Collection Mémoires et thèses électroniques
Accueil À propos Nous joindre

CHAPTER 5: Creation of an Initial Version of the Multiagent Geosimulation Models

Table des matières

In this chapter we present the third step of our proposed method. In this step, we design an initial version of the multiagent geosimulation models of the phenomenon to be simulated and its environment. We also present the illustration of this step by showing the multiagent-based models of customers’ shopping behavior in a mall.

In the previous chapter, we presented and illustrated the first two steps of our method. In the illustrations we defined: the application domain of the geosimulation, the end-users of the shopping behavior geosimulation application, and their needs. We also specified, in detail, the characteristics of the shopping behavior in a mall based upon in-depth studies in several disciplines related to this behavior. It remains that our goal is to provide end-users (mall managers) with an operational geosimulation application that can help them to make decisions. In order to develop a computer-based geosimulation of shopping behavior, we need to model the phenomenon to be simulated (shoppers) and to design the simulation environment (mall). In computer science, it is well-known that the modeling process is very important, because it reduces the complexity of real systems by creating models which can be used in a computer program. Hence, the aim of the third step of our proposed method is to design the geosimulation models which are used to develop the geosimulation prototype.

This chapter is organized as follows: Section 5.2 presents a generic description of the third step of our method. Section 5.3 aims to present the illustration of this step by presenting the shopping behavior geosimulation models, while in Section 5.4, we discuss the issues presented in this chapter and conclude it.

One of the main objectives of this thesis is to benefit from the recent progress made in the multiagent paradigm in geosimulation. For this reason, our basic foundation is the agent technology which is used to design the geosimulation models that represent the phenomenon to be simulated and its environment. Hence, we must use agent-oriented design techniques to create the geosimulation models (multiagent-based models) and represent the entities of the geosimulation. The Agent-Based Unified Modeling Language (AUML) ( provides such techniques. AUML is a graphical modeling language standardized by the Foundation for Intelligent Physical Agents (FIPA) Modeling Technical Committee (Modelling TC) (Odell et al., 2000). AUML was proposed as an extension of the Unified Modeling Language (UML) in order to design agent based systems. AUML contains certain diagrams that can be used to graphically design multiagent based systems. As examples of these diagrams we can mention class diagrams, interaction diagrams, collaboration diagrams, sequence diagrams.

We chose AUML to model our multiagent geosimulation models because it proposes graphical specifications and tools that can be used to design agent-based systems and of course agent-based simulation applications (Peres and Bergmann, 2005). The variety of AUML’s diagrams is sufficient to create the various diagrams that we need to specify multiagent geosimulations.

When designing the multiagent model of the geosimulation, we start from the initial model of the phenomenon designed in the previous step. It is relevant to note that the design of the multiagent geosimulation models must be independent from the simulation tool that will be used to execute these models. Naturally, this tool must be intended for multiagent-based simulations. Since we deal with geosimulation, the spatial aspects of the phenomenon to be simulated and the geographic features of its environment must be considered in the multiagent geosimulation models.

In the multiagent geosimulation model, we can distinguish the following categories of agents:

Abstract and common agents : This category contains the agents which do not represent any element in the simulation. They are, mainly, abstract and they are designed in order to contain common and abstract characteristics that may be shared by the other agents of the simulation models (active and passive agents).

Passive Agents (PA) : This category of agents represents entities that have a structure, without a behavior. Usually, a large part of the elements contained in the simulation environment belongs to this category. Here, we must characterize the spatial and non-spatial structures of the passive agents in both 2D and 3D modes.

Active agents (AA) : This category of agents represents entities, which have structures and behaviors: These entities actively participate in the simulation. For this category, we must specify the data structures of the entities (spatial and non-spatial structures) as well as their behaviors (spatial and non-spatial behaviors).

Each passive or active agent can have one or more profiles . A profile is represented by one or more roles that can be played by the agent in the simulation.

This third step of our proposed method contains the following sub-steps (see Fig 5.1):

(1) Design the common and abstract agents of the geosimulation : It aims to design the abstract and common agents (structure and behavior) of the geosimulation models.

(2) Design the active agents of the geosimulation : When designing these agents, we start from the initial version of the phenomenon model (part of the system to be simulated) (see step 2 of the method). Hence, the factors or variables belonging to the initial version of the model, are represented by what we call ‘ attributes ’ or ‘ property ’ of the active agents and the processes are represented by what we call ‘ methods ’ of the active agents. Of course, since we deal with geosimulation, we must take into account the non-spatial and spatial aspects of the system in the multiagent geosimulation model.

