c3f183ea1f560d5a5098d8e690081548.ppt
- Количество слайдов: 16
ICWS 2006 A Framework for Intelligent Web Services: Combined HTN and CSP Approach Incheon Paik, University of Aizu Daisuke Maruyama, University of Aizu Michael N. Huhns, University of South Carolina Presented by: Jingshan Huang Computer Science and Engineering Dept. University of South Carolina ICWS 2006, Chicago 1
Motivation u Solving real-life problems requires a set of appropriate services to be (1) composed via planning, (2) scheduled, and (3) executed. l l u HTN (Hierarchical Task Network) : Planning CSP (Constraint Satisfaction Problem) : Scheduling Problems of HTN-Only system ― Cannot satisfy scheduling efficiently ― Difficulty in finding an autonomous solution u Suggest a Novel WS Composition Engine: the HTN-CSP combined system ICWS 2006, Chicago 2
Hierarchical Task Network (HTN) Planning u To produce a sequence of actions by task decomposition from large tasks u Similar to classical AI planning u Can produce plans to reach a final goal, but it is not suitable for scheduling u We chose SHOP 2 as our basic planner ICWS 2006, Chicago 3
Planning Domain Example task: preconditions: : Method (transport ? p) : compound task (at ? p ? x) (destination ? p ? y) (available-truck ? t) : primitive task : state subtasks: (dispatch ? t ? x) task: (!load ? t ? p) (!move ? t ? x ? y) (return ? t ? y) (dispatch ? t ? x) ICWS 2006, Chicago (return ? t ? y) subtasks: (!reserve ? t) task: (!move ? t home ? x) (!move ? t ? x home) (!free ? t) 4
HTN-Only System Framework Web Services Hotel Reservation WS plan Train Timetable WS Airplane Timetable WS Input Domain Analyzer ICWS 2006, Chicago … input data HTN Planner planning problem Web Service Executor 5
HTN-Only System Framework plan I'd like to make a trip next week. I want to depart Aizu-wakamatsu at 14: 00 and arrive at San-Francisco. Web Services Hotel Reservation WS Train Timetable WS Input Domain Analyzer ICWS 2006, Chicago … input data Departure Location = Aizuwakamatsu Airplane Timetable WS Departure Time = 14: 00 Departure Date = Feb. 21, 2006 Arrival Location = San-Francisco HTN Planner planning problem Web Service Executor 6
HTN-Only System Framework initial states: Web (departure-location aizu-wakamatsu) Services (departure-time _1400) Hotel Reservation WS (departure-date _20060221) plan (arrival-location san-francisco)) Train Timetable WS goal task: (trip) Airplane Timetable WS Input Domain Analyzer ICWS 2006, Chicago … input data HTN Planner planning problem Web Service Executor 7
HTN-Only System Framework task: (travel ? from ? to) Web Services preconditions: Hotel Reservation WS plan. . . Train Timetable WS subtasks: (travel-by-airplane ? via ? to) data Input Domain Analyzer ICWS 2006, Chicago HTN Planner planning problem Airplane Timetable WS … (travel-by-train ? from ? via) input Web Service Executor (available-train JR-Banetsu-West-Line aizu-wakamatsu _0600 koriyama _0712). . . 8
HTN-Only System Framework plan Input Domain Analyzer ICWS 2006, Chicago … input data (!travel-by-train JR-Banetsu-West-Line aizu-wakamatsu _1413 _20060221 Web Services koriyama _1513 _ 20060221) (!travel-by-train Hotel Reservation WS Tsubasa-186 koriyama _1525 _ 20060221 tokyo _1648 _ 20060221) (!travel-by-train Timetable WS Narita-Express-37 tokyo _1703 _ 20060221 narita _1757 _ 20060221) Airplane Timetable WS (!travel-by-airplane ANA-B 2 narita _2000 _ 20060221 san-francisco _1100 _ 20060221) HTN Planner planning problem Web Service Executor 9
