A Multiagent Framework for Integrating Revenue-Creating Control ProcessesKlaus Weber Lufthansa SystemsJunqing Sun Civil Aviation Institute of ChinaZhaohao Sun Bond University, Australia: A Multiagent Framework for Integrating Revenue-Creating Control Processes Klaus Weber Lufthansa Systems Junqing Sun Civil Aviation Institute of China Zhaohao Sun Bond University, Australia AGIFORS Reservation & Yield Management Study Group Annual Meeting Berlin 16 – 19 April 2002


Who is Who?: Who is Who? Klaus Weber Lufthansa Systems Berlin, Germany Junqing Sun Civil Aviation Institute of China Tianjin, China Zhaohao Sun Bond University Gold Coast, Australia Special Thanks to Johannes J. Bisschop Paragon Decision Technology The Netherlands


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


Motivation: Motivation System performance depends not only on components but also on System structure Interrelationship among the components Integrated optimization of interdependent processes is better than independent optimization Some achievements in vertical integration of business processes However, usual situation at airlines: distributed hardware separated databases different organizational units but many generally identical functions, e.g. forecastint, optimization use same data Key idea


Motivation (cont): Motivation (cont) Software engineering viewpoint Systems integration increases system complexity Object-oriented paradigm is a powerful tool but not the most appropriate to deal with complexity Agent-oriented techniques new means of analyzing, designing and building complex software systems Multiagent technology is a new paradigm for distributed decision making (i.e. problem solving) Here we present Multiagent framework for integration of revenue-creating control processes Fleet assignment, Pricing, Revenue management under one roof


Motivation (cont): Motivation (cont)


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


Revenue-Creating Control ProcessesO&D Revenue Management System: Revenue-Creating Control Processes O&D Revenue Management System Flight Schedule PNR Data Availabilities Actual Data Forecast Interface Database Group Database Optimizer Database Fares Database Forecaster Optimizer Flight Schedule


Revenue-Creating Control ProcessesFleet Assignment: Revenue-Creating Control Processes Fleet Assignment Internal Fare Data ATPCO MIDT Market Size Forecaster Market Demand Flight Schedule Average Revenue Profitability Evaluation Model Objective Function Fleet Assignment Assignment


Revenue-Creating Control ProcessesPricing: Revenue-Creating Control Processes Pricing Internal Fare Data ATPCO MIDT Market Size Forecaster Price Elasticity Model Competitor Fare Actions Market Change Auto Matcher Product Change


Revenue-Creating Control ProcessesDifferences & Commonnesses: Revenue-Creating Control Processes Differences & Commonnesses Not all databases are commonly used. Each system comprises some specific functions Differences


Revenue-Creating Control ProcessesDifferences & Commonnesses (cont.): Revenue-Creating Control Processes Differences & Commonnesses (cont.) Commonnesses Some databases are used by all systems. Each system comprises functions which are part of the other systems as well. Each system is composed of inter-related sub-systems. Each System is hierarchical. Some sub-systems are hierarchical themselves. Relationship between sub-systems varies, e.g. peer, client-server, team. Whether system components can be considered primitive or complex depends on the viewpoint. It is possible to distinguish between interactions among sub-systems and the interaction between sub-systems. Each system is decomposable to some degree. C o m p l e x S y s t e m


Revenue-Creating Control ProcessesComplexity: Revenue-Creating Control Processes Complexity Canonical Complex System [Jennings & Wooldridge 2000] Sub-system Sub-system component Composed of Frequent interaction Infrequent interaction


Revenue-Creating Control ProcessesIntegrated View: Revenue-Creating Control Processes Integrated View Optimization Method 1, Method 2, ..., Method n Forecasting Method 1, Method 2, ..., Method n Parameter Estimation Method 1, Method 2, ..., Method n Data Pre-Processing Method 1, Method 2, ..., Method n Other algorithms Specific (complex) calculations Auxiliary calculations


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


Intelligent Agents and Multiagent SystemsWhat is an Agent?: Intelligent Agents and Multiagent Systems What is an Agent? Agents can perform some activities autonomously. At a minimum, they must be able to carry out some instructions ... without the help of other agents. Additionally, they may be able to make decisions of various levels of complexity on their own. Agents are part of a community. No agent is an island. When agents co-exist in a community, although some may exhibit a very high degree of autonomy, they are never truly independent of the other agents because they share an environment and therefore may compete for resources, whether intentionally or not. [Hayes1999] ... we define an agent as referring to a component of software and/or hardware which is capable of acting exactly in order to accomplish tasks on behalf of its user. ... it is an umbrella term, meta-term or class, which covers a range of other more specific agent types ... [Nwana1996]


Intelligent Agents and Multiagent SystemsWhat is an Agent?: Intelligent Agents and Multiagent Systems What is an Agent? What is Agent-Oriented Software? [Wooldridge1997] ... an agent is an encapsulated computer system that is situated in some environment, and that is capable of flexible, autonomous action in that environment in order to meet its design objectives ... Agents are Clearly identifiable problem solving entities Situated in a particularly environment Designed to fulfill a specific role Autonomous Capable of exhibiting flexible problem solving behavior