(3) Design the passive agents of the simulation : The design of these agents is based upon the initial version of the phenomenon model (part of the simulation environment) (see step 2 of the method). Hence, each entity belonging to the simulation environment is represented by one or more passive agents. The factors or variables belonging to the simulation environment are represented by the ‘ attributes ’ or ‘ properties ’ of the passive agents. Of course, since we deal with geosimulation, we must take into account the non-spatial and spatial aspects of the simulation environment.

In the following sections, we present the illustration of this step using the customers’ shopping behavior in a mall as a case study.

This section aims to present the details of software agents that build multiagent geosimulation models of shopping behavior in a mall. Such agents can be either passive or active. Passive agents only have structure, having no behavior at all, while active agents have both structure and behavior. Since we deal with geosimulations, agents’ structures and behaviors may, or may not, be spatial. Throughout the following sub-sections, we present the agents of the simulation environment, some are passive and other are active (the simulation actors). These actors correspond to the main agents of the simulation: the shopper agents (individuals, groups, and crowd).

In our work, we represent the spatial entities belonging to the geographic simulation environment (mall) by agents. This choice is motivated by the fact that it would be too complex to model and simulate in detail the behaviors of a crowd of shoppers in all the corridors of the mall as well as in all the stores and specific areas that they can visit: we have to set limits to the simulation. Assigning agents to some of the entities of the environment is the way that we have chosen to set these limits. For example, if we concentrate on the study of shoppers’ displacements in the mall corridors, we will not be able at the same time to model what happens in each store of the mall. For this reason, we propose to assign agents to specific spatial entities such as stores or restrooms. These agents set the limits of the geosimulation in the sense that the simulation of what happens in these geographic areas (i.e. stores, restrooms, etc.) can be done using external models such as statistical models. For example, statistical models implemented in the behaviors of the stores can compute the duration of the visit of specific categories of shoppers within the store and the choice of items that a shopper will purchase. Hence, when a shopper agent enters a particular store, the agent associated to the store computes the duration of the visit of the shopper agent and whether it buys some items. After the visit duration computed by the store agent, the shopper agent exists the store and resumes its navigation in the mall’s corridors.

Hence, a designer must carefully choose which elements of the virtual environment are parts of the limits of the geosimulation, which behaviors should be assigned to the agents associated to these elements, taking into account available statistical models used to simulate the relevant activities that take place in these elements.

We saw, in the previous chapter (sub-section 4.2.3), that the mall is characterized by many dimensions. Among these dimensions, we have the physical dimension, which represents the spatial/geographic characteristics of the environment, as well as the atmospheric dimension. In this sub-section, we present those agents related to these two dimensions of the mall.

The agents of the physical environment represent the physical entities that belong to the simulation environment (the mall). The majority of these entities belong to the spatial aspect of the mall’s model, which is presented in the previous chapter (sub-section 4.2.3).

These agents are very important to the simulation because the shopper agents interact with these entities when they accomplish their shopping behavior in the mall. These agents are: Door_Agent, Electronic_Door_Agent, Retail_Agent, Store_Agent, Kiosk_Agent, Room_Agent, Wash_Room_Agent, Cloak_Room_Agent, Desk_Agent, Window_Agent, Product_Agent, Stairs_Agent, Seat_Agent, Notice_Agent, Electronic_Notice_Agent, Escalator_Agent, Elevator_Agent, Area_Agent, Phone_Agent, Fountain_Agent, Slot_Machine_Agent, etc. and the list is not exhaustive. It was shown, in the literature review (Chapter 4), that these physical entities influence the shopping behavior of people in a mall. Details of the structure and behavior of these agents are presented in the Annex B of this thesis.

Since we deal with geosimulation, the spatial features of the agent of the physical environment are important. Hence, it is relevant to design these aspects in the agent’s models. The 2D spatial structure of these agents is designed in a geographic information system (GIS) representing the mall (see Fig 5.2). The 3D spatial structure is designed as a 3D model, generated from the GIS of the mall, using 3D studio max software ( (see Fig 5.3).