Problems with an HTN-Only System u Web service composition by a planner has limitations: Inefficient for autonomous finding solution in planning l Not good for dealing user’s various requests l Weak in maintenance l Not efficient for scheduling l ICWS 2006, Chicago 10
HTN-CSP Combined System Framework Web Services plan Input Domain Analyzer Web Service Executor HTN Planner planning problem ICWS 2006, Chicago Train Timetable WS Airplane Timetable WS … input data Hotel Reservation WS CSP Constructor CSP representation CSP Solver CSP tuple 11
CSP Constructor HTN Planner Goal Variable Representation <Z, D, C> Z: variables D: domains C: constraints Domain Representation Domain Analyzer Sub-Goal Trip Sub-Goal Shopping Trip Domain Knowledge Base Trip Domain Ontology Information needed for constructing CSP tuple Knowledge Base Ontology Train Label Assigner ・ ・ ・ Knowledge Base Which assigner should be applied. ICWS 2006, Chicago Assigner Repository Worldwide Web Service Train Timetable Web Service est Hotel Label Assigner Request Data defined with Ontology equ Check Hotel Reservation Web Service ic R Control Assigner Management UDDI nam Robot with Situation Calculus Information needed for Domain Analysis Web Service Sub-Layer Ontology Dy Master-Layer Knowledge Base Information needed for HTN planning CSP Solver Constraint ・ ・ ・ Constraint Representation Airplane Timetable Web Service ・ ・ ・ Constraint-satisfied Label 12
HTN-CSP Combined System Framework … (riding-duration tokyo narita) = 0: 54 (riding-duration koriyama tokyo) = 1: 23 (riding-duration aizu-wakamatsu koriyama) = 1: 00 Web Services (departure-time tokyo) = 17: 03 (departure-time koriyama) = 15: 25 plan Hotel Reservation WS (departure-time aizu-wakamatsu) = 14: 13 (departure-date tokyo) = 2006/2/21 (departure-date koriyama) = 2006/2/21 Train Timetable WS Web (departure-date aizu-wakamatsu) = 2006/2/21 Service (arrival-time tokyo) = 16: 48 input Executor (arrival-time narita) = 17: 57 Airplane Timetable WS data (arrival-time koriyama) = 15: 13 (arrival-date tokyo) = 2006/2/21 (arrival-date narita) = 2006/2/21 Input (arrival-date koriyama) = 2006/2/21 HTN CSP (departure-time narita) = 20: 00 Domain Planner Constructor (departure-date narita) = 2006/2/21 Solver Analyzer (arrival-time san-francisco) = 11: 00 planning (arrival-date san-francisco) = 2006/2/21 CSP problem ICWS 2006, Chicago representation tuple 13
Performance Comparison Scenario 3 Activity in Composition HTN-Only 3+1=4 Total : 3039 Planning HTN-CSP Combined 3+1=4 Total : 5789 Planning and Scheduling 4+2=6 Total : 2661 Planning HTN-CSP Combined 4+2=6 Total : 3187 Planning and Scheduling HTN-Only 2 Processing Time (ms) HTN-Only 1 Solving Type Number of Web Service Invocation 0 Total : 122 Planning HTN-CSP Combined 0 Total : 2248 Planning and Scheduling ICWS 2006, Chicago 14
Advantages of the HTN-CSP Combined System u Development Efficiency l u Scheduling is already integrated Flexibility l HTN-CSP combined system: ― l HTN-only system: ― u Can represent complex problems more easily Might produce scenarios that are hard to solve Extensibility l HTN-CSP combined system: ― Features can be added easily ― Many algorithms for solving CSPs l available HTN-only system: ― To add new features, the planning domain must be rewritten in many cases ICWS 2006, Chicago 15
Conclusion and Future Work u Suggested a combined HTN-CSP architecture for automatic Web service composition l Efficient for planning and scheduling u Future work Automating HTN – CSP encoding l Fully automated problem solving (service composition) in a Semantic Web environment using this framework l ICWS 2006, Chicago 16
c3f183ea1f560d5a5098d8e690081548.ppt