework as the basis of an automatic solver for PBTP.A set of qualitative preferences is used to describe plan goals and to restrict planning process,with argumentative deduction being used in automatic plannner to achieve plans with best satisfaction.Contributions of this paper lies in four aspects.(1)An extensive temporal plan theory is proposed to support time-constrained relations.Time tag is added to state predicates before time-based state predicates and state database are established to describe a world state with specific time qualification.Operation is expressed as a 6-tuple with premises,effects and time constraints inside.When a state-database matches an operation,an action will be instantiated,the performance of which will lead to an update of the database.The enhanced temporal theory provides foundation for the planning of responsive satellites.(2)A structed language for temporal planning preferencs is established.Qualitative preference for planning is summarized as a hierarchical strucute.Based on the extensive temporal plan theory(ETPT)proposed earlier,trajectory preferences is defined,before preferences with complex grammar are defined via the semantics of comparison,combination and priority.The preference-expression language owns clarified semantics in ETPT,and provides powerful expression ablity over most of the current research.(3)Analysis of the computing complexity,and also the solution for a typical sub-problem is proposed in solving the preference-based temporal planning problem.When preferences and temporal constraints are functioning together,especially when a complex preference set is imposed on a planning problem,planning turns out to be NP-hard.If it can be devided into sub-problems called signle rational-objective-preference planning,combination of the sub-problems' solutuions will result in a satisfactory solution for the original problem.Algorithms for solving the sub-problem is designed as a process with backward search,able to achieve all possible plans at the cost of satisfactory computation complexity.(4)An instantiated argumentation framework is proposed for automatic planning in PBTP.In seeking a planner with best satisfaction for a PBTP problem equipped with a general preference set,an argumentation framework based on practical arguments is established.A practical argument is constructed with a rational-objective-preference as its conclusion,and a partial plan supporting the conclusion.We defined the attack relationship and plan-consist relationship between practical arguments,based on which we propose the notion “plan-consist preferred extension”,which is a set of arguments with outstanding properties,the premises of all arguments can be combined to be a plan with maximum user satisfaction.A diaglue game is designed to generate the plan-consist preferred extensions.Taking the service planning of responsive satellite as the background,which is summarized as the preference-based temporal planning problem,this thesis carries study on planning theory,representation of qualitative preference and automatic planner,resulting in a representation system and an automatic planner utilizing argumentation techniques.Proofs are proposed to show that,i)the planning theory and the argument-based reasoner are correct,ii)the major algorithms are correct and complete.With the establishment of a prototype system and a detailed case study,it shows that the representation is valid and appliable,and the planner is practicable and effective.The result of this thesis paves a way for general application of responsive satellites,and also provides a framework for other preference-based planning problem.