In the two previous sub-sections, we presented the passive agents that belong to the simulation environment. We remind that the main objective of our work is to simulate the shopping behavior of persons in a mall. Hence, these persons are represented by shopper agents which belong to the category of active agents or actors . As shown in the previous chapter (Chapter 4), the shopping activity is social and can be performed individually or in groups. For this reason, in the design of the shopper agents’ models we take into account three levels: individuals, groups, and a crowd of shoppers. The shopper agent represents an individual shopper, the agent group of shoppers represents a group of shoppers who come together to the mall, and the agent crowd of shoppers which represent all the shoppers visiting the mall.

In the following points, we present details of the attributes (structure) and methods (behavior) of these actor agents.

Some shoppers come to the mall in groups (families, friends, colleagues) to accomplish a shopping behavior. In our simulation, a group of shopper is represented by a Shoppers Group Agent. Unfortunately, in the literature review, we did not find any model of the structure or behavior of a mall’s group of shoppers. Therefore, we turned to other disciplines that study groups. Based on the study results related to groups (see Section 4.4), we propose a simple structure for the group of shoppers which is used to record the members of the group. In the following points we discuss the structure and behavior of a group of shoppers.

The structure of the group of shoppers : A group of shopper contains some members. Each member of the group has his/her goals when coming in the mall. The whole group has its own list of goals which is based upon the goals of its members. The members of the group are interconnected by weighted relationships. The relationship can be based on family ties (kinship), affection, or friendship. In each group, each member projects a certain weight inside the group. This weight can give a role to the member, and can influence the decision-making process of the group. This weight represents the degree of influence of each member of the group. The member with the highest weight is called the leader of the group; the other members are called the followers . What’s more, it is possible to find more than one leader in a group.

The behavior of the group of shoppers : The members of a group of shoppers are not independent and, in some situations, they make some collective decisions. Depending upon the objectives (goals) of all the members, we must decide upon the goals of the group as a whole. According to the literature review made in the discipline of group and social psychology, the decision-making process of a group can be based upon several schemes: vote, delegation, polarization, etc. (Rao and Steckel, 1991). In this dissertation, we do not study the decision-making process inside a group. Therefore, we consider the group of shoppers as a whole entity which has its own list of shopping goals to reach.

In our work, the group of shoppers is represented by the Group_Shoppers_Agent. This agent has a special structure that can contain a list of agents. The attributes and methods of the agent Group_Shoppers_Agent are presented in the Annex B of this thesis. The group of shopper agents represents the meso-scale of the simulation model that simulates the shopping behavior in a mall.

In this chapter, we presented the third step of our proposed method. This step aims to design the agent-based models of the geosimulation (the phenomenon to be simulated and its environment). We also presented the illustration of this step using the customers’ shopping behavior as a case study. Hence, we presented the multiagent-based models of the shopping behavior simulation.

There are several works that use the multiagent paradigm to simulate several behaviors. However, among these works, there are very few that simulate the shopping behavior in a geographic space. For example, we can cite the work of (Ben Said et al., 2001), which simulates general consumption behavior. In their work, the authors simulated the consumption behavior of items without considering the spatial aspect of this behavior. Furthermore, this aspect is very relevant for us given that shopping behavior in a mall is often influenced by the mall’s spatial characteristics. (Koch, 2001), (Dijkstra et al., 2001), and (Timmermans et al., 2003) simulated shopping behavior in a spatial context, which can be a city, or a shopping mall. Unfortunately, these scholars did not present the shopping behavior as a pedestrian behavior inside a geographic environment, nor the details of these models as we do in this chapter. Our work presented agent-based shopping behavior models, in detail, based upon a solid literature review, specialized in this subject. We integrated into the multiagent models the majority of factors that influence shopping behavior in a mall (attributes or properties of the agents), as well as the majority of processes that make-up this behavior (methods of the agents).

Moreover, based upon our literature review, we can affirm that there are no research papers involving the simulation of shopping behavior of group in a mall. The majority of works that implicate groups and collective decision-making are restricted to social psychology. In our work, we propose a simple model to represent a group of shoppers. We also propose a simple model representing a crowd of shoppers in the mall. What is important is that we take into account the spatial characteristics of the phenomenon to be simulated and the geographic features of the simulated environment in each scale of the simulation.

Proposing a multi-scale multiagent geosimulation model of shopping behavior in a mall can be considered as a contribution in the computer simulation and multiagent geosimulation fields.

The next chapter aims at presenting our methods for gathering the data which will be used to feed these models.

© Walid Ali, 2006