Intelligent Agents and Multiagent SystemsWhat is a Multiagent System?: Intelligent Agents and Multiagent Systems What is a Multiagent System? What is a multiagent system? [Sun2001] A multiagent system is a group of agents that work together to find answers to problems that are beyond the individual capabilities of knowledge of each agent. A multiagent system (MAS) is a loosely coupled network of agents, (which can be distributed over several computers) that communicate together to accomplish individual and/or common goals, which cannot be solved by one single agent alone. MAS do not have global system control over the agents, usually uses decentralized data. Communication between agents mainly takes place asynchronously.


Intelligent Agents and Multiagent SystemsWhat is a Multiagent System? (cont): Intelligent Agents and Multiagent Systems What is a Multiagent System? (cont) Questions Which agent roles? Which interactions between agents? Which acquaintances between agents? Which system structure? To be cleared later


Intelligent Agents and Multiagent SystemsComplex Systems and Agents: Intelligent Agents and Multiagent Systems Complex Systems and Agents Integration of Revenue Management, Fleet Assignment and Pricing results in a complex system. Are multiagent systems the appropriate paradigm? Principle mechanisms to manage complexity Decomposition divide problem, deal and solve parts in relative isolation Abstraction define simplified model, emphasize some details, suppress others Organisation identify and manage inter-relationships between problem solving components Software Engineering Viewpoint


Intelligent Agents and Multiagent SystemsComplex Systems and Agents (cont.): Intelligent Agents and Multiagent Systems Complex Systems and Agents (cont.) Decomposition sub-systems work together to achieve functionality of their parent system localisation and encapsulation interactions occur at unpredictable times, for unpredictable reasons, between unpredictable components multiple, interacting, autonomous components Abstraction minimize semantic gap between units of analysis and the constructs in the solution paradigm system « sub-system è Decomposition interplay between sub-systems = high level social interaction agents cooperate to achieve common objectives Organization broad variety of relationships peers ® control hierarchies short-term ® ongoing relationships may frequently change organizations are first-class entities in agent systems MAS P MAS P MAS P


Intelligent Agents and Multiagent SystemsMultiagent Systems and OO Software Engineering: Intelligent Agents and Multiagent Systems Multiagent Systems and OO Software Engineering Important differences Objects are generally passive in nature Objects encapsulate state and behavior - they do not encapsulate behavior activation (action choice) Object-orientation fails to provide an adequate set of concepts and mechanisms for modeling such systems Object-oriented approaches provide minimal support for structuring collectives.


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


A Multiagent FrameworkAgent Roles: A Multiagent Framework Agent Roles Main parts in software architecture [Perry1992] Data elements Processing elements Connecting elements Agent roles Coordinating agents Dispatcher agents Control agents Worker agents


A Multiagent FrameworkAgent Interaction: A Multiagent Framework Agent Interaction Dispatcher agents Keep track of idle worker agents Divide computational work effectively Coordinating agents Gather status information from agents below them in the hierarchy To be defined Interdependencies and interrelations of roles Agent interactions: messages Message priorities (e.g. highest priority for user interference) Who? With whom? How? Not: When?


A Multiagent FrameworkSystem Overview: A Multiagent Framework System Overview DB 10 DB 9 DB 8 DB 7 DB 3 DB 4 DB 5 DB 1 DB 2 DB 6 Dispatcher agent Dispatcher agent Dispatcher agent Dispatcher agent Coordinating agent Coordinating agent Coordinating agent Coordinating agent Control agent Control agent Control agent Control agent Control agent Worker agents


A Multiagent FrameworkTools: A Multiagent Framework Tools Several development tools for agent-oriented software available Commercial tools Research prototype http://www.agentbuilder.com/AgentTools Toolbox Libraries including functional agent components Predefined coordination and organisation relations General planning and scheduling mechanisms Editors Visualisation tools


Agenda: Agenda Motivation Revenue-Creating Control Processes Intelligent Agents and Multiagent Systems A Multiagent Framework Conclusions / Outlook


Conclusions / Outlook: Conclusions / Outlook Vertical integration requires adequate software engineering techniques Multiagent Framework Revenue management, pricing, and fleet assignment systems have all characteristics of complex systems From a software engineering viewpoint the new paradigm of multiagent technology is advantageous for development of complex systems. We introduced agents, multiagent systems, agent roles and agent interaction both generally and with respect to revenue management, pricing and fleet assignment systems. Special development tools already exist.


Outlook: Outlook Further characterise agents within the framework Develop a prototype using multiagent technology Study on integration of existing systems using multiagent technology Study on integration of additional processes e.g. processes in airline operations control


Thank You for Your Attention!Any Questions?: Thank You for Your Attention! Any Questions? Klaus.Weber@LHSystems.com