Multiagent meta-model for strategic decision support
Introduction
The suggested meta-model allows the creation of multiagent systems (MAS) [22]. MAS generated is conceived to carry out simulations for strategic decision support for a project or a company. A MAS is adapted to a problem specific to solve. The meta-model does not solve the problems of strategy. It adapts a MAS to a given problem. It is composed of several concepts of learning which act a great role in the problem adaptation and the search for the strategic choices. MAS representation suit well. It is easy to represent a problem of strategy. An agent can be likened to an actor of the problem. The concept of groupe in Aalaadin MAS model (cf. Fig. 10) can represent a coalition. Several coalitions can be carried out by varying the behaviors of the agents without knowing the interaction.
But in MAS, it is not enough to allow the communication between agents. Each agent reacts to the messages of the other agents according to its possibilities and its interests if necessary. In complex situation, we observe overall that several scenarios of cooperation are possible. How can we determine the best scenario to satisfy the best each agent?
The game theory, formalized by Von Neumann and Morgenstern [11] bring suggestions for solutions. It is based on the following principles: one or more decision makers are brought to carry out strategic choices (as in a poker game) according to possibilities of profits and the assumed reactions of the other players. The game theory then consists in formalizing the scenarios according to the choices of each player, then to determine the best strategy for a player. In this theory two families of games appear; in the no cooperative games, the players play the ones against the others; in the cooperative games the players can cooperate between them i.e. to form coalitions. This search for coalition is called bargaining. But the solution of Von Neumann and Morgenstern is very restrictive for the bargaining in the cooperative games. Recent works to improve this theory did not bring decisive progress. Another solution was required.
To answer in a better way to the bargaining problem, the context of the decision-making must be taken into account. The theoretical reflexion on the strategic decision support is based on the expertises of the Centre Européen de Santé Humanitaire (CESH) and the Strataegis company. They cover varied fields from the management of company to the geopolitic field: management of crisis, international health, investment, business-to-businness cooperation, sustainability, plan of intervention, etc [20].
Three reflexions appeared from these expertises. The first reflexion relates to the diversity of the strategic decisions. The strategic decision management of a refugees camp are very different from those of a company which wants to develop its exports. Our model must be able to reduce the space of search for solutions to allow the identification of the decision types organized by a preordering. The second reflexion relates to the formation of the coalition with a multicriteria choice. The formation of a group seldom rests on only one criterion. Another mechanism of bargaining must take into account several criteria. The third reflexion relates to the distribution of the decisions in the time. The development of a strategy also consists in planning decisions. A decision depends on the decision taken at the previous period in the quoted fields. The bargaining is subject to this phenomenon. These three reflexions made it possible to work out concepts to build a prototype of the meta-model of strategic decisions in the REBOL programmation language. This leads to organize our communication in three points: study of the bargaining in the game theory and recently made improvements, presentation of the new approaches of the bargaining and the description of the generating meta-model of MAS.
Section snippets
Search for coalition in game theory
The theoretical reflexion which follows aim is to show the difficulties of finding solutions in the bargaining set. The game theory proposes solutions to find the coalitions in a cooperative game. This is bargaining. The Von Neumann and Morgenstern's solution of the game wants to answer this problem. The insufficiencies of this solution pushed authors recently to propose improvements.
Search for coaliton: new approaches in the cooperatives games
We want to form groups of players who cooperate. The players, called intervening agents in our model, cooperate within a group. The models realized for the search for the coalitions structures use techniques of learning. Two models of learning are proposed: a model of ordinal classification of the nearest neighbors and a model of learning per Markov chain.
Interest of the MAS model
The traditional data-processing models are often based on equations and on relations of cause to effect. They are powerful but however have a weakness to describe complex situations in particular when there are several levels of decisions. A MAS has a centered approach on a direct representation of the individuals and entities. This approach is well adapted to the simulation of the complex systems. Global operation emerges from the actions of the individuals (agents). A model allows to make
Conclusion
The strategy of company utilizes several actors. They can cooperate to achieve a laid down goal. A model equipped with a MAS is appropriate well to represent problems of strategy. The concept of actor can be compared to the concept of agent and a coalition will be represented by a group of agents. It is enough to act on the behavior of each agent to obtain realistic simulations. A theoritical difficulty has appeared in the search for coalitions.
A new approach of the bargaining in the
References (22)
- et al.
Coalition structure generation with worst case guarantees
Artificial Intelligence
(1999) - A. Aner, Y. Mansour, Nearest neighbors algorithm, in: Machine Learning Foundations,...
- V. Conitzer, T. Sandholm, Complexity of determining nonemptiness of the core, AAMAS-02 Workshop on Distributed...
- et al.
Elevator group control using multiple reinforcement learnings agents
Machine Learnings
(1998) - A. Dutech, Apprentissage d'environnement: approches cognitives et comportementales, PhD thesis, ENSAE, Toulouse,...
Reactive Distribued Artificial Intelligence: Principles and Application in Foundations of Distribued Artificial Intelligence
(1996)- L. Gasser, Computationnal Organization Research, in: ICMAS'95,...
- O. Guthnetcht, J. Ferber. A meta-model for the analysis and design of organizations in multi-agent systems, Proceedings...
Workshop on Case-Based Reasoning. DARPA 88—Clearwater
(1988)Case Based Reasoning
(1993)
Cited by (3)
Cross-community interoperation between knowledge-based multi-agent systems: A study on EMERALD and Rule Responder
2012, Expert Systems with ApplicationsCitation Excerpt :The main advantage of this approach is that it provides a safe, generic, and reusable framework for modeling and monitoring agent communication and agreements. EMERALD supported, so far, the implementation of various applications, like brokering (Antoniou, Skylogiannis, Bikakis, Doerr, & Bassiliades, 2007; Benjamins, Wielinga, Wielemaker, & Fensel, 1999), agent negotiations (Fang & Wong, 2010; Governatori, Dumas, Hofstedeter, & Oaks, 2001; Lin, Chen, & Chu, 2011) and bargaining (Kebriaei & Majd, 2009; Muthoo, 1999; Petit & Magaud, 2006), a single issue negotiation between two parties. In order to model and monitor the parties involved in an agent dialogue, a generic, reusable agent prototype for knowledge-customizable agents (KC-Agents), consisted of an agent model (KC Model), a yellow pages service (Advanced Yellow Pages Service) and several external Java methods (Basic Java Library), is deployed (Fig. 3).
Advanced agent discovery services
2012, ACM International Conference Proceeding SeriesResearch on the intellectualized system structure of UCAV mission planning
2009, 2009 Chinese Control and Decision Conference, CCDC 2009