Agent Modeling

Transport of emulsions in porous media is relevant to several subsurface applications. Many enhanced oil recovery (EOR) processes lead to emulsion formation and as a result conformance originating in the flow of a dispersed phase may arise. In some EOR processes, emulsion is injected directly as a mobility control agent. Modeling the flow of emulsion in porous media is extremely challenging due to the complex nature of the associated flows and numerous interfaces. The descriptions based on effective viscosity are not valid when the drop size is of the same order of magnitude as the pore-throat characteristic length scale. An accurate model of emulsion flow through porous media should describe this local change in mobility. The available filtration models do not take into account the variation of the straining and capturing rates with the local capillary number. In this work, we present experiments of emulsion flow through sandstone cores of different permeability and a first step on a capillary network model that uses experimentally determined pore-level constitutive relationships between flow rate and pressure drop in constricted capillaries to obtain representative macroscopic flow behavior emerging from microscopic emulsion flow at the pore level. A parametric analysis is conducted to study the effect of the permeability and dispersed phase droplet size on the flow response to emulsion flooding in porous media. The network model predictions qualitatively describe the oil-water emulsion flow behavior observed in the experiments.

Designing agent behaviors is rarely straight forward, especially if the natural environment of the agent is open and complex (as is the case of the Internet). In this paper, we describe our behavior discriminative agent (BDA) model which aims at supporting designers of such agents. The behaviors are here themselves represented by agents so the BDA agent comprises a multi-agent system, where behavior agents interact with a coordinator agent which decides the actions of the BDA agent. The BDA agent uses the Q-learning technique and learns by sensing the consequences of its actions and updates its belief in the behaviors continually. We implement our agent model in a chess game setting and present some early results.

Electronic institutions are the agents' counterpart of human organizations, which are specifically designed for providing support, trust, and legitimacy in electronic commerce applications. Two approaches have been advocated for the design and modeling of multi agent systems in an environment that is governed by some kind of (social) norms: in coordination strategy, multi agent systems are defined as a set of entities regulated by mechanisms of social order and created by more or less autonomous actors to achieve common goals; in cooperation strategy, agents model specific roles in the society and interact with each other as means to accomplish their goals. In this paper, we argue that there is a relative similarity between the two approaches with respect to their use of norms as constraints on the social behavior of multi agent systems.

In general, the integration of different modeling philosophies is a successful enterprise. This paper presents a concept that integrates two approaches, conjoint analysis and multi-agent simulation. The concept also employs some notions of social network analysis, consumer behavior modeling, and word-of-mouth marketing. These methods are self-complimentary: for example, conjoint analysis may serve as a perfect tool to supply multi-agent models with behavioral data, while multi-agent simulation offers dynamics to the static results of conjoint analysis. The idea of dynamic conjoint modeling is based on the assumption that one can predict an individual's purchases on the basis of his/her measured product preferences, communication habits and market behavior attributes. Dynamic conjoint modeling involves a simulated series of conjoint analyses executed by means of each time-step. This paper outlines a market model to simulate the association between consumers' communication and sales.

The structure and the functionality of a multi-agents based model is presented. The model requirements imposed by interactive techniques are highlighted. The model consists of active entities called agents, and passive entities such as behaviors, trajectories, actions, and conditions. The agents have explicit defined behaviors as spatial and temporal evolutions. The agents use fuzzy rules for an intelligent and adaptive

The idea of using computer aided games (gaming) to design agents in social models is presented. Many characteristics of the games to adapt them to this end are described, particularly the record of the justifica tion of actor decisions. Some possible methods to use game collected data for agent design are considered. A discussion is included about the problems and advan tages of a flexible model, which combines three ways to introduce in the model the decisions of the actors: by human beings (Players), by algorithms (Agents) and by simple assignations of values (Cast), the PAC (Play ers, Agents, Cast) technique. Finally a game for a natio nal economic system is described and preliminary result s are reported.